C-HILS-Based Evaluation of Control Performance, Losses, and Thermal Lifetime of a Marine Propulsion Inverter

Abstract

This paper presents a controller-hardware-in-the-loop simulation (C-HILS) framework for validating models, evaluating control performance, and assessing the thermal lifetime of a tens-of-kilowatt inverter. The real inverter and the C-HILS platform were operated in parallel, and accuracy was quantified using phase-current root mean square error, voltage spectral analysis, and total harmonic distortion (THD). Across a wide range of SVPWM and DPWM cases, deviations remained within 2–5%, confirming close agreement between experiment and simulation. Using the validated C-HILS system, sampling frequency and output power were swept while comparing current tracking, THD, average switching frequency, semiconductor losses, and efficiency. SVPWM achieved lower THD, whereas DPWM reduced average switching frequency and switching losses, improving efficiency. C-HILS waveforms were then applied to a Foster thermal network to reconstruct the junction–temperature trajectory; $T_{\mathrm{j}}\left(t\right)$, and $\mathsf{\Delta }{T}_{\mathrm{j}}$ and $T_{\mathrm{j},\mathrm{m}\mathrm{i}\mathrm{n}}$ were mapped to lifetime using the Bayerer model. For a representative cyclic mission, $\mathsf{\Delta }{T}_{\mathrm{j}}$ decreased from approximately $25.6 °\mathrm{C}$ with SVPWM to about $17.5 °\mathrm{C}$ with DPWM, increasing the estimated lifetime from approximately 1.36 years to 9.14 years. These results demonstrate that the proposed C-HILS framework provides a unified pre-prototype tool for model verification, control strategy comparison, and quantitative thermal reliability assessment of shipboard propulsion inverters.

1. Introduction

Maritime transport accounts for approximately 90% of global trade volume, representing a dominant share of worldwide logistics compared with other transportation modes. Beyond its industrial significance, the environmental impact of the shipping sector has drawn increasing global attention [1]. In response, the International Maritime Organization (IMO) adopted mid-term greenhouse gas (GHG) reduction measures during the 83rd session of the Marine Environment Protection Committee (MEPC) held in April 2025 [2]. The IMO has also established a long-term target of achieving net-zero emissions by 2050 through a 100% reduction in carbon output. Consequently, the development of eco-friendly propulsion technologies to replace conventional fossil fuel-based internal combustion engines has become essential [3,4,5,6,7,8].

Representative eco-friendly ship technologies include alternative fuel propulsion systems using carbon-free fuels such as ammonia and hydrogen, electric propulsion systems powered by batteries or fuel cells, and hybrid electric propulsion systems that combine fuel-based and electrical energy sources. Among these technologies, electric and hybrid electric propulsion vessels provide superior maneuvering performance and improved dynamic response of the main engines compared with conventional internal combustion systems [9,10]. In particular, because their electrical systems can be configured as DC networks that eliminate the need for generator synchronization, they can achieve fuel savings of over 20% relative to AC systems, thereby enhancing energy efficiency and reducing emissions [11].

With the advancement of eco-friendly ship technologies, research has expanded to equipment not previously used on conventional vessels. For internal combustion engine systems, representative developments include sulfur oxide reduction, nitrogen oxide reduction, carbon capture, and waste heat recovery technologies [12,13,14,15]. Research on electric propulsion ships has focused primarily on key components of electric propulsion systems, including marine batteries, inverters, and propulsion motors [16,17,18]. Although these components are widely used in land-based industries, marine applications expose them to harsh conditions such as high humidity, vibration, salinity, and dust, as well as constraints including limited installation space and maintenance challenges. Therefore, redesign and reliability verification that explicitly consider the marine environment are required.

Ships demand a high level of reliability from such emerging technologies, and all equipment must satisfy type-approval tests and relevant standards before installation [19]. Meeting these requirements necessitates extensive testing and validation during the early development stages. However, compared with automobiles and aircraft, testing costs for ships are significantly higher, and continuously changing sea conditions make it difficult to perform repetitive verification under consistent conditions. In particular, sea trials can be conducted only when a vessel under construction reaches the completion stage, rendering validation under actual operating conditions impractical during development. Even when onboard testing is possible, physical separation between the test area and personnel is difficult to achieve, potentially leading to severe financial losses and even casualties in the event of an accident [20], while timely response at sea remains inherently challenging [21]. For electric propulsion ships, verification is required not only for individual components but also for the overall stability of the power system, which includes the power converter, battery, and propulsion motor [22]. Yet, the number of land-based test beds is limited, and constructing motor and battery configurations that satisfy specific operating conditions requires substantial time and cost [23]. Critically, conducting validation across the wide range of operating conditions encountered in real marine environments remains fundamentally difficult [24]. As a result, growing attention has been directed toward alternative testing environments that can pre-validate the reliability of equipment and control strategies during development [25].

To address these limitations, hardware-in-the-loop simulation (HILS) has gained increasing attention [26,27,28]. HILS is a simulation methodology widely used in fields such as automotive, aerospace, and defense, where it serves as an effective tool for validation and pre-testing [29,30,31]. HILS performs real-time simulations by linking a virtual plant model to an actual controller or embedded system, eliminating the need for physical hardware. Because all components except the controller operate within the simulation model, safe and repetitive verification can be performed without the risks associated with connecting real hardware. Moreover, HILS provides high flexibility in system configuration, significantly reducing the cost and time required for testing and development, which is particularly advantageous for ship systems that require substantial resources for fabrication. HILS can also reproduce load variations similar to those encountered during real sea operations, overcoming many physical constraints associated with sea trials. For electric propulsion system components, HILS enables verification of both hardware stability and control algorithm performance under conditions that replicate real marine environments and diverse operational profiles. Therefore, by employing HILS, developers can simultaneously improve cost efficiency and operational safety while overcoming the limitations of physical sea trials, enabling preliminary evaluations under a wide range of operating conditions.

Despite these advantages, research on the application of HILS to ship propulsion systems and electric propulsion vessels remains limited [32,33,34]. Most existing studies rely on model-based simulation validation, which lacks sufficient reliability assessment through comparison with real hardware systems [35,36,37,38,39]. Although HILS has also been applied in other power electronics domains, prior studies have primarily focused on controller operation verification or system modeling rather than on quantitative evaluation of accuracy and reliability through experimental comparison with physical hardware [40,41,42].

