在功率電子與預測性維護領域中，IGBT等功率半導體的劣化預測是確保系統可靠性與運行安全的核心課題。隨著電動車、航空電子系統以及AI資料中心的快速發展，高功率密度與高運算需求使功率半導體元件長時間運行於高溫、高電流與頻繁負載變動的環境中，其健康狀態將直接影響系統能效、設備壽命與整體運行穩定性。因此，建立準確且可信賴的劣化預測模型已成為智慧能源與先進運算基礎設施的重要研究方向。然而，傳統深度學習模型多依賴純數據驅動方法，缺乏對物理規律的約束，導致模型可能產生不符合實際退化機制的預測結果，進而限制其在高安全性與高可靠性應用場景中的部署，例如電動車、航空系統及AI資料中心等關鍵基礎設施。此外，現有模型在特徵感知與融合能力方面仍存在不足，難以有效捕捉退化信號中的多尺度非線性特徵，進而影響剩餘壽命（Remaining Useful Life, RUL）預測的準確性與泛化能力。

為解決上述問題，本研究提出一個結合生物感知邏輯與物理資訊的混合式深度學習架構，旨在實現對功率半導體剩餘壽命的高精度與高穩定性預測。模型首先透過史蒂文斯冪定律（Stevens’ Power Law）與韋伯–費希納定律（Weber–Fechner Law）構建門控感知層，以模擬人類對弱刺激的高靈敏度與對強刺激的壓縮反應，提升早期劣化徵兆的檢測能力與梯度穩定性。接著，利用多尺度正弦位置編碼（Multi-Scale Sinusoidal Encoding）與傅立葉神經運算元（Fourier Neural Operator, FNO）克服頻率偏置問題，捕捉退化過程中的高頻動態與長程依賴特性。為進一步強化特徵交互，本研究引入具可學習權重的雙線性融合層（Learnable Bilinear Fusion Layer），以二階乘法交互形式實現特徵層級的非線性耦合。

在訓練階段，模型被嵌入物理資訊神經網路（Physics-Informed Neural Network, PINN）框架中，藉由單調性與邊界條件約束確保物理一致性。

本研究的貢獻包括：（1）提出結合心理物理學特性的門控感知層以穩定梯度與強化早期退化辨識；（2）融合多尺度頻譜學習與雙線性交互以提升特徵表達能力；（3）透過PINN實現物理一致性的RUL建模，證實其在工業預測性維護中的應用潛力。

In the field of power electronics and predictive maintenance, degradation prediction of power semiconductor devices such as IGBTs is crucial for ensuring system reliability and operational safety. However, traditional deep learning models often rely solely on data-driven methods without incorporating physical constraints, resulting in predictions that may deviate from real degradation behaviors. This limitation restricts their applicability in high-safety industries such as electric vehicles and aerospace systems. Furthermore, existing models exhibit insufficient feature perception and fusion capability, making it difficult to effectively capture multi-scale nonlinear degradation features.

To address these issues, this study proposes a hybrid deep learning architecture that integrates biological perceptual logic and physical information to achieve highly accurate and stable Remaining Useful Life (RUL) prediction for power semiconductors. The model introduces a gated perceptual layer inspired by Stevens’ Power Law and Weber–Fechner Law, simulating human sensory sensitivity to weak stimuli and compressive response to strong stimuli. This design enhances early-stage degradation detection and stabilizes gradient propagation. Subsequently, Multi-Scale Sinusoidal Encoding and the Fourier Neural Operator (FNO) are employed to overcome spectral bias, enabling the model to capture both high-frequency dynamics and long-range dependencies within degradation signals. To further enhance feature interaction, a Learnable Bilinear Fusion Layer is introduced, which captures second-order multiplicative interactions between features for expressive nonlinear coupling at minimal parameter cost.

During the training phase, the proposed network is embedded within a Physics-Informed Neural Network (PINN) framework, incorporating monotonicity and boundary constraints to ensure physical consistency.

The contributions of this work include: (1) proposing a biologically inspired gated perceptual layer that enhances gradient stability and early degradation sensitivity; (2) integrating multi-scale spectral learning with bilinear feature interactions for improved representational capacity; and (3) ensuring physically consistent RUL modeling through PINN, demonstrating the model’s practical potential in industrial predictive maintenance.

