**Electrostatic Charging Tendency Analysis Concerning Retrofilling Power Transformers with Envirotemp FR3 Natural Ester**

### **Maciej Zdanowski**

Faculty of Electrical Engineering, Automatic Control and Informatics, Opole University of Technology, Prószkowska 76, 45-758 Opole, Poland; m.zdanowski@po.edu.pl

Received: 30 July 2020; Accepted: 24 August 2020; Published: 27 August 2020

**Abstract:** Natural and synthetic esters are liquids characterized by insulating properties, high flash point, and biodegradability. For this reason, they are more and more often used as an alternative to conventional mineral oils. Esters are used to fill new or operating transformers previously filled with mineral oil (retrofilling). It is technically unfeasible to completely remove mineral oil from a transformer. Its small residues create with esters a mixture with features significantly different from those of the base liquids. This article presents electrostatic charging tendency (ECT) tests for mixtures of fresh and aged Trafo EN mineral oil with Envirotemp FR3 natural ester from the retrofilling point of view. Under unfavorable conditions, the flow electrification phenomenon can damage the solid insulation in transformers with forced oil circulation. The ECT of the insulating liquids has been specified using the volume density of the *q<sup>w</sup>* charge. This parameter has been determined using the Abedian–Sonin model on the basis of the electrification current measured in the flow system, as well as selected physicochemical properties of the liquids. It was shown that ECT is strongly dependent on the type of insulating liquid and pipe material, as well as the composition of the mixtures. The most important finding from the research is that a small amount (up to 10%) of fresh and aged mineral oil is effective in reducing the ECT of Envirotemp FR3 natural ester.

**Keywords:** insulating liquids; mineral oil; natural ester; synthetic ester; dielectric liquid mixtures; retrofilling of power transformers; streaming electrification; ECT; insulation aging; insulation diagnostics

#### **1. Introduction**

The basic request of liquid dielectrics applied in power transformers is to ensure good electrical insulation and to remove heat effectively. The insulation liquids also improve the strength of cellulose paper, make it easier to extinguish electrical arcs, and protect the solid insulation against moisture and air. Most transformers being manufactured in the world are filled with mineral oils for economic reasons. The key disadvantages of mineral oils are their limited resistance to oxidation, high toxicity and explosiveness, and poor biodegradability [1]. In recent years, due to fire protection regulations and environmental protection considerations, alternative insulating liquids are becoming more popular. Among those, the most important ones include natural and synthetic esters. The physicochemical, electrical, thermal properties, and the environmental impact of mineral oils and esters used in transformers are relatively well known and described in the relevant literature [2–6]. For many years, electrostatic charging tendency (ECT) tests have been performed for mixtures of pure hydrocarbons, mineral oils, and liquid esters [7–14]. The process of removing mineral oil from a transformer and then refilling it with another insulating liquid is called retrofilling [15–18]. It is technically impossible, however, to completely remove mineral oil from a transformer. Its small amount (4–7%) usually remains within the paper insulation, and, to a small extent, in hardly accessible places of the transformer

tank [19]. As a consequence, in the transformer, a mixture of two insulating liquids with unknown properties is formed; they may also vary during the transformer's operation. In many scientific facilities around the world, intensive tests are being performed regarding the features of insulating liquid mixtures in terms of retrofilling. Beroual et al. [20] presented a comparative study of AC and DC breakdown voltage of naphthenic mineral oil, Midel 7131 synthetic ester, Envirotemp FR3 natural ester, and different mixtures based on these liquids. It showed that the ester oils always have a significantly higher breakdown voltage than mineral oil. The addition of natural or synthetic ester considerably increases the breakdown voltage of mineral oil. These authors demonstrated that transformer refilling can be considered with mixtures composed of mineral oil (20%) and ester oil (80%). Yu et al. [21] presented the research results of the physicochemical and dielectric properties of insulating mixtures based on Karamay No. 25 mineral oil and Envirotemp FR3 natural ester. The research showed that with an increase of the natural ester content in the mixture, the dynamic viscosity, acidy, pour point, and AC breakdown voltage increased. The authors observed that the fire point of the mixtures was similar to mineral oil, while the flash point increased by 11.4%. Hamadi et al. [22] presented a comparative study of the electrical and thermal stability behavior of Borak 22 and Midel 7131 synthetic ester mixtures. The authors showed that mixing synthetic ester with mineral oil efficiently reduced the aging rate. Dombek and Gielniak [23] presented the research results of the flash point, fire point, net calorific value, breakdown voltage, relative permittivity, dissipation factor, and conductivity of mixtures of the Nynas Draco mineral oil and Midel 7131 synthetic ester with Envirotemp FR3 natural ester. It was shown that the content of the mixture significantly determined the change of the tested parameters. Zdanowski and Maleska [24] observed a high correlation between the electrification current and the composition of mixtures of Trafo EN mineral oil and Midel 1204 natural ester and Midel 7131 synthetic ester. The authors found big differences in the form of the current characteristics depending on whether the oil used was fresh or aged. Rajab et al. [25] presented research on ECT results of PFAE (palm fatty acid ester) and mineral oil mixtures at various percentage ratios of PFAE. The authors showed that electrostatic charging tendency increases with the percentage ratio of PFAE to 80% but then decreases for the liquid containing only PFAE. The purpose of this paper was to specify the ECT of mixtures which may be formed in a transformer as a result of replacing mineral oil with Envirotemp FR3 natural ester. The most important goal of the work was the most favorable range of admixing Envirotemp FR3 natural ester to Trafo EN mineral oil specified for retrofilling power transformers. In the first stage of the study, the impact of the hydrodynamic conditions and the physicochemical properties of the liquids on selected parameters of the Abedian–Sonin electrification model were analyzed. In the next stage, the electrification current of the liquids in a flow system was measured with pipes made of different materials. Then, the volume density of the *q<sup>w</sup>* charge, designating the ECT of the insulating liquids, was determined.

#### **2. Materials and Methods**

The base liquids used for the research were Trafo EN mineral oil (MO) produced by Orlen Oil (Kraków, Poland) and Envirotemp FR3 natural ester (NE) produced by Cargil (Minneapolis, MN, USA). The mineral oil was subject to accelerated thermal aging in accordance with IEC 1125 standard (Method C: 164 h, 120 ◦C, cooper—1144 cm<sup>2</sup> /kg of oil, air—0.15 l/h). The mixtures of oil and ester were prepared at ambient temperature and then seasoned for a month in tightly sealed bottles with a capacity of 1000 mL. The volumetric composition of the mixtures varied every 10%. Density (ρ) was marked with a universal glass areometer (ISO 3675). Kinematic viscosity (ν*<sup>k</sup>* ) was measured with a Brookfield DV-II+Pro viscometer (ISO 2555). Conductivity (σ) was determined based on the resistivity measurement with a three-terminal capacitor and MR0-4c meter (IEC 60247). Relative dielectric constant (ε*r*) was determined based on the electric capacity measurement with three-terminal capacitor and Hioki 3522-50 LCR HiTester (IEC 60247). Molecular diffusion coefficient (*Dm*) was

determined according to Equation (1) given by Adamczewski [26]. The main characteristics of the insulating liquids used are given in Tables 1 and 2.

$$D\_m = \frac{3.93 \cdot 10^{-14} \cdot T}{\upsilon\_k \rho} \tag{1}$$

where *T*—liquid temperature, ν*k*—liquid kinematic viscosity and ρ—liquid density.


**Table 1.** Properties of Envirotemp FR3 natural ester and fresh Trafo EN mineral oil mixtures (20 ◦C).

**Table 2.** Properties of Envirotemp FR3 natural ester and aged Trafo EN mineral oil mixtures (20 ◦C).


Figure 1 is the diagram of the flow system for measuring the electrification current of insulating liquids. The liquid was electrified as a result of flowing from the top container through the pipe down to the insulated bottom container placed in a Faraday cage. The electrification current was measured with a Keithley 6517A electrometer. The flow speed (0.34–1.75 m/s) was adjustable by changing the gas (nitrogen) pressure in the top tank. The time of flow (120 s) was determined with a solenoid valve. The temperature (20 ◦C) was stabilized using a heater with a thermostat. After the measurement had been completed, the liquid was transported from the bottom container to the top container by means of a pump. The lower container (max. 5 l) was made of acid-resistant steel. The point on the current characteristic is the average of 300 values obtained from five measurement series, carried out during 120 s. Error bars were determined using the electrification current average, standard deviation, and α = 0.05 significance level. The measuring pipes with a length of 400 mm and a diameter of 4 mm were made of aluminum, Tertrans N cellulose paper produced by Tervakoski Oy (Tervakoski, Finland), and Nomex paper produced by Dupont (Wilmington, DE, USA). The measurement process was controlled by means of a dedicated software [27] installed on a portable computer.

**Figure 1.** Flow system for the investigation of electrification current of insulation liquids: 1—upper container with liquid, 2—solenoid valve, 3—measuring pipe, 4—Faraday cage with lower container, 5—Keithley 6517A electrometer, 6—portable computer, N—nitrogen, H—heater with thermostat, and P—pump.

ஶ ௪ ଶ 1 − + 2<sup>ଶ</sup> Insulating liquid electrification in a flow system is a very complex process. The phenomena that take place at the time are described using the electrification model prepared by Abedian and Sonin [28]. The measure of the ECT of liquid dielectrics is the volume density of the *q<sup>w</sup>* charge. The *q<sup>w</sup>* parameter is determined using the dependencies (2) and (3):

$$\frac{I\_{\infty}}{q\_{\text{w}}\pi R^{2}v} = \text{Re}\frac{\tau\_{\text{w}}\lambda^{2}}{\rho v^{2}R^{2}} \left[1 - \frac{\frac{\delta}{\lambda}}{\left[\sinh\left(\frac{\delta}{\lambda}\right)\right]}\right] + \frac{\frac{\delta}{\lambda}}{\sin\ln\left(\frac{\delta}{\lambda}\right)} \left[\frac{2\lambda^{2}}{1 + R\frac{\delta}{2\lambda^{2}}}\right] \tag{2}$$

$$I = I\_{\infty} \left[ 1 - \mathbf{e}^{-\frac{\ell}{L}} \right] \tag{3}$$

 = <sup>௪</sup> = 8 The following are the equations that describe the Reynolds number (4), the shearing stress (5), the laminar sublayer thickness (6), and the Debye length (7):

=

$$\text{Re}\begin{aligned} \text{Re}\begin{aligned} \text{e} & \stackrel{\text{e}}{=} 2\text{Re}\text{v} \\ \text{v} & \text{e} \end{aligned} \tag{4} \end{aligned} \tag{4}$$

$$
\sigma\_w = \frac{8\rho v}{Re} \tag{5}
$$

$$\delta = \frac{A\nu\_k}{S^\delta \left(\frac{\tau\_w}{\rho}\right)^{0.5}}\tag{6}$$

$$
\lambda = \sqrt{\frac{D\_m \varepsilon\_0 \varepsilon\_r}{\sigma}} \tag{7}
$$

− − where *I*∞—electrification current for infinite pipe length, *qw*—volume charge density on the phase border, *R*—pipe radius, *v*—average liquid velocity, *Re*—Reynolds number, τ*w*—shearing stress, λ—Debye length, ν*k*—liquid kinematic viscosity, ρ—liquid density, δ—laminar sublayer thickness, *I*—electrification current for any pipe length, *L*—characteristic length of the pipe, *l*—length of the pipe, *Dm*—molecular diffusion coefficient, ε*0*—vacuum electric permittivity, ε*r*—relative dielectric constant of liquid, *A, C*—constant (*A* = 11.7; *C* = 3), and *S*—Schmidt number (*S* = ν*<sup>k</sup>* /*Dm*).

#### **3. Results**

∞

Based on the data from Tables 1 and 2, it was found that the aging processes did not cause significant changes in the density and relative dielectric constant of the Trafo EN mineral oil (below 1%). It was observed that the kinematic viscosity increased by about 9% and the conductivity by nearly two

orders of magnitude (from 7.97 10−<sup>13</sup> to 1.33 10−11). The molecular diffusion coefficient decreased by about 10%. The change in the composition of the mineral oil and natural ester mixture caused a linear decrease in density, relative dielectric constant, and a non-linear decrease in kinematic viscosity and molecular diffusion coefficient. The conductivity when using fresh mineral oil in the mixture decreased non-linearly. When using aged oil, the conductivity increased non-linearly. From the analysis of physicochemical properties, it can be concluded that the viscosity and conductivity may have the greatest influence on the ECT of the insulating liquids.

The Reynolds number (*Re*), the shearing stress (τ*w*), and the laminar sublayer thickness (δ) are parameters that describe synthetically the impact of the hydrodynamic conditions and the physicochemical properties of the liquids on the occurring electrification processes. A change in the flow rate of fresh Trafo EN oil and Envirotem FR3 natural ester between 0.34–1.75 m/s causes a linear growth in the Reynolds number (Figure 2a), in the shearing stress (Figure 2b), and a non-linear drop in the thickness of the laminar sublayer (Figure 2c). The Debye length (λ) characterizes the distribution of charges in the laminar sublayer. The λ parameter does not depend on the hydrodynamic conditions and only on the relative electrical permittivity, conductivity, and the molecular diffusion coefficient of the liquid (Figure 2d). On the basis of the Reynolds number, the type of flow (laminar or turbulent) is determined. The parameter *Re* for both liquids does not exceed the value of 2300, which indicates laminar flow. The shearing stress determines the thickness of the laminar sublayer, through which the *q<sup>w</sup>* charge is diffused from the electrical double layer area into the volume of the liquid. An increase in the value of parameter τ*<sup>w</sup>* reduces the thickness of the laminar sublayer and, thus, intensifies the process of the electrification current generation. The differing values of the parameters *Re*, τ*w*, and δ result from the difference in the viscosity and density of the mineral oil and the natural ester. *τ δ λ λ τ τ δ*

**Figure 2.** Selected parameters of the Abedian–Sonin model vs. flow velocity of insulating liquids: (**a**) Reynolds number; (**b**) shearing stress; (**c**) laminar sublayer thickness; (**d**) Debye length.

A percentage change in the content of oil and ester in the mixtures results in a non-linear increase in the Reynolds number (Figure 3a), the laminar sublayer thickness (Figure 3c), the Debye length (Figure 3d), and a non-linear drop in the shearing stress (Figure 3b). It results from the model analyses that the hydrodynamic conditions and the physicochemical properties of the liquids substantially affect the parameters of the Abedian–Sonin model in the flow system and, as a consequence, contribute to the generation of the *q<sup>w</sup>* charge, which is the source of the flowing electrification current. Figure 4a shows the electrification current vs. flow time of fresh Trafo EN mineral oil through the aluminum pipe. The tests showed that the electrification current stabilized after about 20 s from the start of the measurement procedure. Figure 4b presents sample dependencies between the electrification current in fresh and aged Trafo EN mineral oil and Envirotemp FR3 natural ester and the speed of flowing (0.34–1.75 m/s) through an aluminum pipe. The registered characteristics are linear. The study demonstrated that natural ester electrified more than mineral oil. Figure 4c presents the impact of the flow rate of the liquids being studied on the change in the volume density of the *q<sup>w</sup>* charge. The experimental tests confirm the assumptions of the Abedian–Sonin model that the *q<sup>w</sup>* parameter does not depend on the hydrodynamic conditions. For this reason, it can be used as a material indicator for determining and comparing the ECT of insulating liquids.