In this study, the reliability of an inverter—one of the key components of electric propulsion ships—was verified using a controller-hardware-in-the-loop simulation (C-HILS) framework. A 100-kW real inverter and its corresponding C-HILS inverter model were operated under identical conditions, and the three-phase current, line-to-line voltage, and fast Fourier transform (FFT) results were compared to evaluate consistency between the physical system and the C-HILS model. Subsequently, space–vector pulse-width modulation (SVPWM) and discontinuous pulse-width modulation (DPWM) methods were applied under the same operating conditions in the C-HILS environment to compare their performance differences and error characteristics across various scenarios. In addition, semiconductor device losses were estimated based on the current and voltage ripple characteristics obtained from the C-HILS experiments.

Experimental results showed that the output deviation between the real inverter and the C-HILS model remained within 5%, verifying the reliability of the proposed C-HILS system. The results also demonstrated that the inverter’s operational characteristics and thermal losses, and consequently its lifetime, varied depending on the control strategy, confirming that control algorithms directly influence system reliability. Accordingly, this paper presents a C-HILS-based reliability assessment methodology for marine propulsion inverters and demonstrates that performance and reliability verification using HILS can be effectively implemented during the development process of electric propulsion ship inverters. Furthermore, the proposed approach can be extended to the performance evaluation and lifetime prediction of other electric propulsion subsystems, including propulsion motors, batteries, and fuel cells.

2. System Overview of Electric Propulsion Inverter for C-HILS Application

2.1. Two-Level Voltage Source Inverter in Electric Propulsion Systems

The propulsion system of an electric propulsion ship consists of power sources such as batteries or generators, power conversion devices, and propulsion motors [43]. Among these components, the inverter is a key device that converts the DC supplied by the power source into AC with the desired voltage and frequency to drive the propulsion motor. Figure 1 shows the circuit configuration of the three-phase, two-level voltage-source inverter and the overall electric propulsion system incorporating this inverter.

The inverter control strategy is critical for determining the quality of the output voltage waveform and the dynamic response of the control system, which directly influences torque ripple and speed stability in the propulsion motor [44,45]. Increasing the switching frequency improves the voltage waveform quality but increases switching losses, whereas reducing the number of switching events improves efficiency but produces higher torque ripple. This trade-off defines the balance between performance and efficiency [46]. Therefore, selecting an appropriate control strategy is a key design parameter. In this section, two representative inverter-control strategies, SVPWM and DPWM, are introduced.

Among commonly used control methods, SVPWM represents the three-phase voltages as space vectors and implements them through an optimized switching sequence. This method minimizes output voltage distortion and maximizes voltage utilization, making SVPWM widely adopted in modern motor-drive systems. In this study, SVPWM is used as the primary control method. Compared with conventional sinusoidal pulse-width modulation (SPWM), SVPWM provides higher voltage utilization, lower voltage harmonic content, and more precise voltage control [47].

The basic principle of SVPWM is to divide the space-vector plane into six sectors. To achieve this, the three-phase voltages are transformed into the α-β stationary reference frame using the Clarke transformation:

$$\left[\begin{array}{c}{v}_{\mathsf{\alpha }}\\ v_{\mathsf{\beta }}\end{array}\right]={\displaystyle \frac{2}{3}}\left[\begin{array}{ccc}0& -{\displaystyle \frac{1}{2}}& -{\displaystyle \frac{1}{2}}\\ 0& {\displaystyle \frac{\sqrt{3}}{2}}& -{\displaystyle \frac{\sqrt{3}}{2}}\end{array}\right]\left[\begin{array}{c}{v}_{\mathrm{a}\mathrm{s}}\\ v_{\mathrm{b}\mathrm{s}}\\ v_{\mathrm{c}\mathrm{s}}\end{array}\right]$$

(1)

Using the transformed voltages $v_{\mathsf{\alpha }}$ and $v_{\mathsf{\beta }}$, the reference voltage vector $V_{\mathrm{r}\mathrm{e}\mathrm{f}}$ is defined as

$$V_{\mathrm{r}\mathrm{e}\mathrm{f}}=v_{\mathsf{\alpha }}+{jv}_{\mathsf{\beta }}$$

(2)

Within the sector where $V_{\mathrm{r}\mathrm{e}\mathrm{f}}$ is located, the average voltage vector is synthesized by allocating duty ratios to two adjacent active vectors and the zero vector. The magnitude and phase angle of $V_{\mathrm{r}\mathrm{e}\mathrm{f}}$ are expressed as

$$V_{\mathrm{r}\mathrm{e}\mathrm{f}}=\sqrt{v_{\mathsf{\alpha }}^2+v_{\mathsf{\beta }}^2}$$

(3)

$${\theta }_{\mathrm{s}}={\mathrm{t}\mathrm{a}\mathrm{n}}^{-1}\left({\displaystyle \frac{v_{\mathsf{\beta }}}{v_{\mathsf{\alpha }}}}\right)$$

(4)

The duty ratios of the two active vectors, $T_1$ and $T_2$, and the zero vector, $T_0$, are

$$T_1={\displaystyle \frac{T_{\mathrm{s}}{V}_{\mathrm{r}\mathrm{e}\mathrm{f}}}{V_{\mathrm{d}\mathrm{c}}}}\mathrm{s}\mathrm{i}\mathrm{n}\left({\displaystyle \frac{\pi }{3}}-{\theta }_{\mathrm{s}}\right)$$

(5)

$$T_2={\displaystyle \frac{T_{\mathrm{s}}{V}_{\mathrm{r}\mathrm{e}\mathrm{f}}}{V_{\mathrm{d}\mathrm{c}}}}\mathrm{s}\mathrm{i}\mathrm{n}\left({\theta }_{\mathrm{s}}\right)$$

(6)

$$T_0=T_{\mathrm{s}}-T_1-T_2$$

(7)

By selecting an appropriate combination of vectors and duty ratios, the inverter generates waveforms with low total harmonic distortion (THD) and high continuity. For propulsion motors, these characteristics reduce torque ripple and maintain stable torque, thereby enhancing propulsion efficiency while reducing noise and vibration [48].

Thus, SVPWM is advantageous for ensuring stable propulsion performance in electric propulsion ships. Figure 2a shows the six active vectors and two zero vectors formed by the eight switching states of the inverter in the hexagonal space-vector diagram. Figure 2b presents the three-phase voltage waveforms synthesized from these vectors.

