An approach for estimating thermoelectric module equivalent properties by thermal resistance and numerical modeling Authors: Minh Quang Pham, Van Luc Nguyen, Tri Quang Vo, Thong Duc Hong Published: 2026 ABSTRACT SummaryThermoelectric material properties are crucial for simulating thermoelectric modules (TEMs); yet, determining them requires complex and costly experiments due to temperature-dependent behavior and parasitic losses. This work proposes an approach for estimating TEM’s equivalent properties from its output characteristics using thermal resistance (TRM) and numerical simulation modeling (NSM) accessibly, cost-effectively, and time-efficiently. Sensitivity analysis and cross-module validation confirm the method’s reliability. The obtained equivalent property sets enable accurate prediction of TEM’s output characteristics for the TRM and NSM compared with manufacturer data, with deviations of open-circuit voltage, maximum power, and conversion rate of 0.87%, 2.53%, and 3.07% for TRM, and 0.91%, 2.51%, and 3.24% for NSM, respectively. This finding implies that a TRM of TEMs can replace NSM in large-scale simulations, saving resources while ensuring accuracy. The proposed approach makes determining TEM properties less challenging; also, the equivalent properties reduce computational efforts required for simulating the thermoelectric generator system characteristics. Introduction Energy issues are increasingly concerning in the context of environmental pollution and rising fuel prices. Powering vehicles accounts for a considerable share of annual oil demand. However, their energy efficiency is still significantly low, and they discharge a large amount of their input energy into the ambient via exhaust gas. This context promoted research into solutions for recycling this heat further to improve the efficiency of internal combustion engine-powered vehicles, helping them continue to develop as electric and hybrid vehicles rise. With numerous advantages, including direct energy conversion, no moving parts, no noise or working fluids, a long lifespan, and the ability to operate in any position with minimal scale effect,1,2 the thermoelectric generator was proposed as a practical solution. Hence, it has attracted attention as a promising solution for retrieving waste heat from various sources, e.g., solar systems,3 geothermal,4 and transportation.5,6 However, until now, the conversion efficiency of thermoelectric generator systems has been significantly low; thus, numerous researchers have focused on developing thermoelectric generator systems, such as proposing new structures7,8,9 and optimizing the structural parameters10,11,12,13 to enhance the system’s thermal uniformity, output power, and pressure loss. Analytical and simulation approaches are essential for investigating, evaluating, and optimizing thermoelectric generator systems in research and design. Several methods have been employed to describe its characteristics, e.g., thermal resistance models (TRMs), computational fluid dynamics models, thermal-electric models, and fluid-thermal-electric multiphysics simulation models. Firstly, the TRM was applied very early and has been improved continuously to examine characteristics of thermoelectric module (TEM) and thermoelectric generator systems, as well as optimize their performances.14,15,16 Second, numerical simulation approaches, such as computational fluid dynamics and thermal-electric, have been widely applied to investigate the effect of thermoelectric generator systems configuration as well as optimize the structure to improve the performance of thermoelectric generator systems.17,18,19,20 In recent years, studies on a comprehensive model that combines computational fluid dynamics and thermal-electric models have attracted greater attention, leading to more complete and accurate predictive models.21,22,23 Overall, the simulation models are increasingly becoming more complex to predict properties more accurately, supporting more in-depth analysis and design. In this development trend, fully and accurately simulating the thermoelectric effects of TEMs is very important in simulating thermoelectric generator systems, due to the variation of the thermal conductivity, electrical resistivity, Seebeck coefficient, heat loss, and contact resistance according to temperature, which greatly affects the local and overall energy balance of the thermoelectric generator systems, affecting the temperature profile and power generation. Building methods to address the challenge of modeling the characteristics of TEMs has been the subject of numerous studies.24,25,26,27,28,29,30 Typically, a nonlinear computational method for the thermoelectric transport equation based on the control volume method was introduced by Zhang et al.26 This approach accounted for the temperature-dependent properties of materials and incorporated electrical and thermal losses at interfaces when calculating performance. Two fully coupled thermal-electric models for a thermoelectric couple, considering all the thermoelectric effects (Seebeck, Peltier, Thomson, Joule, and Brigman), using ANSYS CFX and iterative analytical calculation, were proposed by Wielgosz et al.27 A mathematical model of TEM that accounted for all effects, partial dependencies, and temperature dependencies was proposed by He et al.28 The model employed numerical methods to handle nonlinear properties and analytical methods for those that remain constant; the entire model was solved in MATLAB. A one-dimensional analytical model that fully accounts for the temperature-dependent properties of TEM to obtain accurate results was employed by Xu et al.30 The control volume and finite element methods were combined in iterative loops to compensate for deviations caused by the temperature distribution. The above studies indicated that to simulate TEM characteristics, numerical or combined methods must be used to account for multiple effects simultaneously. At the same time, the heat loss, contact resistance, and the material’s temperature-dependent properties within the TEM need to be fully determined. Although previous studies provided substantial information on these properties, they varied significantly across studies, even when the same type of material or TEM was used. It may be because the material properties do not show a clear trend, as they depend heavily on material type, impurity concentration, and manufacturing process.31 In addition, the manufacturers only provide output data, such as matched-load output power, current, voltage, conversion efficiency, and open-circuit voltage, while structural dimensions and interior properties are unavailable. As a result, the previous studies may have applied different assumptions and corrections to produce reasonable results. These make it difficult and confusing for new researchers to develop simulation models for thermoelectric generator systems. To overcome this problem, some studies have conducted TEM-level experiments to identify equivalent properties that are easier to apply in simulations. A method for determining the Seebeck coefficient from open-circuit voltage