**Figure 3.** Selected parameters of the Abedian–Sonin model vs. mixture content: (**a**) Reynolds number; (**b**) shearing stress; (**c**) laminar sublayer thickness; (**d**) Debye length.

**Figure 4.** (**a**) Electrification current vs. flow time of Trafo EN mineral oil; (**b**) Electrification current and (**c**) volume charge density *q<sup>w</sup>* vs. flow velocity of insulating liquids through an aluminum pipe.

Figure 5a,b presents the impact of the percentage content of different components in the mixtures on the generation of electrification current. In the measurements, a cellulose, an aramid, and an aluminum pipe were used. The flow rate was 0.34 m/s, and the temperature was 20 ◦C. The conducted research has shown that the type of pipe has a large impact on the electrification current. This is due to the type and surface roughness of the material used [9]. In both types of mixtures, a high correlation between the electrification current and the composition of the mixture and the type of the measuring pipe material was observed. In addition, a high correlation between the shape of the current characteristics and the type of mineral oil applied is present (fresh or aged oil). In the former case (Figure 5a), an increase in the concentration of fresh mineral oil in the mixtures decreases the electrification current, and its significant increase takes place. The current characteristics reach the maximum in the case of a mixture that is composed of 80% fresh mineral oil and 20% natural ester. Any further increase in the share of oil in the mixture results in a rapid drop in the electrification current. In the second case (Figure 5b), it was concluded that a small amount of aged mineral oil (up to 10%) significantly reduced the electrification of natural ester. Any further increase in the content of aged oil in the mixtures does not lead to significant changes in the generation of electrification current.

**Figure 5.** Electrification current vs. mixing of Envirotem FR3 natural ester with (**a**) fresh and (**b**) aged Trafo EN mineral oil.

Similarly, Figure 6a,b presents the impact of the mixtures' composition on the volume density of the *q<sup>w</sup>* charge. The differences in the waveform of the characteristics of the electrification current and the *q<sup>w</sup>* charge for both types of mixtures are a result of including the physicochemical properties of the liquids in the electrification model. The study proved that, in order to determine and compare the ECT of insulating liquids, it is necessary to know both their electrification current and their physicochemical properties. In order to visualize better the differences between the electrification current values measured and the *q<sup>w</sup>* charge values calculated from the model, bar charts have been prepared (Figure 7a,b).

**Figure 6.** Volume charge density *q<sup>w</sup>* vs. mixing of Envirotem FR3 natural ester with (**a**) fresh and (**b**) aged Trafo EN mineral oil.

**Figure 7.** (**a**) Electrification current and (**b**) volume charge density *q<sup>w</sup>* vs. pipe material: 1—fresh Trafo EN, 2—aged Trafo EN, 3—Envirotemp FR3, 4—20% (NE)—80% (fresh MO) mixture.

#### **4. Conclusions**

The purpose of this paper was to determine the ECT of mixtures of traditional mineral oil with natural esters in terms of retrofilling power transformers. For the experiment, it was proposed to use fresh and aged Trafo EN mineral oil and Envirotemp FR3 natural ester. Initially, selected parameters of the Abedian–Sonin electrification model were analyzed depending on the flow rate of the fluid and on the mixtures' composition. Then, the electrification current of the liquids was measured in a flow system. The ECT of the liquids was determined on the basis of the volume density of *q<sup>w</sup>* charge results. This study demonstrated that natural ester electrified more intensely than both fresh and aged mineral oil. In addition, it was concluded that the ECT of the liquids was the highest when flowing through an aluminum pipe and the lowest in a cellulose pipe. The ECT of the mixtures depends substantially on the percentage content of different components and the type of mineral oil applied (fresh or aged). When using fresh oil in the mixtures, the characteristic minimum (at 10% of oil) and maximum (at 80% of oil) value of the *q<sup>w</sup>* charge is observed. When aged oil is applied, a non-linear drop in the *q<sup>w</sup>* charge value takes place regardless of the percentage share of both liquids in the mixtures. Comparing the change in physicochemical parameters (Tables 1 and 2) and the electrification current, no significant correlation can be seen. The change in the composition of the mixture causes minima and maxima in the characteristics of the electrification current, which cannot be seen in the case of, e.g., viscosity and conductivity. Therefore, it cannot be clearly stated which property of the insulating liquid has the greatest influence on the ECT. The most important conclusion from the study conducted is the observation that a small amount of fresh or aged mineral oil (up to 10%) significantly reduces the ECT of Envirotemp FR3 natural ester, which is to the advantage of retrofilling, making it possible to increase the efficiency and operational safety of power transformers.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The author declares no conflict of interest.

### **References**


© 2020 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Statistical Analysis of AC Dielectric Strength of Natural Ester-Based ZnO Nanofluids**

**Hidir Duzkaya 1,2,\* and Abderrahmane Beroual <sup>2</sup>**


**Abstract:** Due to environmental concerns and increased energy demand, natural esters are among the alternatives to mineral oils in transformers. This study examines the electrical behavior of natural ester-based ZnO nanofluids at different concentrations in the range of 0.05–0.4 g/L. AC breakdown voltages are measured in a horizontally positioned sphere–sphere electrode system according to IEC 60156 specifications. The measurement data are analyzed using Weibull and normal distribution functions. Breakdown voltages with 1%, 10% and 50% probability are also estimated, these probabilities being of great interest for the design of power electrical components. Experimental results show that AC breakdown voltage increases with the concentration of ZnO nanoparticles, except for the concentration of 0.05 and 0.4 g/L of ZnO. Moreover, breakdown voltages at 1% and 10% probability increase by 22.7% and 13.2% when adding 0.1 g/L ZnO to natural ester, respectively.

**Keywords:** naturel ester oil; nanofluids; zinc oxide; AC breakdown voltage; Weibull distribution; normal distribution

**Citation:** Duzkaya, H.; Beroual, A. Statistical Analysis of AC Dielectric Strength of Natural Ester-Based ZnO Nanofluids. *Energies* **2021**, *14*, 99. https://dx.doi.org/ 10.3390/ en14010099

Received: 30 November 2020 Accepted: 24 December 2020 Published: 27 December 2020

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/ licenses/by/4.0/).

#### **1. Introduction**

Insulating liquids are widely used in insulating systems for high voltage (HV) electrical components such as transformers, cables, power capacitors, reactors, circuit breakers, bushings and tap changers [1]. Insulation and heat transfer are among the main requirements that these liquids have to ensure in HV power transformers, these latter being indispensable components for the transmission and distribution electrical energy systems. In HV transformers, 75% of total faults are caused by insulation problems. These transformer failures reduce the expected life of transformers by almost half [2]. Note that the insulation system consists of insulating oil (transformer oil) and solid insulating (paper and pressboard) [3]. The main functions of transformer oils are electrical insulation, protection of solid insulation against air and moisture, improvement of solid insulation performance by penetrating cellulose, protection against corrosion and cooling [3].

The most commonly used insulating liquids in transformers are mineral oils. These latter have been marketed and used since the end of the 19th century for their relatively low cost, dielectric and cooling properties, and compatibility with cellulose-based solid insulation materials and availability [4]. Despite these common advantages, mineral oils have significant disadvantages such as flammability, low biodegradability, low moisture tolerance and corrosive sulphur compounds [5]. Low flash and fire temperatures raise heat protection problems and therefore require measures such as fire safety, firewalls and deluge systems [6]. Mineral oils consisting of different hydrocarbon compounds are a by-product of the oil industry. Oil resources will eventually be depleted; some estimates predict that oil shortages may emerge in the middle of the twenty-first century [4]. This turns out to be an important threat to the insulating liquids industry, where several billion liters are used [1]. Another environmental problem of mineral oils is their low biodegradability, below 30% for 28 days [7]. In the event of a leak or spill after an accident, it pollutes the soil and groundwater and turns into an important threat to humans and the ecosystem.

In order to avoid the problems related to these disadvantages, many studies of alternative insulating liquids were launched over more than forty years [8]. Among the requirements that these alternative liquids are expected to meet are a high dielectric strength, a good heat transfer, improved fire safety, good sustainability, environmentally friendly and extended service life [2,9]. The protection of the environment, which has become a demand/requirement nowadays, is a deciding criterion for alternative transformer oils [8]. Environmentally friendly transformer oil is defined by high biodegradability and low toxicity [2].

Alternatives to mineral oils include many types of insulating liquids in the categories of high molecular weight hydrocarbons, synthetic and natural esters [1,10]. Natural esters obtained from plants such as rapeseed, soybean, sunflower, olive, palm and jatropha, consist mainly of triglycerides, which contain unsaturated fatty acids [1,4,11]. The advantages of natural esters compared to mineral oils are high flash and fire temperatures, almost completely biodegradability, non-toxicity, high dielectric strength and high moisture tolerance [7,9]. Natural esters with fire temperatures above 300 ◦C can be used in applications where there is a risk of fire without taking special safety precautions [1]. Power system equipment using these less flammable oils can be positioned at lower separation/safety distances [6]. Natural esters that dissolve in nature within 28 days with a rate of over 95% also successfully meet environmental requirements [4]. These oils have higher dielectric strength than mineral oils. Due to this feature, binary mixtures with mineral oils increase the breakdown voltage in power system equipment such as transformers [7]. These properties have made it so that natural esters have been used in transformer and capacitor applications since the early 1990s [11]. The use of natural esters is increasing especially in coastal areas where transformer oil can be contaminated with water and in applications where fire risk can cause great economic damage [1,11].

The major disadvantages of natural esters are increased dielectric dissipation factor (loss factor, tan δ) at high temperatures, high pour temperature, high viscosity and poor oxidation stability [12,13]. The pour temperature is the lowest temperature at which liquid materials can maintain their fluidity properties. High pour temperature turns into a disadvantage in terms of the use of natural esters in cold climatic conditions. These disadvantages can lead to exceeding standard limits in terms of thermal, loss and electrical aspects. For example, Fofana et al. [14] have found that when the ratio of natural ester in the binary mixture with mineral oils is more than 50%, the density and viscosity parameters of the mixture exceed the standard limits. Hermetically sealed applications that prevent contact of natural esters with moisture and oxygen are widely recommended to overcome these problems [13]. In addition, reducing the ratio of unsaturated fatty acids with esterification processes of these oils is another alternative approach [11].

To circumvent and resolve the drawbacks that natural esters present, and in general to improve the thermal and electrical characteristics of insulating liquids, nanoparticles (NP) were introduced into these liquids in the mid-1990s [15]. Originally, nanoparticle-added fluids or nanofluids (NFs) aimed mainly at improving thermal characteristics such as diffusivity, conductivity, convective coefficient and heat transfer [16]. In addition to the fact that they improve thermal properties, some nanoparticles also make it possible to increase the dielectric strength of fluids. Therefore, NFs constitute ideal alternative insulators for oil-filled high voltage applications [17]. The advantages of NP-added transformer oils include better AC, DC and impulse breakdown performances, better partial discharge characteristics, less sensitivity to moisture, prolonged insulation and transformer life, increased thermal conductivity and better cooling of transformers [2].

Different types of NPs are used in the preparation of these nanofluids. Nanoparticles can be classified into three main categories: Conductive, semi-conductive and nonconductive. These classifications, which are defined in terms of electrical behavior, make it easier to define and discuss the breakdown mechanisms [16]. The most commonly

used NPs in transformer oils are titanium dioxide (TiO2), iron oxide (Fe3O4), aluminium oxide (Al2O3), silicon dioxide (SiO2) and zinc oxide (ZnO), respectively [18]. Apart from these, NPs such as copper oxide (CuO), fullerene (C60) and aluminium nitride (AlN) are also studied [1,19]. The AC and positive impulse breakdown voltages of these NP-added transformer oils could be improved by a factor of up to 50% [2,19,20]. This rate of increase in breakdown voltages differs depending on the type, size, shape and concentration of the NPs and the type of transformer oil [17]. The addition of Fe3O<sup>4</sup> to mineral oil can more than double the AC breakdown voltage [16].

Hanai et al. [21] observed that the AC breakdown voltages of ZnO-based mineral oils increased by up to 8.3% compared to pure mineral oil. Bakrutheen et al. [22] found that this increase is 40.6% at 0.075% concentration for a different mineral oil. Chen et al. [20] observed that AC, positive and negative polarity impulse breakdown voltages increased up to 30.2%, 18.9% and 35.8%, respectively, in FR3 oil with 0.4 g/L ZnO.

The breakdown voltage characteristic and mechanism of ZnO, a semiconductor nanoparticle such as TiO2, in insulating liquids has not been widely studied. Considering that almost 30% of the studies on natural ester-based nanofluids in the literature use TiO<sup>2</sup> [18], the examination of semiconductor ZnO nanoparticle doped nanofluids offers a potential innovation.

This study examines the AC breakdown voltage characteristics of natural ester-based ZnO nanofluids at concentrations of 0.05 to 0.4 g/L; the measurements are conducted according to the IEC 60,156 standard. The experimental results are analyzed with Weibull and normal distribution functions, and probabilities of 1%, 10% and 50% breakdown stresses are deduced.

#### **2. Experiment**

#### *2.1. Preparation of Nanofluids*

The natural ester MIDEL eN 1204 transformer oil used in this study is based on rapeseed. The physicochemical properties of this oil are shown in Table 1. The ZnO nanoparticles used in the preparation of the nanofluid are supplied from PlasmaChem Gmbh. The average diameter of these spherical particles is 25 ± 3.5 nm and the density is 5.606 g/cm<sup>3</sup> at 20 ◦C with 99.5% purity.

**Table 1.** Physicochemical properties of MIDEL eN 1204.


Two different methods are used in the preparation of NFs. In the one-step method, nanoparticles are synthesized and dispersed simultaneously in the base fluid. This method does not include the drying, storage and transportation of nanofluids. Therefore, agglomeration is minimized and fluid stability is improved [10]. Due to the high cost of large-scale production with this method, the two-step method is preferred for nanofluids based on transformer oils [2]. In this method, the fluid is firstly mixed with nanoparticles using a magnetic stirrer as depicted in Figure 1. After this step, surfactant is added to this solution and nanofluid is produced by ultrasonication [23]. In order to avoid the agglomeration problem, measurements are taken after the preparation of the nanofluid. Nanofluids prepared by the two-step method exhibit a homogeneous characteristic for several months

without any agglomeration problems [10]. This method can be used on an industrial scale for almost all nanofluids. In this study, oleic acid is used as surfactant.

‐

‐

‐ **Figure 1.** Diagram of two-step method for preparation of nanofluids (NFs).

‐ Due to the high surface energies and attractive/repulsive forces of nanoparticles, the NF can become unstable. Surfactant reduces the surface tension of the fluid and increases the immersion of NPs [2]. This mechanism is also defined as steric stabilization. Oleic acid is the most widely used in NFs prepared with transformer oils. Apart from oleic acid, long-chain hydrocarbons such as hexadecyl trimethyl ammonium bromide (CTAB), sorbitan esters and sodium dodecyl sulphate (SDS) are also used, but rarely [10,23].