Table 1. Space-voltage vectors corresponding to inverter switching states.

Inverter State	Switch State $[{\mathit{S}}_1,{\mathit{S}}_3,{\mathit{S}}_5]$	Pole Voltage $\left[{\mathit{v}}_{\mathbf{a}\mathbf{n}}\right]$	Pole Voltage $\left[{\mathit{v}}_{\mathbf{b}\mathbf{n}}\right]$	Pole Voltage $\left[{\mathit{v}}_{\mathbf{c}\mathbf{n}}\right]$	Space Vector $[{\mathit{V}}_{\mathbf{k}}={\mathit{v}}_{\mathbf{d}}+{\mathit{j}\mathit{v}}_{\mathbf{q}}]$
					${\mathit{v}}_{\mathbf{d}}$	${\mathit{v}}_{\mathbf{q}}$
$V_0$	$[0 0 0]$	$0$	$0$	$0$	$0$	$0$
$V_1$	$[1 0 0]$	${\displaystyle \frac{2}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{2}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$0$
$V_2$	$[1 1 0]$	${\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{2}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{\sqrt{3}}{3}}{V}_{\mathrm{d}\mathrm{c}}$
$V_3$	$[0 1 0]$	$-{\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{2}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{\sqrt{3}}{3}}{V}_{\mathrm{d}\mathrm{c}}$
$V_4$	$[0 1 1]$	$-{\displaystyle \frac{2}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{2}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$0$
$V_5$	$[0 0 1]$	$-{\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{2}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{\sqrt{3}}{3}}{V}_{\mathrm{d}\mathrm{c}}$
$V_6$	$[1 0 1]$	${\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{2}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	${\displaystyle \frac{1}{3}}{V}_{\mathrm{d}\mathrm{c}}$	$-{\displaystyle \frac{\sqrt{3}}{3}}{V}_{\mathrm{d}\mathrm{c}}$
$V_7$	$[1 1 1]$	$0$	$0$	$0$	$0$	$0$

summarizes the relationships among the inverter switching states, pole voltages, and space-voltage vectors.

The SVPWM operating process in a two-level inverter used in an electric propulsion ship is illustrated in Figure 3.

DPWM is a representative technique used to reduce inverter switching losses. In DPWM, only two of the three phases perform switching operations during each modulation period, while one phase remains clamped in either the ON or OFF state for a certain interval. This intentional clamping reduces the number of switching transitions, lowering switching losses and heat generation. Various DPWM types can be implemented depending on the timing and selection of zero vectors [49].

The distribution ratio $\mathsf{\mu }$ for the zero-vector application time is defined as

$$\mathsf{\mu }={\displaystyle \frac{t_{0\_\mathrm{v}7}}{t_{0\_\mathrm{v}0}{+t}_{0\_\mathrm{v}7}}}$$

(8)

where $t_{0\_\mathrm{v}0}$ and $t_{0\_\mathrm{v}7}$ denote the application times of vectors $V_0$ and $V_7$, respectively. When $\mathsf{\mu }=0$ or $\mathsf{\mu }=1$, the modulation corresponds to a DPWM scheme. By alternating the clamping region (where $\mathsf{\mu }=0$ or $\mathsf{\mu }=1$) every 60°, DPWM variants such as DPWM 30°, DPWM 60°, and DPWM 120° can be implemented depending on the clamping position [50].

In general, DPWM techniques offer advantages such as improved efficiency and reduced cooling requirements owing to lower switching frequency and switching losses. These characteristics make DPWM suitable for high-power or long-duration operating conditions where switching losses and thermal management are critical [51]. However, because of the discontinuous operation, the line-to-line voltage waveform exhibits step-like discontinuities, increasing harmonic content and current distortion relative to SVPWM [52]. For propulsion motors, this leads to increased torque ripple, speed fluctuations during low-speed operation, and higher noise and vibration, making DPWM less desirable in applications requiring low noise and low vibration [48].

Figure 4a presents the space-voltage vector configuration of DPWM based on the eight switching states, where six active vectors form a hexagonal structure similar to SVPWM. However, in DPWM, one phase is clamped within each sector. Figure 4b shows the corresponding three-phase voltage waveforms.

The DPWM operating process in a two-level inverter for electric propulsion applications is illustrated in Figure 5.

2.2. C-HILS

A HILS system generally comprises four key components: a physical controller, a plant or system model, a real-time simulator, and an input–output (I/O) interface. The real-time simulator computes the inverter’s dynamic behavior and generates corresponding sensor signals, while the I/O interface exchanges voltage, current, and PWM signals between the physical controller and the simulation. This configuration reproduces the input–output behavior of an actual system [53].

HILS platforms can be categorized into C-HILS and power-hardware-in-the-loop simulation (P-HILS), depending on the intended purpose. In C-HILS, the physical controller is connected to a virtual inverter model rather than physical power hardware, enabling system testing and validation in a fully simulated environment. Because all components other than the controller operate in software, C-HILS allows system verification without risk to actual equipment [54]. It also provides substantial flexibility for system modification and reduces cost and development time [55]. Furthermore, C-HILS facilitates early error detection during initial development and supports safe testing under operating conditions that are difficult or hazardous to reproduce experimentally [56,57].

In contrast, P-HILS is designed to verify full system operation, including the power stage, by incorporating both the controller and power hardware into the experiment [58]. In this study, C-HILS is employed because the objective is to compare and validate the electrical responses, losses, and thermal characteristics of different control strategies in real time. Figure 6 compares the architectures and signal flows of C-HILS and P-HILS, highlighting the structural differences and the types of signals exchanged among their components.

Modern real-time simulators used in power electronic applications can be broadly classified into three categories: Speedgoat, OPAL-RT, and Typhoon HIL. Their main characteristics are summarized in

Table 2. Comparison of real-time simulation platforms for power system applications.