through experiment was introduced by Hsu et al.;32 however, electrical resistivity and thermal conductivity are not mentioned. Kim et al.33 calculated the average thermal conductivity of a TEM by applying a TRM to the internal components of the TEM. Luo et al.31 and Meng et al.34 determined the functions of the thermal conductivity, electrical resistivity, and Seebeck coefficient with the average temperature by combining the experimental data with the simple analytical model. He et al.35 presented a new experimental approach, the quasi-steady-state method, for determining TEM properties. The temperature-dependent properties of the material were extracted by measuring the output characteristics as the temperature difference gradually increased over time and by dividing the thermoelectric leg into several segments based on the computed temperature distribution. However, existing approaches still suffer from some limitations. The methods based on applying calculations using constant material properties with simple average temperatures can cause considerable errors, especially at large temperature differences36,37; hence, experiments need to be performed many times, at many small temperature ranges, as well as applying specialized experimental conditions,31,34,35 requiring a lot of time, workload, and cost to be invested. On the other hand, advanced numerical methods developed by many previous studies,26,28,30 although more accurate, cannot be used to calculate backward from output characteristics to material properties. Therefore, there is still a lack of an accessible method to determine equivalent thermoelectric properties of TEMs, overcoming the limitations associated with constant-property assumptions in inverse calculation, which often lead to increased experimental effort. To address this gap, this study proposes a method that enables the determination of equivalent temperature-dependent properties for TEM from solely manufacturer-provided output characteristics. The method first derives the thermoelectric properties by applying an iterative process that utilizes only the TRM based on output characteristics. Based on that, the thermal resistance-based equivalent temperature-dependent property functions were established through the integrated-temperature average method and least-squares regression. Furthermore, the electrical resistivity and thermal conductivity were adjusted to establish the adjusted equivalent property (AEP) set for the numerical simulation model (NSM), without additional specific experiments. Finally, this procedure was validated across various output characteristics and module types, confirming the effectiveness, accuracy, and stability of the proposed method. This study’s primary contribution is the proposal of a straightforward, cost-effective, and time-efficient method for deriving equivalent properties for TEM from its output characteristics alone, as provided by the manufacturer or experiments with few measurement points and large temperature steps, thereby eliminating the need for several extensive experiments. The second contribution is the proposal of equivalent property sets, taking into account temperature-dependent behaviors and compensating for internal losses of TEM, that can be immediately applied to analytical and numerical models, making the modeling of the TEMs characteristics in the thermoelectric generator system more accessible, as well as reducing the computational time and cost. Results The effectiveness of equivalent properties computed from an iterative process using the thermal resistance model Figure 1 shows the temperature-dependent function of equivalent properties determined by the iterative process using the TRM (TRBEP) and some reference properties from other studies,30,35,37,38 which are also BiTe-based materials. The iterative process and property functions construction are presented in detail in STAR Methods (determination of the average equivalent properties and construction of the temperature-dependent function of properties respectively). The input parameters of the TEM can be found in STAR Methods (parameters of thermoelectric modules). Note that the referenced properties were averaged from the absolute values of P-type and N-type materials for comparison accessibility. The electrical resistivity and Seebeck coefficient values of TRBEP in this work are lower, but the trends are comparable to those observed in almost all of the literature. This phenomenon may be due to the high carrier concentration of the TEM in this study, which creates a lower Seebeck coefficient and electrical resistivity than in other studies39; in addition, because the TRM ignores the contact thermal resistance, the model may generates a higher temperature discrepancy across the thermoelectric couple’s junctions than in reality, leading to a smaller Seebeck coefficient and electrical resistivity. Besides, ignoring heat losses and contact resistances means that the equivalent thermal resistance of the thermoelectric couple includes these losses, leading to a lower equivalent thermal conductivity than the references.Figure 1The temperature-dependent function of equivalent properties was computed based on the TRM of the current work and the properties referenced from other studies(A) Seebeck coefficient.(B) Electrical resistivity.(C) Thermal conductivity. The temperature-dependent function of equivalent properties was computed based on the TRM of the current work and the properties referenced from other studies (A) Seebeck coefficient. (B) Electrical resistivity. (C) Thermal conductivity. In summary, although the TRBEP set differs slightly from other studies, it shows suitable temperature dependence trends within a reasonable range. It should be noted that the estimated equivalent properties include the influence of parasitic loss; therefore, they do not accurately reflect the actual properties of thermoelectric materials, but are only suitable for application in simulating the overall characteristics of TEMs and thermoelectric generator systems. The comparative results only show their agreement with previous studies, demonstrating the reliability and feasibility of the estimation method. Figure 2 presents the output characteristics of the TEM harvested from the thermal resistance and NSMs using TRBEP compared with the manufacturer’s data at Tc = 303 K. The details of the TRM and the numerical model are presented in STAR Methods (thermal resistance model and numerical simulation model). The results reveal that the TRM outputs strongly agree with the manufacturer’s data. The average deviations of open-circuit voltage, maximum power, conversion rate, internal resistance, and hot-side heat flow are 0.66%, 0.71%, 0.26%, 0.63%, and 0.25%, respectively. It demonstrates that determining the equivalent properties using the iteration and integration method is highly accurate and reliable.Figure 2Comparison of TEM’s output characteristics between TRM and NSM using TRBEP and the manufacturer’s data according to different Th at Tc of 303 K(A) Open-circuit voltage.(B) Maximum power.(C) Conversion rate.(D) Internal resistance.