‐ ‐ In this study, natural ester is purified by using a micro membrane filter and vacuum pump. ZnO NPs of five different concentrations ranging from 0.05 to 0.4 g/L are added to this fluid. Then, each sample is mixed with a magnetic stirrer for 30 min. Oleic acid is added to this solution and an ultrasonication process is applied for 2 h with an ultrasonic homogenizer. This process ensures that the NPs are homogeneously dispersed in the fluid and remain stable without aggregations/clusters. Sonics Vibra-cell sonicator used in ultrasonication has 500 W power rating, 20 kHz frequency and 60% amplitude. This ultrasonication process is applied in periods of 20 min with a 10 min waiting time between each to prevent the nanofluid from overheating. In order to eliminate possible humidity and micro air bubbles that develop during the preparation of the nanofluid, the nanofluid is kept in the oven and then under vacuum at a pressure of 1.0 Pa for 24 h.

#### *2.2. AC Breakdown Measurement*

‐ ‐ ‐ ‐ AC breakdown voltages of pure Midel eN-1204 and these natural ester-based ZnO nanofluids are measured using the BAUR DTA 100C, according to the IEC60156 [24] using a 400 mL test cell, 12.5 mm diameter electrodes horizontally positioned at a 2.50 ± 0.05 mm electrode gap. The rest time of each sample is 30 min in order to eliminate gas bubbles in the test cell. The voltage is increased with a rise rate of 2 kV/s until breakdown occurs. The time delay between each breakdown is 2 min and the number of measured breakdown voltages in each set is 6. In order to have sufficient data for statistical analysis, five series of six measurements each, i.e., a total of 30 measurements, are carried out on each type of nanofluid sample [5,8]. After the measurement, the test cell and electrodes are cleaned with ethanol. After this stage, it is washed with hot water at a temperature of 60–80 ◦C and dried in an oven at 60 ◦C for one hour. This procedure is compatible with approaches in similar measurement studies reported in the literature [4,5].

The AC breakdown voltage characteristics are analyzed with Weibull and normal distribution functions, and the withstand voltage levels at 1%, 10% and 50% are determined.

#### **3. Results and Discussions**

The measurement results taken to check the conformity or not of their distribution with the Weibull law and the normal law are presented in Figure 2. The mean and standard deviation of these measurements are calculated using Equations (1) and (2), respectively:

$$\overline{\mathcal{U}} = \frac{1}{n} \sum\_{i=1}^{n} \mathcal{U}\_i \tag{1}$$

‐

$$
\sigma = \sqrt{\frac{1}{n} \sum\_{i=1}^{n} \left( \mathcal{U}\_i - \overline{\mathcal{U}} \right)^2} \tag{2}
$$

**Figure 2.** Distribution of AC breakdown voltages of natural ester (NE) and NFs.

‐ ‐ It is noticed that the AC breakdown voltages of natural ester (NE) are reduced by 2.7% and 8.7% for nanofluids with a concentration of 0.05 and 0.4 g/L ZnO, respectively. This characteristic changes for 0.1, 0.2 and 0.3 g/L ZnO and increases by 5.8%, 5.8% and 5.1%, respectively.

 

 

Khaled and Beroual [8] examined the same natural ester-based Fe3O4, Al2O<sup>3</sup> and SiO<sup>2</sup> NFs and observed that the best improvement in breakdown voltages did not exceed 7%. The breakdown voltage was reduced by about 15% in the 0.05 g/L-added SiO<sup>2</sup> nanofluid [8].

The mean and standard deviation range of these breakdown voltage measurements at different concentrations are given in Figure 3. The ratio of standard deviation to mean breakdown voltage measurements is 5.9% to the maximum for pure naturel ester and 3.1% to the minimum for 0.1 g/L ZnO concentration. The growth of this ratio is linearly related to the difference between the measurements in each concentration set.

**Figure 3.** Average breakdown voltages of NE and ZnO nanofluids for different concentrations.

‐

‐

Histogram charts of NE and nanofluids are given in Figure 4. These charts include breakdown frequency at different voltage levels, mean value of breakdown voltage and standard deviation. In terms of average breakdown voltage, 0.1, 0.2 and 0.3 g/L ZnO NFs perform better than NE.

‐ ‐

‐

‐

α

**Figure 4.** Histograms of NE and ZnO nanofluids for different concentrations.

In order to statistically analyze the probability of breakdown voltage by adhering to these measurements, the Weibull and normal distribution should be tested using a hypothesis. In the test of this hypothesis, which questions the distribution of measurements in the 5% significance level (α = 0.05), Anderson–Darling and Shapiro–Wilk tests are used for Weibull and normal distributions, respectively [7,8].

The Anderson–Darling normality test can examine measurement data without grouping and is very sensitive to distributions in the tail region rather than the median [25]. The Shapiro–Wilk test is a regression-correlation based test using a sequential sample. This test, in which the normality of the samples is tested, is consistent in all alternative datasets up to 50 samples [26]. The *p*-value is the probability of making an error in testing the hypothesis that the measurement data conforms with the statistical law [7].

A hypothesis is accepted if the *p*-value obtained in these tests is greater than the significance level. Under the condition that the hypothesis is accepted, the distribution of the measurements is defined as a statistical distribution and different probability levels can be estimated [27].

W, which is also defined as the test statistic, is evaluated differently for both tests. The hypothesis is rejected due to a too large W value in the Anderson–Darling test. In order for the hypothesis to be accepted, W should be below 1.5786 at the 0.05 significance level [25]. In the Shapiro–Wilk test, in order for the hypothesis to be accepted at the 95% confidence interval, W should be in the range of 0.9303–1.0000 depending on the number of samples used [26]. The test statistics provide the necessary conditions for both tests and the normal distribution hypothesis is accepted.

The Weibull distribution of the measurements given in Figure 2 is examined using the Anderson–Darling test. According to this test, the distributions of measurements for natural ester and natural ester-based ZnO nanofluid samples are within the acceptable significance level, see Table 2. The acceptance of hypothesis tests of conformity for all samples allows the analysis of the breakdown voltage characteristics using the Weibull distribution function.


**Table 2.** Hypothesis test of conformity to Weibull distribution of NE and NFs.

The probability curves due to the Weibull distribution are shown in Figure 5. In the 50% probability region, 0.1 to 0.3 g/L ZnO nanofluids have similar AC breakdown voltage characteristic. In this region, the AC breakdown voltage of 0.05 and 0.4 g ZnO nanofluids worsens. In the 10% probability region, the best breakdown voltage is in the 0.1 g/L ZnO nanofluid sample.

**Figure 5.** Weibull probability of NE and ZnO nanofluids for different concentrations.

α The same sets of measurements are used to examine the compliance with the normal distribution function. The statistical distributions of these measurements are examined with the Shapiro–Wilk test and the hypothesis that the normal distribution is distributed in the significance level (α = 0.05) is accepted in all samples, see Table 3.


**Table 3.** Hypothesis test of conformity to normal distribution of NE and NFs.

‐ ‐ ‐ Using these normal distribution functions in which experimental measurements are analyzed, 1%, 10% and 50% probability breakdown voltages can be estimated, see Table 4. The statistical nature of the breakdown voltages complicates the design of power system equipment. In order to overcome these difficulties, the withstand voltages calculated as a statistical parameter are defined with different possibilities. The withstand voltage of the insulation is not the average value of the breakdown voltage, but as a statistical variable, a low probability of breakdown voltage, such as 1% or 10% [28]. These critical risk levels for design safety are widely studied in the literature [7,28,29]. The 1% probability of

> ‐ ‐

> ‐

breakdown voltage is a safety factor in the design of electrical equipment and is defined as the voltage limit for operation in the safety margin [7,28].


**Table 4.** AC breakdown withstand voltages at different probabilities for NE and NFs.

These breakdown probabilities are also considered as withstand stress levels. NFs other than that with 0.4 g/L ZnO outperform natural ester in terms of 1% withstand voltage probability. Specifically, a 22.7% increase in 0.1 g/L sample significantly improves the withstand voltage performance for this critical parameter. Similarly, this value increases for 0.2 and 0.3 g/L ZnO samples; the rate of increase is 13.2% and 11.1%, respectively. In the withstand voltage where there is a 10% probability of breakdown, the 0.1 g/L sample shows the best performance as in the previous probability level. The increase rate in this withstand voltage for 0.1 g/L nanofluid is 13%.

The electrical insulation characteristics of nanofluids with ZnO additives are improved compared to the natural ester in terms of withstand breakdown voltages at 1% and 10% probabilities.

Mechanisms of breakdown characteristics of nanofluids are not clearly defined and remain the subject of controversy. The electrical conductivity of nanoparticles seems to be an important parameter in the explanation of this mechanism. Conductive nanoparticles capture the rapidly moving electron in the fluid and turn into slow negatively charged nanoparticles. Streamer propagation slows down and therefore the breakdown voltage increases with this mechanism [16]. Conductive and nonconductive nanoparticles trap electrons by charge induction polarization, respectively. These scavenger nanoparticles reduce free electrons moving in the fluid [5].

ZnO is a semiconductor nanoparticle that traps high mobility electrons. It slows down the electrons responsible for streamer development with trapping and de-trapping processes [2]. In nanofluids using semiconductor nanoparticles, the surface trap density and charge dissipation velocity are 2.5 and 4 times higher compared to pure transformer oil, respectively [30]. With the effect of these mechanisms, AC, DC and impulse breakdown voltages can increase by 20% compared to pure oil [30]. The capture of electrons by scavenger nanoparticles increases the initial threshold voltage of the streamer, and therefore more energetic breakdown mechanisms emerge [16,27]. Due to the semiconductor property of the ZnO, the mechanism defined as bridging or tunnelling develops when the nanoparticle density exceeds a certain concentration [19,21]. In this case, layers adjacent to nanoparticles separate insulating oil and these layers act as a conductor in a very high electric field [20]. Approximately 10% reduction of breakdown voltage at a concentration of 0.4 g/L can be explained by this mechanism.

Breakdown voltages in nanofluids using ZnO nanoparticles can increase by 8.3% to 40.6% in different concentrations and fluids [21,22]. The positive and negative impulse breakdown voltage of natural ester-based ZnO nanofluids increased by 19.8% and 35.8%, respectively, for the 0.4 g/L ZnO concentration [20]. Despite the findings in these studies, the AC breakdown characteristics and withstand voltages of natural ester-based ZnO nanofluids are presented in detail with this study. The increase in AC breakdown and withstand voltages can be up to 5.8% and 22.7% for 0.1 g/L ZnO concentration, respectively.

The AC breakdown voltage averages of mineral oils commonly used in power transformers measured using the same method are 38.5 kV [16], 39.0 kV [28] and 51.6 kV [7]. The AC breakdown voltage measurement averages of synthetic esters under the same conditions are 47.0 kV [28] and 60.03 kV [5]. The electrical insulation characteristic of the natural ester, whose average AC breakdown voltage is measured as 66.67 kV, has a better performance than mineral oils and synthetic esters. This insulating performance can be improved with the addition of ZnO nanoparticles and provides a more reliable insulating medium alternative for the transformer.

The flash and fire points of natural esters are higher than mineral oils and synthetic esters. Due to these properties, the performance of power transformers using natural esters continues without deterioration even when exposed to high temperatures [6].

#### **4. Conclusions**

The main findings obtained in this study can be summarized as follows:


**Author Contributions:** Conceptualization, H.D. and A.B.; Data curation, H.D. and A.B.; Formal analysis, H.D. and A.B.; Investigation, H.D. and A.B.; Methodology, H.D. and A.B.; Project administration, A.B.; Resources, H.D. and A.B.; Software, H.D.; Supervision, A.B.; Validation, H.D. and A.B.; Visualization, H.D.; Writing-original draft, H.D.; Writing-review & editing, A.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data is contained within the article.

**Acknowledgments:** This work is supported by The Scientific and Technological Research Council of Turkey (TUBITAK) 2219 grant program.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **Investigation of Survival/Hazard Rate of Natural Ester Treated with Al2O<sup>3</sup> Nanoparticle for Power Transformer Liquid Dielectric**

**Raymon Antony Raj \* , Ravi Samikannu , Abid Yahya and Modisa Mosalaosi**

Department of Electrical, Computer and Telecommunications Engineering, Faculty of Engineering and Technology, Botswana International University of Science and Technology, Private Bag 16, Palapye Plot 10071, Botswana; ravis@biust.ac.bw (R.S.); yahyaa@biust.ac.bw (A.Y.); mosalaosim@biust.ac.bw (M.M.) **\*** Correspondence: raymonhve@gmail.com

**Abstract:** Increasing usage of petroleum-based insulating oils in electrical apparatus has led to increase in pollution and, at the same time, the oils adversely affect the life of electrical apparatus. This increases the demand of Mineral Oil (MO), which is on the verge of extinction and leads to conducting tests on natural esters. This work discusses dielectric endurance of Marula Oil (MRO), a natural ester modified using Conductive Nano Particle (CNP) to replace petroleum-based dielectric oils for power transformer applications. The Al2O<sup>3</sup> is a CNP that has a melting point of 2072 ◦C and a low charge relaxation time that allows time to quench free electrons during electrical discharge. Al2O<sup>3</sup> is blended with the MRO and Mineral Oil (MO) in different concentrations. The measured dielectric properties are transformed into mathematical equations using the Lagrange interpolation polynomial functions and compared with the predicted values either using Gaussian or Fourier distribution functions. Addition of Al2O<sup>3</sup> indicates that 0.75 g/L in MRO has an 80% survival rate and 20% hazard rate compared to MO which has 50% survival rate and 50% hazard rate. Considering the measured or interpolated values and the predicted values, they are used to identify the MRO and MO's optimum concentration produces better results. The test result confirms the enhancement of the breakdown voltage up to 64%, kinematic viscosity is lowered by up to 40% at 110 ◦C, and flash/fire points of MRO after Al2O<sup>3</sup> treatment enhanced to 14% and 23%. Hence the endurance of Al2O<sup>3</sup> in MRO proves to be effective against electrical, physical and thermal stress.

**Keywords:** power transformer; mineral oil; natural ester; interpolation; mathematical modeling

#### **1. Introduction**

The power transformer (PT) is vulnerable and is the most expensive power system network equipment. It provides functions such as stepping up and stepping down the voltage between electrical power generating stations. The PT works at 100% load for 24 h throughout the day whereas the distribution transformer works at 50% or 70% of the full load [1]. Increase in the power demand across the world is persistent and it is expected to be double in the forthcoming years as per the predicted data of International Energy Associations [2]. This demand has caused the rise in PT numbers in the power system network. The report released by allied market research indicates that countries in the Asia-Pacific (APAC) are the biggest market for the future transformer industries [3]. Here, a projected growth of USD 3 million is expected by 2025, which almost strikes a peak of 6.9% Compound Annual Growth Rate (CAGR) from the year 2020 [4]. This growth is brought about by the huge demand for power, rapid urbanization, and PT's replacement. The report published by the Allied Market Research estimated that the consumption of MO in 2014 was nearly 1437.8 million litres [5]. It further predicted that MO's sale was to reach 3.4 billion United States dollar by 2020 which represents a 6.3% increase from 2015. Upon extrapolating the data with the same growth rate up to 2030, the consumption of MO

**Citation:** Raj, R.A.; Samikannu, R.; Yahya, A.; Mosalaosi, M. Investigation of Survival/Hazard Rate of Natural Ester Treated with Al2O<sup>3</sup> Nanoparticle for Power Transformer Liquid Dielectric. *Energies* **2021**, *14*, 1510. https://doi.org/10.3390/en14051510

Academic Editor: Pawel Rozga

Received: 10 February 2021 Accepted: 1 March 2021 Published: 9 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

will be 2815.56 million litres. MO will dominate the insulating oil market up to the year 2025. Even though there are several alternatives to MO in the market, bio-based insulating fluids have their shortcomings on the dielectric properties [6]. Insulating oil is an integral part of the power utility with applications in power and distribution transformers, power cables, bushings, and circuit breakers. Natural ester proves to be a suitable candidate with higher moisture tolerance than MO [7]. The use of mineral oil in the transformer as a coolant and insulator is well documented [8,9]. MO's availability is limited since it is extracted from a non-renewable source. Its lack of biodegradability has also led to a debate regarding its usage in recent times [10]. Mineral oil is a fossil fuel (crude oil) made up of hydrocarbons and compounds which are refined through a distillation process by boiling and refining for suitable use in transformers [11]. Numerous oils, like paraffin, naphtha, aromatic, to name a few, are used in various ratios and components [12–14]. Silicones are inert synthetic liquids, known as polydimethylsiloxane, are thermally stable with a composition exhibiting electrical properties like mineral oils with a similar structure as the methyl organic group [15,16]. High molecular coolants, either natural or synthetic are classified by National Electric Code (NEC) as less inflammable with a fire-point below 300 ◦C [5]. Synthetics are generally polymerized olefins popularly known as polyalphaolefins or PAOs. Both silicones and high molecular coolants are comparable in performance with MO. However, they differ in their temperature endurance, viscosity, and pour point [16,17].

