**Performance Enhancement of Hybrid Solid Desiccant Cooling Systems by Integrating Solar Water Collectors in Taiwan**

**Win-Jet Luo 1,\*, Dini Faridah 1, Fikri Rahmat Fasya 2, Yu-Sheng Chen 2, Fikri Hizbul Mulki <sup>2</sup> and Utami Nuri Adilah <sup>3</sup>**


Received: 30 July 2019; Accepted: 4 September 2019; Published: 9 September 2019

**Abstract:** A hybrid solid desiccant cooling system (SDCS), which combines a solid desiccant system and a vapor compression system, is considered to be an excellent alternative for commercial and residential air conditioning systems. In this study, a solar-assisted hybrid SDCS system was developed in which solar-heated water is used as an additional heat source for the regeneration process, in addition to recovering heat from the condenser of an integrated heat pump. A solar thermal collector sub-system is used to generate solar regeneration water. Experiments were conducted in the typically hot and humid weather of Taichung, Taiwan, from the spring to fall seasons. The experimental results show that the overall performance of the system in terms of power consumption can be enhanced by approximately 10% by integrating a solar-heated water heat exchanger in comparison to the hybrid SDCS system. The results show that the system performs better when the outdoor humidity ratio is large. In addition, regarding the effect of ambient temperature on the coefficient of performance (COP) of the systems, a critical value of outdoor temperature exists. The COP of the systems gradually rises with the increase in ambient temperature. However, when the ambient temperature is greater than the critical value, the COP gradually decreases with the increase in ambient temperature. The critical outdoor temperature of the hybrid SDCS is from 26 ◦C to 27 ◦C, and the critical temperature of the solar-assisted hybrid SDCS is from 27 ◦C to 30 ◦C.

**Keywords:** hybrid solid desiccant cooling system; regeneration process; solar thermal collector; coefficient of performance

#### **1. Introduction**

Heating, ventilating and air conditioning (HVAC) systems are designed to maintain specific indoor conditions, which vary depending on the application. The main factors that influence the thermal comfort of occupants are metabolic rate, clothing insulation, air temperature, mean radiant temperature, air velocity and relative humidity [1]. The main purpose of the HVAC system is to provide good indoor air quality that meets the criteria for hygienic air conditions and satisfies the thermal comfort of occupants or products in a building or space. Moreover, the working environment can influence the productivity of workers, which is an economic reason for installing HVAC systems in buildings [2,3].

Generally, HVAC systems require a large amount of energy, especially in large capacity applications. HVAC systems make a significant contribution to carbon-based energy consumption and greenhouse

gases emissions [4]. In the USA, nearly 50% of the energy consumption of buildings is used for HVAC systems [5]. Furthermore, conventional HVAC systems cause pollution, not only due to consuming a large amount of energy, but also due to the use of hydro-chlorofluorocarbon (HCFC), hydrofluorocarbon (HFC) and other refrigerants that produce greenhouse gases [6].

Cooling-based dehumidification systems are the most popular systems of recent decades [7,8]. Such systems provide cooling and dehumidification by utilizing a single vapor-compression unit. Cooling coils are used to cool the air below its dew point so moisture can be removed from the air. Therefore, low humidity and low temperature air can be generated and a reheat coil is required to avoid overcooling, which consumes a large amount of energy and is difficult to control. Current studies are developing new approaches that are more energy efficient and more environmentally friendly.

Recently, novel modern air conditioning systems have been proposed, most notably utilizing split processes of cooling and dehumidification. Instead of the cooling coil unit used in conventional cooling dehumidification systems, a single unit that can handle latent heat is applied, so unnecessary energy use can be avoided. One developed method is a solid desiccant cooling system (SDCS) in which refrigerants of HCFCs are unnecessary and low-grade thermal energy can be used for regeneration [9,10]. In an SDCS system, solid desiccants, such as silica gel, activated carbon, molecular sieves, alumina gel and other materials with strong hygroscopic ability, are used to dehumidify the process air. The solid desiccant itself has many pores, and the inner surface of each pore is concave. When the process air passes through the solid desiccant, because of the lower partial pressure of water vapor on the concave surface with a small radius of curvature, the vapor may migrate from the air to the concave surface of the pores. Then, the vapor condenses on the concave surface and releases adsorption heat to the desiccant. The most widely used solid desiccant dehumidification equipment is the rotary dehumidification wheel, in which solid desiccant is coated on the surface of wheel. The rotary dehumidification wheel can realize continuous dehumidification and regeneration through periodical rotation. The high-humidity process air passes through a portion of the desiccant wheel, and the vapor in the process air is absorbed by the solid desiccant of the wheel due to the vapor pressure difference between the air and the desiccant. Then, the temperature of the process air rises due to adsorption heat, and humidity is reduced after passing through the wheel. When the desiccant in the process air stream absorbs enough vapor from the process air, the vapor adsorption portion of the desiccant wheel is rotated to the high temperature regeneration air stream. While the regeneration air passes through the portion of the desiccant wheel containing a greater amount of vapor in the desiccant, the heat from the high temperature air stream leads to higher vapor pressure in the desiccant in comparison to the vapor pressure in the air stream. Then, vapor in the desiccant is ejected into the regenerated air stream until the lower vapor pressure condition in the desiccant is attained. The desiccant wheel periodically and dynamically rotates between the process and regeneration air streams. Dehumidification and regeneration processes are conducted in the corresponding air streams, and a low-humidity process air stream can be continuously obtained using the rotary wheel. Narayanan et al. [11] analyzed the performance characteristics of a solid-desiccant evaporative cooling system with TRNSYS software. The results show that the system could provide thermal comfort, however, the capability of the system to provide suitable air temperature and humidity depends on the performance of the evaporative cooling, energy-recovery, and heat-generation systems. Therefore, the availability of a cheap or waste-heat source is essential in making this system economically viable. Narayanan et al. [12] numerically investigated the dehumidification potential of a solid desiccant-based evaporative cooling system with an enthalpy exchanger operating in subtropical and tropical climates with TRNSYS software. The results show that in hot and humid conditions, the system thermal comfort capability drops to around 54% to 63%.

Solar energy utilization is a new approach developed in recent years specifically for space heating applications. The solar energy source is limitless and safer for the environment. Recently, heat from solar energy was developed for desiccant cooling systems. The heat generated by a solar thermal collector can be used for the regeneration process of the dehumidification wheel. Guidara et al. [13] proposed the use of evaporative coolers for pre-cooling and re-cooling of the process air before and after the dehumidification wheel in order to satisfy the load demand of an air-conditioned space. From their numerical analysis, they indicated that the pre-cooling design is suitable to be applied in drier ambient conditions. Enteria et al. [14] investigated the performance of a solar-desiccant cooling system with a silica gel and titanium dioxide desiccant wheel in East Asia. From numerical simulations, it was found that using solar desiccant cooling systems has great potential for East Asian countries. The study pointed out that, in the tropical region, a larger area and capacity of the solar thermal collector and greater regeneration air flow rate are required, and the operational performance of the system is in the range of 1.5 to 3. The major operational energy loss of the proposed system comes from solar collectors, water pipes, electric heaters and thermal storage tanks. Speerforck et al. [15] numerically investigated the performance of a solar-desiccant cooling system incorporating borehole heat exchangers for direct cooling and solar energy for desiccant regeneration. They indicated that the proposed system allows electricity saving of 50% and reduces CO2 equivalent emissions by 91%. White et al. [16] proposed a solar-assisted SDCS incorporating direct and indirect evaporative coolers in a serial arrangement to cool process air after the dehumidification wheel. From numerical analysis, it was indicated that the sensible heat removal capability of the process air can be enhanced by a two-stage evaporative cooling design, which is suitable for cities with low-humidity climatic conditions.

Environmental conditions in different regions may affect the performance of a desiccant cooling system (DCS). In hot and humid regions, the use of a hybrid SDCS that is integrated with a heat pump is suggested, instead of a conventional SDCS that utilizes an evaporative cooler and can generate a more humid air supply. Jani et al. [17] proposed a hybrid cooling system integrated with a heat pump (hybrid SDCS), and indicated the proposed system has good performance in hot and humid climate conditions. The dehumidification performance of the system is also highly sensitive to the outdoor ambient conditions. The performance of the hybrid SDCS is better for cases where ambient humidity is high [18]. It was found that the use of a hybrid SDCS can decrease the total power consumption by 20–30% and increase the cooling capacity by 40–60% [19]. Jani et al. [20] used the numerical software TRNSYS to simulate the performance of a solid desiccant-assisted hybrid space cooling system. The results show that the system achieves a good performance in hot humid climates. The humidity ratio of a room's process air was substantially lowered, from 0.014 kg of water/kg of dry air to 0.006 kg of water/kg of dry air, by use of a solid desiccant-based rotary dehumidifier.

In addition to environmental conditions, regeneration temperature is one of the main parameters used to determine the performance of a SDCS. The performance of an SDCS is sensitive to changes in humidity and regeneration temperature [21,22]. However, if the regeneration temperature in the hybrid SDCS is too high, the system performance may be reduced due to high condensation pressure and an increased amount of work performed by the compressor. Several heat sources can be used for the regeneration process, such as an electric heater, gas, a low-grade thermal energy such as solar energy, and waste heat. A further cogeneration system can also be considered as the heat source for the regeneration process.

Beccali [23,24] developed a new solar-assisted hybrid SDCS in which the condensing heat from the condenser of the incorporated heat pump is recovered to preheat the regeneration air stream before solar heating. Long-term measurements of the developed system were conducted in southern Italy. Five control modes based on the temperature and humidity of the outdoor air conditions were designed. From their performance analysis, it was pointed out that preheating of the recovered heat for the regeneration process can reduce the required heat from the solar collector by approximately 30%. In other words, the required area of the solar collector can also be moderately reduced. Fong et al. [25,26] developed a solar-assisted SDCS incorporating an adsorption chiller and analyzed the performance of the system by numerical analysis. The outdoor air was dehumidified by a rotary dehumidification wheel and passed through a radiant cooling coil in an air-conditioned space. The chilled water from the adsorption chiller was provided to the radiant cooling coil for handling the sensible heat of the air-conditioned space. Both required regeneration heating for the rotary dehumidification wheel

and the adsorption chiller supplied by the solar collector. Compared with the traditional centralized air conditioning system, the energy saving potential of the integrated system could reach 36.5%. As alternatives to silica gel, Bareschino et al. [27] proposed other hygroscopic materials for desiccant wheels, MIL101@GO-6 (MILGO) and Campanian ignimbrite, in conjunction with an air-conditioning system driven by evacuated tube solar collectors equipped with a desiccant wheel. The numerical simulations were carried out by means of TRNSYS 17® (version 17, Thermal Energy System Specialists, Madison, WI, USA) to dynamically assess the energy flows in the considered plants and compared with that of a conventional system. The results demonstrate that primary energy savings of approximately 20%, 29%, and 15% can be reached with silica-gel, MILGO and zeolite-rich tuff desiccant wheel-based air handling units, respectively. Li et al. [28] investigated a two-stage rotary desiccant cooling/heating system driven by evacuated glass tube solar air collectors. The results show that the major advantage of the two-stage desiccant cooling system was that moisture removal reached 6.68–14.43 g/kg in hot and humid climate conditions. Solar heating with desiccant humidification can improve indoor comfort significantly. A solar hybrid SDCS is a good substitute for traditional vapor compression air conditioning systems, especially in hot and humid climates, since solar energy can result in energy savings in the range of 40–45% [29]. Rambhad [30] indicated regeneration temperatures of hot water in the range of 54.3 ◦C to 68.3 ◦C can be achieved in solar SDCS systems by simulation.

In previous studies, the solar-assisted hybrid SDCS system was considered a replacement of the refrigerant vapor compression air conditioning system due to its higher energy efficiency. However, most studies of the solar-assisted hybrid SDCS have been conducted using numerical analysis. Research into the system's long-term practical operations and analysis of its performance under the effects of different ambient temperatures and humidity ratios is less common, especially in hot and humid environments. The performance of the solar-assisted hybrid SDCS has not been investigated experimentally in detail. In this study, the effect of ambient humidity and temperature on the performance of a solar-assisted hybrid SDCS was investigated under the high humidity and high temperature ambient conditions of Taichung, Taiwan. The performance analysis and comparisons of solar-assisted hybrid SDCS, hybrid SDCS and solar SDCS systems were conducted through long-term experiments in order to understand their characteristics and operational ranges in terms of environmental conditions.

#### **2. Experimental Configuration and Methods**

The experiments were conducted in the typical hot and humid weather of Taichung. The city has four seasons with an average temperature of 23.3 ◦C. The monthly average temperature, average relative humidity (*RH*) and the solar irradiation of Taichung from April to October are listed in Table 1. In the solar hybrid SDCS system, the heat from solar collectors and the condenser of an integrated heat pump sub-system are used for regeneration.