	Speedgoat	OPAL-RT	Typhoon HIL
Toolchain Integration	Works natively with MATLAB/Simulink and Simulink Real-Time. Models can be directly deployed, tested, and logged in the same environment.	Uses RT-LAB to connect with Simulink/Modelica. Flexible, but the workflow is more complex.	Uses mainly its own IDE. Simulink connection is possible but mostly through indirect/bridge methods.
Real-Time Performance	Fixed-step real-time execution. Easy parameter tuning and logging. Best for medium/high-speed control and controller testing (C-HIL).	Combines multi-core CPU and FPGA. Strong for large power system/EMT simulations with small time steps.	Ultra-low latency and very short time steps, very good for accurate inverter switching tests.
FPGA/CPU Architecture	Direct link with HDL Coder/SOC Blockset. Easy to implement FPGA from Simulink models and control I/O timing.	Provides strong FPGA offload options (e.g., OP4xxx series). Flexible to add custom IP.	Optimized specifically for power electronics switching. Less flexible for general models.
Model Compatibility	Easily reuses existing Simulink models. Automatic code generation works consistently.	Supports many modeling tools (Simulink, Modelica, etc.) and is highly portable.	Mainly focused on power electronics libraries. General models need extra work to port.
Scalability & Expansion	Easy to expand with more I/O slots and racks. Good for step-by-step growth from CHIL to PHIL.	Can scale up with chassis/cluster setups. Very strong for large-scale simulations.	Scales well for specific power electronics use cases but is limited for general-purpose scaling.

.

OPAL-RT is well suited for large-scale power system studies; however, its resource structure is unnecessarily complex for C-HILS applications involving inverters in the tens-of-kilowatts range [59,60,61,62]. Typhoon HIL systems are primarily used for real-time verification of small-scale converters and power electronic systems of several kilowatts and therefore exhibit limitations in computational performance when modeling complex systems with multiple interacting components [63,64,65]. In contrast, Speedgoat offers computational and real-time processing capabilities appropriate for tens-of-kilowatts power electronic applications [66].

Because this study aims not only to analyze the inverter’s electrical characteristics but also to perform lifetime estimation based on thermal losses, the inverter model is implemented in MATLAB/Simulink, which supports real-time multi-physics simulation [67,68,69]. Speedgoat provides high compatibility and seamless integration with MATLAB/Simulink, enabling controller development and real-time validation within a single software environment [70,71,72].

OPAL-RT is generally well suited for grid-level HILS applications, whereas Typhoon HIL is more appropriate for device-level HILS. In contrast, this study focuses on HILS implementation at the level of a single inverter system, for which the Speedgoat platform is well suited. In addition, since this work investigates both the power electronic behavior and the thermal performance of the inverter, the C-HILS framework was developed using MATLAB/Simulink, and Speedgoat provides the highest level of compatibility with this environment.

3. Methodology

The experimental procedure consists of three stages: feasibility verification, performance evaluation, and reliability assessment. The overall experimental configuration is shown in Figure 7.

In the first stage, feasibility verification was performed to validate the accuracy of the proposed C-HILS system by comparing the experimental results of the real inverter with those obtained from the C-HILS platform. Quantitative comparison indices included the root-mean-square (RMS) values of the phase currents and the FFT- and THD-based analysis of the line-to-line voltages. The C-HILS model was considered valid when the difference between the two sets of results remained within 5% over the same time interval.

The second stage, performance evaluation, involved comparing the current-tracking behavior, THD, switching losses, and overall efficiency of two modulation strategies—SVPWM and DPWM—under identical operating conditions in the C-HILS environment.

The third stage, reliability assessment, calculated the conduction and switching losses of the inverter’s semiconductor devices using the current and voltage ripple characteristics, along with the switching behavior obtained from the C-HILS experiments. These losses were applied to a Foster RC thermal network to reconstruct the junction-temperature trajectory $T_{\mathrm{j}}\left(t\right)$ and determine the junction-temperature fluctuation $\mathsf{\Delta }{T}_{\mathrm{j}}$. A Bayerer lifetime model was then used to estimate the allowable cycle count and projected lifetime, providing a quantitative evaluation of the thermal reliability of each control method.

The configuration and external appearance of the C-HILS system and the three-phase, two-level voltage-source inverter used in this study are shown in Figure 8. The inverter was modeled in MATLAB/Simulink using identical parameters and operated in real time within the C-HILS experiments while interfaced with the physical controller.

3.1. Validation of C-HILS

In this study, the power loss $P_{loss}$, calculated based on the current $i$ and DC-link voltage $V_{dc}$, is used as an input to a Foster RC–based thermal model to calculate the junction temperature. The resulting temperature variation is then applied to a power-cycling lifetime model to estimate the inverter lifetime, and the overall configuration is presented in Figure 9.

C-HILS experiments were conducted under the same operating conditions as the real inverter tests to verify the applicability and reliability of the proposed simulation platform. For quantitative comparison, the RMS values of the three-phase currents were computed over the same time interval $[t_1, t_2]$. FFT analysis was performed using identical window sizes and frequency resolutions to calculate the THD and harmonic-component errors.

The RMS values of the three-phase currents for the real inverter and C-HILS experiments were computed using

$$\mathrm{R}\mathrm{M}\mathrm{S}(x)=\sqrt{{\displaystyle \frac{1}{T}}{\int }_{{\mathrm{t}}_1}^{{\mathrm{t}}_2}{x}^2\left(t\right)dt}, T=t_2-t_1$$

(9)

$${\epsilon }_{\mathrm{x}-\mathrm{R}\mathrm{M}\mathrm{S}}(\mathrm{\%})={\displaystyle \frac{\mid {\mathrm{R}\mathrm{M}\mathrm{S}}_{\mathrm{E}\mathrm{x}\mathrm{p}}-{\mathrm{R}\mathrm{M}\mathrm{S}}_{\mathrm{S}\mathrm{i}\mathrm{m}}\mid }{{\mathrm{R}\mathrm{M}\mathrm{S}}_{\mathrm{E}\mathrm{x}\mathrm{p}}}}\times 100, x=a, b, c$$

(10)

The frequency-domain error of the line-to-line voltage was quantified using the normalized spectrum root-mean-square error (Spec-NRMSE):

$$\mathrm{S}\mathrm{p}\mathrm{e}\mathrm{c}-\mathrm{N}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}(\mathrm{\%})={\displaystyle \frac{\sqrt{\frac{1}{K-1} \sum _{k=2}^K(V_{\mathrm{k},\mathrm{E}\mathrm{x}\mathrm{p}}-V_{\mathrm{k},\mathrm{S}\mathrm{i}\mathrm{m}}{)}^2}}{V_{1,\mathrm{E}\mathrm{x}\mathrm{p}}}}\times 100$$

(11)

Here, $K$ denotes the maximum harmonic order, set to the 40th order, and the FFT used an identical frequency interval $∆f$.