Esters are alternatives to the mineral oils and are generally either synthesized chemically from organic precursors or natural oils [18–20]. Researchers further added alternative insulating fluids from natural esters, food-grade synthetic (Butylated Hydroxy Toluene (BHT), Butylated Hydroxy Anisole (BHA), Propyl Gallate (P.G.)), natural antioxidants (α, β-Tocopherols (αT, βT)) and synergists (Citric Acid (C.A.), Ascorbic Acid (A.A.), Rosemary extracts (R.A.)) which contribute significantly [21]. Among the several synthetic esters, Pentaerythritol, a polyol, is found to have suitable dielectric properties, good thermal stability, low-temperature tolerance and more biodegradable as compared to mineral oils [22,23]. Natural esters compose of chemically stable fatty acids with high viscosity [24]. The commercialization of vegetable oil as coolant resulted from significant research to find a suitable, fully biodegradable insulating medium. The use of vegetable oil as biodegradable, moisture absorbent, coolant and insulator in transformers started as a research project as early as the 1990s. Vegetable oils are known to be highly biodegradable (>95%), less toxic, high flash and fire points (>300 ◦C) as compared to mineral oils [24]. Vegetable oils possess very high moisture and viscosity than MO. Vegetable oil is not stable like MO when subjected to temperature stresses.

The process of extracting vegetable oil starts with the process of seeding. The seeds are refined, bleached, and deodorized to form pure vegetable oils. The first step is to refine the oil using alkaline refinement to remove fatty acids, followed by bleaching to eliminate the coloring materials and the deodorization. The volatile and odoriferous materials are removed by using vacuum steam distillation at high temperatures [25]. Vegetable oils such as sunflower, rice bran, rapeseed, palm, coconut, and soybean, have been studied extensively on their dielectric nature for transformer insulation and cooling. Natural oil age monitoring in the transformers should be based on various dielectric properties such as dielectric strength, dissipation factors, acidity, dissipation factor [5,17]. The degradation rate of vegetable oil can be compared to that of mineral oil. The relationship between hottest spot temperature and long-term operations for both mineral oil and vegetable oil-based transformer are determined under medium voltages.

The work discusses statistical approach to understand the performance of the insulating fluid that is normally required when the field testing is complicated. Such analysis is meaningful in performing reliability analysis to determine the time to failure. Moreover, the statistical analysis helps to realize the temperament of insulating fluid under accelerated heating conditions. This approach used in the research is simple and it derives mathematical equations using interpolation function, which helps the power engineer to recognize the actual condition of the insulating oil without the need to test the oil at field

conditions. The functions are used to derive the bath-tub curves used in the reliability analysis, but in our case, we focus on the response of oils to nanoparticle and their stability to accelerated ageing.

#### **2. Key Materials of Research**

#### *2.1. Marula Oil (MRO)*

Marula or Sclerocarya birrea is a drought-resistant plant species, found in Southern Africa, Botswana. The tree can grow a diameter of 2 m and produces fruits [26]. The nuts are crushed to produce MRO, which has alcohol content range from 1% to 7% [27]. The oil contains amino acids, fatty acids, oleic and linoleic, flavonoids, catechins, and procyanidin and high amounts of antioxidants [27]. Oleic acid in MRO has a melting point of 360 ◦C, which helps the antioxidants to resist high-temperature stress. MRO naturally contains oleic acid, which shows high temperament against temperature stress. In addition, the oil contains vitamin-E (tocopherol), which naturally inhibits free electrons during the discharge phenomena [24]. Natural esters containing antioxidants helps to prevent formation of valence electrons during electron avalanche. Thus, natural ester retards ionization delays breakdown in oil. On the other hand, the procyanidin in the oil acts as an excellent antioxidant and inhibits reactive oxygen, thereby preventing the oil from ionization.

#### *2.2. Conductive Nanoparticle (CNP)*

The theory of nanoscience has its application as a drug carrier in various fields of engineering and health. These nano-sized particles (10 nm) are highly stable, and it is complicated to predict their physio-chemical characteristics. The application of CNP in the transformer liquid dielectric starts with the nanoparticle's ability to capture free electrons during the discharge or streamer application [28]. Discharge or streamer is initiated by the ionization; a process by which valence electrons are separated from shells and start bridging a conductive path between electrodes. This establishes a short-circuited path in liquid dielectric and results in dielectric breakdown [21,24]. The CNP traps the fast-moving electrons during the discharge mechanism and slows the process either by neutralizing or slowing down the ionization. This increases the breakdown strength of the liquid dielectric and reduces ageing. Unlike magnetic and semi-conductive nanoparticles, CNP is influenced by the direction of the electric field. This leads to CNP coming into action whenever there exists a strong field between the two electrodes [29]. Al2O<sup>3</sup> shows very low charge relaxation time, metal chelation, and electron scavenging properties. This retards free electrons that are responsible for electron avalanche, reducing energy density of avalanche chain. It is also known that CNP improves the breakdown voltage at a concentration of 0.5 g/L in the host fluid as mentioned by Raymon et al. [28]. Al2O<sup>3</sup> possesses amphoteric structure and it does not react with water in the presence of Natural Esters. Since Al2O<sup>3</sup> is slower to saturate in water, it is used for adsorption purification of oils as adsorbent to capture hydrocarbon impurities from the air. The effect of Al2O<sup>3</sup> has been well utilized in this study to understand the behavior in MRO and MO, and its suitability as an additive in liquid dielectric coolant for the transformer.

#### **3. Dielectric Behavior of Host Fluids**

MRO and inhibited MO are collected from Botswana and India, respectively. Both MRO and MO are considered as host fluids, and their dielectric characteristics are measured as per IEC (International Electrotechnical Commission) [30] and ASTM (American Society for Testing and Materials) International [31–34] standards and presented in Table 1.

The dielectric characteristics like breakdown voltage, kinematic viscosity and flash/fire point are considered in the statistical study. These parameters are much likely to influence other parameters of the liquid dielectric; hence, it is necessary to investigate the behavior analytically. The analysis involves the estimation of survival rate and hazard rate by populating the data points of the dielectric characteristics of base fluids as density and cumulative probability distributions.


**Table 1.** Dielectric characteristics of host fluids.

#### *Experimental Procedure of the Investigated Dielectric Properties*

The study makes use of the dielectric parameters such as breakdown voltage, kinematic viscosity, flash point, fire point to validate them with hazard function and survival function. Breakdown voltage of the liquid dielectric is defined as the maximum withstand voltage capacity of the liquid dielectric under standard room temperature and pressure. Usually, the breakdown voltage is measured by filling the measuring cup (test cup) with 500 mL of the liquid, spherical electrodes fully submerged in the liquid. Gap spacing between electrodes is 2.5 mm and 50 Hz ac test voltage is increased using the control knob at a rate of 2 kV/s until breakdown. The voltage is recorded as breakdown voltage of the sample. Between the successive measurements, a 2 min relaxation time is given, and the test cup is inspected for the carbon formation, then the oil is slowly stirred to avoid formation of bubbles. The test is repeated according to IEC 60156 standards and measurements are recorded [30]. Similarly, the kinematic viscosity is measured using a Redwood viscometer. Kinematic viscosity is defined as the shear stress of the liquid against its flow. The test cup of the Redwood viscometer is filled with the liquid and placed on the water bath, which is heated using the electric heater. The liquid temperature is measured using the thermometer setup and the orifice is controlled manually to drain the oil to a vessel placed under the viscometer. The time of flow is measured using the stopwatch and kinematic viscosity is calculated by considering the pipette and burette constants according to ASTM D445 standards [32]. On the other hand, the fire point and flash point are measured using the Pensky–Martens closed cup apparatus according to ASTM D93 standards [33]. The test cup is filled with the oil and placed in a water bath then heated using an electric heater. The temperature of the liquid is measured using a thermometer and flame is introduced to a small opening in the top surface of the test cup. At some temperature, the liquid begins producing vapour that can be a combustible source when oxygen is introduced to the source in the presence of a fire source. This phenomenon is termed as flash point (smoke point). Likewise, when the combustion becomes continuous for at least 5 s, it is called as fire point (ignition point). Although there are no direct connections between the dielectric properties, they can be individually affected by factors like concentration of the nanoparticles or temperature, which is used to describe the changes in the dielectric properties.

#### **4. Assessment of Breakdown Voltage of Host Fluids**

The dielectric breakdown voltage of MRO and MO is distributed as a density plot shown in Figure 1. Readings were taken two times a day for 30 successive days. During this period, the samples are stored in a dark container to protect the fluid from photo oxidation and dust particles. Density plot is a continuous plot of variables that are random in space. The breakdown strength of MRO is higher than that of MO, with MRO showing a wider range of breakdown voltage as compared to MO. MRO's breakdown voltage limits are from 36 kV to 47 kV while MO falls between 22 kV and 33 kV. The maximum strength of MRO breakdown voltage lies between 45 kV to 47 kV. On the other hand, MO shows endurance between 22 kV to 25 kV and 30 kV to 33 kV. This indicates that most of the breakdown strength's measured values lie within these indicated regions for both MRO and MO. Breakdown endurance is due to the presence of CNP that retards free electrons detached from valence shell and then reduces the energy density of the electron chain

between electrodes. Figure 2 illustrates the cumulative probability of random breakdown voltage of MRO and MO that occur at different voltage instances (sequential events). The sequential events are independent of each other, and it is critical not to have two events occur simultaneously [35].

**Figure 1.** Showing the different ranges of breakdown voltage at which, the MRO and MO have its % density.

**Figure 2.** Cumulative probability of breakdown instances of MRO showing wider range and MO showing linear range.

A wider range of breakdown voltage of the MRO indicates that it experiences a lesser effect on the successive breakdown events. The density plot also shows a quick recovery between breakdown events. Such recovery is due to the presence of natural antioxidants in MRO. Hence, it is hypothetical and practical to conclude that MRO has a lesser propensity to electrical stress. Reliability analysis such as survival function and hazard function are used in this study to understand and compare the endurance of the liquid dielectrics. Survival function determines the number of data points that survive over time or in other words the data points that fail in the expected duration of time. Survival function is determined using the Kaplan–Meier function. A simple way to calculate the distribution of the survival function is through a hypothesis which evaluates the risk over a constant time

i.e., λ(t) = λ. From the expression of distribution function; F(t) = 1 − exp(−λt), the survival function is; S(t) = exp(−λt), can be expressed as S(t) = 1 − F(t). Likewise, hazard function indicates the rate at which the data points experience hazard over time or in other words it is the likelihood or frequency of failure per unit time. The above survival and hazard functions are used to determine the reliability of the liquid dielectric. Such mathematical analysis is therefore essential to realize the bathtub function of the liquid dielectric. When the distribution becomes exponential, the hazard function becomes; λ(t) = F(t)/S(t) [36]. According to Figure 3, the proportion of MRO surviving 40 kV is 70%. On the other hand, 80% of the time MO survives only 26 kV and 20% of the time it survives 30 kV. MRO shows 50% survivalat 44 kVwhile MO is at 28 kV. This indicates that MRO has a superior survival rate than MO. From Figure 4, at 50% cumulative hazard, MRO is less likely to fail until it reaches a breakdown voltage of 45 kV. It is also evident that MO experiences a 50% hazard at 32 kV and above voltages. Both λ(t) and H(x) from Figures 3 and 4 shows some data exceeding the confidence bounds of MO. This clearly indicates that MRO is likely to follow the upper survival bounds. Most of the time, it is less likely to fail above 40 kV. Here, the Weibull parameters are obtained from the generated Figures 3 and 4 with a shape parameter "K > 1". This shows the rate of failure increases with time as ageing goes on with time. Similarly, the scale parameter "λ" is assumed to be greater than "0" to spread the distribution evenly.

Figure 5 shows the failure probability of MRO and MO for 30 days from the breakdown voltage tests as per IEC standard [30]. The probability of failure of the dielectric fluid is the ability of the fluid to show a constant breakdown strength during successive testing. The probability of failure is measured at the breakdown voltages of MRO (47 kV) and MO (32 kV). Typically, MRO shows no changes upto 5 days, while MO shows 40% probability to breakdown from day 1 of testing. By day 15, MO has 87% breakdown probability which is a 47% rise from day 1 while MRO has a 60% rise for the same period. There is an inflexion point on day 23, where MRO shows the same rate of failure as MO. MO has shown slower growth in failure rate even though its probability of failure was initially 40% higher than that of MRO. This indicates that electrical stress has a greater impact on MRO than MO from the inflexion point on day 23. In this case, MRO needs enhancement by adding a CNP such as Al2O3.

**Figure 3.** Survival function showing the cumulative distribution of MRO and MO, for any positive number "T" surviving over the time "t".

**Figure 4.** Cumulative hazard of MRO and MO showing the integral of hazard function H(x) = −ln(1 − f(x)) which shows the probability of failure at time x has given survival until time x.

**Figure 5.** Showing the endurance of breakdown voltage and probability of failure of MRO and MO for 30 days.

#### **5. Preparation of the Nanofluids**

The nanofluids are prepared by amalgamation of MRO and MO with nano sized Al2O<sup>3</sup> in different concentrations from 0.1 g/L to 2 g/L with concentration increase interval of 0.25 g/L. They are heated up to 100 ◦C and treated in an ultrasonication bath with 30 kHz as mixing frequency. The oil moisture is removed during the heating process. The nanofluid is collected in a closed container after the procedure and kept at room temperature out of reach of sunlight to avoid ultraviolet and photo oxidation. The nanofluid prepared using MRO shows a fine dispersion of Al2O<sup>3</sup> since it contains oleic acid which naturally allows the additive to float in the host fluid. Comparatively, the MO-based nanofluid shows agglomeration after 5 h and needs repeated dispersion using the ultrasonication bath.

