Breakthrough in Real-Time Thermal Stress Monitoring for EV Power Modules
As electric vehicles (EVs) continue to gain momentum across global markets, the demand for enhanced reliability and longevity of core components has never been higher. Among the most critical yet vulnerable parts in an EV’s powertrain is the power module, particularly the insulated-gate bipolar transistor (IGBT) or silicon carbide (SiC) MOSFET-based units responsible for high-efficiency power conversion. These modules operate under extreme thermal cycling conditions, where repeated heating and cooling lead to material fatigue, micro-cracking, and eventual failure. Traditionally, assessing the health and remaining lifespan of these modules has relied on offline data analysis, a method that, while accurate, fails to deliver real-time insights necessary for predictive maintenance and dynamic reliability management.
Now, a groundbreaking study conducted by researchers from Zhejiang University and Viridi E-Mobility Technology (Ningbo) Co., Ltd. has introduced a high-precision, real-time method for extracting thermal fatigue stress data directly from vehicle-mounted power modules. Published in the prestigious Proceedings of the CSEE, the research led by Xiang Enyao, Lu Yiping, Luo Haoze, Yang Huan, Wang Haibing, and Ruan Ou presents a novel online algorithm that overcomes the limitations of conventional offline rainflow counting methods, enabling continuous, on-the-fly assessment of thermal stress cycles without compromising accuracy or computational efficiency.
The significance of this advancement cannot be overstated. In the fast-evolving world of electric mobility, where vehicle performance, safety, and lifecycle costs are paramount, the ability to monitor and predict the health of power electronics in real time represents a major leap forward. Unlike mechanical components such as engines or transmissions, semiconductor-based power modules degrade silently and cumulatively. Their failure is often sudden and catastrophic, leading to vehicle immobilization and expensive repairs. Preventing such failures requires a deep understanding of the thermal stress history these modules endure during daily operation—data that has historically been difficult to capture and analyze in real time.
The research team’s innovation lies in a reimagined approach to the rainflow counting algorithm, a well-established method in fatigue analysis used to convert complex, variable-amplitude load histories into equivalent stress cycles. While the standard offline version requires complete load data and significant computational resources, making it unsuitable for embedded systems in vehicles, the new method introduces a series of intelligent optimizations tailored for real-time execution on resource-constrained automotive microcontrollers.
One of the key features of the proposed algorithm is its use of a variable filter window width that adapts based on the material properties and life stage of the power module. This dynamic filtering allows the system to distinguish between meaningful thermal stress cycles and insignificant temperature fluctuations or measurement noise. By setting different thresholds for crack initiation, linear propagation, and rapid crack growth phases—determined through thermal resistance monitoring—the algorithm effectively filters out low-amplitude cycles that contribute negligibly to fatigue damage. This not only improves accuracy but also reduces unnecessary data processing, a crucial factor for real-time applications.
The core of the method is an improved real-time rainflow counting algorithm based on a “three-point double-variable range” principle. Instead of processing the entire temperature history at once, the algorithm continuously evaluates the latest three extreme points (peaks and valleys) in the junction temperature data stream. By comparing the amplitude differences between consecutive points, it can instantly identify full and half stress cycles, updating the damage accumulation in real time. This approach drastically reduces computational load while maintaining a level of accuracy that closely matches the gold-standard offline method.
To further enhance practicality, the researchers implemented a discrete normalization process that clusters and standardizes the counted stress cycles. Rather than storing every individual cycle with its precise amplitude and mean temperature, the algorithm maps each cycle into predefined bins based on amplitude and temperature ranges. This discretization significantly reduces memory usage and data redundancy, making it feasible to store long-term stress history on the vehicle’s embedded controller. The result is a lightweight, efficient system capable of running continuously without overwhelming the vehicle’s computational resources.
The feasibility and accuracy of the proposed method were rigorously tested through both simulation and real-world experiments. In simulation, the algorithm was applied to a complex junction temperature profile spanning 100 seconds with rapid fluctuations between 25°C and 120°C. The results showed perfect agreement with the standard offline rainflow counting method, capturing all full and half cycles without omission. In contrast, two existing real-time methods from prior literature missed one and two half-cycles, respectively, highlighting the superior precision of the new approach.
More compelling was the real-world validation conducted on a passenger electric vehicle equipped with the algorithm running on its onboard microcontroller. Over a 26-kilometer test route encompassing urban streets, highways, and mountainous terrain, the system continuously processed junction temperature data sampled via the module’s built-in NTC (negative temperature coefficient) thermistor. The real-time damage increment was monitored and compared against post-drive offline analysis. The results were striking: the cumulative damage calculated by the online method deviated by less than 1% from the offline benchmark, demonstrating exceptional fidelity under actual driving conditions.