3.2. Lifetime Estimation of the Inverter for SVPWM and DPWM

The lifetime of the inverter was estimated using a power-cycling approach applied to the switching devices within the C-HILS environment. The lifetime of each device was computed using the Bayerer lifetime model:

$$N_f=A(\mathsf{\Delta }{T}_j{)}^{{\beta }_1}exp({\displaystyle \frac{{\beta }_2}{T_{j,min}+273}})t_{on}^{{\beta }_3}{i}_B^{{\beta }_4}{V}_c^{{\beta }_5}{d}_b^{{\beta }_6}$$

(12)

where $N_{\mathrm{f}}$ denotes the number of allowable cycles to failure, $\mathsf{\Delta }{T}_{\mathrm{j}}$ is the junction temperature variation within a single cycle, $T_{\mathrm{j},\mathrm{m}\mathrm{i}\mathrm{n}}$ represents the minimum junction temperature in a cycle, $t_{\mathrm{o}\mathrm{n}}$ is the duration of the high-load interval, $i_{\mathrm{B}}$ indicates the current per bond wire, $V_{\mathrm{c}}$ is the rated voltage divided by 100, and $d_{\mathrm{b}}$ represents the bond wire diameter.

The junction temperature variation per cycle is computed from the loss and Foster RC thermal network models, as follows:

$$\mathsf{\Delta }{T}_{\mathrm{j}}=T_{\mathrm{j},\mathrm{m}\mathrm{a}\mathrm{x}}-T_{\mathrm{j},\mathrm{m}\mathrm{i}\mathrm{n}}=\mathrm{m}\mathrm{a}\mathrm{x}\left(T_{\mathrm{j}}\left(t\right)\right)-\mathrm{m}\mathrm{i}\mathrm{n}\left(T_{\mathrm{j}}\left(t\right)\right)$$

(13)

The corresponding device lifetime $L_{sw}$ was obtained using

$$\begin{array}{c}{L}_{\mathrm{s}\mathrm{w}}=N_{\mathrm{f}} T_{\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{f}\mathrm{i}\mathrm{l}\mathrm{e}}\end{array}$$

(14)

where $T_{\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{f}\mathrm{i}\mathrm{l}\mathrm{e}}$ is the period of the applied load profile.

In this study, a Foster RC thermal network was used to model heat dissipation from the semiconductor devices. The model analyzes the transient thermal response from junction to case by applying thermal resistance $R$ and thermal capacitance $C$ values associated with each material layer. The thermal path is represented by an equivalent RC network, enabling dynamic temperature estimation. The instantaneous device power loss $P_{\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}}\left(t\right)$ served as the input to the model. Unless otherwise specified, the case temperature $T_c$ was set to $50 °\mathrm{C}$, which reflects conditions representative of a marine environment. Four RC branches were used, as shown in Figure 10.

The parameter values used for the Foster RC thermal network are listed in

Table 3. Parameters of the Foster RC network model for thermal behavior modeling.

Thermal Resistance	Value [k/W]	Time Constant	Value [s]
$R_1$	0.008	${\tau }_1$	$1.0\times {10}^{-4}$
$R_2$	0.020	${\tau }_2$	$5.0\times {10}^{-3}$
$R_3$	0.030	${\tau }_3$	$5.0\times {10}^{-2}$
$R_4$	0.032	${\tau }_4$	$5.0\times {10}^{-1}$

.

The transient thermal impedance between the junction and case is expressed as

$$Z_{\mathrm{t}\mathrm{h},\mathrm{J}\mathrm{C}}\left(t\right)={\sum }_{\mathrm{i}=1}^4{R}_{\mathrm{i}}\left(1-e^{-{\displaystyle \frac{t}{{\tau }_{\mathrm{i}}}}}\right), {\tau }_{\mathrm{i}}=R_{\mathrm{i}}{C}_{\mathrm{i}}$$

(15)

The junction temperature is then given by

$$T_{\mathrm{j}}\left(t\right)=T_{\mathrm{c}}+\left(Z_{\mathrm{t}\mathrm{h},\mathrm{J}\mathrm{C}}·P_{\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}}\right)\left(t\right)$$

(16)

The electrical characteristics of the IGBT used in the C-HILS experiments were obtained from the device datasheet and applied using interpolation. The characteristic curves used in this study included the saturation voltage–current relationship $V_{\mathrm{C}\mathrm{E}\left(\mathrm{s}\mathrm{a}\mathrm{t}\right)}-I_{\mathrm{C}}$ and the switching-energy curves $E_{\mathrm{o}\mathrm{n}},E_{\mathrm{o}\mathrm{f}\mathrm{f}}- I_{\mathrm{C}}$, based on a reference DC-link voltage of $V_{\mathrm{D}\mathrm{C}}=600$. The total switching-device loss was computed as the sum of conduction and switching losses. The characteristic curves are shown in Figure 11.

The Bayerer lifetime model was applied to the junction-temperature swing $\mathsf{\Delta }{T}_{\mathrm{j}}$ and the mean junction temperature $T_{\mathrm{mean}}$ for each thermal cycle, extracted using the rainflow counting method. Cumulative damage $D_{\mathrm{r}\mathrm{e}\mathrm{c}}$ for each cycle was calculated using Miner’s rule. The annual damage hours $H_{yr}$ and annual damage rate $D_{\mathrm{y}\mathrm{r}}$ were then obtained. The expected inspection and replacement interval $I$ for a specified cumulative damage threshold $D_{\mathrm{t}\mathrm{h}}$ is given by

$$I={\displaystyle \frac{D_{\mathrm{t}\mathrm{h}}}{D_{\mathrm{y}\mathrm{r}}}}[\mathit{year}]$$

(17)

In this study, the threshold was conservatively set to $D_{\mathrm{t}\mathrm{h}}=1$.

4. Experimental Validation and Results

Real inverter tests and C-HILS experiments were conducted under identical operating conditions to verify the applicability of the C-HILS approach and the reliability of the inverter model. The overall configuration and experimental setup for the comparative validation between the real inverter and the C-HILS system are shown in Figure 12. In both experiments, the same control board and control algorithm were used to ensure identical operating conditions. Current and voltage waveforms were measured for both the SVPWM and DPWM control strategies, which were generated by the same controller in each test.