#### *5.1. Nanoparticle Impact on the Breakdown Voltage*

The impact of nanoparticle on the breakdown voltage of MRO and MO is presented for varying concentrations starting from 0.1 g/L, and 0.25 g/L increment upto 2 g/L. The addition of Al2O<sup>3</sup> has resulted in improved breakdown voltage of the host fluids as seen from Table 2.


**Table 2.** Enhancement of Host Fluids using Al2O<sup>3</sup> Nanoparticle.

There is a constant rise in the breakdown voltage when Al2O<sup>3</sup> is added from the concentration of 0.1 g/L to 0.75 g/L and it begins to decline after further additions of Al2O3. This variation is seen from Table 2 where the highest enhancement of breakdown voltage is measured at 0.75 g/L with 26.6% for MRO and with 31.2% for MO. A composite cross-section of liquid-nanoparticle helps to reduce the random nature of breakdown of fluid by energy reduction achieved by field grading. This implies that breakdown is an extreme-value process. For further additions of Al2O<sup>3</sup> at 1.75 g/L, enhancement is lost for MRO, showing a decline of 4.4% from the host's original value. A similar behaviour is measured for MO at 2 g/L with a 6.7% decline. Both oils do not experience a major stochastic change in the breakdown voltage with the addition of CNP. % enhancement from Table 2 indicates that addition of nanoparticles increases the electron trapping density of the liquid when an ac breakdown voltage is applied between the electrodes. The increase in the kinematic viscosity by the addition of CNP above 0.75 g/L distracts the capture crosssection of the liquid molecules. As a result, the mean free path is very short for breakdown to occur, which gradually decreases breakdown voltage after 0.75 g/L concentration of CNP. A polynomial interpolation function is developed by considering the breakdown voltage of MRO and MO with the concentration of the Al2O3. The Lagrange interpolation polynomial is found for a given set of different breakdown voltage readings. The polynomial function *P*(*x*) is calculated for points '*x<sup>j</sup>* ' which is given in Equations (1) and (2).

$$P(\mathbf{x}) = \sum\_{j=1}^{n} P\_j(\mathbf{x}) \tag{1}$$

$$P\_j(\mathbf{x}) = \left. y\_j \prod\_{k=1, k \neq j}^n \frac{\mathbf{x} - \mathbf{x}\_k}{\mathbf{x}\_j - \mathbf{x}\_k} \right. \tag{2}$$

The Lagrange Interpolation helps to find the polynomial which takes certain values at arbitrary points. When more data points are used for developing a polynomial function, greater data turbulence is observed between data points. The breakdown voltage data is used to develop a mathematical function by concentrating on four limits for reducing the polynomial and data turbulence order. Equations (3) and (4) represent the Lagrange polynomial interpolation functions of MRO and MO, respectively, as calculated for the breakdown voltage variations for the addition of Al2O3. Here, *U*(*x*) is breakdown voltage at the instant and *U* is the maximum withstand capacity of oil. The generated generalized form of polynomial function for MRO and MO is given in Equation (5). Equation (5) is then used to estimate arbitrary values of the breakdown voltage of the host fluids transformed using the nanoparticle.

$$\frac{\mathcal{U}(\mathbf{x})}{\mathcal{U}} = 0.3977\mathbf{x}^4 - 1.567\mathbf{x}^3 + 1.4324\mathbf{x}^2 + 0.1364\mathbf{x}^1 + 1\tag{3}$$

$$\frac{V(\mathbf{x})}{\mathcal{U}} = 0.1873\mathbf{x}^4 - 0.8746\mathbf{x}^3 + 0.8906\mathbf{x}^2 + 0.1874\mathbf{x}^1 + 1\tag{4}$$

$$\frac{dI(x)}{dI} = \int\_{x=0}^{x=2} K1x^4 + K2x^3 + K3x^2 + K4x^1 + \mathcal{C} \, dx \tag{5}$$

$$f(\mathbf{x}) = a\mathbf{1} \* \exp\left(-\left(\frac{\mathbf{x} - b\mathbf{1}}{c\mathbf{1}}\right)^2\right) \tag{6}$$

A Gaussian function shown in Equation (6) is a function for predicting arbitrary real constants. It is a probability density function that calculates the value of the distribution function for the specified concentrations between a range. Here, the concentration is a random value "*x*", which is usually distributed as seen from Figure 6a,b. Though the data points are typically distributed, from Table 3, the measured maximum breakdown voltage is higher than the predicted voltages. The predicted values face deviation of 0.82 kV and 0.24 kV for MRO and MO, respectively, with low root mean square error (RMSE). This shows that there is no data disorder for the confidence interval of 95%. Lagrange interpolation is useful in finding arbitrary points and the development of the mathematical functions. At 95% confidence, a maximum conformity (R<sup>2</sup> ) is achieved in Lagrange interpolation polynomial function. Another reason for using the Lagrange interpolation polynomial is that for a given set of points (with no two values equal), the function assumes the lowest degree at each value the corresponding value. This helps the functions to coincide with each other at each point. Other interpolation functions seem to be not fitting well with the arbitrary points.

**Figure 6.** Measured breakdown voltage distribution using Lagrange polynomial function is compared with the predicted breakdown voltage distribution using Gaussian function. (**a**) MRO, (**b**) MO.

**Table 3.** Variation of Gaussian prediction of the breakdown voltage of MRO and MO with 95% confidence bound with conformity more than 98%.


Assessment of Survival and Hazard Functions of Breakdown Voltage

The density plot of the breakdown voltage for the fluids with varying concentrations of Al2O<sup>3</sup> is presented in Figure 7a,b. From Figure 7a, the frequency of breakdown voltage for MRO is highest when 0.75 g/L of the nanoparticle is added, and density (%) begins to rise for the breakdown voltage from 59 to 67 kV.

**Figure 7.** % Density function of the breakdown voltage in different concentration of the Al2O3, (**a**) MRO, (**b**) MO.

Similarly, without the addition of nanoparticles MRO has the lowest density of 10% with a range of 33 to 47 kV while a maximum of 6% density is observed for the 0.5 g/L addition with a range of 45 to 60 kV. The least density is obtained with 2 g/L addition for the range of 32 to 45 kV. This explains that MRO's breakdown voltage with the addition of 0.75 g/L Al2O<sup>3</sup> has the highest density. The MO's highest density from Figure 7b is observed at 0 g/L, showing the breakdown voltage range of 32 to 34 kV. Unlike MRO, MO shows a wider density range for the concentration of 0.75 g/L of Al2O3, covering the range 30 to 42 kV with less than 15% probability that can be seen from Figure 8a,b. Like in MRO, the addition of 0.75 g/L of Al2O<sup>3</sup> in MO shows a reduction in the dielectric breakdown voltage. At 50% of cumulative probability, MRO's data points with 0 g/L Al2O<sup>3</sup> are likely to fail at 40 kV, showing a wider range from Figure 8a compared to MO observed from Figure 8b. Similarly, MO's breakdown voltage is 32 kV for 2 g/L, which is very close to the 30 kV for the 0 g/L concentration. At 0.5 g of Al2O3, MRO shows higher strength. A maximum of 62 kV obtained within the range of 59 to 62 kV, breakdown is evident with the addition of 0.75 g/L Al2O3, which shows a narrow probability for MRO. For the 0.75 g concentration, MO shows 37 kV to 42 kV with 50% cumulative probability as seen from Figure 8b, below which the MO is very likely to fail. At 0.75 g/L, MRO shows a 26.56% breakdown voltage rise from Table 3 compared to the MO's breakdown voltage. Above 0.75 g/L concentration of Al2O3, the fluids' breakdown voltage begins to decrease.

**Figure 8.** Breakdown voltage density function in different concentration of the Al2O<sup>3</sup> , (**a**) MO, (**b**) MRO.

The cumulative hazard function for the breakdown voltage of MRO and MO under varying concentrations of Al2O<sup>3</sup> is presented in Figure 9a,b. There are two to three outbounds of the cumulative hazard as seen in Figure 9a for Al2O<sup>3</sup> concentrations of 0 g/L, 0.5 g/L, and 0.75 g/L. For the 2 g/L concentration as seen from Figure 9a,b, with both overreaching upper and lower limits of Weibull distribution, the hazard rate is 70%. The data for MRO at 0.5 and 0.75 g/L concentrations has closely followed the Weibull distribution as shown in Figure 9a.

**Figure 9.** Cumulative breakdown voltage hazard function in different concentration of the Al2O<sup>3</sup> with Weibull fit showing the 95% confidence, (**a**) MRO, (**b**) MO.

Moreover, 0.75 g/L addition shows a narrow distribution which is less likely to survive than the other concentrations. In the event of cumulative breakdown voltage, the hazard rate for MO is higher as seen in Figure 9b for the 2 g/L Al2O<sup>3</sup> concentration. Only in the 0.5 g/L and 0.75 g/L concentrations of Al2O<sup>3</sup> the hazard functions follow the Weibull distribution with wider extremities out of the 95% confidence limit. It is also observed that the survivors of MRO are higher than those of MO from Figure 10a,b. The data of MO is completely out of the 95% confidence limits for 0 g/L and 2 g/L concentrations of Al2O3. From observation, it can be concluded that the wider the range of the survivor breakdown voltage, the lesser the probability of survival as can be seen from Figures 9b and 10b. However, the trend is narrow for MRO, especially for the 0.75 g concentration, which shows the highest data survival from Figures 9a and 10a.

**Figure 10.** Survivor breakdown voltage function in different concentration of the Al2O<sup>3</sup> with Weibull fit showing the 95% confidence, (**a**) MRO, (**b**) MO.

#### *5.2. Nanoparticle Impact on Kinematic Viscosity*

The kinematic viscosity of MRO and MO for the various concentrations of Al2O<sup>3</sup> and temperature is studied and presented in Table 4. The kinematic viscosity of MRO and MO is measured at temperatures of 0 ◦C, 30 ◦C, 60 ◦C, 90 ◦C, and 110 ◦C while increasing the concentration of Al2O<sup>3</sup> to assess the molecular relaxation of the fluid at the same time. The results show a promising reduction in kinematic viscosity at 0.75 g/L for all temperatures with a reduction of 40% at 0 ◦C, 66% at 30 ◦C, 54% at 60 ◦C, 66% 90 ◦C, and 62% at 110 ◦C for MRO. The interpolation polynomial function for the kinematic viscosity under varying temperatures is shown in Equations (7)–(11). Here, *V*(*x*) is kinematic viscosity at the instant and *V* is the actual measured kinematic viscosity. The generalized interpolation function for the kinematic viscosity under various temperatures is presented in Equation (12).

$$\frac{V(\mathbf{x})}{V} = -4.805\mathbf{x}^5 + 14.7782\mathbf{x}^4 - 15.3986\mathbf{x}^3 + 6.4599\mathbf{x}^2 - 1.2938\mathbf{x}^1 + 1\tag{7}$$

$$\frac{V(\mathbf{x})}{V} = -14.4195\mathbf{x}^5 + 45.106\mathbf{x}^4 - 48.207\mathbf{x}^3 + 20.385\mathbf{x}^2 - 3.196\mathbf{x}^1 + 1\tag{8}$$

$$\frac{V(\mathbf{x})}{V} = -9.8904\mathbf{x}^5 + 30.303\mathbf{x}^4 - 31.3381\mathbf{x}^3 + 12.7129\mathbf{x}^2 - 2.0508\mathbf{x}^1 + 1\tag{9}$$

$$\frac{V(\mathbf{x})}{V} = -9.8904\mathbf{x}^5 + 30.303\mathbf{x}^4 - 31.3381\mathbf{x}^3 + 12.7129\mathbf{x}^2 - 2.0508\mathbf{x}^1 + 1\tag{10}$$

$$\frac{V(\mathbf{x})}{V} = -12.3566 \mathbf{x}^5 + 39.168 \mathbf{x}^4 - 41.3506 \mathbf{x}^3 + 17.6939 \mathbf{x}^2 - 3.0946 \mathbf{x}^1 + 1 \tag{11}$$

$$\frac{V(\mathbf{x})}{V} = \int\_{\mathbf{x}=0}^{\mathbf{x}=2} \mathbf{K} \mathbf{1} \mathbf{x}^5 + \mathbf{K} \mathbf{2} \mathbf{x}^4 + \mathbf{K} \mathbf{3} \mathbf{x}^3 + \mathbf{K} \mathbf{4} \mathbf{x}^2 + \mathbf{K} \mathbf{5} \mathbf{x}^1 + \mathbf{C} \, d\mathbf{x} \tag{12}$$

$$f(\mathbf{x}) = \mathbf{a}\_0 + \mathbf{a}\_1 \cos(\mathbf{x}\mathbf{w}) + \mathbf{b}\_1 \sin(\mathbf{x}\mathbf{w})\tag{13}$$


**Table 4.** Variation of kinematic viscosity (cSt) of MRO under the influence of Al2O<sup>3</sup> concentration and Temperature with 95% confidence.

> The concentration at which the kinematic viscosity changes with temperature is estimated using Fourier distribution with Equation (13) and its parameters are presented in Table 5. The interpolation function is used to determine the arbitrary points between the concentrations of Al2O3. The data is used to determine the Fourier distribution through which the predicted curve is generated. The measured and predicted distribution of MRO's kinematic viscosity is presented in Figure 11a–e.

**Table 5.** Factors associated with Fourier distribution for MRO with 95% confidence.


The dispersion of the nanoparticle is well examined up to a concentration of 0.75 g/L Al2O3. Above 0.75 g/L, MRO begins to show viscosity rise. The dispersion of nanoparticles in the host fluid never agglomerate due to the presence of oleic acid [37,38] in MRO. The presence of oleic acid acts as surface coating for the CNP and reduces the density in liquid. Moreover, the surface coating protects the CNP from thermal stresses and releases CNP at the time of ionization process. Both low and high temperatures affect the viscosity of MRO. The negative sign (from Table 5) in the decrement clearly shows that viscosity reduction is significant when compared to the host fluid at 0 or 0.1 g/L concentration of Al2O3. The positive sign indicates the increment or saturation effect compared to the host viscosity at 0 or 0.1 g/L concentration of Al2O3. No change can be seen in MRO's sample with 2 g/L concentration of Al2O<sup>3</sup> where the viscosity is already saturated. At 0.75 g/L of Al2O3, the curve shows good conformity with the host fluid.