**Table 1.** Monthly average temperature, relative humidity (*RH*) and solar irradiation of Taichung [31].

#### *2.1. Proposed System Configuration*

In general, the proposed solar-assisted hybrid SDCS is divided into two major sub-systems. The first is a hybrid SDCS and the second is a solar-heated water system. The solar collector is installed

at a fixed tilt angle of 27◦ facing the south-east direction. Several configurations, including a hybrid SDCS system, solar-assisted SDCS systems and a solar-assisted hybrid SDCS system, were studied to investigate the effect of using a solar regeneration water system on system performance.

#### 2.1.1. Hybrid SDCS Configuration

The hybrid SDCS configuration was the first configuration investigated. In this configuration, the only heat source for the regeneration process is the condenser of the integrated heat pump, as shown in Figure 1. In this configuration, chilled water is generated by the evaporator and stored in a chilled water tank. The chilled water is pumped to pre-cooling and cooling coils in order to cool process air from the environment. In the integrated heat pump sub-system, an additional condenser installed outside the air handling unit can eject excessive condensation heat to the surroundings and avoid the over-loading of the compressor due to higher condensation temperature in the refrigeration cycle. The return air stream gains heat from the condenser inside the air handling unit, then heats the upper part of the solid desiccant wheel and causes the vapor pressure in the solid desiccant to be higher than the vapor pressure in the air stream. The moisture in the solid desiccant is ejected to the surroundings with the air stream. The moisture removal from the desiccant wheel by additional heat in the return air stream is called the regeneration process.

**Figure 1.** Hybrid solid desiccant cooling system (SDCS) configuration.

#### 2.1.2. Solar-Assisted SDCS Configuration

The solar-assisted SDCS configuration is shown in Figure 2. The configuration only uses solar-heated water from a heat exchanger as the heat source for regeneration. Therefore, the regeneration temperature of the system will not be as high as the regeneration temperature of the hybrid SDCS configuration. In terms of process air, latent heat is handled by the desiccant wheel. However, since the heat pump is turned off, the temperature gradually increases with operation time due to adsorption heat in the dehumidification process. Therefore, the supply air temperature will be higher compared to other cases.

#### 2.1.3. Solar-Assisted Hybrid SDCS Configuration

In the solar-assisted hybrid SDCS configuration, the condensing heat of the integrated heat pump and the heat of solar-heated water are used as heat sources for the regeneration process. In the regeneration air stream, the returning air is pre-heated by solar-heated water from a heat exchanger. Then, the returning air passes the condenser of the integrated heat pump to recover the dissipating heat from the condenser again in order to increase the regeneration temperature. In terms of process air,

the air handling process concept is the same as in the hybrid SDCS configuration. The corresponding system configuration is shown in Figure 3.

**Figure 2.** Solar-assisted SDCS configuration.

**Figure 3.** Solar-assisted hybrid SDCS configuration.

The system specifications are as follows:


In terms of air flow rates, the flow rates of the process air and regeneration airstreams are not the same due to a slight leakage in the system. However, the gap between the two flow rates is small. The flow rate of the process stream is 465 m3/hour, while the flow rate of the regeneration stream is 400 m3/hour. The flow rates of both air streams are obtained by adjusting the frequency of the fans. In addition, the flow rate of the solar-heated water heat exchanger is 40 L/min. The thermal effectiveness of the solar-heated water heat exchanger is 62.03%.

#### *2.2. Theoretical Analysis*

The system's performance is analysed using theoretical analysis. The coefficient of performance, *COPhvac* in Equation (1) is the ratio of the total cooling capacity, *Qc* (kW), to the total power consumption of the system, *Etot* (kW):

$$\text{COP}\_{\text{lrnc}} = \text{Q}\_{\text{c}} / \text{E}\_{\text{tot}} \tag{1}$$

The total cooling capacity of the system, *Qc* (kW), is calculated by Equation (2):

$$Q\_{\mathcal{L}} = \dot{m}\_p \left( h\_{\alpha \alpha} - h\_{sa} \right) \tag{2}$$

where *hoa* (kJ/kg) is the specific enthalpy of outdoor air and *hsa* (kJ/kg) is the specific enthalpy of supply air. The total power consumption in Equation (3) consists of the fan power, *Pfan* (kW), compressor power, *Pcomp* (kW), water pump power, *Ppump* (kW) and other electrical components, *Pother* (kW):

$$E\_{tot} = P\_{\text{fan}} + P\_{\text{comp}} + P\_{\text{pump}} + P\_{\text{other}} \tag{3}$$

In the supply air stream, the latent heat performance, *COPlt*, of the system in Equation (4) is the ratio of latent heat capacity, *Qlt* (kW), to the total power consumption of the system:

$$
\mathbb{C}OP\_{lt} = \mathbb{Q}\_{lt}/\mathbb{E}\_{tot} \tag{4}
$$

The latent heat capacity, *Qlt* (kW), of the system is calculated by Equation (5):

$$Q\_{lt} = \dot{m}\_{\text{P}} \ (h\_{\text{av}} - h') \tag{5}$$

where *h'* (kJ/kg) is the specific enthalpy of air with the temperature of the outside air and the humidity ratio of the supply air.

The specific moisture removal, *SMR* (kg/kgda), is the humidity ratio difference between the outdoor air, ω*oa* (kg/kgda), and the supply air, ω*sa* (kg/kgda), along the process air stream. The *SMR* value can be calculated by Equation (6):

$$SMR = \alpha \iota\_{\rm tot} - \alpha \iota\_{\rm sat} \tag{6}$$

The effectiveness of the desiccant wheel, ε, is defined by Equation (7):

$$
\varepsilon = \text{SMR} / (\omega\_{\text{ca}} - \omega\_{\text{i}}) \tag{7}
$$

where the ideal specific moisture, ω*i*, is assumed to be 0 kg/kgda; thus, the denominator is the maximum moisture that can be removed by the desiccant wheel system. The moisture removal rate, *MRR* (kg/h), can be calculated by Equation (8):

$$MRR = \dot{m}\_p \left(\omega\_{\text{ar}} - \omega\_{\text{sa}}\right) \text{ 3600} \tag{8}$$

where *m˙ <sup>p</sup>* is the mass flow rate of the process air stream (kg/s). The solar fraction, *SF*, is determined by Equation (9):

$$SF = E\_{sol} / E\_{tot.m} \tag{9}$$

where *Esol* (MJ) is the total useful solar thermal energy available in a month, which can be obtained from the solar collector capacity *Qsol* (kW). *Qsol* (kW) can be calculated by Equation (10).

$$Q\_{\rm sol} = \eta\_{\rm soc} \, A\_{\rm scc} \, R \tag{10}$$

where η*soc* is the solar thermal collector overall efficiency, *Asoc* (m2) is the gross area of the solar thermal collector, and *R* (kW/m2) is the total incident solar radiation. In this study, the efficiency, η*soc*, and the total area, *Asoc*, of the solar thermal collector are 94.5% and 11.699 m2, respectively. The total energy input in a month, *Etot.m* (MJ), can be obtain from the total power *Ptot* (kW). *Ptot* (kW) can be calculated by Equation (11):

$$P\_{tot} = I\_{tot} \text{ V} \tag{11}$$

where *Itot* and *V* are the total current and voltage of the system, respectively.

#### **3. Results and Discussion**

#### *3.1. Comparison of Average Temperature Declination and Specific Moisture Removal*

A comparison of the temperature declination (*Td*) and *SMR* of each system in different months in 2018 is shown in Table 2, where *Td* is the temperature drop between the average outdoor air temperature (*Toa*) and the average supply air temperature (*Tsa*).

According to the table, the temperature declination of the hybrid SDCS and solar-assisted hybrid SDCS are almost the same, while the solar-assisted SDCS does not have a cooling effect because the heat pump is turned off. The cooling and moisture removal of the process air by the solar-assisted hybrid SDCS are the highest among the three configurations. It can be seen that, for the solar-assisted SDCS, the average temperature declination is in the range of 3.65 ◦C to 6.79 ◦C, which gradually increases from April to June and then gradually decreases until October. Regarding moisture removal, the average *SMR* value is in the range of 0.0033 kg/kgda to 0.0075 kg/kgda, which also gradually rises from April to June then declines until October.


**Table 2.** Comparison of *Td* and specific moisture removal (*SMR*) for different system configurations.

#### *3.2. Comparison of Regeneration Temperatures for Each System Configurations*

A comparison of the regeneration temperature of each system during one day is shown in Figure 4. The regeneration temperature samples shown are for the system operating in better ambient conditions with enough sunlight intensity. As shown in Figure 4, the peak regeneration temperature value is generally reached between 14:00 and 16:00 each day. The regeneration temperature of each system is affected by the ambient temperature, especially for the solar-assisted SDCS and solar-assisted hybrid SDCS, in which the heat of solar-heated water is used as an additional heat source.

**Figure 4.** Comparison of regeneration temperature in each system configuration.

It can be observed from Figure 4 that, for a given time, the regeneration temperature of the solar-assisted hybrid SDCS is the highest among the three configurations; the temperature gradually rises from 57 ◦C at 06:00 to a peak value of 79 ◦C at 14:00, then gradually decreases to a lowest value of about 55 ◦C. In the case of the hybrid SDCS, the regeneration temperature distribution is similar to that of the solar-assisted hybrid SDCS, which is in a range of 49 ◦C to 65 ◦C. A peak value of 65 ◦C is attained at about 16:00. In the case of the solar-assisted SDCS, the regeneration temperature is in a range of 32 ◦C to 43 ◦C, and the peak value of 43 ◦C is attained at 16:00. These tendencies also occurred

on other days. In general, the peak regeneration temperature is reached in the time range from 14:00 to 16:00.

Solar fraction (*SF*) is an important technical indicator to assess the feasibility of solar cooling systems. The higher the value of *SF*, the greater the contribution of solar energy to the system. *SF* is the ratio of the solar energy contribution to the total energy input needed to drive the solar cooling system. The total useful solar energy was described in the previous section and the total energy input of the system is the total energy input for the system's operation per month. The average *SF* value of the solar-assisted hybrid SDCS for each month is provided in Figure 5. As shown in Figure 5, it can be observed that the solar collector system shows a high contribution in July. This is because it is the peak summer season at this time, so the energy gained from the solar system is higher. On the other hand, the total energy required to operate the system is approximately constant in each month. The *SF* of the system in July is 0.82 and the lowest *SF* occurs in March with the value of 0.673.

**Figure 5.** Solar fraction (*SF*) of solar-assisted hybrid SDCS.

#### *3.3. Comparison of Specific Moisture Removal and Moisture Removal Rate*

The effect of increasing relative humidity (*RH*) on *SMR* for each configuration is shown in Figure 6, where the outdoor air temperatures are 26 ◦C, 28 ◦C and 30 ◦C, respectively. However, the maximum *RH* attained by the three configurations at 28 ◦C and 30 ◦C is only 80% due to the effect of the ambient temperature of the seasons during the experiments. For a constant temperature, the value of relative humidity, *RH* (%), determines the humidity and specific vapor pressure of the process air. At the same process air temperature, higher *RH* values result in higher humidity ratios and vapor pressure. The higher vapor pressure value results in higher pressure differences between the process air and solid desiccant material, which can enhance the adsorption effect of the desiccant material. Thus, the *SMR* value of each configuration tends to increase with the increase in the *RH* value. At the same relative humidity value, the higher humidity ratio of the process air with higher temperature leads to a higher vapor pressure of the process air; thus, the *SMR* value of each configuration is also better for the higher outdoor temperature of 30 ◦C, as shown in Figure 6. In the solar-assisted SDCS, the *SMR* of the system is relatively low. The system only has a dehumidification effect if the *RH* value is greater than 65%. The *SMR* value of the configuration at an outdoor temperature of 26 ◦C and 85% *RH* is 0.00262 kg/kgda.

The *SMR* of the hybrid SDCS is greater than the solar-assisted SDCS for any *RH* value, with the highest value of 0.0048 kg/kgda at an outdoor temperature of 26 ◦C and 85% *RH*. The *SMR* of the solar-assisted hybrid SDCS is the largest under all relative humidity conditions, with a maximum value of 0.0063 kg/kgda under the same operational conditions. The reason for this phenomena is that the solar-assisted hybrid SDCS has the highest regeneration temperature, which can enhance the adsorption effect of the desiccant material.