Quantitative comparisons were performed over the same time interval $[t_1, t_2]$ by analyzing the RMS values of the three-phase currents and the FFT of the line-to-line voltages obtained from both experiments. The target error margin for the comparison was set to within $\pm 5\%$.

The resistance and inductance values used to emulate voltage, current, and motor characteristics under identical conditions are summarized in

Table 4. Experimental parameters common to both the real inverter and C-HILS tests.

Category	Value
$\mathrm{V}$	DC $100 \left[\mathrm{V}\right]$ Using DC Supply Equipment (Toyotech Co., Ltd., Yokohama, Japan)
$\mathrm{I}$	${\mathrm{I}}_{\mathrm{q}}$: $10 \left[\mathrm{A}\right]$
$\mathrm{R}$	$500 \left[\mathrm{W}\right]$, $1 \left[\mathsf{\Omega }\right]$
$\mathrm{L}$	3-Phase Reactor: $380 \left[\mathrm{V}\right]$, max 40$\left[\mathrm{A}\right]$, $7.32 \left[\mathrm{m}\mathrm{H}\right]$ (BEOMHAN CO., LTD., Anyang, Republic of Korea)

.

4.1. Quantitative Validation of C-HILS Through RMS and THD Analysis

Figure 13 compares the phase-current and line-to-line voltage waveforms from the real inverter experiment and the C-HILS experiment. The phase currents $i_{\mathrm{a}},i_{\mathrm{b}},i_{\mathrm{c}}$ and the line-to-line voltages exhibit close agreement in the time domain. The lower portion of the figure shows the FFT of the line-to-line voltage, demonstrating that the magnitudes of the fundamental and dominant harmonic components closely match between the two results.

Table 5. Comparison of three-phase current and line-to-line voltage errors by control method.

Control Method	${\mathit{\epsilon }}_{\mathbf{a}\mathbf{-}\mathbf{R}\mathbf{M}\mathbf{S}}$	${\mathit{\epsilon }}_{\mathbf{b}\mathbf{-}\mathbf{R}\mathbf{M}\mathbf{S}}$	${\mathit{\epsilon }}_{\mathbf{c}\mathbf{-}\mathbf{R}\mathbf{M}\mathbf{S}}$	$\mathbf{N}\mathbf{R}\mathbf{M}\mathbf{S}\mathbf{E}(\mathbf{\%})$
SVPWM	1.97	2.73	2.35	4.12
DPWM	1.97	4.12	4.63	4.08

provides a quantitative comparison between the C-HILS and real inverter results for both control strategies. The RMS error of the phase currents was approximately 2–5%, and the RMS and THD errors of the line-to-line voltages were within 2–4%. These results confirm that the C-HILS model accurately reproduces the operating behavior of the real inverter.

4.2. Time-Domain Waveform and Algorithm Analysis of SVPWM and DPWM

Both SVPWM and DPWM were implemented in the C-HILS platform under identical hardware configurations, and their time-domain waveforms were compared. The C-HILS results included the three-phase currents, $d$-$q$ axis currents, reference and common-mode voltages, and switching signals. Under steady-state conditions, the magnitude and phase of the three-phase currents remained balanced, and both $i_{\mathrm{d}}$ and $i_{\mathrm{q}}$ closely followed their reference values. During transient conditions, $i_{\mathrm{q}}$ converged rapidly to its reference, and the voltage and gate commands were generated as expected, verifying stable implementation of both control algorithms.

Figure 14a illustrates that SVPWM produces a modified reference voltage by adding a third-harmonic component to the original reference, resulting in continuous switching across the operating range. In contrast, Figure 14b shows that DPWM introduces an offset voltage to suppress switching during high-current intervals, resulting in discontinuous switching in specific regions.

Figure 15 compares the current THD for both control strategies. Figure 15a shows the SVPWM results, and Figure 15b shows those of DPWM. In both the real inverter and C-HILS experiments, SVPWM exhibited lower THD across all phases than DPWM, indicating lower current distortion. The C-HILS trends closely matched the experimental results, demonstrating that the simulation accurately replicated the system behavior.

4.3. Switching Loss and Efficiency Comparison

Figure 16 compares the average switching frequencies measured in the C-HILS and real inverter tests. Because DPWM employs discontinuous switching intervals, its average switching frequency is approximately 23% lower than that of SVPWM. The C-HILS results closely matched those of the real inverter.

Figure 17 compares the switching and conduction losses of the six switching devices (S1–S6) for each control strategy. Figure 17a shows the results for SVPWM, and Figure 16 shows the results for DPWM. Owing to its switching-suppression intervals, DPWM exhibits lower overall losses than SVPWM.

Figure 18 presents the overall inverter efficiency for both control strategies. The reduced switching losses of DPWM result in higher efficiency compared to SVPWM.

Together, these results confirm that the C-HILS experiments accurately reproduced the time-domain behavior of both control methods. Quantitative indicators—including THD and average switching frequency—also matched the real inverter results closely, validating the capability of the C-HILS system to support comparative performance analysis of inverter control strategies.

4.4. Parametric Trend Analysis for Sampling Frequency and Output Power

Figure 19 compares the output characteristics of SVPWM and DPWM under varying sampling frequencies. Figure 19a shows that current THD decreases as the sampling frequency increases, whereas Figure 18 indicates that the average switching frequency increases accordingly. Figure 19c shows that total power loss rises with higher sampling frequency. As shown in Figure 19d, under low output power conditions, the efficiency is relatively low due to the higher proportion of fixed losses associated with the switching devices and gate drive circuits. As the output power increases, the relative impact of these losses decreases, resulting in an improvement in overall efficiency. SVPWM delivers superior current quality with lower THD, while DPWM maintains a lower average switching frequency because of its discontinuous switching behavior, resulting in reduced losses and improved efficiency.

Figure 20 compares the output characteristics of both control strategies at different output power levels. Figure 20a shows that current THD decreases with increasing output power, while Figure 20b shows minimal variation in the average switching frequency. Figure 20c shows that total power loss increases at higher power, and Figure 20d shows that efficiency also increases. SVPWM showed higher losses and lower efficiency because of its higher switching frequency, whereas DPWM maintained lower losses and higher efficiency over the full output power range.