(E) Hot-side heat flow. Comparison of TEM’s output characteristics between TRM and NSM using TRBEP and the manufacturer’s data according to different Th at Tc of 303 K (A) Open-circuit voltage. (B) Maximum power. (C) Conversion rate. (D) Internal resistance. (E) Hot-side heat flow. Regarding the results of the NSM, the open-circuit voltage also achieved high agreement with the manufacturer, with an average relative error of only 0.78%, since the Seebeck effect solely depends on the thermoelectric couple’s cold-side and hot-side temperatures. Hence, the equivalent Seebeck coefficient function can be applied to the NSM without further correction. On the other hand, the maximum power and conversion rate exhibit large errors, especially at large temperature differences, with average and maximum errors in the output power of 7.81% and 14.76%, respectively, and in the conversion rate of 5.73% and 10.44%, respectively. In the integrated-temperature average method, when constructing the property function,37 the influence of the temperature distribution along the thermoelectric couples was not considered. In contrast, this influence is included in the NSM. As a result, the internal resistance of the NSM is significantly higher than that of the TRM and the manufacturer’s; the average and maximum errors are 9.65 and 14.82%, respectively, while the hot-side heat flow is lower than the manufacturer’s data, the average and maximum deviations are 2.74% and 4.21%, respectively. These results are similar to previous studies;40,41 the TRM produces lower internal resistance, leading to higher power than the simulation with the same material properties. Therefore, resistivity and thermal conductivity adjustments are conducted to find an equivalent property set that can be applied to NSM. The effectiveness of equivalent properties adjusted by the numerical simulation model Figure 3 shows the variation in equivalent electrical resistivity and thermal conductivity functions across the adjustment process. The adjustment process is clearly presented in STAR methods (adjustment of the electrical resistivity and thermal conductivity functions). The results show that the electrical resistivity function fluctuates strongly around the convergent value during the first two adjustments, especially in the high-temperature region, and then gradually converges after the third adjustment. Due to the internal resistance being calculated from the integrated value of electrical resistivity, its error has a cumulative nature; hence, the adjustment of the property in the low temperature region also affects the deviation of the internal resistance in the high temperature region. Consequently, three adjustment times are needed to eliminate errors and harvest an appropriate function. On the other hand, with a small initial error, the thermal conductivity only needs to be adjusted once to achieve the required accuracy. The variations of the internal resistance ratio and Fourier heat ratio with different temperatures are presented in Figure S1.Figure 3The AEP functions according to temperature across the adjustment process(A) Electrical resistivity functions.(B) Thermal conductivity functions.The variations of the internal resistance ratio and Fourier heat ratio with different temperatures are presented in Figure S1. The AEP functions according to temperature across the adjustment process (A) Electrical resistivity functions. (B) Thermal conductivity functions. The variations of the internal resistance ratio and Fourier heat ratio with different temperatures are presented in Figure S1. Figure 4 depicts the TEM’s output characteristics obtained from the NSM using the AEPs compared with the manufacturer’s data at Tc = 303 K. After adjustment, the output characteristics obtained from the NSM were consistent with the manufacturer’s data with insignificant deviations. The average errors of open-circuit voltage, maximum power, conversion rate, internal resistance, and hot-side heat flow are 0.81%, 0.88%, 0.42%, 0.68%, and 0.46%, while their maximum errors are 1.27%, 1.62%, 0.89%, 1.77%, and 0.69%, respectively. The distribution of errors with temperature was not uniform due to noise in the input characteristics provided by the manufacturer, especially at low temperatures. Hence, the required error of the output characteristics in the adjustment process is kept at 2%. This high consistency shows the accuracy and reasonableness of the adjustment method.Figure 4Comparison of TEM’s output characteristics between NSM using AEP and the manufacturer’s according to different Th at Tc = 303 K(A) Open-circuit voltage.(B) Maximum power.(C) Conversion rate.(D) Internal resistance.(E) Hot-side heat flow. Comparison of TEM’s output characteristics between NSM using AEP and the manufacturer’s according to different Th at Tc = 303 K (A) Open-circuit voltage. (B) Maximum power. (C) Conversion rate. (D) Internal resistance. (E) Hot-side heat flow. Sensitivity analysis of the input parameter Figure 5 presents the temperature-dependent function of TRBEPs and AEPs obtained from the estimation process using different characteristics provided by the manufacturer at different Tc. Combining these properties, two sets of combined equivalent properties, combined-TRBEP and the combined-AEP, are established using polynomial fitting methods. The results show that the trends of the properties at all three cold-side temperatures are consistent. The error of the Seebeck coefficient is negligible, with an average deviation of only 0.89%. The electrical resistivity also shows minor deviations, with average errors of 1.93% and 2.94% for the combined-TRBEP and combined-AEP, respectively. However, the thermal conductivity presents significant discrepancies between cold-side temperatures, especially at high temperature differences. The combined-TRBEP and combined-AEP average errors are 10.30% and 10.68%, respectively. These results may be due to errors in the heat measurement process at different cold-side temperatures,24,35,42 leading to remarkable errors in the manufacturer’s conversion rate characteristics.Figure 5The temperature-dependent function of TRBEP and AEP obtained from the estimation process using characteristics provided by the manufacturer at three different Tc of 303, 323, and 353 K(A) Seebeck coefficient.(B) Electrical resistivity.