**Figure 6.** Effect of outdoor relative humidity (*RH*) on *SMR*.

Regarding the humidity ratio effect on *MRR* for each configuration, Figure 7 shows the distributions of moisture removal rate (*MRR*) of the three configurations with the increase in the humidity ratio at an outdoor operational temperature of 26 ◦C. However, the humidity ratio range of the solar-assisted SDCS and hybrid SDCS is narrower than the humidity ratio range of the solar-assisted hybrid SDCS due to the weather conditions during the experiments. As shown in Figure 7, larger humidity ratios lead to better *MRR* values. The solar-assisted SDCS has the lowest *MRR* value due to lower regeneration temperature. The *MRR* value of the hybrid SDCS and solar-assisted hybrid SDCS is better in comparison to the solar-assisted SDCS configuration. This indicates the regeneration temperature of both systems is sufficient to reactivate the desiccant material. Thus, the *MRR* values of the two configurations are also larger. When the humidity ratio of outdoor air reaches 0.018 kg/kgda, the *MRR* can attain to a high value of 3.5 kg/h in the case of the solar-assisted hybrid SDCS.

**Figure 7.** Effect of outdoor humidity ratio on moisture removal rate (*MRR*).

#### *3.4. E*ff*ect of Relative Humidity to Desiccant Wheel E*ff*ectiveness*

The specific moisture removal (*SMR*) value of a solid desiccant configuration determines the moisture removal capability of the system. It can represent the vapor mass of process air absorbed by the solid desiccant material to be ejected to the surroundings. The moisture removal effectiveness of solid desiccant is also determined by the *SMR* value. On the other hand, the *SMR* value is mostly affected by the relative humidity of the outdoor air. The effect of the relative humidity of outdoor air on solid desiccant effectiveness at constant air temperatures of 26 ◦C and 28 ◦C is shown in Figure 8. As shown in Figure 8, the effectiveness of solid desiccant increases with the rise in outdoor air relative humidity. At the outdoor temperature of 26 ◦C and outdoor relative humidity of 50%, the desiccant effectiveness is 0.274, and this value continues to increase as relative humidity increases. The effectiveness value at a relative humidity of 85% is 0.36. The increase in the relative humidity value at the same outdoor air temperature leads the humidity ratio of outdoor air, ωoa, rises, together with the *SMR* value. In the case of this study, the solar-assisted hybrid SDCS increment value of *SMR* in all configurations is greater than the increment of the humidity ratio of outdoor air. Thus, from the definition of desiccant effectiveness, the effectiveness value gradually increases with the increase in relative humidity at the same outdoor air temperature. However, at the same relative humidity value, the effectiveness of solid desiccant wheel in the case with higher process air temperature is less than that with lower process air temperature. The effectiveness of the desiccant wheel is the ideal specific moisture, ωi, which is assumed to be 0 kg/kgda, so the denominator is the maximum moisture that can be removed by the desiccant wheel system. Increasing outdoor temperature with the same relative humidity causes the process air to have a higher humidity ratio. However, with the increase in outdoor air temperature, the increase in the *SMR* value is lower relative to the ideal moisture able to be removed by the system. Therefore, the effectiveness decreases with the temperature of the process air at the same relative humidity ratio.

**Figure 8.** Effect of outdoor *RH* on desiccant effectiveness.

#### *3.5. E*ff*ect of Regeneration Temperature to Specific Moisture Removal and Latent Heat Performance*

It was mentioned in the previous section that the regeneration temperature impacts the moisture removal ability of the SDCS. The effect of regeneration temperature on specific moisture removal (*SMR*) of the SDCS is shown in Figure 9. The present investigation shows the regeneration temperature effect of each system. The regeneration temperature of the solar-assisted SDCS is in the range of 35–45 ◦C, while the regeneration temperature of the hybrid SDCS ranges from 50 to 65 ◦C and the regeneration temperature of the solar-assisted hybrid SDCS ranges from 55 to 70 ◦C. The regeneration temperatures can then be drawn as a single line as shown in Figure 9. The effect of regeneration temperature on specific moisture removal and *COPlt* are thus investigated in this section. In addition to ambient temperature, the regeneration temperature also depends mostly on the capacity of the heat source. In Figure 9, the results are shown for a relative humidity ratio range of 58–62%. As shown in Figure 9, it can be concluded that a higher regeneration temperature leads to a higher *SMR* value. Consequently, the supply air condition will also be drier. If the regeneration temperature is in the range of 35–45 ◦C, the *SMR* is less than 0.002 kg/kgda, which is not sufficient for the supply air condition. The average regeneration temperature of the solar-assisted SDCS in this study is about 40.4 ◦C. Even the dehumidification effect will not be apparent if the regeneration temperature is less than 35 ◦C. However, for the hybrid SDCS, the average regeneration temperature is about 53.9 ◦C. The higher average regeneration temperature of the hybrid SDCS provides a better *SMR* value than the solar-assisted SDCS. In addiition, the solar-assisted hybrid SDCS possesses the highest regeneration temperature, with an average value of 63.7 ◦C. Therefore, the *SMR* value of the solar-assisted hybrid SDCS is higher and the dehumidification performance is better among the three configurations.

**Figure 9.** Effect of regeneration temperature on *SMR*.

The effect of regeneration temperature on latent heat performance (*COPlt*) can be observed in Figure 10. In this case, the outside air relative humidity (*RH*) is also in the range of 58–62%. It can be seen that the *COPlt* value increases as the regeneration temperature rises from 35 ◦C to 70 ◦C. The regeneration temperature depends on the heat generated by heat sources. In the solar-assisted SDCS with the regeneration temperature range of 30 ◦C to 45 ◦C, the *COPlt* of the system is comparatively low. Therefore, the dehumidification effect is also low. The *COPlt* of the system at a regeneration temperature of 45 ◦C is 0.556.

**Figure 10.** Effect of regeneration temperature on latent heat performance (*COPlt*).

In the hybrid SDCS, in which the regeneration temperature is in the range of 50 ◦C to 65 ◦C, the dehumidification performance is better in comparison to the previous case. The increase in regeneration temperature shows a performance improvement in the system. The *COPlt* of the hybrid SDCS under an average regeneration temperature of 55 ◦C is 0.643. The *COPlt* value shows a massive inclination in the range from 55 ◦C to 70 ◦C. It indicates that the optimum regeneration temperature of the desiccant cooling system is 70 ◦C or higher. However, in this case, the maximum temperature that can be generated by the system is 70 ◦C. This condition occurs while the system utilizes the solar-assisted hybrid SDCS. The heat from the solar water heat exchanger can effectively increase the regeneration temperature of the system. The *COPlt* of the solar-assisted hybrid SDCS at the average regeneration temperature of 65 ◦C is 0.694 and at the maximum regeneration temperature 70 ◦C is 0.88.

#### *3.6. Comparison of System Total Performance for Di*ff*erent Outdoor Air Temperatures*

Figure 11 shows a comparison of *COPhvac* for each configuration at different outdoor temperatures.

**Figure 11.** Comparison of the coefficient of performance (*COPhvac*) at different outdoor temperatures.

The humidity ratio is maintained at 0.016 kg/kgda and 0.017 kg/kgda. However, the *COPhvac* of the solar-assisted SDCS is not investigated since it does not have a cooling effect. Figure 11 indicates the *COPhvac* is greater for the solar-assisted hybrid SDCS than for the hybrid SDCS, especially at high outdoor temperatures. At low outdoor temperatures, the solar sub-system does not have a significant effect on the system, since the heat from solar-heated water and the regeneration temperature are smaller. At outdoor temperatures from 26 to 30 ◦C, the performance enhancement of the solar-assisted hybrid SDCS is obvious. At a humidity ratio of 0.016 kg/kgda, the maximum *COPhvac* of 1.52 can be reached by the solar-assisted hybrid SDCS at an outdoor temperature of 28 ◦C, and the maximum *COPhvac* of 1.58 can be reached by the system at an outdoor humidity of 0.017 kg/kgda.

The solar-assisted hybrid SDCS has better performance at high outdoor temperatures than the hybrid SDCS. However, if the outdoor temperature is too high, system performance is degraded because the condensation temperature rises with the increase in outdoor temperature, which leads to an increase in the power consumption of the compressor. The maximum outdoor temperature in the solar-assisted hybrid SDCS is 30 ◦C, and in the hybrid SDCS is 26 ◦C.

#### *3.7. Comparison of System Total Performance for Di*ff*erent Outdoor Humidity*

Figure 12 shows a comparison of *COPhvac* for each system configuration at different outdoor humidities. The outdoor air temperatures in this figure are 26 ◦C and 32 ◦C, respectively. The *COPhvac* of the configuration at an outdoor temperature of 32 ◦C and humidity less than 0.013 kg/kgda are not provided due to operational environment limitations. The result shows an increase in the outdoor humidity leads to better system performance. The *COPhvac* inclination of solar assisted SDCS in every 1 g/kgda humidity ratio is about 6.17%, which is higher than that of the hybrid SDCS with the vaue of 5.4%. The *COPhvac* of the solar-assisted hybrid SDCS is approximately 3.7% greater than that of the hybrid SDCS. The reason for this is that the solar-assisted hybrid SDCS has better *MRR* and dehumidification effects. The maximum *COPhvac* at a humidity ratio of 0.018 kg/kgda and a temperature of 26 ◦C is 1.53 in the solar-assisted hybrid SDCS and 1.49 in the hybrid SDCS. The *COPhvac* of systems at an outdoor temperature of 32 ◦C is approximately 18% lower than at 26 ◦C due to the temperature effect. At an outdoor temperature of 32 ◦C, the *COPhvac* of the solar-assisted hybrid SDCS system is 7% greater than that of the hybrid SDCS. The maximum *COPhvac* of the solar-assisted hybrid SDCS at an outdoor temperature of 32 ◦C is 1.369 for a humidity ratio of 0.0175 kg/kgda, while the maximum *COPhvac* of the hybrid SDCS is 1.218.

**Figure 12.** Comparison of *COPhvac* at different outdoor humidity ratios.

#### **4. Conclusions**

In this study, a solar-assisted hybrid SDCS system was developed in which solar-heated water is used as a heat source for the regeneration process in addition to heat from the condenser of an integrated heat pump. A solar thermal collector sub-system is used to generate the solar regeneration water. Experiments were conducted in the typical hot and humid weather of Taichung, Taiwan, from the spring to fall seasons. According to the experiment results, several points can be concluded from the study, as follows:


**Author Contributions:** Research concept were proposed by W.-J.L., D.F. and F.R.F.; data processing and the manuscript preparation were conducted by Y.-S.C., F.H.M. and U.N.A.; data analysis and interpretation were implemented by W.-J.L., D.F. and F.R.F.; manuscript editing was performed by W.-J.L., and F.H.M.

**Funding:** This research was funded by the Ministry of Science and Technology of Taiwan under grant number MOST 106-2221-E-167-026 and MOST 104-2221-E-167-026-MY2.

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

#### **Nomenclature**


#### **Subscripts**


#### **References**


© 2019 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 (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Condition Monitor System for Rotation Machine by CNN with Recurrence Plot**

#### **Yumin Hsueh, Veeresha Ramesha Ittangihala, Wei-Bin Wu, Hong-Chan Chang and Cheng-Chien Kuo \***

Department of Electrical Engineering, National Taiwan University of Science and Technology, Taipei City 10607, Taiwan

**\*** Correspondence: cckuo@mail.ntust.edu.tw

Received: 2 August 2019; Accepted: 19 August 2019; Published: 21 August 2019

**Abstract:** Induction motors face various stresses under operating conditions leading to some failure modes. Hence, health monitoring for motors becomes essential. In this paper, we introduce an effective framework for fault diagnosis of 3-phase induction motors. The proposed framework mainly consists of two parts. The first part explains the preprocessing method, in which the time-series data signals are converted into two-dimensional (2D) images. The preprocessing method generates recurrence plots (RP), which represent the transformation of time-series data such as 3-phase current signals into 2D texture images. The second part of the paper explains how the proposed convolutional neural network (CNN) extracts the robust features to diagnose the induction motor's fault conditions by classifying the images. The generated RP images are considered as input for the proposed CNN in the texture image recognition task. The proposed framework is tested on the dataset collected from different 3-phase induction motors working with different failure modes. The experimental results of the proposed framework show its competitive performance over traditional methodologies and other machine learning methods.