A similar effect from Table 6 can be seen in MO with an impressive reduction at 0.75 g/L at all temperatures with the drop of 46.67% at 0 ◦C, 53.85% at 30 ◦C, 50% at 60 ◦C, 42.86% at 90 ◦C, and 75% at 110 ◦C. Unlike MRO, MO shows saturation at all temperatures above 0.75 g/L concentration of Al2O3. The interpolation polynomial function for various temperatures is presented in Equations (14)–(18). Here the trend shows better conformity (R<sup>2</sup> ) than in MRO, with a low RMSE. The generalized interpolation function for the kinematic viscosity under various temperatures is presented in Equation (19). The predicted curve is generated using the Fourier distribution given in Equation (20), and its parame-

ters are presented in Table 7. Comparison of measured and predicted curves for MO is presented in Figure 12a–e.

$$\frac{V(\mathbf{x})}{V} = 2.433\mathbf{x}^4 - 2.6938\mathbf{x}^3 + 0.2383\mathbf{x}^2 - 0.11537\mathbf{x}^1 + 1\tag{14}$$

$$\frac{V(\mathbf{x})}{V} = -5.694\mathbf{x}^4 + 14.7804\mathbf{x}^3 - 11.6376\mathbf{x}^2 + 2.3692\mathbf{x}^1 + 1\tag{15}$$

$$\frac{V(\mathbf{x})}{V} = -10.0209\mathbf{x}^4 + 25.052\mathbf{x}^3 - 19.15003\mathbf{x}^2 + 4.0884\mathbf{x}^1 + 1\tag{16}$$

$$\frac{V(\mathbf{x})}{V} = -2.4335\mathbf{x}^4 + 10.4507\mathbf{x}^3 - 10.2361\mathbf{x}^2 + 2.44708\mathbf{x}^1 + 1\tag{17}$$

$$\frac{V(\mathbf{x})}{V} = 6.0244\mathbf{x}^4 - 4.2292\mathbf{x}^3 - 3.0693\mathbf{x}^2 + 1.5797\mathbf{x}^1 + 1\tag{18}$$

$$\frac{V(\mathbf{x})}{V} = \int\_{\mathbf{x}=0}^{\mathbf{x}=2} \mathbf{K} \mathbf{1} \mathbf{x}^4 + \mathbf{K} \mathbf{2} \mathbf{x}^3 + \mathbf{K} \mathbf{3} \mathbf{x}^2 + \mathbf{K} \mathbf{4} \mathbf{x}^1 + \mathbf{C} \, d\mathbf{x} \tag{19}$$

$$\mathbf{x}^{(1)} = \begin{array}{c} \dots \\ \dots \end{array} \tag{10}$$

$$f(\mathbf{x}) = \mathbf{a}\_0 + \mathbf{a}\_1 \cos(\mathbf{x} \mathbf{w}) + \mathbf{b}\_1 \sin(\mathbf{x} \mathbf{w}) \tag{20}$$

**Figure 11.** Comparison of measured and predicted kinematic viscosity of MRO using Lagrange polynomial interpolation function and Fourier distribution function. The predicted value showing deviation of ±2 cSt and 0.6 ± 0.2 g/L from the measured value, (**a**) at 0 ◦C, (**b**) 30 ◦C, (**c**) 60 ◦C, (**d**) 90 ◦C, (**e**) 110 ◦C.

**Figure 12.** Comparison of measured and predicted kinematic viscosity of MO using Lagrange polynomial interpolation function and Fourier distribution function. The predicted value showing deviation of ±1 cSt and 0.6 ± 0.2 g/L from the measured value, (**a**) at 0 ◦C, (**b**) 30 ◦C, (**c**) 60 ◦C, (**d**) 90 ◦C, (**e**) 110 ◦C.

**Table 6.** Variation of kinematic viscosity (cSt) of MO under the influence of Al2O<sup>3</sup> concentration and temperature with 95% confidence.



**Table 7.** Factors associated with Fourier distribution for MO with 95% confidence.

Assessment of Survival and Hazard Functions of Kinematic Viscosity

The density plots from Figures 13 and 14 show the temperature effect on MRO and MO's kinematic viscosity at various concentrations of Al2O3. The concentration of data points of viscosity at high temperatures for MRO is populated within the range of less than 25 cSt while MO has a wider range up to 32 cSt. It can be seen that at concentration of 1 g/L of Al2O3, the viscosity of MO increases by 22% from the host fluid viscosity value. From Figures 15 and 16, the cumulative probability of MRO and MO at various concentrations of Al2O<sup>3</sup> indicates that above 50% probability the temperature has reduced the viscosity of the fluids. There exists inverse relationship between temperature and oil's viscosity according to the above results and results of [21,24]. According to that, when temperature increases, the viscosity begins to reduce. Moreover, temperature rise increases the volume of the oil and reduces consumption of oil in transformers. It can also be observed that at 50% probability, MRO and MO at 110 ◦C have a viscosity less than 11 cSt and 7 cSt, respectively.

It is evident from Figures 15 and 16 that molecular relaxation is highly effective in MRO than MO. At a temperature of 90 ◦C from Figure 15, the kinematic viscosity begins to shift from 23 cSt at 60 ◦C to 12 cSt, which is not realized in MO. Moreover, both Figures 15 and 16 show saturation of viscosity above the 70% probability with a much steeper increase than the previous curve or inflexion points. The MO's saturation in Figure 16 supports the argument that irrespective of temperature, the concentration of Al2O<sup>3</sup> begins to increase the viscosity followed by agglomeration.

**Figure 13.** Distribution of kinematic viscosity of MRO as % density function.

**Figure 14.** Distribution of kinematic viscosity of MO as % density function.

**Figure 15.** Cumulative probability density function of MRO showing variation of kinematic viscosity in different temperature level.

**Figure 16.** Cumulative probability density function of MO showing variation of kinematic viscosity in different temperature level.

However, viscosity of MRO from Figure 15 shows similar behavior to that in MO, with slower saturation above the 0.75 g/L concentration of Al2O3. The survival function of MRO and MO's kinematic viscosity is shown by the data quantile or inverse cumulative survival rate as seen from Figures 17–22. MRO and MO have a better survival rate at different temperatures within an 80% probability from Figures 19 and 20.

**Figure 17.** Inverse cumulative probability of MRO showing the probability of data items with different survival rate.

**Figure 18.** Inverse cumulative probability of MO showing the probability of data items with different survival rate.

**Figure 19.** Kinematic viscosity survival function of MRO showing the confidence of 95% fitted to the Weibull.

The nanoparticle concentration effect on kinematic viscosity reverses after 0.75 g/L in MO as seen from Figure 20 with a 20% probability. However, the impact on MRO from Figure 18 is slower and shows that at 30 ◦C MRO experiences saturation at 0 ◦C with 20% probability. Whenever a steeper rate is observed from Figures 19 and 20, there is less survival within the 95% confidence bounds, which is a risk for both MRO and MO. There is

more significant data turbulence in the steeper curve with insufficient proximity to the Weibull curve, which is higher for MO as seen in Figure 22 than MRO in Figure 23. The cumulative hazard, on the other hand, supports the conclusion devised from the survival rate of MRO and MO. From Figures 21 and 22, 20% of the data points experience hazards whenever the saturation starts. It is also desirable to maintain a minimum concentration of Al2O<sup>3</sup> to avoid agglomeration and thermal conductivity alteration that results in local hot spots.

**Figure 20.** Kinematic viscosity survival function of MO showing the confidence of 95% fitted to the Weibull.

**Figure 21.** Cumulative hazard of MRO showing the data points of kinematic viscosity under varying concentration of Al2O<sup>3</sup> in different temperature levels.

**Figure 22.** Cumulative hazard of MO showing the data points of kinematic viscosity under varying concentration of Al2O<sup>3</sup> in different temperature levels. 107

**Figure 23.** Comparison of measured and predicted flash point and fire point of MRO and MO using Lagrange polynomial interpolation function and Fourier distribution function. The predicted value showing deviation of ±3 ◦C and ±0.2 g from the measured value, (**a**) flash point of MRO, (**b**) fire point of MRO, (**c**) flash point of MO, (**d**) fire point of MO.

#### *5.3. Nanoparticle Impact on Flash Point and Fire Point*

MRO and MO's thermal characteristics are well discussed by calculating the absolute difference and relative difference between the flash and fire points. At various concentrations of Al2O<sup>3</sup> in MRO and MO, the flash and fire points are modified as seen from Table 8. Al2O<sup>3</sup> is an excellent heat conduction additive that significantly alters the fluids thermal characteristics. Although Al2O<sup>3</sup> increases the viscosity of the fluids above 0.75 g/L concentration, it also affects the flash and fire points. The addition of nanoparticles increases the viscosity and reduces thermal conduction. Even though viscosity decreases with an increase in temperature, it increases above 0.75 g/L addition, resulting in a rise in thermal conduction. Effectively, this leads to a drop-in flash and fire points above 0.75 g/L concentration. Comparing the absolute and relative difference of MRO and MO's flash and fire points from Table 8, there is a significant absolute and relative difference of 34 ◦C and 12.41% observed for MRO. MRO absolute difference is almost four times higher than MO. On the other hand, MO has shown 8 ◦C and 4.79%, which need rapid extinction whenever smoke is identified. The drop is well predictable in MO as it shows clear evidence of the temperature drop below the base values of flash point and fire point. It is also evident to have a minimum of 7 to 10% relative difference between the flash and fire points reported by Raymon et al. [5]. The Lagrange interpolation polynomial function for flash point and fire point for MRO and MO is presented in Equations (21)–(24). Here, *FLP*(*x*) and *FRP*(*x*) is fire point and flash point measured at the instant. The generalized Lagrange function and Fourier function (predicted) for the varying concentration of Al2O<sup>3</sup> is given in Equations (25) and (26). The parameter for Fourier function is presented in Table 9. The graphs obtained using Lagrange and Fourier functions shown in Figure 23a–d show lower data deviation for MRO and higher data deviation for MO, which is a clear

indication that the actual parameter is closely predicted with a deviation of ±3 ◦C and ±0.2 g Al2O<sup>3</sup> concentration.

$$\frac{FLP(\mathbf{x})}{FLP} = -5.43768\mathbf{x}^5 + 11.7452\mathbf{x}^4 - 8.7423\mathbf{x}^3 + 2.7637\mathbf{x}^2 - 0.200152\mathbf{x}^1 + 1 \tag{21}$$

$$\frac{FRP(\mathbf{x})}{FRP} = -10.4497\mathbf{x}^5 + 14.7782\mathbf{x}^4 - 15.6351\mathbf{x}^3 + 4.4423\mathbf{x}^2 - 0.18219\mathbf{x}^1 + \tag{22}$$

$$\frac{FLP(\mathbf{x})}{FLP} = -2.9763\mathbf{x}^5 + 7.1620\mathbf{x}^4 - 6.2319\mathbf{x}^3 + 2.3427\mathbf{x}^2 - 0.2441\mathbf{x}^1 + 1 \tag{23}$$

$$\frac{FRP(\mathbf{x})}{FRP} = -0.14713\mathbf{x}^5 - 0.0372\mathbf{x}^4 - 0.1507\mathbf{x}^3 + 0.0390\mathbf{x}^2 + 0.0579\mathbf{x}^1 + 1\tag{24}$$

$$\frac{f(\mathbf{x})}{f} = \int\_{\mathbf{x}=0}^{\mathbf{x}=2} \mathbf{K} \mathbf{1} \mathbf{x}^4 + \mathbf{K} \mathbf{2} \mathbf{x}^3 + \mathbf{K} \mathbf{3} \mathbf{x}^2 + \mathbf{K} \mathbf{4} \mathbf{x}^1 + \mathbf{C} \, d\mathbf{x} \tag{25}$$

$$f(\mathbf{x}) = \mathbf{a}\_0 + \mathbf{a}\_1 \cos(\mathbf{x} \mathbf{w}) + \mathbf{b}\_1 \sin(\mathbf{x} \mathbf{w}) \tag{26}$$

**Table 8.** Flash and fire points of MRO and MO under various concentration of Al2O<sup>3</sup> .


**Table 9.** Factors associated with the predicted Fourier distribution function of MRO with 95% confidence.


Assessment of Survival and Hazard Functions of Flash Point and Fire Point

The density (%) and cumulative curves for the flash point and fire point of MRO and MO are presented in Figure 24a,b and Figure 25a,b. From Figure 25a, MRO has the highest data traces covering 230 to 291 ◦C, which is 58% higher than MO as seen in Figure 25a. The density of MO shows the highest residues in the range 150 to 163 ◦C from Figure 24a. Figure 25b shows that its probability to maintain 160 ◦C flash points is only 40%. MRO is likely to keep 240 ◦C with a probability of 40%, which is almost an 80 ◦C difference to ensure the safety of solid transformer insulation and core components. Figure 25b shows that MRO's fire point within the 40 to 60% probability range is nearly 258 to 268 ◦C. Similarly, for MO at 40 to 60% probability, the range is between 158 to 160 ◦C. This in particular shows

that the temperature difference between the flash point and fire point of MRO and MO is 10 ◦C and 2 ◦C, respectively, which indicates that MRO is comparatively safer than MO.

Therefore, the determination of survival and hazard for the flash point is adequate to discuss because the relative percentage between the flash point and fire point is already discussed. The survival rate and cumulative hazard rate of MRO and MO's flash point characteristics are presented in Figure 26a,b. At 50% probability, the data points of fire point of MO leading to survival is 159 ◦C and MRO is 242 ◦C. At 90% probability, MO begins to show a poor survival rate and increased hazard, while MRO shows less hazard rate. Moving from the 50% to 90% probability, MO can survive at a maximum of 140 ◦C. On the other hand, MRO shows survival of up to 222 ◦C. There is a difference in the temperature drop from 50% probability to 90% probability observed for MO and MRO which is 19 ◦C and 20 ◦C, respectively. The decline is mainly due to the concentration rise of Al2O<sup>3</sup> above 0.75 g/L, thereby affecting saturation in the fluid's dielectric properties.

**Figure 24.** Distribution of data point of flash points of MRO and MO, (**a**) showing the % density plot, (**b**) showing the cumulative probability with survival rate between the range of 40–60% probability.

**Figure 25.** Flash point of MRO and MO with (**a**) survival, (**b**) hazard function with 95% confidence interval fitted to the Weibull.

**Figure 26.** Flash point of MRO and MO with (**a**) survival, (**b**) hazard function with 95% confidence interval fitted to the Weibull.

#### **6. Conclusions**

The results of statistical analysis shows the treatment of natural esters like Marula oil with Al2O<sup>3</sup> enhances the oil's electrical, physical, and thermal characteristics. The survival function and hazard function enable analysis of the host fluids' dielectric properties treated with Al2O3, as well as the electron scavenging properties of Al2O3. An oil's stability is accessed by subjecting the oil to accelerated heating, it also shows the temperament of the oil after accelerated ageing. At the same time, it helps to derive mathematical function that actually fits the linear relationship of the parameter, which are then used for prediction of the future events using the supervised and unsupervised methods. The rate of increase in breakdown voltage at 0.75 g/L concentration of Al2O<sup>3</sup> shows the Al2O<sup>3</sup> nature helps to capture the free electrons generated during the inception of ionization. The addition of CNP is effective at electron trapping, reducing the mean free path between the two electrons. The increase in the valence electron trap density particularly with the CNP is moreover reduced with a small concentration (0.75 g/L) of Al2O<sup>3</sup> than the higher concentration. Thus, the addition of Al2O<sup>3</sup> delays ionization and improves the breakdown voltage of the MRO. The presence of oleic acid in MRO ensures a fine dispersion without the need to coat the CNP with oleic acid and at the same time improves the dielectric breakdown voltage of the insulating fluid. MRO treated with CNP is thermally stable and show endurance for an extended time. A higher absolute and relative difference is seen in MRO than MO. This indicates that the fluid possesses high heat resistance and immunity to temperature stress. The statistical reliability analysis helps to do reliability analysis (time to failure) and enables one to evaluate the lifetime behavior of the insulating oil when subjected to ageing. According to the observations recorded for the different samples, MO become no longer useful and loses its endurance. Such modeling is normally required when actual testing is complicated or impractical. The addition of Al2O<sup>3</sup> to MRO acts as a superior nano-ester insulating fluid and it will perform well in actual transformer conditions. The superior nano-ester insulating fluid can serve various power apparatus like circuit breakers, reactors, capacitors, and cables. Thus, maintaining the life of the equipment for an extended period without the risk of failure.