(C) Thermal conductivity. The temperature-dependent function of TRBEP and AEP obtained from the estimation process using characteristics provided by the manufacturer at three different Tc of 303, 323, and 353 K (A) Seebeck coefficient. (B) Electrical resistivity. (C) Thermal conductivity. Figure 6 presents the TEM’s output performance results at different cold-side temperatures, obtained from the TRM using the combined-TRBEP and the NSM using the combined-AEP, compared with the manufacturer’s data at different Tc. The open-circuit voltages of the two models are well aligned with the reference data, with average errors of 0.87% and 0.91%, respectively. The output powers are also considerably consistent with the reference data. Although the maximum errors reach 4.72% and 5.77% at the largest temperature difference, the average errors are 2.53% and 2.51% for the TRM and NSM, respectively. On the other hand, the conversion rates exhibit considerable deviations due to fluctuations in the equivalent thermal conductivity. The maximum errors of the TRM and NSM are 7.67% and 6.73%, respectively, while the average errors are 3.07% and 3.24%, respectively. This result shows the substantial impact of input parameters on the determination process for equivalent properties. However, the TEM’s output characteristics of TRM and NSM using combined equivalent properties are pretty consistent with the manufacturer’s data, especially in terms of electrical characteristics. This result reinforces the accuracy and validity of the proposed method.Figure 6Comparison of TEM’s output characteristics between the TRM using combined-TRBEP, NSM using combined-AEP, and the manufacturer’s data according to Th under different Tc(A) Open-circuit voltage.(B) Maximum power.(C) Conversion rate. Comparison of TEM’s output characteristics between the TRM using combined-TRBEP, NSM using combined-AEP, and the manufacturer’s data according to Th under different Tc (A) Open-circuit voltage. (B) Maximum power. (C) Conversion rate. On the other hand, considering the results from the two models, using two different sets of equivalent properties for each model, the TRM and NSM achieve high agreement, with average errors of only 0.1%, 0.51%, and 1.06% for voltage, power, and conversion rate, respectively. This result is significant in simulating the large-scale or comprehensive thermoelectric generator systems. The TRM with equivalent properties can be applied to achieve high accuracy and minimize computational time and resources. Application to other modules The authors proceeded with two modules of TEHP1-1263-1.5 (TEM-2) and TELBP1-12656-0.45 (TEM-3) with the manufacturer’s characteristics at Tc = 303 K, to further verify the effectiveness of the developed approach. Among them, module TEM-2 is reported to have BiTe-based material, similar to TEHP1-1264-0.8 (TEM-1), while the material of TEM-3 is PbTe&BiTe-based. The determined property functions of these two modules are presented in Figure 7. Generally, these two modules’ harvested equivalent properties are also in the appropriate range for thermoelectric materials. The equivalent properties of TEM-2 tend to be very similar to those of TEM-1, since they have the same type of material. The difference in values may be caused by differences in the thermoelectric couple’s size, heat losses, and contact resistances. In contrast, because of the discrepancies in materials, the equivalent thermal conductivity of TEM-3 is significantly different from those of TEM-1 and TEM-2. The comparison of output characteristics of TEM-2 and TEM-3 between the NSM and the manufacturer’s specifications is presented in Figure S2. In addition, the comparison of error variation across the adjustment process between different TEMs is presented in Figure S3.Figure 7The temperature-dependent function of TRBEP and AEP of TEHP1-1263-1.5 (TEM-2) and TELBP1-12656-0.45 (TEM-3)(A) Seebeck coefficient.(B) Electrical resistivity.(C) Thermal conductivity. The temperature-dependent function of TRBEP and AEP of TEHP1-1263-1.5 (TEM-2) and TELBP1-12656-0.45 (TEM-3) (A) Seebeck coefficient. (B) Electrical resistivity. (C) Thermal conductivity. Discussion This study proposes an approach for estimating the equivalent temperature-dependent properties of TEMs using only their output characteristics. By combining the TRM and NSM, two sets of equivalent properties (Seebeck coefficient, electrical resistivity, and thermal conductivity) were determined without requiring additional specific experiments. The results demonstrate that the obtained equivalent properties enable accurate prediction of TEM output characteristics for both models, with average deviations of open-circuit voltage, maximum power, and conversion rate of 0.87%, 2.53%, and 3.07% for the TRM, and 0.91%, 2.51%, and 3.24% for the NSM, respectively. The sensitivity analysis of input parameters and module types indicates that the proposed method is highly accurate, reliable, and feasible within the provided data range. These findings indicate that the proposed approach is straightforward to implement and can be applied to can be applied to various TEMs with different sizes, structures, and materials to determine the equivalent sets of temperature-dependent properties with only the manufacturer’s output specification or experimental data with fewer measurement points and larger temperature difference intervals, significantly improving the time and cost savings. The obtained equivalent properties can be directly applied to TRMs or NSMs, reducing the requirements for input parameters and avoiding the need to model parasitic losses explicitly. This result simplifies the modeling process and makes the simulation of TEMs and thermoelectric generator systems more accessible, especially when limited experimental data are available. Furthermore, the use of different sets of equivalent properties ensures consistency between the TRM and NSM, resulting in negligible discrepancies in predicted output characteristics. This finding is particularly important for large-scale or multiphysics simulations of thermoelectric generator systems, where the TRM can replace the numerical model of TEMs to achieve comparable accuracy while substantially reducing computational cost. Based on this work, future studies can further investigate the application of these equivalent properties in system-level and multiphysics simulations to evaluate their performance under more complex operating conditions. Limitations of the study It should be noted that the equivalent properties in this work are established to be intended for accurate simulation of the overall output characteristics of TEMs, particularly suitable for simulations at the thermoelectric generator system level. Due to their equivalence, these properties cannot completely replace the actual thermoelectric material properties and are not suitable for detailed material-level analysis or thermoelectric couple structural optimizations. In addition, the current study utilizes the TEM’s specifications provided by the manufacturer, which are limited in the amount of data and exhibit considerable errors across different cold-side surface temperatures. As a result, the equivalent thermal conductivity varies substantially. The next research direction is to perform more precise experiments to further validate the proposed method in comparison with other estimating methods. Furthermore, mathematical models that consider the influence of temperature distribution along the thermoelectric legs can be incorporated into the estimating process, replacing the numerical model, to shorten the estimating time. Resource availability Lead contact •Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Thong Duc Hong (hongducthong@hcmut.edu.vn). •Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Thong Duc Hong (hongducthong@hcmut.edu.vn). Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Thong Duc Hong (hongducthong@hcmut.edu.vn). Materials availability •This study did not generate new unique reagents. •This study did not generate new unique reagents. This study did not generate new unique reagents. Data and code availability •All data are available in the article and in the supplemental information.•The code related to the developed method is publicly available at https://doi.org/10.5281/zenodo.19600340.•Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request. •All data are available in the article and in the supplemental information. All data are available in the article and in the supplemental information. •The code related to the developed method is publicly available at https://doi.org/10.5281/zenodo.19600340. The code related to the developed method is publicly available at https://doi.org/10.5281/zenodo.19600340. •Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request. Acknowledgments We acknowledge Ho Chi Minh City University of Technology (10.13039/100017533HCMUT), 10.13039/501100004692VNU-HCM for supporting this study. Author contributions Conceptualization, M.Q.P. and T.D.H.; methodology, M.Q.P. and T.D.H.; investigation, M.Q.P. and T.Q.V.; writing – original draft, M.Q.P. and T.Q.V.; writing – review and editing, V.L.N. and T.D.H.; resources, T.D.H.; supervision, V.L.N. and T.D.H. Declaration of interests The authors declare no competing interests. Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work, the authors used ChatGPT (GPT-5 Thinking, OpenAI) to assist with language editing and clarity. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERSoftware and algorithmsANSYS 2025R2ANSYS, Inchttps://www.ansys.com/MATLAB 2024bMathworkshttps://ch.mathworks.com/products/matlab.htmlCode for thermoelectric equivalent property estimationThis paperhttps://doi.org/10.5281/zenodo.19600340OtherThermoelectric modulesThermonamic Electronics Corp. Ltdwww.thermonamic.com.cn REAGENT or RESOURCESOURCEIDENTIFIERSoftware and algorithmsANSYS 2025R2ANSYS, Inchttps://www.ansys.com/MATLAB 2024bMathworkshttps://ch.mathworks.com/products/matlab.htmlCode for thermoelectric equivalent property estimationThis paperhttps://doi.org/10.5281/zenodo.19600340OtherThermoelectric modulesThermonamic Electronics Corp. Ltdwww.thermonamic.com.cn Experimental model and study participant details This study does not involve any experimental models or study participants. Method details Parameters of thermoelectric modules A TEHP1-1264-0.8 module43 of the Thermonamic Electronics Corp. Ltd was chosen to conduct the estimating equivalent properties process in this study. The thermoelectric module (TEM) information and dimensions are provided in Table S1, referenced from Luo and Wang.31 In addition, the ceramic plate and copper electrode parameters are considered constant for different modules and are referenced from Luo et al.,38 as presented in Table S2. Figure S4 presents the specification of TEM retrieved from the specification data sheet supplied by the manufacturer, including: open-circuit voltage, VOC,MFR, matched load voltage, Vm,MFR, current, Im,MFR, power, Pmax,MFR, and conversion rate at maximum power, ηMFR, according to various hot-sdie and cold-side temperature, Th and Tc, which were used to estimate the equivalent properties of TEMs. From these output characteristics, the internal resistance and the heat flow across the TEM can be extracted from their relationship. In the manufacturer’s specifications, the output characteristic at different Tc was provided; however, measuring heat flow at different Tc is highly sensitive due to the poor insulation conditions in the experiment, which are unpredictable.24,35,42 Using input data at different Tc to estimate equivalent thermal-dependent properties will lead to significant errors and make it challenging to construct reasonably equivalent properties. Hence, the set of TEM’s output characteristic under Tc = 303 K, which yields the largest capacity, was chosen to present the estimation process, with an aim of covering the entire maximum operating range of TEM. The sensitivity of the input characteristics under different Tc will be discussed in the discussion section. Furthermore, the two other TEMs, TEHP1-1263-1.543 and TELBP1-12656-0.45,43 are used to verify the practicality of the proposed estimating method. The dimensions of these modules are supplied in Kramer e al.,44 and Massaguer et al.45 respectively. Thermal resistance model Figure S5 presents the scheme of the heat flow of a typical thermoelectric couple in TEM. In Figure S5A, first, the fundamental effects include the Fourier, Peltier, Joule, and Thomson effects, which generate conductive heat along the thermoelectric legs (QF,n and QF,p), pumping heat from the hot to cold junction (QPnh, QPph, QPnc, and QPpc), heat generation (QJn and QJp), and heat absorption or dissipation in the side of the legs (QT). Also, electrical and thermal contact resistances will appear in the solder layers connecting the ceramic plates, copper electrode, and thermoelectric legs (Rc and TRc).24 Furthermore, conduction, convection, or radiation will occur within the air or filler layers between the sidewalls as well as the thermoelectric legs of the TEM, causing heat losses (Qconv, Qair/filler).28,35,45 Considering all these effects when simulating TEM characteristics is essential to obtain accurate results; however, it is resource- and time-consuming, making it difficult to estimate TEM properties. Therefore, a simple heat flow scheme that only includes the main TEM effects of Fourier, Peltier, and Joule, as shown in Figure S5B, is used. These main effects, along with the equivalent thermal conductivity, electrical resistivity, and Seebeck coefficient, are expected to account for almost all effects within the TEM in the thermal resistance model (TRM) in this study. The assumptions for the thermal resistance model to be applied to determine the equivalent system are as follows:•The materials of P-type and N-type legs are isotropic with the same equivalent thermal conductivity and resistivity, while their Seebeck coefficient are equal in magnitude value but opposite signs.35•One-dimensional heat transfer occurs only between the ceramic plates, copper electrodes, and thermoelectric couple, meaning that the conductive heat driven by the thermoelectric legs’ equivalent thermal conductivity accounts for the conductive heat, thermal contact resistance between surfaces, and heat losses through air/filler. Also, the Thomson effect is ignored due to its negligible impact on the overall performance of TEM.46 In addition, the side walls are considered adiabatic because the heat losses are already taken into account by the equivalent thermal conductivity in this assumption.