**Keywords:** induction motor; convolutional neural networks (CNN); recurrence plots (RP); time-series data (TSD)

#### **1. Introduction**

Induction motors are electromechanical devices used in most industrial applications. Due to their simple design and well-developed manufacturing technologies, induction motors are considered relatively reliable and robust [1]. However, the motors fall into failure mode and seriously affect industrial operations. Eventually, this leads to failure of the entire operating system if the failure condition is not identified or if it is neglected. Several types of faults related to winding, stator, rotor, and bearing can be observed in an induction motor [2,3]. There are mainly four types of fault diagnosis methods such as signal-based, model-based, knowledge-based, and active/hybrid methods. Signal-based methods use the measured signal to extract the features and make a diagnostic decision based on the prior knowledge of the diagnostic process. Signal-based methods can be classified into a time-domain signal-based method, frequency-domain signal-based method, and time-frequency signal-based method. Model-based methods can be categorized into deterministic fault diagnosis methods, fault diagnosis methods for discrete-events and hybrid systems; stochastic fault diagnosis; and fault diagnosis methods for distributed and network systems, which are categorized by the model type used [4]. Hybrid methods are studied as a combination of two or more fault diagnosis methods. For example, He et al. [5] diagnosed plastic bearing faults by combining the signal-based and data-driven methods. By combining signal-based and knowledge-based techniques, a fault diagnosis method was studied to detect the inter-turn faults in induction motors, in which wavelet transform is applied to extract the features from the collected vibration signals, and the principal component

analysis (PCA) and neural network (NN) were used as classifiers to classify healthy from faulty motors [6]. Active methods are studied as a system in which a suitably designed input signal is injected into a dynamic process during a test period to distinguish accurately and quickly the faulty modes from the normal modes. Stochastic active fault diagnosis and deterministic active fault diagnosis are the two active methods studied [7–9]. For example, Campbell et al. [10], used two candidate models one as a normal system and another as a fault system in a multimodel system, and an auxiliary signal was designed to detect the correct model under a given interval of test time. As an extended study, an active fault detection method for multiple faults generated simultaneously or sequentially [11].

Knowledge-based methods, also considered as data-driven methods, are the most commonly used methods for analyzing signals such as vibration, temperature, electrical tension, and current. These methods require a huge amount of historical data to find the patterns in the given signal. However, data signals can be captured using sensors [12,13]. Signal-based features are extracted, and feature selection methods are applied to reduce feature dimensions and also to avoid the repeated information, which in turn improves the performance by holding the significant features [14]. The extracted features are used for fault diagnosis by various traditional machine learning methods [15–17]. Traditional machine learning methods have achieved prominent results. However, feature extraction depends significantly on diagnostic knowledge and signal processing expertise. Furthermore, traditional methods are incapable of extracting discriminative features from raw data and always require a process to extract the feature from the signal [16,18–21].

In spite of the advanced development in machine learning, deep learning (DP) has become the most effective study that can significantly overcome the drawbacks of traditional machine learning methods for fault diagnosis. DP can automatically extract and learn abstract features from raw data, and avoid manual feature extraction [18]. Many deep learning models, such as deep belief network (DBN) [22], stacked sparse auto-encoder [19], sparse auto-encoder [21], denoising auto-encoder [23] and sparse filtering [24] have been studied to diagnose the faults, and very significant results have been achieved. One of the most effective used deep learning model, called convolutional neural network (CNN), has been used to learn hierarchical feature representation from raw data and has delivered promising results [20–25].

Most pattern recognition tasks deal with time-series data signals. Financial data (stock and currency exchange rates), video processing, music mining, weather and forecasting, biometric data, and biomedical signal processing are few examples, in which time-series data have been studied [26–28]. Likewise, electrical industrial devices also often work with time-series data such as measurements of voltage, current, temperature, and vibration signals. One-dimensional CNN has been studied and applied to time-domain machinery data signals to diagnose the faults in induction motors [29]. In a few cases, machinery data can also be represented in two-dimensional (2D) model, such as the time-frequency domain using the wavelet transform technique [30]. In addition, time-series data can be represented in 2D texture images using the concept called recurrence plot (RP) [31]. Image representation of time-series data provides a different set of features that are not available for 1D signals. Therefore, 2D texture images can be used for classification [32].

In this study we use a time-series data signals converted as recurrence plots (RP) to feed the proposed deep CNN for fault diagnosis. Current signals with a phase difference were collected from 3-phase induction motors, and each phase current was used as an input data sample. Our results show that RP provides an effective method for viewing trajectory periodicity over a phase space, allowing us to discover specific elements of the m-dimensional phase space trajectory using a 2D figure. The main contributions of this paper are illustrated as follows. First, the raw current signals are represented as RP images. Second, an efficient deep CNN model is studied and applied on RP images to extract the multi-level features for fault classification. Lastly, the proposed deep CNN based framework achieves significant results compared to other deep learning methods.

The rest of the paper is arranged in the following way: Section 2 discusses the related works. Section 3 explains the proposed framework—time-series data to 2D texture image conversion—and

proposed deep CNN models are discussed. Section 4 presents the experimental results followed by the conclusions and future work presented in Section 5.

#### **2. Related Works**

This section briefly reviews recent deep learning contributions on induction motors' fault diagnosis. Several types of signal processing methodologies have been studied in the time-domain, frequency-domain, and time-frequency domain to extract and learn the features in order to classify the working condition of the motor. Lee et al. [33] studied the convolutional deep belief network (CDBN) to classify audio signals. They converted time-domain into frequency-domain data to learn the features form audio signals. A multi-channel CNN has been studied to handle multi-variate time-series data [34]. A separate CNN is used to learn the features from individual time-series data, and result from all the CNNs are combined and classified using a fully connected multilayer perceptron (MLP) classifier. Audio signals are transformed into a time-frequency domain to feed into CNN for classification [35]. The Gramian Angular Field (GAF) and Markov Transition Field (MTF) are used to convert time-series signals as images. A tiled CNN is used to classify time-series images [36,37].

Ngaopitakkul et al. [15] explain a decision algorithm based on ANN for fault diagnosis. Pandya et al. [17] propose an efficient KNN (k-Nearest Neighbours) classifier using the asymmetric proximity function for fault diagnosis. Yang et al. [16] constructed an SVM (Support Vector Machine)-based method to diagnose the fault patterns of the roller bearings. Jia et al. [38] propose a fault diagnosis method based on deep neural networks using an auto-encoder. Deep learning models such as deep autoencoder (DAE), deep belief network (DBN), and CNN have been discussed for fault diagnosis [19,20,22]. Ince et al. [26] propose a one-dimensional (1D) CNN to diagnose faults using real-time motor data. Abdeljaber et al. [39] studied 1D CNN to detect real-time structural damages. A deep CNN was used to analyze multichannel time-series data signals for human activities [40]. However, these models only used a small amount of low-level features in hidden layers. However, in this paper, we study a deep CNN method to automatically learn the useful texture features in order to classify faults. 1D raw current signals were converted to 2D images, and the proposed CNN model was able to successfully capture the temporal and spatial dependencies in the images by applying relevant filters. Furthermore, the proposed model was able to extract and learn high-level features from these images along with the low-level features. The performance of fault classification improved by the combined implementation of feature extraction and the CNN classifier.

#### **3. Proposed Study and Framework**

This section explains the proposed framework based on RP images and deep CNN for fault diagnosis. It consists of two subsections: (1) the time-series data signals are converted into 2D texture images, and (2) a deep convolutional neural network model is discussed to learn features from texture images for fault classification.

#### *3.1. Time-Series Data to 2D Texture Images*

The time-series data can be categorized using a unique recurring behavior such as periodic and irregular cyclic aspects. Moreover, time-series data are generated as the repetition of states, which is a normal phenomenon for ever-changing irregular systems or random processes. RP [31,32] is a tool for visualizing and investigating the m-dimensional phase space trajectory using a 2D representation of its repetitive occurrences. The primary idea of RP is to disclose trajectory movements from the current state to the previous state and it can be formulated as:

$$R\_{i,j} = \theta(\epsilon - \|\overrightarrow{\mathbf{s}\_l} - \overrightarrow{\mathbf{s}\_m}\|), \ \overrightarrow{\mathbf{s}}\,(.) \in \mathfrak{R}^n, \ l, m = 1, 2, \dots, K$$

where *K* is the number of states <sup>→</sup> *<sup>s</sup>* , is a threshold value of distance, . is the norm and θ(.) is the Heaviside function. The recurrence matrix (*R*) comprises two sets of values called texture and typology. The texture information belongs to individual dots, sloping lines, perpendicular lines, and horizontal lines, whereas the typology information categorized by uniform, regular, shift, and interrupted. Obviously, in RP, there are patterns and information that are not easily visually seen and interpreted. The detailed explanation can be found in [32].

Raw current signals are collected from 3-phase induction motors for the fault analysis every 5 s with a sampling rate of 10,000 samples per second. Data samples taken for different periods of time from 1 s to 5 s were investigated with recurrence plots. Five seconds of data samples gave the most distinguishable patterns in the recurrence plots. The collected raw current signals from the operating induction motor are represented as recurrence plot shown in Figure 1.

**Figure 1.** Feature representation. Time-series data into a 2D (Two dimensional) texture images (recurrence plot).

As shown in the Figure 2, nondistinguishable recurrence plots were generated for two different motors operating with different modes of failure, when the raw current signal values were used to generate recurrence plots. Even though, the motors working with different modes such as faulty or healthy, it can be clearly seen that it is almost impossible to find a distinguishable pattern in these recurrence plots—they look exactly the same with no difference in any color or pattern. However, to find distinguishable patterns in RP images, an effective preprocessing technique called Max–Min difference was used in this study, and it is implemented as follows:

**Figure 2.** Recurrence plots before signal preprocessing. Left: recurrence plot for fault condition (bearing axis deviation) motor. Right: recurrence plot for healthy condition motor.

Step 1: maximum and minimum peaks of the current signal are collected for each one full cycle. Step 2: difference between the maximum and minimum peak value is then used to generate the recurrence plot for the whole signal.

Step 3: the above two steps are repeated for all types of faults, and healthy motor signals generate distinguishable recurrence plot.

As shown in the Figure 3, after applying the preprocessing technique to raw current signal values, the generated recurrence plots are well distinguishable and can be considered for classification of

different faults and healthy conditions of the motors by CNN. Initially, the 2D recurrence texture images were generated by raw one dimensional (1D) current signals and then classifier automatically learned the features from texture images to classify the motors' fault condition.

**Figure 3.** Recurrence plots with (Max–Min) preprocessing technique. Left: recurrence plot for fault condition (bearing axis deviation) motor. Right: recurrence plot for healthy condition motor.

#### *3.2. The Methodology's Architecture*

The methodology's architecture consists of two parts. Part one explains the architecture used to train the CNN model and part two belongs to motor testing using the trained model.

The relevant data were collected for a total of five conditions of the induction motor. A setup of four induction motors for the following four faults and one motor for the healthy condition were used. The four faults were: (1) bearing axis deviation, (2) stator and rotor friction, (3) rotor end ring break, and (4) poor insulation.

As shown in Figure 4, the training setup had two stages. The first stage setup was done in one of the lab servers (lab server). Data for all conditions of the motor were collected in this server in CSV format. The dataset comprised of 3-phase current signals. Data preprocessing was applied to the raw current signals to generate the recurrence plots. Recurrence plots' 2D texture images were stored into the S3 database hosted on the cloud platform (EI-PaaS).

**Figure 4.** The architecture used while training the CNN (Convolutional Neural Network) model for motor fault diagnosis. Left: the setup used in one of the lab servers, where the historical motor data are collected and stored for different types of faults and healthy conditions. Right: this setup is on the cloud platform called EI-PaaS, where the S3 database and the CNN model are designed.

The second stage setup is in the cloud. The Edge Intelligent Platform as a Service (EI-PaaS) has the analytical framework service, where the CNN model was implemented along with the S3 database. While training the CNN model, the images were maintained in the temporary directory structure. Figure 5 illustrates the setup used for the deployment/testing phase of the application. The architecture has a 3-phase induction motor connected to a data acquisition system (DAQ), which sends motor-related data such as current signal values in binary format to an Edge device. The Edge device reads the binary data and converts the data into decimal format and stores them as a CSV file. The data stored in the CSV file are used to generate the relevant RP image. The generated RP texture image is fed to the well-trained CNN model to diagnose and classify the motor condition as one of the four faults or healthy condition.

**Figure 5.** The architecture of the framework setup used in testing/deployment of the CNN model application for motor fault diagnosis. Left: this setup has an induction motor connected to a data acquisition system and Edge device for data preprocessing. Right: this setup has S3 database to store RP, the trained CNN model for motor fault diagnosis in the cloud platform (EI-PaaS).

#### *3.3. Architecture of the Proposed CNN Model*

The proposed deep CNN model has a three-stage structure. Each stage representing a feature learning stage with different feature-levels and it includes convolution, activation, and pooling layers.

As shown in Figure 6, the proposed deep CNN model has three convolutional layers with 32–3 × 3 filter, 64–3 × 3 filter, and 128–3 × 3 filter, respectively. In addition, three max-pooling layers of pooling size 2 × 2 were used. Type of layers, output shape of each layer, along with the number of trainable parameters are listed in Table 1.

**Figure 6.** The proposed deep convolutional neural network architecture diagram. The RP images are resized to 64 × 64 and fed as input to the CNN model. The architecture consists of one input and two hidden layers followed by pooling layers and a fully connected layer.