**Author Contributions:** Conceptualization, R.A.R., A.Y.; methodology, R.A.R.; data curation, R.A.R.; writing—original draft, R.A.R.; writing-review and editing, R.A.R., A.Y., M.M.; validation. R.A.R., A.Y., M.M.; data curation, R.S.; writing—review and editing, R.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** The APC was funded by Botswana International University of Science and Technology (BIUST), Private Bag 16, Palapye, Botswana through the GRAND REF: DVC/RDI/2/1/7 V(57) and PROJECT CODE: S00299.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**

MRO—Marula Oil; MO—Mineral Oil; CNP—Conductive Nanoparticle; PT—Power Transformer; APAC—Asia-Pacific, CAGR—Compound Annual Growth Rate; NEC—National Electric Code; PAO—Polyalphaolefins; BHT—Butylated Hydroxy Toluene; BHA—Butylated Hydroxy Anisole, P.G.—Propyl Gallate; αT—Alpha Tocopherol; β-T—Beta Tocopherol; C.A.—Citric Acid; A.A.—Ascorbic Acid; R.A.—Rosemary extracts; Al2O3—Aluminium-III-Oxide; g/L—grams per Litre; IEC—International Electrotechnical Commission; ASTM—American Society for Testing and Materials; kV/s—kilo Volts per Second; mm—millimeter; Hz—Hertz; mL—milli Litre; kV—kilo Volts; W/m-K—Watts per meter Kelvin; ppm—parts per million; mgKOH//g—milligrams of Potassium Hydroxide per gram; %—Percentage; R2—Conformity; RMSE—Root Mean Square Error; cSt—Centistokes; ◦C—degree Celsius; Temp—Temperature.

#### **References**


### *Article* **Energy Distribution of Optical Radiation Emitted by Electrical Discharges in Insulating Liquids**

### **Michał Kozioł**

Faculty of Electrical Engineering, Automatic Control and Informatics, Opole University of Technology, Proszkowska 76, 45-758 Opole, Poland; m.koziol@po.edu.pl

Received: 26 March 2020; Accepted: 29 April 2020; Published: 1 May 2020

**Abstract:** This article presents the results of the analysis of energy distribution of optical radiation emitted by electrical discharges in insulating liquids, such as synthetic ester, natural ester, and mineral oil. The measurements of optical radiation were carried out on a system of needle–needle type electrodes and on a system for surface discharges, which were immersed in brand new insulating liquids. Optical radiation was recorded using optical spectrophotometry method. On the basis of the obtained results, potential possibilities of using the analysis of the energy distribution of optical radiation as an additional descriptor for the recognition of individual sources of electric discharges were indicated. The results can also be used in the design of various types of detectors, as well as high-voltage diagnostic systems and arc protection systems.

**Keywords:** optical radiation; electrical discharges; insulating liquids; energy distribution

#### **1. Introduction**

One of the characteristic features of electrical discharges is the emission to the space in which they occur, an electromagnetic wave with a very wide range. Such typical ranges of emitted radiation include ionizing radiation, such as X-rays, optical radiation, acoustic emission, and radio wave emission. Based on most of these emitted ranges, diagnostic methods were developed, which enables the detection and location of the source of electrical discharges, which is a great achievement in the diagnostics of high-voltage electrical insulating devices [1–4]. These methods are constantly being improved and modified in terms of increasing their effectiveness and speed of operation. Parallel to these activities, research was also carried out in the field of basic studies aimed at learning new possibilities of using the physicochemical properties of electrical discharge forms [5–10]. Examples of not fully understood areas are X-ray radiation and optical radiation emitted by electrical discharges [11–14].

The research topic discussed in this article is focused in particular on the analysis of optical radiation emitted by electrical discharges, which is usually interpreted using a designated spectrum. For this study, the optical radiation range from 200 nm to 1100 nm was assumed. The radiation spectrum represents the visual form of electromagnetic radiation distributed over the individual components of the wavelengths. Using the radiation spectrum, information about the range of waves that are involved in the analyzed radiation is presented, but their quantitative values were not determined. The dependence of the quantitative size on the occurring wavelength component was represented by the spectral distribution. Spectral distribution, in addition to the range of wavelengths of occurring radiation, most often shows the intensity value of individual components of wavelengths.

Registration of optical radiation is a particularly difficult task in the case of emissions in insulating liquids where there is a large attenuation of the optical signal [15–17]. In addition, there was also an environment with high electric field strength. Therefore, to record radiation in such conditions, it required the use of advanced measuring devices that enabled transmission and processing of optical signals without interference. An additional problem was the correct positioning of the measuring probe

(optical fiber) in the expected location of the electrical discharge, so that the emitted optical radiation can be introduced and transmitted by means of an optical fiber. Currently, effective measuring probes have not yet been developed, and all measurements carried out in this area are of an experimental nature. effective measuring probes have not yet been developed, and all measurements carried out in this area are of an experimental nature. Conducted and published studies were mainly focused on the possibility of recording

emitted optical radiation can be introduced and transmitted by means of an optical fiber. Currently,

*Energies* **2020**, *13*, x FOR PEER REVIEW 2 of 10

conditions, it required the use of advanced measuring devices that enabled transmission and processing of optical signals without interference. An additional problem was the correct positioning

Conducted and published studies were mainly focused on the possibility of recording discharges and determining spectral distributions on their basis [18–21]. However, there is much less work devoted to the development of useful descriptors which, determined on the basis of the obtained spectral distributions, could be used to identify the forms of electrical discharges in various insulation systems (both gas and liquid). Such an approach was presented in the work [22], where a group of descriptors for identifying forms of electrical discharges in insulating oil were developed. discharges and determining spectral distributions on their basis [18–21]. However, there is much less work devoted to the development of useful descriptors which, determined on the basis of the obtained spectral distributions, could be used to identify the forms of electrical discharges in various insulation systems (both gas and liquid). Such an approach was presented in the work [22], where a group of descriptors for identifying forms of electrical discharges in insulating oil were developed. With regard to the already conducted research related to the registration and analysis of optical

With regard to the already conducted research related to the registration and analysis of optical radiation emitted by electric discharges, in terms of the possibility of using their results in high-voltage diagnostics, the author proposed a new approach to the interpretation of recorded spectral distributions. This solution is based on the analysis of the optical spectrum in terms of the share of individual ranges of optical radiation and their use as a descriptor to recognize single-source forms of electrical discharges. radiation emitted by electric discharges, in terms of the possibility of using their results in highvoltage diagnostics, the author proposed a new approach to the interpretation of recorded spectral distributions. This solution is based on the analysis of the optical spectrum in terms of the share of individual ranges of optical radiation and their use as a descriptor to recognize single-source forms of electrical discharges.

#### **2. Method of Measuring Optical Spectra 2. Method of Measuring Optical Spectra**

The tests were carried out on two electrode systems, on which single-source forms of electrical discharges were generated. The first system consisted of needle–needle electrodes, where a high voltage was applied to one of the electrodes and the other was earthed. The second system consisted of a needle electrode, and a solid dielectric was used to generate surface discharges. Both systems can be used as models of potential damage in the high power insulating liquid filled transformers, where the needle–needle electrode system was a model of damage to neighboring transformer windings, while the surface discharge system was a model of damage at the contact between the solid and liquid dielectric. The electrode systems were subsequently immersed in insulating liquids, and the measurements were carried out in identical metrological conditions for each variant. Figure 1 shows the general scheme of the measuring system. The tests were carried out on two electrode systems, on which single-source forms of electrical discharges were generated. The first system consisted of needle–needle electrodes, where a high voltage was applied to one of the electrodes and the other was earthed. The second system consisted of a needle electrode, and a solid dielectric was used to generate surface discharges. Both systems can be used as models of potential damage in the high power insulating liquid filled transformers, where the needle–needle electrode system was a model of damage to neighboring transformer windings, while the surface discharge system was a model of damage at the contact between the solid and liquid dielectric. The electrode systems were subsequently immersed in insulating liquids, and the measurements were carried out in identical metrological conditions for each variant. Figure 1 shows the general scheme of the measuring system.

**Figure 1.** Diagram of the measuring system: Ro—protective water resistor (1.1 MΩ); POF—Polymer Optical Fiber; HR4000—optical spectrophotometer; L1 and L2—control signalling; S1—voltage switch; kV—voltmeter; and U—mains voltage 230 V. **Figure 1.** Diagram of the measuring system: Ro—protective water resistor (1.1 MΩ); POF—Polymer Optical Fiber; HR4000—optical spectrophotometer; L1 and L2—control signalling; S1—voltage switch; kV—voltmeter; and U—mains voltage 230 V.

The schematic diagram and general view of the spark gap for generating electric discharges in a needle–needle system is shown in Figure 2. Two identical electrodes with the following dimensions were used: total length, 35 mm; base diameter, 20 mm; apex angle, 32°; and needle head diameter, 0.8 The schematic diagram and general view of the spark gap for generating electric discharges in a needle–needle system is shown in Figure 2. Two identical electrodes with the following dimensions were used: total length, 35 mm; base diameter, 20 mm; apex angle, 32◦ ; and needle head diameter,0.8 mm. Distance between the electrodes in the needle–needle system was constant for all cases and was 2 cm. The electrodes were made of copper, and their surfaces were electroplated with nickel. The

galvanic coating of the copper electrode with nickel improves its mechanical resistance and thermal strength, which also allows multiple uses of the same electrode. galvanic coating of the copper electrode with nickel improves its mechanical resistance and thermal strength, which also allows multiple uses of the same electrode. strength, which also allows multiple uses of the same electrode.

2 cm. The electrodes were made of copper, and their surfaces were electroplated with nickel. The

galvanic coating of the copper electrode with nickel improves its mechanical resistance and thermal

*Energies* **2020**, *13*, x FOR PEER REVIEW 3 of 10

mm. Distance between the electrodes in the needle–needle system was constant for all cases and was

*Energies* **2020**, *13*, x FOR PEER REVIEW 3 of 10

**Figure 2.** Needle–needle electrode system: schematic diagram side view (**a**) and view from above (**b**). **Figure 2.** Needle–needle electrode system: schematic diagram side view (**a**) and view from above (**b**). **Figure 2.** Needle–needle electrode system: schematic diagram side view (**a**) and view from above (**b**).

A system of two metal electrodes with a solid dielectric between them was used to generate electrical discharges in the surface system. A schematic diagram and general view of the spark gap in the surface system is shown in Figure 3. The supplying electrode was a needle electrode, and the grounded electrode was a flat cylindrical plate with a base diameter of 69 mm and thickness of 9 mm, which was made of metal. The spark gap electrodes were separated by a solid dielectric, which was a plate made of sodium glass, with external dimensions of 90 mm × 90 mm and a thickness of 10 mm. A system of two metal electrodes with a solid dielectric between them was used to generate electrical discharges in the surface system. A schematic diagram and general view of the spark gap in the surface system is shown in Figure 3. The supplying electrode was a needle electrode, and the grounded electrode was a flat cylindrical plate with a base diameter of 69 mm and thickness of 9 mm, which was made of metal. The spark gap electrodes were separated by a solid dielectric, which was a plate made of sodium glass, with external dimensions of 90 mm × 90 mm and a thickness of 10 mm. A system of two metal electrodes with a solid dielectric between them was used to generate electrical discharges in the surface system. A schematic diagram and general view of the spark gap in the surface system is shown in Figure 3. The supplying electrode was a needle electrode, and the grounded electrode was a flat cylindrical plate with a base diameter of 69 mm and thickness of 9 mm, which was made of metal. The spark gap electrodes were separated by a solid dielectric, which was a plate made of sodium glass, with external dimensions of 90 mm × 90 mm and a thickness of 10 mm.

**Figure 3.** Surface discharge system: schematic diagram side view (**a**) and view from above (**b**). **Figure 3.** Surface discharge system: schematic diagram side view (**a**) and view from above (**b**). **Figure 3.** Surface discharge system: schematic diagram side view (**a**) and view from above (**b**).

The three most frequently used electroinsulating liquids in power engineering were used for testing natural ester Midel 1204, synthetic ester Midel 7131, and mineral oil Orlen Trafo EN. All liquids were brand new and free of any contamination. The temperature of the insulating liquids were the same in all examined cases and was 20 °C. Due to the experimental nature of basic research, The three most frequently used electroinsulating liquids in power engineering were used for testing natural ester Midel 1204, synthetic ester Midel 7131, and mineral oil Orlen Trafo EN. All liquids were brand new and free of any contamination. The temperature of the insulating liquids were the same in all examined cases and was 20 °C. Due to the experimental nature of basic research, the influence of liquid temperature on the obtained measurement results was not analyzed. The three most frequently used electroinsulating liquids in power engineering were used for testing natural ester Midel 1204, synthetic ester Midel 7131, and mineral oil Orlen Trafo EN. All liquids were brand new and free of any contamination. The temperature of the insulating liquids were the same in all examined cases and was 20 ◦C. Due to the experimental nature of basic research, the influence of liquid temperature on the obtained measurement results was not analyzed.

the influence of liquid temperature on the obtained measurement results was not analyzed. The optical spectrophotometer, HR4000 from Ocean Optics (Dunedin, FL, USA) was used to record the optical radiation emitted by electrical discharges. The applied spectrophotometer recorded optical radiation in the ultraviolet, visible, and near-infrared range (UV–VIS–NIR spectral range from 200 nm to 1100 nm). The device is equipped with a 3648-element linear silicon CCD array and an optical resolution of 0.47 nm FWHM (Full Width at Half Maximum). This enabled the detection of The optical spectrophotometer, HR4000 from Ocean Optics (Dunedin, FL, USA) was used to record the optical radiation emitted by electrical discharges. The applied spectrophotometer recorded optical radiation in the ultraviolet, visible, and near-infrared range (UV–VIS–NIR spectral range from 200 nm to 1100 nm). The device is equipped with a 3648-element linear silicon CCD array and an optical resolution of 0.47 nm FWHM (Full Width at Half Maximum). This enabled the detection of 3648 components of the recorded optical spectrum in the range of 200 nm to 1100 nm. The optical spectrophotometer, HR4000 from Ocean Optics (Dunedin, FL, USA) was used to record the optical radiation emitted by electrical discharges. The applied spectrophotometer recorded optical radiation in the ultraviolet, visible, and near-infrared range (UV–VIS–NIR spectral range from 200 nm to 1100 nm). The device is equipped with a 3648-element linear silicon CCD array and an optical resolution of 0.47 nm FWHM (Full Width at Half Maximum). This enabled the detection of 3648 components of the recorded optical spectrum in the range of 200 nm to 1100 nm.