The experiment also revealed important insights into how different driving patterns affect thermal stress. For instance, uphill and downhill segments showed sharp spikes in damage accumulation, corresponding to rapid temperature changes during acceleration and regenerative braking. Highway driving, despite higher speeds, generated less thermal stress due to more stable operating conditions. Conversely, city driving, with its frequent stops and starts, produced a steady accumulation of small but damaging temperature cycles. These findings underscore the value of real-time monitoring in understanding the true impact of driving behavior on component lifespan.
A critical aspect of the study was the investigation of temperature sampling frequency on damage calculation accuracy. The team tested sampling periods ranging from 20 milliseconds to 5 seconds. They found that sampling intervals between 20 ms and 500 ms had minimal impact on accuracy, with errors remaining below 1%. However, beyond 1 second, the error increased rapidly, reaching nearly 60% at 5 seconds. This sensitivity arises because longer sampling intervals can miss rapid temperature transients, leading to undercounting of high-frequency stress cycles. The conclusion: a sampling period of 500 ms or shorter strikes the optimal balance between computational efficiency and measurement accuracy, making it ideal for deployment in production vehicles.
The implications of this research extend far beyond academic interest. For automakers and tier-one suppliers, integrating such a real-time health monitoring system into electric drivetrains opens new possibilities for predictive maintenance, warranty optimization, and customer service. Imagine a vehicle that can alert the driver or service center when a power module is approaching end-of-life, allowing for proactive replacement before failure occurs. Or a fleet operator using aggregated stress data to optimize driving patterns and extend the lifespan of their vehicles. The technology could also feed into digital twin models, enabling virtual testing and lifecycle simulation based on real-world usage data.
Moreover, the method’s compatibility with existing vehicle hardware is a major advantage. It leverages the NTC sensors already present in most power modules, eliminating the need for costly additional instrumentation. The algorithm’s low computational footprint means it can run on standard automotive microcontrollers without requiring high-performance processors or external computing units. This makes it a highly scalable solution, suitable for mass-market EVs as well as high-performance models.
From a broader industry perspective, this work addresses a critical gap in the current state of EV reliability engineering. While much attention has been paid to battery management systems, the power electronics that control energy flow between the battery, motor, and grid have received comparatively less focus in terms of real-time health monitoring. This research helps to level the playing field, bringing the same level of sophistication to power module diagnostics as exists for battery state-of-charge and state-of-health estimation.
The success of this project is a testament to the growing synergy between academic research and industrial application. Zhejiang University, a leading institution in power electronics and control systems, collaborated closely with Viridi E-Mobility Technology, a company deeply embedded in the EV supply chain. This partnership ensured that the theoretical innovations were grounded in real-world engineering constraints and practical deployment scenarios. The support from the Zhejiang Province “Pioneer” and “Leading Goose” R&D Program and the Ningbo City Major Scientific and Technological Tasks Research Project further highlights the strategic importance of this work in China’s push for technological leadership in the EV sector.
Looking ahead, the research team envisions several directions for future development. One is the integration of the stress data with advanced lifetime prediction models that account for material aging, environmental factors, and manufacturing variability. Another is the extension of the method to other types of power electronic systems, such as onboard chargers, DC-DC converters, and renewable energy inverters. Additionally, the possibility of cloud-based aggregation and analysis of stress data from entire vehicle fleets could enable large-scale reliability studies and continuous improvement of component design.
In conclusion, the work by Xiang Enyao, Lu Yiping, Luo Haoze, Yang Huan, Wang Haibing, and Ruan Ou represents a significant milestone in the evolution of electric vehicle technology. By transforming a traditionally offline, post-mortem analysis technique into a real-time, embedded monitoring system, they have opened a new frontier in vehicle health management. Their method not only enhances the reliability and safety of EVs but also contributes to the overall sustainability of the transportation sector by extending component lifespans and reducing waste. As the automotive industry continues its transition to electrification, innovations like this will be essential in building the intelligent, resilient, and user-friendly vehicles of the future.
The study, published in the Proceedings of the CSEE, is a prime example of how cutting-edge research can directly address real-world engineering challenges. With its high accuracy, low computational cost, and seamless integration into existing vehicle architectures, the proposed method is poised to become a standard feature in next-generation electric drivetrains. As more automakers recognize the value of real-time component health monitoring, this technology could soon become as ubiquitous as anti-lock braking or electronic stability control—silent guardians of vehicle reliability, operating behind the scenes to ensure a safer, more efficient driving experience.
Xiang Enyao, Lu Yiping, Luo Haoze, Yang Huan, Wang Haibing, Ruan Ou, Zhejiang University and Viridi E-Mobility Technology (Ningbo) Co., Ltd., Proceedings of the CSEE, DOI: 10.13334/j.0258-8013.pcsee.231914