**Table 1.** Proposed CNN model summary.

Total params: 242,661, Trainable params: 242,661, Non-trainable params: 0.

The activation function Leaky ReLU (Rectified Linear Units) was applied to introduce nonlinearity into each stage, allowing CNN to learn complex models. A specific reason for adding Leaky ReLU was to avoid and attempt to fix the problem of dying ReLUs. It has proven to be more effective than the logistic sigmoid function. However, during the training, ReLU units can die and this could occur when large gradient flows through a ReLU neuron. It causes the weights to update such that the neuron will never activate again on any data point. Leaky ReLU makes an attempt to solve this problem [41,42]. Pooling layers were introduced to reduce the resolution of the input image by the process of subsampling and the max-pooling was applied in the proposed model.

At the end of the three stages, the feature maps were flattened into a column vector. The flatted output vector supplied to a feed-forward neural network and backpropagation was employed to every iteration of training. During training, the proposed model was able to distinguish among the dominating and also low-level features in texture images and classify by a fully connected layer for five types of faults. To estimate the parameters of the proposed model, one of the gradient-based optimization (backpropagation algorithm) methods was used. Adam optimizer was used to update the parameters to achieve faster convergence [43].

#### **4. Experimental Results and Discussion**

To assess the performance of the proposed methodology, raw current signals from an experimental setup of a total of five induction motors with the same configuration were used. One healthy and four fault types of raw current data signals were collected from the experimental setup. The different five current signals were studied and analyzed for the healthy condition of the motor, as well as for the following four faulty conditions of the motors [44]. The raw signal to image conversion method and the CNN model implementation were written in python 3.6 with TensorFlow and run on the Windows 64 bit operating system.

#### *4.1. Healthy and Fault Conditioned Motors*

1. Bearing axis deviation: this condition is considered as class 'Fault 1'. This happens due to the offset of centers on both sides of coupling when the motor is connected to load.


The proposed framework uses induction motor raw current signal values to generate recurrence plots for respective conditions. Generated recurrence plots are used as input images to the CNN model for further classification on the fault conditions of motors.

#### *4.2. Dataset*

The dataset was collected from a lab setup for four fault scenarios and one healthy scenario. The setup included five motors each for four faults and one healthy condition operating at full load. The dataset consisted of 750 samples and 150 samples for each type of scenario as described in Table 2. At 5 s of sampling rate, 50,000 raw current signal data points were collected per sample.



The dataset was divided mainly into three parts as described in the Table 3. Sixty percent of the dataset (450 data samples for training) and 15% of the dataset (112 data samples for validation) were used simultaneously to train the CNN model. The remaining 20% (188 data samples) were used to test the trained CNN model.


**Table 3.** Dataset samples used for evaluation.

#### *4.3. Performance Results of the Proposed CNN*

The proposed deep CNN model was trained over 100 epochs to learn the multi-level features for one healthy and for each type of the motors' faulty condition. The deep CNN model was trained to automatically train and learn the robust features from 450 samples of training data and simultaneously validated against 112 data samples during each iteration. Then, the trained deep CNN model was tested against the 188 samples of the testing dataset. The proposed deep CNN model was trained and verified with a batch of 16, 32, and 64. The deep CNN model achieved the best results with a batch of size 32.

The accuracies and losses were collected for each iteration while training the proposed deep CNN model, and plotted as shown in the Figure 7. The CNN model was able to learn the complex features efficiently with reaching almost 100% accuracy, and ~97–98% validation accuracy. Over the 100 epochs, the proposed deep CNN model was able to learn the generalized features from the recurrence plot texture images to diagnose the induction motor faults and classify the motors as faulty or healthy.

**Figure 7.** Accuracy and loss curves over 100 epochs of CNN training.

The performance of the proposed deep CNN was evaluated using a testing dataset consisting of 188 data samples. The performance results of the trained deep CNN model were very prominent with an average accuracy of 98% as described in Table 4. The classification report clearly explains how well the proposed deep CNN model was able to extract and learn the features from the testing data samples and classify the features into respective classes. In the classification report, the average values of the evaluation metrics such as precision, recall, and f1 score on test dataset were impressive, with 98% of accuracy. The proposed model was able to classify Fault 3 and Fault 0 (healthy) conditions accurately. However, the proposed model misclassified other faults, but with an acceptable margin. The confusion matrix for the test data was calculated using the trained deep CNN model (Figure 8). Almost all test samples were accurately classified with very few misclassifications.

In order to evaluate the performance of the proposed framework relative to traditional and other deep learning methods, the proposed framework was compared and the performance metrics were collected in terms of accuracy. The proposed framework was compared to support vector machine SVM [45], artificial neural network (ANN) [22], adaptive deep convolutional neural network (ADCNN) [46], sparse filter [24], and deep belief network (DBN) [22].


**Table 4.** Classification report on the test dataset.

**Figure 8.** Confusion matrix for the testing dataset (Fault 0: healthy, Fault 1: bearing axis deviation, Fault 2: stator and rotor friction, Fault 3: rotor end ring break, and Fault 4: poor insulation).

The prediction accuracy of these methods was calculated (Table 5). Comparing the performance results explains the significant results achieved by the proposed deep CNN framework compared to the other methods listed above. A prediction accuracy of 99.81%, shows the significant performance from the proposed deep CNN model.


**Table 5.** Comparison results.

SVM: Support Vector Machine, DBN: Deep Belief Network, ANN: Artificial Neural Network, ADCNN: Adaptive Deep Convolutional Neural Network.

#### **5. Conclusions and Future Work**

In this study, we investigate and discuss a novel framework to diagnose the faults in 3-phase induction motors based on recurrence plots and the deep CNN model. The important contributions of this paper are: proposing a method to transform a time-series data signal to 2D texture images (recurrence plots) and applying the proposed deep CNN model to learn the features from the recurrence plots to classify the 2D texture images for the fault diagnosis. The proposed framework is implemented for four types of faults including bearing axis deviation, stator and rotor friction, rotor end ring break, and poor insulation, and achieved a very prominent accuracy of 99.81%. The proposed framework

outperforms other traditional and deep learning models due to its ability to learn both high-level and low-level features. The proposed framework demonstrates promising results by considering a single variable as an input feature compared to the rule-based diagnosis methods that require multiple features for fault diagnosis.

The limitations of the proposed methodology are discussed as follows. First, the dataset collected for the experiment is comparatively small and a huge amount of data samples is needed for different load conditions such as no load, half load, or full load. Second, data from the motors with different specifications are needed to extract and learn more generalized features. Therefore, important future work should focus on motors working with different load conditions to collect more data samples, and investigating data to generate more generalized features for CNN model training. Furthermore, future work includes the review of transfer learning to avoid any unnecessary time required to train the model and utilize the model to learn other feature types.

**Author Contributions:** Y.H. has generated the data and analyzed the faults in Induction motors. W.B.W. validated the data for each kind of faults. V.R.I. performed data preprocessing to train the CNN model and evaluated the trained model for fault diagnosis. Y.H. and V.R.I. analyzed the experimental results with the guidance from C.C.K.; H.C.C. and C.C.K. revised the manuscript for submission.

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

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

#### **References**


© 2019 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 (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Design of a Logistics System with Privacy and Lightweight Verification**

#### **Chin-Ling Chen 1,2,3, Dong-Peng Lin 4, Hsing-Chung Chen 5,6,7,\*, Yong-Yuan Deng 3,\* and Chin-Feng Lee 4,\***


Received: 29 April 2019; Accepted: 27 June 2019; Published: 8 August 2019

**Abstract:** Presently, E-commerce has developed rapidly as a result of many services and applications integrating e-commerce technologies offered online. Buyers can buy goods online and sellers can then deliver the goods to them. Logistics therefore plays an important role in online e-commerce applications, with a focus on rapid delivery, the integrity of goods, and the privacy of personal information. Previous studies have proposed secure mechanisms for the transfer of electronic cash and digital content, in which only the sender and the receiver know the secret information hidden in the signature. However, they did not consider requirements such as the anonymous and lightweight verification in the logistics architecture. Therefore, this study designs a secure logistics system, with anonymous and lightweight verification, in order to meet the following requirements: Mutual authentication, non-repudiation, anonymity, integrity and a low overhead for the logistics environment. A buyer could check the goods and know if the parcel has been exchanged by a malicious person. Moreover, the proposed scheme not only presents a solution to meet the logistics system's requirements, but also to reduce both computational and communication costs.

**Keywords:** mutual authentication; privacy; logistics system; ECC; ban logic

#### **1. Introduction**

#### *Background*

In recent years, with the rapid development of e-commerce, online shopping has become a current trend and many shopping and financial transactions can now be completed via online shopping. These activities include online orders and online payments [1]. As buyers and sellers interact online, the purchase of goods is divided into digital and physical products. If a product is physical, the seller will entrust their logistics to deliver the goods to the buyer. As these logistics requirements grow, greater focus is required, not only on rapid delivery, but also on ensuring the integrity of goods and the privacy of personal information [2–7].

Unfortunately, the current process of goods delivery and online shopping does not entail an immediate physical exchange of goods. There is therefore a risk of counterfeiting and fraud, in addition to a risk that goods may be lost due to human error, and this may be compounded by information errors, which could mean that it cannot be determined where the goods were lost. In 2016, Liu and Wang [6] noted that the means of preventing the loss of goods has become a very important issue in this field.

However, during the goods transportation process, the logistics provider will copy both the buyer and seller's personal information on an order detail and paste the order detail on the packages. That is, the goods can only be accurately delivered to the buyer, but this process includes the risk of private personal information being leaked, which may result in improper use or theft of that personal information. The delivery verification can also include the risks of identity impersonation, parcel exchange, and the loss of packages. Since there is no reliable mechanism for the buyer and seller to identify each other, it is impossible to know who has the goods or when goods are lost.

In 2006, Aijaz et al. [1] classified various attacker behaviors as active and passive attackers, internal and external personnel, and malicious and rational attackers. Active attackers tamper with shopping information, while passive attackers do not actively participate in tampering with information, but rather eavesdrop on shopping information. The stolen information may then be forwarded to other attackers. Internal attackers are very dangerous in the transmission process. As a consequence of their good understanding of items and personal information, internal personnel can cause various kinds of complex attacks. External personnel are not members of the transaction process, so they are much less harmful than internal personnel. The main goal of malicious attackers is to steal or tamper with information and cause the loss of property.

This paper proposes a logistics method using the public key crypto system to protect the personal privacy and the shopping information of buyers, sellers, and logistics companies during the transmission process to prevent information from being stolen. In addition, lightweight encryption technology is used to protect personal tag information to prevent personal information from being leaked during the delivery process.

In 2016, Liu and Wang [6] published papers on an NFC-based security-enhanced express delivery systems, in which the individuals' personal information was hidden in tags and only authorized people could get permission to access that information, thus protecting personal information from being stolen and achieving fast identity authentication. Digital signatures are then used by buyers, sellers, and logistics companies to achieve non-repudiation. The proposed system thus achieves mutual authentication, lightweight and fast verification, cost savings, the anonymity of personal information, non-repudiation in the transaction, and the completeness of the product.

The remainder of this paper is arranged as follows. Section 2 presents the system architecture. Section 3 presents the proposed secure package logistics system, based on protecting personal information anonymously by tag. Section 4 presents a security analysis and then illustrates the computation cost, communication cost, and performance analysis of the proposed scheme. Section 5 offers conclusions.

#### **2. Related Works and Requirement**

#### *2.1. Related Works*

In 2016, Liu and Wang noted that digital tags may not be able to perform complex encryption and decryption operations due to computation limitations [6]. In general, current logistics schemes lack face to face package checking procedures and rely on buyers to ensure that packages are intact upon receipt. In addition, the security issues of RFID systems are not completely suitable for the scheme proposed in this study [8]. Instead, this study uses ECC (elliptic curve cryptography) to generate session keys that are used to secure data transmissions and the BAN logic model [9] to prove the correctness of the proposed scheme with mutual authentication. Recently, many authentication schemes have applied BAN logic to prove the correctness of authentication and key establishment. The following is the notation of BAN logic.


#### *2.2. Requirements*

In order to achieve a good logistics system, the following security requirements must be met and known attacks must be prevented:


There are several common malicious attack patterns that can target package logistic systems [15,17–19], as follows:


#### **3. The Proposed Scheme**

#### *3.1. System Architecture*

The system consists of the following parties: Seller (S), logistics (L), buyer (B), and deliverer (D). The architecture and information flow are shown in Figure 1. The four parties in the scheme, in detail, are the following:


**Figure 1.** Logistics system architecture.

The eight steps in the scheme, in detail are as follows:


*Energies* **2019**, *12*, 3061

#### *3.2. Notations*

The notations used in this paper are listed below:


#### *3.3. Initialization Phase*

During the initialization phase, the Certificate Authority (CA) issues the public key and private key, and selects a large prime, P, and elliptic curve, E, over a finite field for each party.