•The thermoelectric leg equivalent electrical resistance includes the copper electrode and contact resistance. •The materials of P-type and N-type legs are isotropic with the same equivalent thermal conductivity and resistivity, while their Seebeck coefficient are equal in magnitude value but opposite signs.35 The materials of P-type and N-type legs are isotropic with the same equivalent thermal conductivity and resistivity, while their Seebeck coefficient are equal in magnitude value but opposite signs.35 •One-dimensional heat transfer occurs only between the ceramic plates, copper electrodes, and thermoelectric couple, meaning that the conductive heat driven by the thermoelectric legs’ equivalent thermal conductivity accounts for the conductive heat, thermal contact resistance between surfaces, and heat losses through air/filler. Also, the Thomson effect is ignored due to its negligible impact on the overall performance of TEM.46 In addition, the side walls are considered adiabatic because the heat losses are already taken into account by the equivalent thermal conductivity in this assumption. One-dimensional heat transfer occurs only between the ceramic plates, copper electrodes, and thermoelectric couple, meaning that the conductive heat driven by the thermoelectric legs’ equivalent thermal conductivity accounts for the conductive heat, thermal contact resistance between surfaces, and heat losses through air/filler. Also, the Thomson effect is ignored due to its negligible impact on the overall performance of TEM.46 In addition, the side walls are considered adiabatic because the heat losses are already taken into account by the equivalent thermal conductivity in this assumption. •The thermoelectric leg equivalent electrical resistance includes the copper electrode and contact resistance. The thermoelectric leg equivalent electrical resistance includes the copper electrode and contact resistance. From Figure S5B, the heat flow Qh enters the hot-side junction, and heat flow Qc is rejected from the cold-side junction according to Fourier’s law as follows38:(Equation 1)Qh=Th−Th,legTRhce+TRcowhere TRhce and TRco denote the thermal resistances of the hot ceramic plate and copper electrode, respectively. Th,leg depicts the thermoelectric legs’ hot junction temperature.(Equation 2)Qc=Tc,leg−TcTRcce+TRcowhere TRcce denotes the cold ceramic plate thermal resistance. Tc,leg presents the cold junction temperature.(Equation 3)TRhce/cce=δhce/cceAhce/cceλhce/ccewhere Ahce/cce, λhce/cce, and δhce/cce present the hot/cold ceramic plate’s area, thermal conductivity, and thickness, respectively.(Equation 4)TRco=δcoAcoλcowhere Aco, λco, and δco, present the copper electrode’s area, thermal conductivity, and thickness, respectively. Next, the combined effect of the Fourier, Seebeck, Peltier, and Joule occurs at the thermoelectric couples. The heat flow entering and leaving them is described as follows38:(Equation 5)Qh=N(α¯p−α¯n)ITh,leg+Th,leg−Tc,legTRlegs−12I2Rint(Equation 6)Qc=N(α¯p−α¯n)ITc,leg+Th,leg−Tc,legTRlegs+12I2Rintwhere α¯ depicts the average Seebeck coefficient. Rint denotes the TEM’s internal resistance, which is defined by Equation 7. TRlegs is the thermoelectric legs’ thermal resistance, determined by Equation 8. I is the electric current, described by Equation 9. N is the thermoelectric couple quantity.(Equation 7)Rint=N(σp−1¯+σn−1¯)δlegAleg(Equation 8)TRlegs=δlegNAleg(λp¯+λn¯)(Equation 9)I=Nα¯(Th,leg−Tc,leg)Rin+RLwhere δleg and Aleg denote the thermoelectric leg thickness and area. RL depicts the external load resistance. σp/n−1¯ and λp/n¯ present the average electrical resistivity and thermal conductivity of the thermoelectric legs. The average values of the properties are determined by using the integrated-temperature average method as follows:(Equation 10)−α¯n=α¯p=∫Tc,legTh,legα(T)dTTh,leg−Tc,leg(Equation 11)σp−1¯=σn−1¯=∫Tc,legTh,legσ−1(T)dTTh,leg−Tc,leg(Equation 12)λp¯=λn¯=∫Tc,legTh,legλ(T)dTTh,leg−Tc,leg The cold-side heat flow is subtracted from the hot-side heat flow to determine the output power.(Equation 13)P=Qh−Qc=Nα¯I(Th,leg−Tc,leg)−I2Rint The voltage in the model is determined by:(Equation 14)V=VOC−IRint=Nα¯(Th,leg−Tc,leg)−IRint The expression for the open-circuit voltage is as follows:(Equation 15)VOC=Nα¯(Th,leg−Tc,leg) The conversion rate of TEM is defined as:(Equation 16)η=PQh These equations can be solved with certain values of Th, Tc, and RL by using the iterative method in a common coding program, such as MATLAB. Numerical simulation model Numerical model descriptions Figure S6 shows a three-dimensional geometric computational mesh and boundaries of the TEM in the numerical simulation model (NSM), including ceramic plates, copper electrodes, 126 thermoelectric couples, and an external load resistance, whose detailed dimensions and known physical properties are presented in Tables S1 and S2. The boundary conditions included both electrical connections and temperature constraints. The TEM’s hot-side and cold-side surfaces were subjected to constant, uniform hot and cold surface temperatures, respectively, to establish the temperature boundary. For the electrical boundary, the external load resistance is linked to the TEM to fulfill the electrical circuit, namely, one side of the load and the copper electrode linked to the N-type leg were set to ground voltages; meanwhile, the opposite electrode of the TEM was linked with the remaining side of the load via a voltage coupling boundary. The load temperature was maintained at the environment temperature of 293.15 K. Adiabatic conditions were applied to the side walls of the thermoelectric legs, as convection was assumed to be absent. In each simulation case, the hot-side temperature, cold-side temperature, and electrical resistivity of the external load resistance are defined according to the desired simulation condition. The numerical simulation model used a structured grid discretized by Solid226 hexahedral elements for the thermoelectric legs and copper electrode, and Solid179 for the ceramic plates. A grid sensitivity examination was conducted to verify the simulation’s accuracy and maintain grid independence. The operating conditions of RL 1.67 Ω, Tc = 303 K, and Th = 573 K were applied. The element quantity was progressively raised by decreasing the element size while retaining constant model conditions. Five mesh levels with 62000, 98000, 166000, 249000, and 461000 elements were obtained. Once the total element reached 166000 with an element size of 0.3 mm, the change in the TEM’s output power was below 0.5%. To balance model accuracy and computational cost, this grid was chosen for the numerical simulation. In this study, the numerical simulation was performed using the ANSYS Thermal–Electric module under steady-state conditions. The coupled thermal and electrical governing equations were solved simultaneously using the finite element method implemented in the Mechanical APDL solver. A tolerance of 10−3 was applied to all convergence measures. The solution was considered converged only when all criteria for temperature, voltage, heat flow, and current were simultaneously satisfied. The heat transfer equation combines the effects of heat conduction, Joule heating, Thomson heating along the thermoelectric legs, and the Peltier effect at the interface of P-type and N-type legs47:(Equation 17)∇·(λp(T)∇Tp)=∇αp(T)J→Tp−σp−1(T)J→2(Equation 18)∇·(λn(T)∇Tn)=∇αn(T)J→Tn−σn−1(T)J→2where the first term on the right side depicts the Thomson heat in the legs, or the Peltier heat at the couple’s junction, and the second term is the Joule