3648 components of the recorded optical spectrum in the range of 200 nm to 1100 nm. Polymer optical fiber 600SR (POF) manufactured by Ocean Optics was used as the measuring head, and one of its endpoints was placed near the electrode system. The basic parameters of the optical fiber were presented in Table 1. During the emission of optical radiation by electrical discharge, the light beam was introduced into the optical fiber and sent to spectrophotometer. The spectrophotometer converts the light beam into a parallel stream with a spectral range of 200 nm to 1100 nm and counts the number of emitted photons for each wavelength. The integration time of the spectrophotometer (matrix exposure time) was the same in all cases and was set to 1 s. Obtained data were presented in the form of spectral characteristics, where the intensity corresponded to the number of counts for wavelengths in the analyzed range. 1100 nm and counts the number of emitted photons for each wavelength. The integration time of the spectrophotometer (matrix exposure time) was the same in all cases and was set to 1 s. Obtained data were presented in the form of spectral characteristics, where the intensity corresponded to the number of counts for wavelengths in the analyzed range. **Table 1.** Basic parameters of the optical fiber.

spectrophotometer converts the light beam into a parallel stream with a spectral range of 200 nm to

*Energies* **2020**, *13*, x FOR PEER REVIEW 4 of 10

Polymer optical fiber 600SR (POF) manufactured by Ocean Optics was used as the measuring head, and one of its endpoints was placed near the electrode system. The basic parameters of the optical fiber were presented in Table 1. During the emission of optical radiation by electrical


**Table 1.** Basic parameters of the optical fiber. **Parameter Value** 

Figure 1 shows how the fiber was placed in the electrode system area. The optical fiber head was placed at a distance of 2.5 cm from the expected source of optical radiation emission. This distance was determined on the basis of the fiber acceptance cone parameter and due to the metal elements of the fiber head. placed at a distance of 2.5 cm from the expected source of optical radiation emission. This distance was determined on the basis of the fiber acceptance cone parameter and due to the metal elements of the fiber head. All measurements were made in a darkened laboratory room, separated from external sources

Figure 1 shows how the fiber was placed in the electrode system area. The optical fiber head was

All measurements were made in a darkened laboratory room, separated from external sources of optical radiation. Before each measurement series, the spectrophotometer was calibrated to determine the minimum background level. This operation aimed to eliminate interference resulting from the process of converting the optical signal to digital form. The background calibration function is available in the device software. Spectral calibration of the spectrophotometer with a dedicated POF was performed by the manufacturer. The supply voltage of electrode systems was from 25 to 50 kV (RMS voltage) of 50 Hz alternating current (AC), with gradation every 5 kV. In order to limit the discharge current, a water resistor (Ro) of 1.1 MΩ was used, which limited the current to mA range (about 100 mA for this system). For each supply voltage, 10 measurement tests were performed. of optical radiation. Before each measurement series, the spectrophotometer was calibrated to determine the minimum background level. This operation aimed to eliminate interference resulting from the process of converting the optical signal to digital form. The background calibration function is available in the device software. Spectral calibration of the spectrophotometer with a dedicated POF was performed by the manufacturer. The supply voltage of electrode systems was from 25 to 50 kV (RMS voltage) of 50 Hz alternating current (AC), with gradation every 5 kV. In order to limit the discharge current, a water resistor (Ro) of 1.1 MΩ was used, which limited the current to mA range (about 100 mA for this system). For each supply voltage, 10 measurement tests were performed.

#### **3. Measurement Results 3. Measurement Results**  Figure 4 presents examples of registered optical spectra emitted by electrical discharges

discharges.

**4. Optical Radiation Energy** 

Figure 4 presents examples of registered optical spectra emitted by electrical discharges generated on the system of needle–needle electrodes for each of the tested insulating liquids. generated on the system of needle–needle electrodes for each of the tested insulating liquids.

**Figure 4.** Example results of measurements generated on a needle–needle spark gap at 35 kV supply voltage for the following insulating liquids: Midel 1204 (**a**); Midel 7131 (**b**); and Mineral oil (**c**). **Figure 4.** Example results of measurements generated on a needle–needle spark gap at 35 kV supply voltage for the following insulating liquids: Midel 1204 (**a**); Midel 7131 (**b**); and Mineral oil (**c**).

mainly includes visible light and, to a small extent, near infrared and ultraviolet. Figure 5 presents examples of registered optical spectra emitted by electrical discharges generated on the surface

(**a**) (**b**)

(**c**) **Figure 5.** Example results of measurements generated on a surface discharge system at 35 kV supply voltage for the following insulating liquids: Midel 1204 (**a**); Midel 7131 (**b**); and Mineral oil (**c**).

By comparing the obtained spectral characteristics in both analyzed electrode systems and for the same supply voltage level, significant differences in their shapes and spectral ranges can be observed. For the needle–needle electrode system, the spectral range is mainly in visible light, and a small extent in the near-infrared and ultraviolet range. In turn, the spectra obtained for the surface discharge system contained a higher proportion of ultraviolet radiation. This showed the potential possibility of using optical spectra analysis for the recognition of single-source forms of electrical

Based on the obtained characteristics of spectral distributions and using the quantum description where optical radiation is described as a photon stream, the share of emitted energy can

discharge system for each of the tested insulating liquids.

Obtained optical spectra in the needle–electrode system shows some similarity in the shape of the spectral characteristics in all three analyzed liquids. The spectral range of the characteristics mainly includes visible light and, to a small extent, near infrared and ultraviolet. Figure 5 presents examples of registered optical spectra emitted by electrical discharges generated on the surface discharge system for each of the tested insulating liquids. Obtained optical spectra in the needle–electrode system shows some similarity in the shape of the spectral characteristics in all three analyzed liquids. The spectral range of the characteristics mainly includes visible light and, to a small extent, near infrared and ultraviolet. Figure 5 presents examples of registered optical spectra emitted by electrical discharges generated on the surface discharge system for each of the tested insulating liquids.

(**c**) **Figure 4.** Example results of measurements generated on a needle–needle spark gap at 35 kV supply

*Energies* **2020**, *13*, x FOR PEER REVIEW 5 of 10

**Figure 5.** Example results of measurements generated on a surface discharge system at 35 kV supply voltage for the following insulating liquids: Midel 1204 (**a**); Midel 7131 (**b**); and Mineral oil (**c**). **Figure 5.** Example results of measurements generated on a surface discharge system at 35 kV supply voltage for the following insulating liquids: Midel 1204 (**a**); Midel 7131 (**b**); and Mineral oil (**c**).

By comparing the obtained spectral characteristics in both analyzed electrode systems and for the same supply voltage level, significant differences in their shapes and spectral ranges can be observed. For the needle–needle electrode system, the spectral range is mainly in visible light, and a small extent in the near-infrared and ultraviolet range. In turn, the spectra obtained for the surface discharge system contained a higher proportion of ultraviolet radiation. This showed the potential possibility of using optical spectra analysis for the recognition of single-source forms of electrical By comparing the obtained spectral characteristics in both analyzed electrode systems and for the same supply voltage level, significant differences in their shapes and spectral ranges can be observed. For the needle–needle electrode system, the spectral range is mainly in visible light, and a small extent in the near-infrared and ultraviolet range. In turn, the spectra obtained for the surface discharge system contained a higher proportion of ultraviolet radiation. This showed the potential possibility of using optical spectra analysis for the recognition of single-source forms of electrical discharges.

#### discharges. **4. Optical Radiation Energy**

**4. Optical Radiation Energy**  Based on the obtained characteristics of spectral distributions and using the quantum description where optical radiation is described as a photon stream, the share of emitted energy can Based on the obtained characteristics of spectral distributions and using the quantum description where optical radiation is described as a photon stream, the share of emitted energy can be estimated. Each wavelength of emitted radiation corresponds to a specific photon energy which can be determined from the relation:

$$E = n \cdot h \cdot \upsilon \tag{1}$$

where: *E*—total energy of the photon stream (J), *n*—number of photons counted (-), *h*—Planck constant 6.626 <sup>×</sup> <sup>10</sup>−<sup>34</sup> (J·s), and <sup>υ</sup>—wave frequency (1/s).

The frequency of the waveform is expressed by the formula:

$$\nu = \frac{c}{\lambda} \tag{2}$$

where: <sup>υ</sup>—wave frequency (1/s); *<sup>c</sup>*—phase speed of the wave, speed of light in vacuum 2.998 <sup>×</sup> <sup>10</sup><sup>8</sup> (m/s); and λ—wavelength (nm).

Table 2 presents examples of the calculated energy values of optical radiation emitted by electric discharges generated in the tested electrode systems. In order to better present the determined total energy values, the physical unit of energy description in the form of electronvolts (eV) was used. They were calculated from a simple relationship resulting from the definition of eV, where 1 J <sup>≈</sup> 6.241509126(38) <sup>×</sup> <sup>10</sup><sup>18</sup> eV. The calculated energy was not the total energy radiated by electrical discharges. The presented values were estimated based on the registered optical radiation by the spectrophotometer. This stage of research does not include an attempt to prepare energy balance, but only presents the possibility of applying energy distribution analysis to recognize the form of electrical discharges.


**Table 2.** Examples of optical radiation energy values.

#### *Analysis of the Variability of Optical Energy Shares in the UV–VIS–NIR Range*

The analysis of optical radiation variability in the UV–VIS–NIR range was carried out using traditional statistical methods. The main aim of the variability analysis was to determine the share of the energy of optical radiation, in particular spectral ranges for recorded optical spectra emitted by electric discharges generated in the analyzed electrode configurations. The results of the variability analysis are presented in Table 3. The results presented in Table 3 were determined on the basis of 10 registered characteristics for each variant of the measuring system.



Natural ester

**Type of Liquid** 


**Table 3.** *Cont*. NIR 1.70 0.08 0.29 17.06 1.41 < Xtyp < 1.99

**needle–needle system** 

**2**

*Energies* **2020**, *13*, x FOR PEER REVIEW 7 of 10

**Table 3.** Comparison of the share of optical radiation emitted by electric discharges.

**Variance Standard** 

**Deviation** 

UV 0.70 0.01 0.06 8.57 0.64 < Xtyp < 0.76 VIS 97.60 0.62 0.79 0.81 96.81 < Xtyp < 98.39

**Coefficient of Variation** 

 **s Vs Xtyp**

**Range of Typical Values of Radiation Energy Share** 

Figures 6–8 present the percentage share of energy in optical radiation ranges emitted by electrical discharges generated on the system of needle–needle electrodes and surface discharge systems in the insulating liquids adopted for testing. The UV radiation was less than 1% of the total energy radiation and poorly detectable in all tested insulating liquids. However, it was different in the case of electrical discharges generated in a surface discharge system, where the proportion of ultraviolet radiation was much higher. Figures 6–8 present the percentage share of energy in optical radiation ranges emitted by electrical discharges generated on the system of needle–needle electrodes and surface discharge systems in the insulating liquids adopted for testing. The UV radiation was less than 1% of the total energy radiation and poorly detectable in all tested insulating liquids. However, it was different in the case of electrical discharges generated in a surface discharge system, where the proportion of ultraviolet radiation was much higher.

**Figure 6.** Percentage of optical radiation energy for electrical discharges generated in natural ester, Midel 1204, on electrode systems: needle–needle (**a**) and for surface discharges (**b**). **Figure 6.** Percentage of optical radiation energy for electrical discharges generated in natural ester, Midel 1204, on electrode systems: needle–needle (**a**) and for surface discharges (**b**). *Energies* **2020**, *13*, x FOR PEER REVIEW 8 of 10

**Figure 7.** Percentage of optical radiation energy for electrical discharges generated in natural ester, Midel 7131, on electrode systems: needle–needle (**a**) and for surface discharges (**b**). **Figure 7.** Percentage of optical radiation energy for electrical discharges generated in natural ester, Midel 7131, on electrode systems: needle–needle (**a**) and for surface discharges (**b**).

**Average Energy (%Ec)** 

**Range of Optical Radiation** 

(**a**) (**b**) **Figure 8.** Percentage of optical radiation energy for electrical discharges generated in mineral oil,

This is probably due to various energy transformations accompanying the phenomenon of electrical discharges, which occurs in the used electrode systems. In the electrode system of the needle–needle type, electric discharges mostly occur in the form of a spark, while in the system for surface discharges, the phenomena occur continuously. Radiation in the ultraviolet range, in the case

The spectral distribution of the optical radiation emitted by electrical discharges in insulating liquids differed according to the electrode geometry. Needle–needle electrodes had < 1% UV radiation in all analyzed cases. In contrast, surface discharges had 7% or more UV radiation, depending on the type of electrical insulating liquid. This result might allow identification of the discharge type from the radiation spectrum and might be incorporated in expert diagnostic systems used in various technical areas. The results also justify further research, in terms of the applicability of the proposed indicator, for recognizing forms of electrical discharges in high-voltage insulation

**Funding:** This research was co-funded by the National Science Centre, Poland (NCN) as a part of the Preludium

**Conflicts of Interest:** The author declares no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to

Orlen Trafo EN, on electrode systems: needle–needle (**a**) and for surface discharges (**b**).

of discharges in the needle–needle system, is short-lived and effectively suppressed.

**5. Conclusions** 

systems.

publish the results.

Research Project No. 2017/25/N/ST8/00590.

(**a**) (**b)**

**Figure 7.** Percentage of optical radiation energy for electrical discharges generated in natural ester,

*Energies* **2020**, *13*, x FOR PEER REVIEW 8 of 10

**Figure 8.** Percentage of optical radiation energy for electrical discharges generated in mineral oil, Orlen Trafo EN, on electrode systems: needle–needle (**a**) and for surface discharges (**b**). **Figure 8.** Percentage of optical radiation energy for electrical discharges generated in mineral oil, OrlenTrafo EN, on electrode systems: needle–needle (**a**) and for surface discharges (**b**).

This is probably due to various energy transformations accompanying the phenomenon of electrical discharges, which occurs in the used electrode systems. In the electrode system of the needle–needle type, electric discharges mostly occur in the form of a spark, while in the system for surface discharges, the phenomena occur continuously. Radiation in the ultraviolet range, in the case of discharges in the needle–needle system, is short-lived and effectively suppressed. This is probably due to various energy transformations accompanying the phenomenon of electrical discharges, which occurs in the used electrode systems. In the electrode system of the needle–needle type, electric discharges mostly occur in the form of a spark, while in the system for surface discharges, the phenomena occur continuously. Radiation in the ultraviolet range, in the case of discharges in the needle–needle system, is short-lived and effectively suppressed.

#### **5. Conclusions 5. Conclusions**

The spectral distribution of the optical radiation emitted by electrical discharges in insulating liquids differed according to the electrode geometry. Needle–needle electrodes had < 1% UV radiation in all analyzed cases. In contrast, surface discharges had 7% or more UV radiation, depending on the type of electrical insulating liquid. This result might allow identification of the discharge type from the radiation spectrum and might be incorporated in expert diagnostic systems used in various technical areas. The results also justify further research, in terms of the applicability of the proposed indicator, for recognizing forms of electrical discharges in high-voltage insulation The spectral distribution of the optical radiation emitted by electrical discharges in insulating liquids differed according to the electrode geometry. Needle–needle electrodes had < 1% UV radiation in all analyzed cases. In contrast, surface discharges had 7% or more UV radiation, depending on the type of electrical insulating liquid. This result might allow identification of the discharge type from the radiation spectrum and might be incorporated in expert diagnostic systems used in various technical areas. The results also justify further research, in terms of the applicability of the proposed indicator, for recognizing forms of electrical discharges in high-voltage insulation systems.

systems. **Funding:** This research was co-funded by the National Science Centre, Poland (NCN) as a part of the Preludium Research Project No. 2017/25/N/ST8/00590.

**Funding:** This research was co-funded by the National Science Centre, Poland (NCN) as a part of the Preludium Research Project No. 2017/25/N/ST8/00590. **Conflicts of Interest:** The author declares no conflict of interest. The funders had no role in the design of the **Conflicts of Interest:** The author declares no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to

#### publish the results. **References**


© 2020 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