#### *3.4. Session Key Generation Phase and Order Request Phase*

In the session key generation phase and order request phase, the buyer provides shopping information to the seller. The seller sends the buyer and the seller's information to the logistics and asks for the goods to be delivered. The logistics generates the transaction number and the tag for the seller and sends the transaction numbers to the buyer. Figure 2 presents the session key generation and order request phase of the proposed scheme.

Step 1: The buyer selects a random *rB* and computes *RB* as follows:

$$R\_B = r\_B " P\_\prime \tag{1}$$

The buyer signs the *(RB,IDB)* with the private key *PrkB*, as follows:

$$S\!\!\!g\_{BS} = S\_{prkB}(R\_B, lD\_B),\tag{2}$$

The buyer then sends (*RB,IDB,SigBS*) to the seller.

Step 2: The seller selects a random number *rS* and then computes *RS* and signs the *(RS,IDS)* with the private key *PrkS*, as follows:

$$R\_S = r\_S \, ^\ast \mathcal{P} \, ^\ast \tag{3}$$

$$Sig\_{SL} = S\_{prkS}(R\_S, ID\_S). \tag{4}$$

The seller then sends *(RS,IDS,SigSL)* to logistics. The seller then verifies the *SigBS* with the public key *PukB* to determine whether the signagture is legal or not, as follows:

$$\left(V\_{pukB}\left(S i\_{\mathcal{GS}}\right)\right)^{?} = \stackrel{?}{=} \left(R\_{B\prime} I D\_B\right). \tag{5}$$

If it passes the verification, the seller computes session key *SKBS*, as follows:

$$SK\_{BS} = h((r\_s \, ^\ast R\_B) \| ID\_B \| ID\_S), \tag{6}$$

uses the *SKBS* to encrypt *(RB, IDB)* with *SKBS*, as follows:

$$\mathbf{C}\_1 = E\_{SK\_{BS}}(R\_B, ID\_B), \tag{7}$$

and signs the *(RS,IDS)* with the private key *PrkS*, as follows:

$$S \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\/\!\!\!\/\!\!\!\/\!\!\/\!\!\/\!\!\/\/\tag \!\!\!\!\!\!\/\/\tag \!\!\!\/\/\,\tag \!\!\!\/\/\,\tag \!\!\/ \,\tag \!\!\/ \,\tag \!\!\/ \,\tag \!\/\,\tag \!\/\,\tag \!\/\,\tag \!\/\,\tag \!\/\,\tag \!\/\,\tag \!\,\tag \!\,\tag \!\,\tag \!\,\tag \!\,\tag \!\,\tag \!\,\tag \!\,\tag \!\,\tag \!\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\tag \,\,\}$$

The seller then sends *(RS, IDS,C*1*,SigSB)* to the buyer.

Step 3: The logistics selects a random number *rL*, and computes *RL*, as follows:

$$R\_L = r\_L \, ^\ast \text{P.}\tag{9}$$

Logistics then verifies the *SigSL* with the public key *PukS* to determine whether the signagture is legal or not, as follows:

$$\left(V\_{pukS}(S\text{ig}\_{SL})\right) \stackrel{?}{=} (\mathcal{R}\_{\mathcal{S}\prime}ID\_{\mathcal{S}})\_{\prime} \tag{10}$$

If it holds, logistics computes session key *SKSL*, as follows:

$$SK\_{SL} = h((r\_L "\mathcal{R}\_S) \| ID\_S \| ID\_L). \tag{11}$$

Then the logistics encrypts *(RS,IDS)* with *SKSL*, as follows:

$$C\_3 = E\_{SK\_{SL}}(R\_{\rm S}, ID\_{\rm S}).\tag{12}$$

Next, logistics signs the *(RL,IDL)* with the private key *PrkL*, as follows:

$$\text{Si}\_{\text{S}}\text{Si}\_{\text{LS}} = \text{S}\_{prkL}(\text{R}\_{L}, \text{ID}\_{L}), \tag{13}$$

and sends *(RL,IDL,C*3*,SigLS)* to the seller.

Step 4: The buyer verifies the *SigSB* with the public key *PukS* to determine whether the signagture is legal or not, as follows:

$$(V\_{\text{pushS}}(S\text{ig}\_{SB})\stackrel{?}{=}(R\_{\text{S}}/ID\_{\text{S}}).\tag{14}$$

The buyer then computes session key *SKBS*, as follows:

$$SK\_{BS} = h((r\_B \, ^\ast R\_S) \| ID\_B \| ID\_S) \, \tag{15}$$

and uses the *SKBS* to decrypt *C*1, as follows:

$$(R\_{B^\*} \! \! \_\prime D\_B \! \! \_\ast) = D\_{\hat{S} \! \! K\_{\hat{B}}} (\! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \_\text{I} \! \! \! \_\text{$$

and determines whether *(RB,IDB)* is equal or not, as follows:

$$\left(\left(\mathbb{R}\_{\rm B}/ID\_{\rm B}\right)\right)^{?} = \left(\mathbb{R}\_{\rm B}/ID\_{\rm B}\right)\*.\tag{17}$$

The seller then encrypts *(RS,IDS,IDB,MB,Mproduct)* with *SKBS*, as follows:

$$C\_2 = E\_{SK\_{BS}} \Big( R\_{S\prime} \, ID\_{S\prime} \, ID\_{B\prime} M\_{B\prime} M\_{product} \Big) \, \tag{18}$$

Then buyer then sends *(IDB,C*2*)* to the seller.

Step 5: The seller decrypts *C*<sup>2</sup> with *SKBS*, as follows:

$$\left(R\_{S^\*}\!\_{\prime}ID\_{S^\*}\!\_{\prime}M\_{B}, M\_{B}, M\_{\text{product}}\right) = D\_{SK\_{\text{BS}}}\left(C\_{\text{2}}\right),\tag{19}$$

and then gets *(RS\*,IDS\*)*, and determines whether *(RS,IDS)* is equal or not, as follows:

$$(R\_{S\prime}ID\_S) \stackrel{?}{=} (R\_{S^\*\prime}ID\_{S^\*}).\tag{20}$$

The seller verifies the *SigLS* with the public key *PukL* to determine whether the signagture is legal or not, as follows

*VpukL*(*SigLS*) ? <sup>=</sup> (*RL*, *IDL*). (21)

If it passes the verification, the seller computes *SKSL*, as follows:

$$SK\_{SL} = h((r\_S " R\_L) \| ID\_S \| ID\_L), \tag{22}$$

and decrypts *C*<sup>3</sup> with *SKSL*, as follows:

$$(R\_{\mathbb{S}^\bullet}, ID\_{\mathbb{S}}) = D\_{\mathbb{S}K\_{\mathbb{S}\mathbb{L}}}(\mathbb{C}\_{\mathbb{S}}).\tag{23}$$

The seller gets *(RS\*,IDS\*)*, determines whether *(RS,IDS)* is equal or not, as follows:

$$(R\_S, ID\_S) \stackrel{?}{=} (R\_{S^\*} !\_{\prime} ID\_{S^\*}),\tag{24}$$

If it holds, the seller encrypts *(RL,IDL,IDS,MS,IDB,MB)* with *SKSL*, as follows:

$$\mathbf{C}\_4 = E\_{SK\_{\rm SL}}(R\_{L\prime}ID\_{L\prime}ID\_{\rm S\prime}M\_{\rm S\prime}ID\_{\rm B\prime}M\_{\rm B}),\tag{25}$$

and then sends *(IDS,C*4*)* to logistics.

*(RS,IDS,SigSL) (RS\*,IDS\*,IDB,MB,Mproduct) =*ܦௌಳೄ*(C) (RS,IDS)* <sup>ǫ</sup> ൌ*(RS\*,IDS\*) VpukL(SigLS)* <sup>ǫ</sup> ൌ*(RL,IDL) SKSL = h((rS\*RL)||IDS||IDL) (RS\*,IDS\*) =* ܦௌೄಽ*(C) (RS,IDS))* <sup>ǫ</sup> ൌ*(RS,IDS)\* C =* ܧௌೄಽ*(RL,IDL,IDS,MS,IDB,MB) VpukS(SigSB)* <sup>ǫ</sup> ൌ*(RS,IDS) SKBS = h((rB\*RS)||IDB||IDS) (RB\*,IDB\*) =* ܦௌಳೄ*(C1) (RB,IDB)* <sup>ǫ</sup> ൌ*(RB\*,IDB\*) C =* ܧௌಳೄ*(RS,IDS,IDB,MB,Mproduct) Logistics*  6HOHFWDUDQGRPQXPEHU *rL RL = rL\*P VpukS(SigSL)* <sup>ǫ</sup> ൌ*(RS,IDS) SKSL = h((rL\*RS)||IDS||IDL) C3 =* ܧௌೄಽ*(RS,IDS) SigLS = SprkL(RL,IDL) Buyer Seller*  6HOHFWDUDQGRPQXPEHU*rB RB = rB\*P SigBS = SprkB(RB,IDB)*  6HOHFWDUDQGRPQXPEHU *rS RS = rS\*P SigSL = SprkS(RS,IDS) VpukB(SigBS)* <sup>ǫ</sup> ൌ*(RB,IDB) SKBS = h((rs\*RB)||IDB||IDS) C =* ܧௌಳೄ*(RB,IDB) SigSB = SprkS(Rs,IDs) (RB,IDB,SigBS) (RL,IDL,C3,SigLS) (RS,IDS,C1,SigSB) (IDB,C2) (IDS,*&*) ((RL\*,IDL\*,IDS,MS,IDB,MB) =*ܦௌೄಽ*(C) (RL,IDL)* <sup>ǫ</sup> ൌ*(RL\*,IDL\*)*  \*HQHUDWH *TID TagDB = IDD(IDB,MB) C5 =* ܧௌೄಽ*(TagDB,TID) (IDL,C) (TagDB,TID) =* ܦௌೄಽ*(C5) C =* ܧௌಳೄ*(TID) (IDS,C) (TID) =* ܦௌಳೄ*(C)*  \*HW*TID* 

Step 6: The logistics decrypts *C*<sup>4</sup> with *SKSL*, as follows:

$$D(R\_{L} \ast\_{\prime} ID\_{L} \ast\_{\prime} ID\_{S\prime} M\_{\overline{S}\prime} ID\_{\overline{B}\prime} M\_{\overline{B}}) = D\_{\text{SK}\_{\overline{S}\mathbb{L}}}(\mathbb{C}\_{\mathsf{4}}),\tag{26}$$

and then gets *(RL\*,IDL\*)* and determines whether *(RL,IDL)* is equal or not, as follows:

$$(\mathcal{R}\_L, ID\_L) \stackrel{?}{=} (\mathcal{R}\_L \star\_\prime ID\_L \star)\_\prime \tag{27}$$

Logistics generates *TID* and *TagDB,* and computes the following:

$$\text{Tag}\_{DB} = \text{ID}\_{D} \oplus (\text{ID}\_{B}, \text{M}\_{B}). \tag{28}$$

and then uses the *SKSL* to encrypt *(TagDB,TID)*, as follows:

$$\mathbf{C\_{\sf S}} = \mathbf{E\_{SK\_{S\sf L}}} (Tag\_{\sf DB}, TID), \tag{29}$$

then sends *(IDL,C*5*)* to the seller.

Step 7: The seller decrypts *C*<sup>5</sup> with *SKSL*, as follows:

$$(T \text{ag}\_{\text{DB}}, TID) = D\_{\text{SK}\_{\text{SL}}}(\mathbb{C}\_5). \tag{30}$$

The seller encrypts *(TID)* with *SKBS*, as follows:

$$\mathbf{C}\_6 = E\_{SK\_{\rm BS}}(TID),\tag{31}$$

then sends (IDS,C6) to the buyer.

Step 8: The buyer decrypts C6 with SKBS, as follows:

$$TID = D\_{\text{SK}\_{\text{BS}}}(\text{C}\_6), \tag{32}$$

and then gets *TID*.

#### *3.5. Package Collection Phase*

The logistics sends the tag containing the seller information to the deliverer. The deliverer decrypts the tag and goes to the seller's house. After verifying the delivery identity, the seller transmits their signature to give the goods and the buyer's tag to the deliverer. The package collection phase is illustrated in Figure 3.

Step 1: The logistics signs *(IDD,IDL,TID)* with private key *PrkL*, as follows:

$$\text{Sig}\_{LS} = \text{S}\_{prkl.} (\text{ID}\_{\text{D}} \text{ID}\_{\text{L}} \text{TID}), \tag{33}$$

The logistics uses *SKSL* to encrypt *(IDD,IDL,TID)*, as follows:

$$C\_{\mathcal{T}} = E\_{\text{SK}\_{\text{SL}}} \left( ID\_{\mathcal{D}\_{\text{L}}} ID\_{\text{L}}, TID \right), \tag{34}$$

then generates *TagDS*, as follows:

$$\text{Tag}\_{DS} = ID\_D \oplus (ID\_S M\_S)\_\prime \tag{35}$$

and sends *(IDL,TagDS,C*7*,SigLS)* to the deliverer.