These findings confirm that the C-HILS approach is an effective tool for evaluating and comparing inverter control strategies for propulsion motors used in electric ships.

5. Lifetime Estimation

This section presents a quantitative estimation of the lifetimes of the inverter switching devices based on the current and voltage waveforms and the thermal responses obtained in the previous analysis. The Foster RC thermal network model was applied using the switching loss and thermal cycling characteristics as inputs, and the resulting junction temperature variation $\mathsf{\Delta }{T}_{\mathrm{j}}$ and estimated lifetime were compared for each control strategy.

5.1. Power-Cycling-Based Lifetime Prediction

The reference load condition consisted of a high-load interval of $1.66 \mathrm{s}$ at $25 \mathrm{k}\mathrm{W}$, and a low-load interval of $1.66 \mathrm{s}$ at 18 kW, yielding a total cycle time of $T_{\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{f}\mathrm{i}\mathrm{l}\mathrm{e}}=3.33 \mathrm{s}$. The estimated lifetimes were converted from seconds to years using the relation $1 year=3600\times 24\times 365$ s, and the shortest lifetime among all cycles was conservatively adopted. The variables used in the analysis are listed in

Table 6. Parameters used in the IGBT lifetime model.

Parameter	Value	Parameter	Value
$A$	$1.2\times {10}^{15}$	${\beta }_6$	$-0.50$
${\beta }_1$	$-4.70$	$t_{\mathrm{o}\mathrm{n}}$ [s]	$5$
${\beta }_2$	$1.30\times {10}^3$	$i_{\mathrm{B}}$ [A]	$10$
${\beta }_3$	$-0.50$	$V_{\mathrm{c}}$ [V]	$12$
${\beta }_4$	$-0.70$	$d_{\mathrm{b}}$ [µm]	$400$
${\beta }_5$	$-0.80$	$T_{\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{f}\mathrm{i}\mathrm{l}\mathrm{e}}$ [s]	$10$

. The parameters $A$ and ${\beta }_1$–${\beta }_6$ were assigned representative values for an IGBT aluminum wire-bond module and calibrated through regression based on the manufacturer’s power-cycling test data. $A$ and ${\beta }_1$–${\beta }_6$ are dimensionless model constants, while the remaining parameters represent operating and geometric conditions.

Figure 21 shows the C-HILS results for the junction temperature $T_{\mathrm{j}}$ of the phase-A device and the three-phase output current under load conditions ranging from 25 kW (maximum load) to 18 kW (minimum load). Simulations were carried out under identical operating conditions for SVPWM and DPWM. For SVPWM, $T_{\mathrm{j}}$ increased to approximately 100.3 °C during the high-load interval and decreased to approximately 74.8 °C during the low-load interval, giving a junction temperature variation of $\mathsf{\Delta }{T}_{\mathrm{j}}\approx 25.6$ °C. Under DPWM, the maximum and minimum junction temperatures were approximately 82.3 °C and 64.8 °C, respectively, yielding a reduced $\mathsf{\Delta }{T}_{\mathrm{j}}\approx 17.5$ °C.

These results were applied to a Bayesian lifetime model to estimate the expected lifetimes of the switching devices. The values of $T_{\mathrm{j},\mathrm{h}\mathrm{i}\mathrm{g}\mathrm{h}}$, $T_{\mathrm{j},\mathrm{l}\mathrm{o}\mathrm{w}}$, $\mathsf{\Delta }{T}_{\mathrm{j}}$, and the resulting lifetime estimates are summarized in

Table 7. Lifetime estimation for the predefined load cycle.

Control Method	${\mathit{T}}_{\mathbf{j}\mathbf{,}\mathbf{h}\mathbf{i}\mathbf{g}\mathbf{h}}$ [°C]	${\mathit{T}}_{\mathbf{j}\mathbf{,}\mathbf{l}\mathbf{o}\mathbf{w}}$ [°C]	$\mathsf{\Delta }{\mathit{T}}_{\mathbf{j}}$ [°C]	Lifetime Estimation [Year]
SVPWM	100.34	74.75	25.59	1.36
DPWM	82.26	64.78	17.48	9.14

. The estimated lifetime for SVPWM was approximately 1.36 years, whereas DPWM yielded a significantly longer lifetime of approximately 9.14 years—An improvement of about 6.7-fold. This improvement can be attributed to the reduced number of switching events and consequently lower average switching frequency $f_{\mathrm{s}\mathrm{w}}$ in DPWM, which decreases switching losses and reduces $\mathsf{\Delta }{T}_{\mathrm{j}}$.

These results demonstrate that the C-HILS framework facilitates quantitative estimation of thermal responses and device lifetime based on the calculated losses, validating its applicability for thermal reliability assessment of inverter systems.

5.2. Lifetime Estimation Under Marine Load Conditions

Figure 22 compares the SVPWM and DPWM strategies under identical conditions based on C-HILSs using a ship load profile. Figure 22A presents the three-phase currents, $i_{\mathrm{d}}$, propulsion motor speed, output power, and reference voltage for SVPWM; Figure 22B shows the corresponding DPWM results.

Both control strategies accurately followed the commanded speed and reflected the step changes in output power. Under low-speed and low-load conditions, the amplitudes of the phase currents and $i_{\mathrm{d}}$ increased, and ripple characteristics differed, indicating the influence of the modulation strategy. The reference voltage waveforms also clearly show the modulation behavior of each method, confirming that spikes and ripples during transient intervals produced differences in power losses and junction temperature responses.

Figure 23 shows the power-cycling test results obtained by periodically applying a maximum power of 46 kW and an average power of 14.6 kW. The junction temperature variation $\mathsf{\Delta }{T}_{\mathrm{j}}$ increased stepwise with load transitions, and DPWM consistently exhibited lower $\mathsf{\Delta }{T}_{\mathrm{j}}$ than SVPWM under the same conditions. These results indicate that DPWM can provide a longer device lifetime than SVPWM in marine applications.

The values of $T_{\mathrm{j},\mathrm{h}\mathrm{i}\mathrm{g}\mathrm{h}}$, $T_{\mathrm{j},\mathrm{l}\mathrm{o}\mathrm{w}}$, $\mathsf{\Delta }{T}_{\mathrm{j}}$, and lifetime estimates are summarized in

Table 8. Comparison of lifetimes estimated with SVPWM and DPWM under the ship load profile.