heat, while the left side represents the Fourier heat. J→ denotes the current density vector. In the copper electrode and ceramic plates, the heat transfer equations are defined as follows47:(Equation 19)∇·(λco∇T)=−σco−1J→2(Equation 20)∇·(λce∇T)=0 The current density vector and the electric field vector were defined as follows47:(Equation 21)E→=−∇ϕ+αp/n∇T(Equation 22)J→=E→/σp/n−1where E→ and ϕ denotes the electric field vector and the electric potential, respectively. The current density is conserved and continuous throughout the entire system47:(Equation 23)∇·J→=0 Figure S7 presents the results of the NSM using the ANSYS Thermal-Electric program in this work, including contours of temperature, voltage, and current density at Th = 573 K and Tc = 303 K. From these results, the current is determined by integrating the current density over a cross-sectional area. The output power is defined by multiplying the voltage and current, which is equal to the joule heat produced by the load resistance. The voltage-current curve was employed to compute the internal resistance as well as the open-circuit voltage. The cold-side and hot-side heat flow is the heat reaction on the cold-side and hot-side boundaries. Numerical simulation model validation To ensure reliability, the numerical simulation model was validated against experimental data from previous studies of Luo et al.31 The TEM used in the prior work was TEG 127020, with detailed structural parameters and properties provided in the study.31 The validation process was performed at hot-side temperatures ranging from 330 to 570 K, with the external resistance kept constant at RL = 4 Ω. Simulation settings and the computational grid were implemented as described in STAR Methods (numerical model descriptions). Figure S8 depicts the output power and output voltage harvested from the current study’s simulation model and experimental data. It can be observed that both output power and voltage of the simulation are highly consistent with the experimental results, with mean errors of 1.01% and 2.11%, respectively. Minor discrepancies may arise from experimental errors and small dimension deviations during the modeling process. These results indicate that the established NSM is reliable for determining the equivalence coefficient. A method for estimating equivalent thermoelectric properties Figure S9 shows the overall procedure for determining the equivalent property set proposed in this study. First, the TRM was constructed, and an iterative process was used to derive the average material properties from the output characteristic provided by the manufacturer. The initial average electrical resistance and thermal conductivity can be referenced from other works. The average Seebeck coefficients were calculated from the open-circuit voltage characteristic; the average electrical resistivities were computed from the maximum power characteristic; and the average thermal conductivities were determined from the conversion rate at the maximum power characteristic. The process is converged as the average errors of properties between two iterations are less than 0.5%. Then, the temperature-dependent property functions were constructed by solving the system equations containing the function’s coefficients with the average properties at various temperature differences. After this step, a set of equivalent properties appropriate for the TRM, which is called the thermal resistance-based property (TRBEP), was determined. Finally, the average electrical resistivity and thermal conductivity were adjusted by the internal resistance ratio between the manufacturer and NSM, and the Fourier heat ratio between the TRM and NSM, respectively; and then the adjusted property functions were reconstructed based on the updated average properties. The adjustment process is repeated until the error between the output characteristics of the numerical simulation and the manufacturer’s specifications is less than 2%, and the most suitable property set for the NSM, which is called the adjusted equivalent property (AEP), is found. In the following sections, the detailed processes for determining the average properties, the method of constructing the temperature-dependent function, and the property adjustment are presented in more detail. Determination of the average equivalent properties Figure S10 presents the iterative process to determine the equivalent average property set of the TEM by using TRM. The manufacturer’s output characteristics, which are presented in STAR Methods (parameters of thermoelectric modules), were employed to extract the average properties in this step of the estimation process. There are three sub-processes. The first sub-process was built with the Seebeck coefficient as a variable, varying in the range from 50 to 250 μV/K referenced from many other works. The initial average thermal conductivity and electrical resistivity values are referenced from other modules with the same material. The TRM was employed at open-circuit conditions, meaning the electric current equals zero. With each Seebeck coefficient, the open-circuit voltage, VOC,TRM, was computed by Equation 15 and compared with the manufacturer’s open-circuit voltage, VOC,MFR. In the second sub-process, the electrical resistivity was considered a variable, ranging from 0.5 to 2 (10−6) Ωm. The average Seebeck coefficient was updated from the previous process, while the average thermal conductivity remained referenced. The maximum power, Pmax,TRM, was calculated when the internal resistance was equal to the load resistance27 and compared with the manufacturer’s maximum power, Pmax,MFR. In the third sub-process, the thermal conductivity was treated as a variable, with a range of 0.5 to 2 W/mK, while the average electrical resistivity and Seebeck coefficient were updated from the previous processes. The hot-side heat flow of TEM, Qh,TRM, was computed at the maximum power condition using Equation 5 and compared with the manufacturer’s hot-side heat flow, Qh,MFR, which was obtained from the conversion rate and maximum power provided by the manufacturer. Finally, the errors between two iterations were checked to ensure the reliability of the obtained properties. The new electrical resistivity and thermal conductivity were reused as initial parameters for determining the Seebeck coefficient in the next round. To limit cumulative error in the subsequent determination steps, the convergence criteria, including the average error of output characteristics between the TRM and manufacturer and the average error of properties between two iterations, were set to 0.5%. From this step, the sets of average Seebeck coefficient, electrical resistivity, and thermal conductivity according to different hot-side and cold-side temperatures were obtained, serving as a basis for constructing and adjusting the property functions in the next steps. Construction of the temperature-dependent function of properties The next step is to construct the temperature-dependent functions of the properties. First, an appropriate averaging method needs to be chosen. Several popular averaging methods can be