Step 2: The deliverer computes the following formula:

$$\text{Tr}(ID\_{\text{S}}M\_{\text{S}}) = \text{Tag}\_{\text{DS}} \oplus ID\_{D\_{\text{V}}} \tag{36}$$

and the deliverer can then get (*IDS,MS*).

Step 3: The deliverer sends *(IDD,IDL,TagDS,C*7*,SigLS)* to the seller for verification and the seller computes *IDD\** as follows:

$$ID\_D{}^\* = \text{Tag}\_{DS} \oplus (ID\_S M\_S), \tag{37}$$

and verifies whether *IDD* is equal or not, as follows:

$$\stackrel{?}{D}\stackrel{?}{D}\_D \stackrel{?}{=} \stackrel{!}{ID\_D}.\tag{38}$$

**Figure 3.** Package collection phase of the proposed scheme.

The seller decrypts *C*<sup>7</sup> with *SKSL*, as follows:

$$(ID\_{\mathcal{D}\_{\mathcal{T}}}ID\_{\mathcal{L}}TID)\* = D\_{SK\_{\mathcal{S}\mathcal{L}}}(\mathbb{C}\_{\mathcal{T}}).\tag{39}$$

The seller verifies the *SigLS* with the public key *PukL* to determine whether the signagture is legal or not, as follows:

$$\left(\mathcal{V}\_{\text{pukL}}(\text{Sig}\_{\text{LS}})\right)^{?} = \left(\text{ID}\_{\text{D}\prime}\text{ID}\_{\text{L}\prime}\text{TID}\right) \* \text{ .} \tag{40}$$

If it passes the verification, the seller signs the *(IDS,IDD,IDL,TID)* with the private key *PrkS*, as follows:

$$\text{Sig}\_{\text{SL}} = \text{S}\_{\text{prkS}} \text{ (ID}\_{\text{S}}, \text{ID}\_{\text{D}}, \text{ID}\_{\text{L}}, \text{TID}), \tag{41}$$

and uses *SKSB* to encrypt (*IDS, DD, IDL,TID*), as follows:

$$C\_8 = E\_{SK\_{\mathcal{BS}}}(ID\_{S\prime}ID\_{D\prime}ID\_{L\prime}TID). \tag{42}$$

The seller then gives the goods and *(IDS,TagDB,C*8*,SigSL)* to the deliverer.

Step 4: The deliverer computes as following formula:

$$\text{Tr}(ID\_{B\prime}M\_B) = \text{Tag}\_{DB} \oplus ID\_{D\prime} \tag{43}$$

and gets (*IDB,MB*).

#### *3.6. Product Transfer Phase*

The deliverer decrypts the tag and sends the goods to the buyer's address. After verifying the deliverer's identity, the buyer obtains the goods and sends a signature to the deliverer. The deliverer takes the signatures of the buyer and the seller. The deliverer then returns to the logistics for confirmation and completes the transaction. The product transfer phase is illustrated in Figure 4.

Step 1: The deliverer sends the goods and (*IDD, TagDB, C*8) to the buyer to verify the identity, using the following formula:

$$ID\_D\text{\*} = \text{Tag}\_{DB} \oplus (ID\_B, \mathcal{M}\_B),\tag{44}$$

Once the deliverer has *IDD\**, they determine whether the *IDD* is equal or not, as follows:

$$\text{ID}\_{\text{D}} \* \stackrel{?}{=} \text{ID}\_{\text{D}}.\tag{45}$$

The buyer decrypts *C*<sup>8</sup> with *SKSB*, as follows:

$$(ID\_{\mathcal{S}\prime}ID\_{\mathcal{D}\prime}ID\_{\mathcal{L}\prime}TID) = D\_{\mathcal{S}\mathcal{K}\_{\mathcal{S}\mathbb{B}}}(\mathbb{C}\_{\mathbb{B}}),\tag{46}$$

and then gets *TID\** and determines whether the *TID*, which is stored in the session key generation and order request phase, is equal or not, as follows:

$$\text{TID} \bullet \begin{array}{c} ? \\ = \\ \end{array} \text{TID} . \tag{47}$$

The deliverer uses *PrkB* to sign *(IDB,IDS,IDD,IDL,TID)*, as follows:

$$\text{Sig}\_{\text{BL}} = \text{S}\_{prkB}(ID\_{\text{B}}, ID\_{\text{S}}, ID\_{\text{D}}, ID\_{\text{L}}, TID), \tag{48}$$

The buyer sends *(SigBL, IDB)* to the deliverer.

Step 2: The deliverer returns to the logistics. Logistics verifies *SigS* with public key *PukS*, as follows:

$$\left(\boldsymbol{V}\_{\text{pukS}}(\text{Sig}\_{\text{SL}})\right) \stackrel{?}{=} (\boldsymbol{ID}\_{\text{S}}, \boldsymbol{ID}\_{\text{D}}, \boldsymbol{ID\_{L}}, \boldsymbol{TID}), \tag{49}$$

and then determines whether the signagture is legal or not. Logistics verifies *SigB* with public key *PukB*, as follows:

$$(V\_{pukB}(S \text{igBL}) \stackrel{?}{=} (ID\_{\text{B}} \, ID\_{\text{S}\prime}ID\_{\text{D}} \, ID\_{\text{L}\prime} \, TD),\tag{50}$$

and determines whether the signagture is legal or not.

If it passes the verification, the transaction is completed.


**Figure 4.** Product transfer phase of the proposed scheme.

#### **4. Security Analysis and Discussion**

#### *4.1. Mutual Authentication Issue*

This study uses BAN logic to prove that the proposed scheme achieves mutual authentication in each phase. In the session key generation and order request phases of the proposed scheme, the main goal is to determine whether the data has been modified between the buyer and seller, or the seller and the logistics provider.

The notation of BAN logic is described below:

*P*|≡*X P believes X, or P would be entitled to believe X. P*-*X P sees X, someone has sent a message containing X to P, who can read and repeat X. P*|*~X P once said X. P at some time sent a message including X. P*| *K* →*X P has X as a public key. P k* ↔*X P and X may use the session key K to communicate P*|⇒*X P has jurisdiction over x. #(X) The formula X is fresh. {X}K The formula X encrypted by K.*

The main goals of the scheme must be achieved in order to verify that the transmitted data has not been modified between buyer and seller, or between the seller and the logistics provider. These goals are listed below:


According to the purchase phase, BAN logic is used to produce an idealized form, as follows:


In order to analyze the proposed improved scheme, this study makes the following assumptions:



According to these assumptions and the rules of BAN logic, this study shows the session key generation and order request phases of the proposed scheme as follows:

a. Seller *S* authenticates Buyer *B* By *M*1 and the seeing rule, derive the following:

$$\mathcal{S} \lhd \mathcal{S} \lhd (\mathcal{l} \mathcal{R}\_{\mathcal{B}} \mathcal{I} \mathcal{D}\_{\mathcal{B}} \mathcal{J} \mathcal{R}\_{\mathcal{S}} \mathcal{I} \mathcal{D}\_{\mathcal{S}} \mathcal{I} \mathcal{D}\_{\mathcal{B}} \mathcal{M}\_{\mathcal{B}} \mathcal{M}\_{\mathcal{P}} \mathcal{M}\_{\mathcal{P}} \mathcal{M}\_{\mathcal{P}} \mathcal{M}\_{\mathcal{B}}).\tag{\text{Statement 1}}$$

By *A*1 and *A*2 and the freshness rule, derive the following:

$$|S\rangle \equiv \# (\langle R\_B, ID\_B \rangle \text{prk}\_B \langle R\_S, ID\_S, ID\_B, M\_B, M\_{\text{product}} \rangle \text{SK}\_{BS}). \tag{\text{Statement 2}}$$

By (Statement 1), *A*9, *A*13, and *A*14 and the message meaning rule, derive the following:

$$|S| \equiv \#(|R\_B, ID\_B\rangle \text{puk}\_B \langle R\_S, ID\_S, ID\_B, M\_B, M\_{\text{product}} \rangle \text{SK}\_{BS}).\tag{\text{Statement 3}}$$

By (Statement 2), (Statement 3), and the verification rule, derive the following:

$$|S| \equiv (|R\_B, ID\_B\rangle \text{puk}\_B \langle R\_S, ID\_S, ID\_B, M\_B, M\_{\text{puduct}} \rangle \text{SK}\_{\text{BS}}.\tag{\text{Statement 4}}$$

By (Statement 4) and the belief rule, derive the following:

$$|S| \equiv |B| \equiv S \stackrel{SK\_{\rm BS}}{\leftrightarrow} B. \tag{\text{Statement 5}}$$

By (Statement 5), *A*21, and the jurisdiction rule, derive the following:

$$|S| \equiv S \stackrel{SK\_{BS}}{\leftrightarrows} B. \tag{\text{Statement 6}}$$

By (Statement 6) and the belief rule, derive the following:

$$|S| \equiv B | \equiv ID\_{B.} \tag{Statement } \tag{Statement } \mathcal{T})$$

By (Statement 7), *A*25, and the belief rule, derive the following:

$$|S| \equiv ID\_B. \tag{\text{Statement 8}}$$

b. Buyer *B* authenticates Seller *S* By *M*2 and the seeing rule, derive the following:

*B*-*({RS,IDS}prkS,{RB,IDB}SKBS).* (Statement 9)

By *A*3, *A*4, and the freshness rule, derive the following:

$$B \mid \equiv \# (I R\_S, I D\_S \rfloor \text{prk}\_S \langle R\_B, I D\_B \rangle S K\_{BS}). \tag{\text{Statement 10}}$$

By (Statement 9), *A*10, *A*15, *A*16, and the message meaning rule, derive the following:

$$|B| \equiv \text{S} |\text{-}\# (\langle R\_S, ID\_S \rangle puk\_S \langle R\_B, ID\_B \rangle \text{S}K\_{BS}). \tag{\text{Statement 11}}$$

By (Statement 10), (Statement 11) and the verification rule, derive the following:

$$|B| \equiv \mathcal{S}| \equiv (\langle R\_{\mathcal{S}}, ID\_{\mathcal{S}} \rangle prk\_{\mathcal{S}} \langle R\_{\mathcal{B}}, ID\_{\mathcal{B}} \rangle \mathcal{S}K\_{\mathcal{B}\mathcal{S}}).\tag{\text{Statement 12}}$$

By (Statement 12) and the belief rule, derive the following:

$$|B| \equiv |S| \equiv S \stackrel{SK\_{BS}}{\leftrightarrow} B. \tag{\text{Statement 13}}$$

By (Statement 13), *A*22 and the jurisdiction rule, derive the following:

$$|B| \equiv S \stackrel{SK\_{\mathbb{B}^S}}{\rightsquigarrow} B. \tag{\text{Statement 14}}$$

By (Statement 14) and the belief rule, derive the following:

$$|B| \equiv S | \equiv I | \equiv I | S \tag{Satement 15}$$

By (Statement 15), *A*26 and the belief rule, derive the following:

$$|B| \equiv ID\_S. \tag{Statement 16}$$

By (Statement 6), (Statement 8), (Statement 14), and (Statement 16), it is proved that buyer B and seller S authenticate each other in the proposed scheme. The seller authenticates the buyer by (5). If it passes the verification, the seller authenticates the legality of the buyer and then the buyer authenticates the seller by (14).

c. Logistics L authenticates Seller S By *M*3 and the seeing rule, derive the following:

$$L \lhd (\{R\_{\mathcal{S}}, \operatorname{ID}\_{\mathcal{S}}\} \operatorname{prk}\_{\mathcal{S}}, \{R\_{\mathcal{L}}, \operatorname{ID}\_{\mathcal{L}}, \operatorname{ID}\_{\mathcal{S}}, \operatorname{M}\_{\mathcal{S}}, \operatorname{ID}\_{\mathcal{B}}, \operatorname{M}\_{\mathcal{B}} \} \operatorname{SK}\_{\operatorname{SL}}) . \tag{\text{Statement 17}}$$

By *A*5, *A*6, and the freshness rule, derive:

$$\mathbb{L}|\text{tr}\#(\{R\_{\text{S}}, \text{ID}\_{\text{S}}\} \text{prk}\_{\text{S}}, \{R\_{\text{L}}, \text{ID}\_{\text{L}}, \text{ID}\_{\text{S}}, \text{M}\_{\text{S}}, \text{ID}\_{\text{B}}, \text{M}\_{\text{B}}\} \text{SK}\_{\text{SL}}).\text{}\tag{\text{Statement 18}}$$