Control Method	${\mathit{T}}_{\mathbf{j}\mathbf{,}\mathbf{h}\mathbf{i}\mathbf{g}\mathbf{h}}$ [°C]	${\mathit{T}}_{\mathbf{j}\mathbf{,}\mathbf{l}\mathbf{o}\mathbf{w}}$ [°C]	$\mathsf{\Delta }{\mathit{T}}_{\mathbf{j}}$ [°C]	Lifetime Estimation [Year]
SVPWM	97.68	70.21	27.47	1.02
DPWM	76.81	65.72	11.09	56.43

.

5.3. Uncertainty and Sensitivity Analysis of the Lifetime Prediction Model

The uncertainty and sensitivity of the proposed C-HILS–loss–thermal–lifetime integrated framework were evaluated quantitatively. Figure 24a shows the lifetime distribution obtained from a Monte Carlo simulation that accounts for coefficient uncertainty in the Bayesian lifetime model. Assuming an annual operating time of 1000 h, the median lifetime for SVPWM was approximately 0.17 years with a 95% confidence interval of 0.06–0.42 years. DPWM exhibited a median lifetime of approximately 0.15 years and a 95% confidence interval of 0.06–0.37 years. In this context, the values of 0.17 years and 0.15 years do not represent the actual service lifetime of the components; rather, they indicate the amount of lifetime consumed per year under the given thermal stress conditions. In other words, a calculated lifetime value of 0.15 years implies that approximately 15% of the thermal fatigue lifetime is consumed under the corresponding operating conditions. Because the confidence intervals for both strategies overlapped significantly, their relative superiority cannot be determined solely based on absolute lifetime values. This outcome indicates that the Bayerer lifetime model is influenced not only by $\mathsf{\Delta }{T}_{\mathrm{j}}$ but also by the mean junction temperature $T_{\mathrm{mean}}$ and the model coefficients. Consequently, the relative lifetime rankings may vary depending on the cooling or operating conditions.

In this case, although DPWM resulted in a lower $T_{\mathrm{mean}}$, its median lifetime was slightly shorter than that of SVPWM owing to differences in thermal cycle frequency and parameter uncertainty, yielding an inspection and replacement interval of $I=0.5$ years.

Figure 24b presents tornado sensitivity analysis results. For both strategies, the model scale factor $a$, reference temperature $T_{\mathrm{ref}}$, and activation energy $E_{\mathrm{a}}$ were the most influential parameters, whereas the exponent $b$ had negligible impact under the analyzed conditions. These findings indicate that performance evaluation of inverter control strategies must consider not only $\mathsf{\Delta }{T}_{\mathrm{j}}$ but also the thermal cycle frequency, dwell time, and the uncertainty of model parameters. This analysis confirms the effectiveness of the proposed framework for integrated performance–loss–thermal–lifetime analysis and optimization.

Table 9. Summary of maintenance indicators based on the C-HILS–Bayerer–Miner framework.

Control Method	$\mathbf{Annual} \mathbf{Damage} \mathbf{Rate} \mathit{X}$ [Damage/Year]	$\mathbf{Maintenance} \mathbf{Interval} \mathit{I}$ [Year, Policy-Based]	Lifetime (Monte Carlo Median) [95% CI]	$\mathbf{Mean} \mathbf{Junction} \mathbf{Temperature} {\mathit{T}}_{\mathbf{j}}$ [°C]
SVPWM	6.008	0.5	0.17 [0.06–0.41]	83.95
DPWM	6.932	0.5	0.14 [0.06–0.36]	71.27

summarizes the annual damage rate $X$, expected inspection and replacement interval $I$, median lifetime, 95% confidence interval, and mean junction temperature $T_{\mathrm{j}}$ for each control strategy under the representative operating profile.

6. Results and Discussion

This study conducted a parallel validation of experimental measurements and C-HILSs for a tens-of-kilowatt marine propulsion inverter. Through this process, the reliability of the C-HILS model was verified, and the current-tracking performance, waveform quality, average switching frequency, losses, and efficiency of the SVPWM and DPWM control strategies were systematically compared under identical conditions. The validation results showed that the RMS errors of the phase currents and the FFT/THD errors of the line-to-line voltages between the C-HILS and experimental measurements were generally within 2–5%, confirming that the C-HILS model accurately reproduced the quantitative behavior of the real system.

Using the junction temperature $T_{\mathrm{j}}\left(t\right)$ obtained from the C-HILSs as input to the Foster thermal model and the Bayerer lifetime model, it was confirmed that DPWM reduced the junction temperature fluctuation $\mathsf{\Delta }{T}_{\mathrm{j}}$ from approximately $25.6 °\mathrm{C}$ to $17.5 °\mathrm{C}$. This reduction resulted from decreased losses associated with the lower average switching frequency $f_{\mathrm{s}\mathrm{w}}$, leading to an estimated lifetime improvement of approximately 6.7-fold. When the proposed framework was applied to a representative ship operating load profile, the C-HILS results accurately reproduced the current and voltage waveforms as well as the thermal response trends observed in the experiments. These findings demonstrated consistent correlations among the control strategy, losses, thermal response, and device lifetime across a range of operating conditions.

Furthermore, the Monte Carlo-based lifetime uncertainty analysis and the tornado sensitivity analysis showed that the dominant factors affecting lifetime prediction included not only $\mathsf{\Delta }{T}_{\mathrm{j}}$ but also the model scale factor $a$, reference temperature $T_{\mathrm{ref}}$, and activation energy $E_a$. These results highlight that appropriate configuration of the operating conditions and thermal boundary design is critical for improving the reliability of lifetime predictions and enhancing thermal robustness.

In summary, SVPWM is more suitable in applications where current quality and torque ripple minimization are prioritized, whereas DPWM is preferable when reducing switching losses, improving efficiency, and extending thermal lifetime are the primary objectives. Overall, the proposed C-HILS-based integrated framework was shown to be an effective and practical analytical tool for consistently comparing, predicting, and optimizing the performance–loss–thermal–lifetime characteristics of power converters. Future work will examine the applicability of the proposed model to megawatt-class marine propulsion systems and analyze the limitations of the thermal model and lifetime prediction associated with increasing power ratings.