used, including the integrated-temperature average, the function value at the average temperature, the average of the function values at the boundary temperatures, the weighted temperature average, and a spatial average of temperature-dependent functions.37 Among these methods, the last provides the most accurate results, but it is too complicated and cannot be used to reconstruct property functions from average properties. Hence, the integrated-temperature average method, which has an acceptable error, was chosen. The average property in a certain temperature difference is determined as follows:(Equation 24)p¯=1ΔTleg∫Tc,legTh,legp(T)dTwhere p(T) presents any function of the thermoelectric properties α, σ -1, and λ; and the average properties p¯ are their average values in the range from Tc,leg to Th,leg. Secondly, in many previous studies, second, third, and fourth-order functions,30,36,48 and even sixth-order functions27 were used to determine the property function. Although the higher-order function yields a stronger correlation, it may lead to overfitting when the input data is noisy, distorting the predicted function, especially in the extrapolation region, making it difficult to control the trend of the coefficients during adjustment and causing significant errors across cases. Therefore, to balance the stability of the material property function trend and high accuracy and reliability, the second-order function was chosen to predict the Seebeck coefficient function, while the third-order function was used to predict and adjust the resistivity and thermal conductivity functions, which are presented respectively as follows:(Equation 25)α(T)=aαT2+bαT+cα(Equation 26)σ−1(T)=aσ−1T3+bσ−1T2+cσ−1T+dσ−1(Equation 27)λ(T)=aλT3+bλT2+cλT+dλwhere aα, bα, and cα, are the function coefficients of Seebeck coefficient. aσ−1, bσ−1, cσ−1, and dσ−1 are the function coefficients of electrical resistivity. aλ, bλ, cλ, and dλ are the function coefficients of thermal conductivity. Substituting the functions in (Equation 25), (Equation 26), (Equation 27) into Equation 24 and expanding the antiderivatives, in turn, the system equations with unknowns being the function coefficients at various known average values and temperature differences are established as follows:(Equation 28){(aαT33+bαT22+cαT)|Tc,leg1Th,leg1=α¯1×ΔTleg1...(aαT33+bαT22+cαT)|Tc,legjTh,legj=α¯j×ΔTlegj(Equation 29){(aσ−1T44+bσ−1T33+cσ−1T22+dσ−1T)|Tc,leg1Th,leg1=σ−1¯1×ΔTleg1...(aσ−1T44+bσ−1T33+cσ−1T22+dσ−1T)|Tc,legjTh,legj=σ−1¯j×ΔTlegj(Equation 30){(aλT44+bλT33+cλT22+dλT)|Tc,leg1Th,leg1=λ¯j×ΔTlegj...(aλT44+bλT33+cλT22+dλT)|Tc,legjTh,legj=λ¯j×ΔTlegj Finally, the function coefficients can be computed using least-squares regression with various Tc,leg, Th,leg, and averaged properties obtained from the iterative process in STAR Methods (determination of the average equivalent properties). The coefficient of determination between the regression results and the input data for average properties was higher than 0.997, and the relative error was lower than 1%, ensuring the appropriateness and reliability of the established function. It should be noted that the high coefficient of determination and relatively small error achieved may be due to the characteristics provided by the manufacturer, which are smooth. In the case of using the experimental characteristic with a high noise level, the input characteristics should be smoothed so that the method can be applied more easily and accurately. Adjustment of the electrical resistivity and thermal conductivity functions Ponnusamy et al.37 have proven that the average electrical resistivity and thermal conductivity depend significantly on the thickness and temperature distribution of the thermoelectric legs. Also, according to Xu et al.,30 temperature distributions are unexpectedly impacted by temperature-dependent material properties. Therefore, constructing functions that rely solely on temperature will lead to unpredictable errors across different materials and temperatures. Thus, the electrical resistivity and thermal conductivity were adjusted through the numerical simulation. The electrical resistivity was adjusted by the internal resistance ratio θR, determined as the ratio between the internal resistance computed by NSM and the manufacturer’s data. At each hot-side and cold-side temperature, the θRj was defined as follows:(Equation 31)θRj=Rint,MFRjRint,NSMjwhere θR is the internal resistance ratio, Rint,MFR, and Rint,NSM are the internal resistances obtained from the manufacturer’s data and NSM. Next, θRj was used to directly modify the corresponding previous average electrical resistivity by Equation 32 because it is proportional to the internal resistance:(Equation 32)σp/n−1¯newj=σp/n−1¯oldjθRj Similarly, the thermal conductivity was adjusted by the Fourier heat ratio θF, defined as the ratio between the Fourier heats obtained from the NSM and TRM. At each hot-side and cold-side temperature, the θFj was defined as follows:(Equation 33)θFj=QF,TRMjQF,NSMjwhere θF is the Fourier heat ratio, QF,TRM, and QF,NSM are the Fourier heat computed by the TRM and NSM. The Fourier heat in the TRM was calculated by Equation 34, with the average thermal conductivity obtained from the previous step. Since the TRM’s heat flow has been calibrated to match the manufacturer’s, this Fourier heat can be considered the manufacturer’s data. Meanwhile, the Fourier heat of the NSM was indirectly determined by subtracting the Peltier and Joule heats from the hot-side heat flow, as defined by Equation 35.(Equation 34)QF,TRMj=Th,legj−Tc,legjTRlegsj(Equation 35)QF,NSMj=Qh,NSMj−QPh,NSMj+0.5QJ,NSMjwhere Qh,NSM, QPh,NSM, and QJ,NSM are the hot-side heat flow, hot-side Peltier heat, and Joule heat in NSM. Next, the θFj was used to directly modify the corresponding previous average thermal conductivity using Equation 36, because it is proportional to the Fourier heat.(Equation 36)λ¯newj=λ¯oldjθFj The new sets of average electrical resistivity and thermal conductivity were computed by modifying the old average values of electrical resistivity and thermal conductivity at different hot-side and cold-side temperatures using the corresponding internal resistance ratios and Fourier heat ratios. Then, the least-squares regression process, which was presented in STAR Methods (construction of the temperature-dependent function of properties), was continuously used to reconstruct the new electrical resistivity and thermal conductivity functions by their new average values. It should be noted that the new functions are constructed using the integrated-temperature averaging method, which still has an error; hence, it is not possible to achieve good results after a single modification. Therefore, the adjustment must be repeated until an acceptable error of output characteristics between the NSM and manufacturer data is achieved, which is set to 2% in this study. Quantification and statistical analysis No statistical hypothesis testing was performed in this study. The regression procedure used to construct temperature-dependent property functions was evaluated using the coefficient of determination and relative error.