Abuelnaga, A., Narimani, M., & Bahman, A. S. (2021). A review on IGBT module failure modes and lifetime testing. IEEE Access, 9, 9643–9663.https://doi.org/10.1109/ACCESS.2021.3049738
Anderson, T. L. (2017). Fracture mechanics: Fundamentals and applications (4th ed.).CRC Press. https://doi.org/10.1201/9781315370293
Cai, B., Liu, Y., Zio, E., & Tian, Z. (2016). Remaining useful lifetime estimation for thermally stressed power MOSFETs based on on-state resistance variation. IEEE Transactions on Industry Applications,52(3), 2554–2563.https://doi.org/10.1109/TIA.2016.2518127
Celaya, J. R., Saxena, A., Saha, S., & Goebel, K. F. (2011). Prognostics of power MOSFETs under thermal stress accelerated aging using data-driven and model-based methodologies [Paper presentation]. Annual Conference of the Prognostics and Health Management Society 2011, Montréal, QC, Canada.https://papers.phmsociety.org/index.php/phmconf/article/view/1749
Chandra, A., Koch, M., Pawar, S., Panda, A., Azizzadenesheli, K., Snippe, J., Alpak, F. O., Hariri, F., Etienam, C., Devarakota, P., Anandkumar, A., & Hohl, D. (2025). Fourier Neural Operator based surrogates for CO2 storage in realistic geologies. arXiv. https://doi.org/10.48550/arXiv.2503.11031
Chen, S., Zhang, J., Wang, S., Wen, P. & Zhao, S.(2022)Circuit parameter identification of degrading DC-DC converters based on physics informed neural network. In 2022 Prognostics and Health Management Conference (PHM-2022 London) 260–268. https://doi.org/10.1109/PHM20 22-Londo n52454.2022.00053
Gao, Y., Beijbom, O., Zhang, N., & Darrell, T. (2016). Compact bilinear pooling. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 317–326. https://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Gao_Compact_Bilinear_Pooling_CVPR_2016_paper.pdf
He, C., Yu, W., Zheng, Y. & Gong, W. (2021). Machine learning based prognostics for predicting remaining useful life of IGBT: NASA IGBT accelerated ageing case study. In 2021 IEEE 5th Information Technology, Networking, Electronic and Automation Control Conference (ITNEC), vol. 5, 1357–1361.https://doi.org/10.1109/ITNEC52019.2021.9587236
Noorizadegan, A., Hon, Y. C., Young, D. L., & Chen, C. S. (2024). Highway networks for improved surface reconstruction: The role of residuals and weight updates. arXiv. https://doi.org/10.48550/arXiv.2407.08134
Huang, W., Zhou, T., Ma, J., & Chen, X. (2024). An ensemble model based on fusion of multiple machine learning algorithms for remaining useful life prediction of lithium battery in electric vehicles. Innovations in Applied Engineering and Technology, 4(1), 12–25. https://doi.org/10.62836/iaet.v4i1.319
Hubel, D. H., & Wiesel, T. N. (1959). Receptive fields of single neurones in the cat’s striate cortex. The Journal of Physiology, 148, 574–591.https://pmc.ncbi.nlm.nih.gov/articles/PMC1363130/pdf/jphysiol01298-0128.pdf
Kahramani, C. L., Roman, D., Kirschbaum, L., Flynn, D., & Swingler, J. (2024). Machine learning pipeline for power electronics state of health assessment and remaining useful life prediction. IEEE Access, 12, 136733–136746. https://doi.org/10.1109/ACCESS.2024.3460177
Kandel, E. R., Markram, H., Matthews, P. M., Yuste, R., & Koch, C. (2013). Neuroscience thinks big (and collaboratively). Nature Reviews Neuroscience, 14, 659–664. https://doi.org/10.1038/nrn3578
Kendall, A., Gal, Y., & Cipolla, R. (2018). Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 7482–7491.http://openaccess.thecvf.com/content_cvpr_2018/papers/Kendall_Multi-Task_Learning_Using_CVPR_2018_paper.pdf
Krishna, G., Singh, R., Gehlot, A., Almogren, A., Altameem, A., Rehman, A. U., & Hussen, S. (2024). Advanced battery management system enhancement using IoT and ML for predicting remaining useful life in Li-ion batteries. Scientific Reports, 14, 30394. https://doi.org/10.1038/s41598-024-80719-1
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444. https://doi.org/10.1038/nature14539
Lehmann, F., Gatti, F., & Clouteau, D. (2024). Multiple-Input Fourier Neural Operator (MIFNO) for source-dependent 3D elastodynamics [Preprint]. arXiv. https://arxiv.org/abs/2404.10115
Lei, Y., Li, N., Guo, L., Li, N., Yan, T., & Lin, J. (2018). Machinery health prognostics: A systematic review from data acquisition to RUL prediction. Mechanical Systems and Signal Processing, 104, 799–834.https://doi.org/10.1016/j.ymssp.2017.11.016
Li, Z., Kovachki, N., Azizzadenesheli, K., Liu, B., Bhattacharya, K., Stuart, A., & Anandkumar, A. (2020). Fourier neural operator for parametric partial differential equations. arXiv. https://doi.org/10.48550/arXiv.2010.08895
