**1. Introduction**

Insulating nanofluids have attracted the attention of researchers for the last twenty years. The dispersion of nanoparticles inside insulating oils can conditionally increase the breakdown voltage and thermal conductivity of matrix oil. Insulating fluids used in transformers exhibit high dielectric strength but their thermal conductivity is low and this introduces limitations on the power rating of the transformers, and, as a consequence, an increase in their size. Thus, the motivation for the insertion of nanoparticles within transformer oil was initially the increase of thermal conductivity. The idea of adding particles in order to increase thermal conductivity goes back to Maxwell in 1873 [1] and Choi et al. [2] were the first to add magnetic nanoparticles Fe3O4 to pure transformer oil and to develop the first nanofluid that demonstrated better thermal conductivity. However, several researchers soon realized that the dispersion of some nanoparticles increased the breakdown voltage at the same time [3–17] and their interest also turned to this direction.

Researchers in [3–10] studied iron oxide nanoparticles, Fe3O4 and Fe2O3, and found they enhanced dielectric and the cooling performance of the constitutive base fluids when they were dispersed in either mineral or natural ester oils. The dispersion of two-dimensional nanoparticles, such as boron nitride (BN) or graphene nanoparticles, also enhanced the cooling capacities of transformer oil [3,11]. After these promising results, different nanofluids were developed using nanoparticles such as TiO2, SiO2, nanodiamond-Ni nanoparticles, and boron nitride (BN) [7,11–19] and they were tested regarding their dielectric performance.

Semi-conductive TiO2 nanoparticles were added to natural ester oil and mineral oil matrix [13,14] resulting in nanofluids with increased AC breakdown voltage (BDV) and lightning impulse withstand capability as compared to that of the matrix oils. In [15–17] high concentrations of SiO2 nanoparticles in mineral oil resulted in increased AC breakdown voltage and positive lightning impulse withstand capability in comparison with the matrix dielectric liquid. On the other hand, the negative lightning impulse voltage withstand capability was decreased. In addition, they highlighted the strong influence of moisture on the performance of nanofluids. Particularly, the higher the presence of humidity in the aforementioned SiO2 nanofluid the better its performance was. However, thermal conductivity seemed to be only slightly affected by the addition of SiO2 nanoparticles.

Aluminum nitride (AIN), graphene oxide, and BN nanoparticles, as dispersants in mineral oil, were also studied [6,11,17,18]. The addition of AIN nanoparticles proved to increase positive lightning impulse voltage withstand capability and partial discharge ignition voltage, as well as the thermal conductivity of nanofluid as compared with pure oil, and simultaneously, AC breakdown voltage was decreased. Similar results were demonstrated with the addition of BN and graphene oxide nanoparticles where both the AC BDV and the thermal conductivity were improved as compared with the matrix oil. A satisfactory explanation for the higher AC BDV of insulating nanofluids with higher conductivity nanoparticles, with respect to the matrix oils, was given by the "electron traps" theory in [20]. According to this, electrons are very rapidly captured by conductive nanoparticles being transmuted into heavy negatively charged nanoparticles. As a consequence, the streamer speed was reduced resulting in an increased breakdown voltage. Experimental results concerning the space charge in both matrix oil and nanofluid oils [21] were in compliance with the proposed theory.

However, the aforementioned theory fails to adequately explain the superior performance of nanofluids with semi-conductive and non-conductive nanoparticles which have been observed with the conductive ones. This can be ascribed, according to a theory [22], to the different conductivity or permittivity between nanoparticles and their surrounding oil and, according to others [12,23,24], to the interfacial region shaped on the surface of the nanoparticles (NPs).

In this study the AC BDV of two completely different natural ester oil matrix nanofluids will be examined. Conductive in situ surface modified colloidal magnetic iron oxides NPs (MIONs) and SiO2 nanoparticles were scattered into natural ester dielectric liquid (FR3®). Different nanoparticle concentrations were tested with respect to the optimal concentration in terms of AC dielectric strength performance. The theoretical model proposed in [22] was realized in order to interpolate the experimental results.

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

During the preparation of the nanofluids, natural ester dielectric liquid Envirotemp FR3® was used as base liquid which was selected due to its biodegrability and high temperature flash point.

At first, the matrix oil was triple filtered and dried as analytically described in [10]. In brief, low vacuum filtration was used by applying a 30 μm filter for the first level, a 1 μm glass microfiber filter for the second level, and a 0.8 μm membrane for the last one. Thereafter, the matrix liquid was dehumidified by means of a rotary evaporator in series with a vacuum pump, while it was positioned in a water bath heated at 80 ◦C for at least two days.