By (Statement 17), *A*11, *A*17, *A*18, and the message meaning rule, derive the following:

$$L|\text{int}S| \text{-} \# (I R\_S, I D\_S | p u k\_{\\$,\prime} \, \{ R\_L, I D\_L, I D\_S M\_{\\$,\prime} \\ L D\_B M\_B | S K\_{\\$,\prime}) . \tag{\text{Statement 19}}$$

By (Statement 18), (Statement 19), and the verification rule, derive the following:

$$|L| \equiv \mathbb{S} | \text{tr} (\boldsymbol{l} \boldsymbol{R}\_{\mathcal{S}} \boldsymbol{I} \boldsymbol{D}\_{\mathcal{S}} \boldsymbol{l} \boldsymbol{p} \boldsymbol{u} \boldsymbol{k}\_{\mathcal{S}} , \boldsymbol{l} \boldsymbol{R}\_{\mathcal{L}} \boldsymbol{I} \boldsymbol{D}\_{\mathcal{S}} \boldsymbol{M}\_{\mathcal{S}} \boldsymbol{M}\_{\mathcal{S}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} | \boldsymbol{S} \boldsymbol{K}\_{\mathcal{S} \mathcal{L}} \boldsymbol{I} \boldsymbol{D}\_{\mathcal{S}} \boldsymbol{M}\_{\mathcal{S}} \boldsymbol{M}\_{\mathcal{S}} \boldsymbol{I} \boldsymbol{K}\_{\mathcal{S} \mathcal{L}} \boldsymbol{I} \boldsymbol{D}\_{\mathcal{S}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}} \boldsymbol{M}\_{\mathcal{B}} \boldsymbol{N}\_{\mathcal{B}$$

By (Statement 20) and the belief rule, derive the following:

$$|L| \equiv |S| \equiv \mathbb{S} \stackrel{SK\_{RS}}{\leftrightarrow} B. \tag{\text{Statement 21}}$$

By (Statement 21), *A*23, and the jurisdiction rule, derive the following:

$$|L| \equiv S \stackrel{\text{SK}\_{\text{BS}}}{\leftrightarrows} B. \tag{\text{Statement 22}}$$

By (Statement 22) and the belief rule, derive the following:

$$L \vert \equiv \text{S} \vert \equiv \text{ID}\_S. \tag{\text{Statement 23}}$$

By (Statement 23), *A*27, and the belief rule, derive the following:

$$L \| \equiv ID\_S. \tag{Statement 24}$$

d. Logistics *L* authenticates Seller *S* By *M*4 and the seeing rule, derive the following:

$$\mathcal{S} \lhd \mathcal{\smile} (\{ \mathcal{R}\_{L\prime} \mathcal{I} \mathcal{D}\_{L\prime} \mathcal{l} \mathcal{r} k\_{L\prime} \mathcal{I} \mathcal{R}\_{S} \mathcal{I} \mathcal{S} \mathcal{K}\_{\mathcal{B}S}).\tag{\text{\textquotedblleft} statement 25})$$

By *A*7, *A*8, and the freshness rule, derive the following:

$$|S| = \# (\langle \mathcal{R}\_L, \mathcal{I}D\_L \rangle prk\_{L'} \langle \mathcal{R}\_S, \mathcal{I}D\_S \rangle SK\_{BS}). \tag{\text{Statement 26}}$$

By (Statement 25), *A*12, *A*19, *A*20 and the message meaning rule, derive the following:

$$|S| \equiv \mathcal{L}|\text{-}\#(\ell R\_L, \text{ID}\_L \text{|}\text{prk}\_L \text{ (\$R\_S, \text{ID}\_S\$/SK}\_{\text{BS}}\$).\tag{\text{Statement 27}}$$

By (Statement 26), (Statement 27), and the verification rule, derive the following:

$$|S| = \mathcal{L}| \equiv (\mathcal{R}\_L \mathcal{I} D\_L | \text{prk}\_{L'} \{ \mathcal{R}\_S \mathcal{I} D\_S | \text{SK}\_{\text{BS}} \}. \tag{\text{Statement 28}}$$

By (Statement 28) and the belief rule, derive the following:

$$|S| \equiv L| \equiv S \stackrel{SK\_{SL}}{\leftrightarrow} L. \tag{\text{Statement 29}}$$

By (Statement 29), *A*24, and the jurisdiction rule, derive the following:

$$|S| \equiv S \stackrel{SK\_{SL}}{\leftrightarrows} L. \tag{\text{Statement 30}}$$

By (Statement 30) and the belief rule, derive the following:

$$|S| \equiv L \vert \equiv ID\_L. \tag{Statement 31}$$

By (Statement 31), *A*28, and the belief rule, derive the following:

$$|S| \equiv ID\_L. \tag{Statement 32}$$

By (Statement 22), (Statement 24), (Statement 30), and (Statement 32), it is proved that logistics L and seller S authenticate each other in the proposed scheme. The logistics authenticates the seller by (14): If it passes the verification, the logistics provider authenticates the legality of the seller and then the buyer authenticates the logistics as (21).

#### *4.2. Non-Repudiation Issue*

The proposed scheme uses digital signatures to achieve non-repudiation between the parties in each phase. The sender uses their private key to sign the transmitted message and then the receiver verifies the received message. The receiver uses their private key to sign the response message. Table 1 shows the non-repudiation of the proposed scheme.


**Table 1.** Non-repudiation of the proposed scheme.

#### *4.3. Anonymity Issue*

All personal information, *TagDS* = *(IDS,MS)*⊕*IDD* and *TagDB* = *(IDB,MB)*⊕*IDD*, is protected so that only the legal identities *IDD*, *IDS*, and *IDB* can read the content. Therefore, the contents will not disclose any information about buyer or seller.

#### *4.4. Low Overhead Issue*

In the package collection phase and the product transfer phase, this study uses exclusive operation or encryption to quickly verify and reduce the verification cost. This study also uses session keys to substitute public key encryption to enhance calculation speed, thus meeting the low overhead requirement.

#### *4.5. Data Integrity Issue*

This study uses digital signatures to ensure data integrity. A malicious attack can be detected using digital signatures to verify the integrity of the data, even if an attacker has tampered with the transmitted data. Thus, attackers cannot tamper with the transmitted data without being detected. Therefore, the proposed scheme achieves data integrity.

#### *4.6. Security Against Known Attacks*

#### 4.6.1. Modification Attack

In the information transmission process, encryption is performed using session keys, preventing the modification of transmitted data:

(1) The session key generation and order request phase is as follows:


(2) Package collection phase:

$$C\_7 = E\_{S\mathcal{K}\_{SL}} \text{(ID}\_{\text{D'}} \text{ID}\_{\text{L'}} \text{TID)},\tag{34}$$

$$C\_8 = E\_{SK\_{50}} \text{(ID}\_{\text{S}}, \text{ID}\_{\text{D}}, \text{ID}\_{\text{L}}, \text{TID} \text{)}. \tag{42}$$

#### 4.6.2. Impersonation Attack

In the session key generation and order request phase, package collection phase, and product transfer phase of information transmission, digital signatures cannot be disguised.

(1) The session key generation and order request phase is as follows:

*SigBS* = *SprkB(RB,IDB),* (2) *SigSL* = *SprkS(RS,IDS),* (4) *SigSB* = *SprkS(RS,IDS),* (8) *SigLS* = *SprkL(RL,IDL).* (13)

(2) Package collection phase:

*SigSL* = *SprkS (IDS,IDD,IDL,TID).* (41)

(3) Product transfer phase:

*SigBL* = *SprkB(IDB,IDS,IDD,IDL,TID),* (48)

4.6.3. Man-in-the-Middle Attack

The proposed scheme uses signature mechanisms *SigBS* = *SprkB(RB,IDB), SigSL* = *SprkS(RS,IDS)*, and *SigLS* = *SprkL(RL,IDL)* to prevent modification of the RB, RS, and RL, and uses those variables to generate session keys *SKBS* = *h((rs\*RB)*||*IDB*||*IDS)* and *SKSL* = *h((rL\*RS)*||*IDS*||*IDL)*. The session key encryption/decryption offers security against man-in-the-middle attacks.

#### 4.6.4. Clone Attack

In the package collection phase and the product transfer phase, the deliverer must give their own information, *IDD* and *TagDS,* and the seller can then execute the exclusive-or operation or encrypt the *TagDS* and verify the identity of the deliverer *IDD\** = *TagDS*⊕*(IDS,MS)*. In the product transfer phase of the proposed scheme, the deliverer must give their own information, *IDD* and *TagDB*, and the buyer can then execute the exclusive-or operation or encrypt the *TagDB* and verify the identity of the deliverer *IDD\** = *TagDB*⊕*(IDB,MB)*, thus preventing a clone attack.

#### *4.7. Computation Cost*

Table 2 shows the computation costs of the proposed scheme.


**Table 2.** Computation costs of the proposed scheme.

#### Notes:



In Table 2, the proposed scheme's computation costs are analyzed for the buyer, seller, logistics, and deliverer in each phase. Due to the insignificant comparison operation impacts, they are not considered. For the highest computation cost reduction in the session key generation and order request phase, a buyer needs three asymmetrical signatures/verifying a signature, three symmetrical encryption/decryption operations, one hash function operation, and one multiplication operation. A seller needs four asymmetrical signatures/verifying a signature, six symmetrical encryption/decryption operations, two hash function operations, and two multiplication operations. The logistics provider needs two asymmetrical signatures/verifying a signature, three symmetrical encryption/decryption operations, one hash function operation, one exclusive-or operation, and one multiplication operation.

#### *4.8. Communication Cost*

Table 3 shows the communication cost of the proposed scheme.


**Table 3.** Communication cost of the proposed scheme.

#### Notes:

*Tsig* The time required to transmit a signature (1024 bits).

*Tsys* The time required to transmit a symmetric encryption/decryption ciphertext (256 bits).

*Txor* The time required to transmit an exclusive-or operation (80 bits).

From Table 3, the communication cost of the proposed scheme during the transaction process of each phase was analyzed and, since other operations have little impact, they were not considered in the communication cost. For the highest communication cost reduction in the session key generation and order request phase, four signature operations and six symmetric encryption/decryption operations must be transmitted. It thus requires 1024 × 4 + 256 × 6 = 5632 bits. In a 3.5G environment, the maximum transmission speed is 14 Mbps, which only takes 0.402 ms to transfer all messages. In a 4G environment, the maximum transmission speed is 100 Mbps and the transmission time is reduced to 0.056 ms (ITU 2016).

#### *4.9. Storage Cost*

Table 4 shows the storage cost of the proposed scheme.



Notes:

*Tasy* The space required to storage an asymmetrical signature (1024 bits). *Tsys* The space required to storage a symmetrical encryption/decryption ciphertext (256 bits). *Th* The space required to storage a one-way hash function calculated message (256 bits). *Tmul* The space required to storage a multiplication calculated message (160 bits).

*Tother* The space required to storage other messages (80 bits).

In Table 4, the storage cost of the proposed scheme was analyzed for the buyer, seller, logistics and deliverer in each phase. For the highest storage cost in the session key generation and order request phase, a seller needs two asymmetrical signatures storage space, three symmetrical encryption/decryption ciphertexts storage space, two one-way hash function calculated messages storage space, three multiplication calculated messages storage space, and five other messages storage space. It thus requires 1024 × 2 + 256 × 3 + 256 × 2 + 160 × 3 + 80 × 5= 4208 bits storage space.

#### **5. Conclusions**

In recent years, e-commerce services have prospered and online shopping has become a current trend. The security of personal information exchanged when purchasing a product online has thus become an important issue. This paper proposes a tag-based protection of personal information and a non-repudiable logistics system. The proposed scheme can effectively provide the secure transmission of personal information transmitted by items.

In the session key generation and order request phases, digital signatures are used to transmit data from the sender to the receiver, which ensures that the data cannot be modified. In the package collection phase and product transfer phase, tags containing hidden personal information are used to prevent personal information being leaked and to speed up the verification of the deliverer for buyers and sellers. The proposed scheme offers a reduction of computation costs, compared to other related works. The logistics can use the proposed system to achieve non-repudiation and to complete transactions by examining the digital signatures of the buyer and seller.


Future work will include the payment flow and applying block-chain technology to track the stream of and to prevent the loss of goods.

**Author Contributions:** Supervision and methodology, C.-L.C.; writing—original draft, D.-P.L.; validation, Y.-Y.D.; surveyed related work, H.-C.C. and C.-F.L.

**Funding:** This research was funded by the Ministry of Science and Technology, Taiwan, ROC, under contract number MOST 108-2221-E-324-013.

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

#### **References**


© 2019 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 (http://creativecommons.org/licenses/by/4.0/).