Lu, Y., & Christou, A. (2019). Prognostics of IGBT modules based on the approach of particle filtering. Microelectronics Reliability, 92, 96–105. https://doi.org/10.1016/j.microrel.2018.11.012
Lu, Z., Guo, C., Liu, M., & Shi, R. (2023). Remaining useful lifetime estimation for discrete power electronic devices using physics-informed neural network. Scientific Reports, 13, 10167. https://doi.org/10.1038/s41598-023-37154-5
Mildenhall, B., Srinivasan, P. P., Tancik, M., Barron, J. T., Ramamoorthi, R., & Ng, R. (2020). NeRF: Representing scenes as neural radiance fields for view synthesis. arXiv. https://arxiv.org/abs/2003.08934v2
Mountcastle, V. B. (1957). Modality and topographic properties of single neurons of cat’s somatic sensory cortex. Journal of Neurophysiology, 20, 408–434.https://journals.physiology.org/doi/pdf/10.1152/jn.1957.20.4.408
Paris, P., & Erdogan, F. (1963). A critical analysis of crack propagation laws. Journal of Basic Engineering, 85(4), 528-534. https://doi.org/10.1115/1.3656900
Połap, D., Jaszcz, A., Wawrzyniak, N., & Zaniewicz, G. (2023). Bilinear pooling with poisoning detection module for automatic side scan sonar data analysis. IEEE Access, 11, 72477–72484. https://doi.org/10.1109/ACCESS.2023.3295693
Raissi, M., Perdikaris, P., & Karniadakis, G. E. (2018). Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. arXiv.https://arxiv.org/abs/1711.10561v1
Shigemoto, K. (2003). Weber-Fechner's law and demand function. arXiv:physics /0303118v2. https://arxiv.org/abs/physics/0303118v2
Srivastava, R. K., Greff, K., & Schmidhuber, J. (2015). Highway networks (arXiv:1505.00387v2). arXiv. https://arxiv.org/abs/1505.00387
Stevens, S. S. (1957). On the psychophysical law. Psychological Review, 64(3), 153–181. https://doi.org/10.1037/h0046162
Torres, E., Schiefer, J., & Niepert, M. (2025). Adaptive physics-informed neural networks. arXiv. https://doi.org/10.48550/arXiv.2503.18181
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30.https://proceedings.neurips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf
Vo, H.-T., Nguyen Thien, N., & Chau Mui, K. (2023). Tomato disease recognition:Advancing accuracy through Xception and bilinear pooling fusion. International Journal of Advanced Computer Science and Applications, 14(8), 1–11.https://dx.doi.org/10.14569/IJACSA.2023.01408113
Wang, C., Jiang, W., Shi, L., & Zhang, L. (2023). Rolling bearing remaining useful life prediction using deep learning based on high-quality representation. Scientific Reports, 13, 21675. https://doi.org/10.1038/s41598-025-93165-4
Wang, Y., Li, Z., Yuan, Z., Peng, W., Liu, T., & Wang, J. (2024). Prediction of turbulent channel flow using Fourier neural operator-based machine-learning strategy. arXiv. https://doi.org/10.48550/arXiv.2403.03051
Wang, G., Pan, L., Peng, S., Liu, S., Xu, C., Miao, Y., Zhan, W., Tomizuka, M., Pollefeys, M., & Wang, H. (2024). NeRFs in robotics: A survey. arXiv.https://doi.org/10.48550/arXiv.2405.01333
Xiong, Y., Cheng, X., Shen, Z., Mi, C., Wu, H., & Garg, V. (2008). Prognostic and warning system for power-electronic modules in electric, hybrid electric, and fuel-cell vehicles. IEEE Transactions on Industrial Electronics, 55(6), 2268–2276. https://doi.org/10.1109/TIE.2008.918399
Xu, G., Wang, X., Wu, X., Leng, X., & Xu, Y. (2024). Development of skip connection in deep neural networks for computer vision and medical image analysis: A survey. arXiv. https://doi.org/10.48550/arXiv.2405.01725
Zhang, S., Lu, J., & Zhao, H. (2024). Deep network approximation: Beyond ReLU to diverse activation functions. Journal of Machine Learning Research, 25(1), 1–39. http://jmlr.org/papers/v25/23-0912.html
Zhang, Z., Zhang, S., Xiong, X., Zhao, Y., & Yang, X. (2025). Physics-Informed Neural Networks and Neural Operators for Parametric PDEs: A Human-AI Collaborative Analysis. arXiv. https://doi.org/10.48550/arXiv.2511.04576
Zhou, H., Du, J., Zhang, Y., Wang, Q., Liu, Q.-F., & Lee, C.-H. (2021). Information fusion in attention networks using adaptive and multi-level factorized bilinear pooling for audio-visual emotion recognition. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 29, 2617–2629. https://doi.org/10.1109/TASLP.2021.3096037
Zhou, Q., Maximov, M., Litny, O., & Leal-Taixé, L. (2024). The NeRFect Match: Exploring NeRF Features for Visual Localization. arXiv. https://doi.org/10.48550/arXiv.2403.09577