The preparation of the nanofluids (NFs) with oleate-coated colloidal magnetic iron oxide nanocrystals (colMIONs) required particular care because magnetic iron oxide nanoparticles, such as Fe2O3 and Fe3O4, have a tendency to indicate residual NPs after their dispersion in the matrix liquid, forming sedimentation within a few days. In an attempt to overcome agglomeration and increase

the long-term stability of the nanofluid, oleate-coated colloidal magnetic iron oxide nanocrystals (colMIONs) were manufactured in the laboratory according to previous reports [9,25]. Initially, 3.62 g of iron-oleate complex and 3.4 g of oleic acid were dissolved in 1-octadence at a temperature of 25 ◦C. For 1 h, the mixture was magnetically agitated at room temperature, then it was kept under agitation (350 rpm) for 30 min at a temperature of 100 ◦C and, finally, it was further heated at 318 ◦C for another hour [26]. Afterwards, the commixture was cooled down and 8 mL of dichloromethane was added. Subsequently, acetone as a dissolver was introduced to decrease the solubility of the MIONs, as well as separate the reacting agents from them. This manufacturing method was repeated until the oleate-coated MIONs obtained a purity level higher than 80%. The colMIONs had a final diameter of approximately 10 nm with a very narrow dimensional distribution [9], and thereafter, they were introduced into the natural ester oil and the final liquid was ultrasonicated for at least 30 min. Six samples, having a concentration range from 0.004 to 0.014% *<sup>w</sup>*/*<sup>w</sup>*, in step 0.002% were prepared for further study.

Nanoparticles SiO2 are well dispersed inside transformer oil and they do not indicate agglomeration effect. On the contrary, they may absorb humidity during the introduction of atmospheric air within 3–4 s. Therefore, the procedure took place in a shielded AtmosBag with dedicated gloves (Aldrich® AtmosBag) while inside N2 was introduced. After the addition of 12 nm average diameter SiO2 nanoparticles to the matrix oil, the final liquid was ultrasonicated for 30 min. Six different samples with concentrations ranging from 0.008–0.024% *w*/*w* in step 0.004% were also prepared.

The measurements of AC dielectric breakdown strength carried out for the samples of the two nanofluids according to IEC 60156 in [27] were enriched with additional ones. The measurement device used was a Baur DTA 100 C, measuring up to 100 kV, Rogowski electrodes based on IEC 60156 [28] with a gap distance of 2.5 mm, a voltage rise of 2 kV/s was adjusted at 50 Hz power system, the breakdown event is calculated based on the level of the current conduction (mA range). The distribution of the breakdown voltage of the experimental results is calculated based on the normal distribution [10]. Before each experimental set, the brass electrodes were polished and cleaned thoroughly. For the matrix oil, 150 breakdown tests were implemented for every sample, whereas, after every 50 successful breakdowns the sample under test was replaced in order to limit the degradation and its effect on the measurements.

Dielectric relaxation spectroscopy was studied by means of a Novocontrol Alpha analyzer (10−1–106 Hz) monitored from a Novocontrol Quatro Cryosystem. The understudied dielectric liquid was studied in a custom-made cylindrical capacitor which consisted of two plane plates at 1–1.2 mm distance. The dielectric relaxation spectroscopy study was adopted at 20–100 ◦C with a 20 ◦C temperature increase rate at the frequency range of 10−1–106 Hz.

Dynamic light scattering (DLS) was performed on oil dispersions of < 0.01% *w*/*v* in Fe2O3 using the viscosity of the oil 32.03 mm<sup>2</sup>/s. A Malvern Instrument ZetaSizer Nano was used, equipped with a 4 mW He-Ne laser, operating at a wavelength of 633 nm. Scattered light was collected at a fixed angle of 173◦. Diameter distribution was reported as number-based results. Transmission electron microscope images were collected with a JEOL 2100 on 200 kV. It should be noted that the DLS technique systematically provides a higher mean size of the crystallite size than the TEM technique, because in the former case the size is evaluated in the dispersion state and takes into account the organic coating of the magnetic crystallites, their solvation sphere, and possible formation of dyads between particles. In Figure 1 the size distribution of the colloidal MIONs in the oil matrix is depicted with a peak value of 23 nm, accordingly, in the same figure the colloidal stability of the nanoparticles in the matrix oil after 2 months of storage is depicted. In Figure 2 a detailed TEM micrograph from the oleate-coated MIONs, as manufactured through the thermolytic route [9], is shown, clearly demonstrating their proper dispersion in absence of agglomeration.

**Figure 1.** Size distribution diagram based on dynamic light scattering measurement of the colloidal magnetic iron oxides nanoparticles (MIONs) in the oil matrix and digital image colloidal nanofluid (colNF) after 2 months of storage (inset).

**Figure 2.** TEM micrograph from the oleate-coated MION colloids.
