現行啟動馬達品質檢測多仰賴人工聽診與動力計量測進行合格與不合格判定，難以進一步辨識缺陷類型與產品品質差異。因此，本研究提出一套結合生成式AI與數位孿生觀測器之啟動馬達多級品質檢測架構。
本研究建置資料擷取平台，量測電壓、電流、轉速與轉矩訊號，並利用數位孿生模型生成模擬訊號作為預訓練資料，結合Transformer Autoencoder建立品質篩選模型，以實現良品與非良品之判別。其次，將多通道時序訊號轉換為時頻圖，並採用擴散模型（Diffusion Model）進行資料擴增，再利用視覺轉換器（Vision Transformer）進行缺陷類型識別。最後，導入數位孿生觀測器估測健康指標，建立良品品質分級機制。
實驗結果顯示，所提出之架構能有效完成品質篩選、缺陷類型識別與良品品質分級，並提供具物理意義之品質評估結果，可作為再製產業導入智慧品檢系統之參考。

Current starter motor quality inspection methods primarily rely on manual auditory inspection and dynamometer measurements to make pass/fail judgments, making it difficult to identify specific defect types or distinguish product quality differences. Therefore, this study proposes a multi-grade quality inspection framework for starter motors based on generative AI and digital twin observers.
A data acquisition platform was developed to collect voltage, current, speed, and torque signals. A digital twin model was employed to generate simulated signals as pre-training data, and a transformer autoencoder was adopted to develop a quality screening model that distinguishes qualified from non-qualified products. Subsequently, multi-channel time-series signals were transformed into time-frequency representations; the representations were augmented using a diffusion model and classified by a vision transformer for defect type identification. Finally, digital twin observers were introduced to estimate health related parameters and establish a quality grading mechanism for qualified products.
Experimental results demonstrate that the proposed framework can effectively accomplish quality screening, defect type identification, and quality grading, while providing physically meaningful quality assessment results. The proposed approach shows potential for intelligent quality inspection in the remanufacturing industry.

[1] S. Bobba, P. Tecchio, F. Ardente, F. Mathieux, F. M. dos Santos, and F. Pekar, "Analysing the contribution of automotive remanufacturing to the circularity of materials," Procedia CIRP, vol. 90, pp. 67-72, 2020, doi: 10.1016/j.procir.2020.02.052.
[2] J. W. Sutherland, D. P. Adler, K. R. Haapala, and V. Kumar, “A comparison of manufacturing and remanufacturing energy intensities with application to diesel engine production,” CIRP Ann. Manuf. Technol., vol. 57, no. 1, pp. 5–8, 2008, doi: 10.1016/j.cirp.2008.03.004.
[3] R. Issa, G. Clerc, M. Hologne-Carpentier, R. Michaud, E. Lorca, C. Magnette, and A. Messadi, “Review of fault diagnosis methods for induction machines in railway traction applications,” Energies, vol. 17, no. 11, Art. no. 2728, Jun. 2024, doi: 10.3390/en17112728.
[4] A. Choudhary, D. Goyal, and S. S. Letha, “Infrared thermography-based fault diagnosis of induction motor bearings using machine learning,” IEEE Sens. J., vol. 21, no. 2, pp. 1727–1734, Jan. 2021, doi: 10.1109/JSEN.2020.3015868.
[5] D. Verstraete, A. Ferrada, E. López Droguett, V. Meruane, and M. Modarres, “Deep learning enabled fault diagnosis using time-frequency image analysis of rolling element bearings,” Shock Vib., vol. 2017, Art. no. 5067651, pp. 1–17, Oct. 2017, doi: 10.1155/2017/5067651.
[6] A. Widodo and B.-S. Yang, “Support vector machine in machine condition monitoring and fault diagnosis,” Mech. Syst. Signal Process., vol. 21, no. 6, pp. 2560–2574, Aug. 2007, doi: 10.1016/j.ymssp.2006.12.007.
[7] Z. Cui, X. Kong, and P. Hao, “Few-shot learning for rolling bearing fault diagnosis based on residual convolutional neural network,” in Proc. 4th Int. Conf. Artificial Intelligence and Big Data (ICAIBD), Chengdu, China, 2021, pp. 320–324, doi: 10.1109/ICAIBD51990.2021.9459024.
[8] M. Sakurada and T. Yairi, “Anomaly detection using autoencoders with nonlinear dimensionality reduction,” in Proc. Workshop on Machine Learning for Sensory Data Analysis (MLSDA), Gold Coast, Australia, Dec. 2014, pp. 4–11, doi: 10.1145/2689746.2689747.
[9] Q. Wen, L. Sun, F. Yang, X. Song, J. Gao, X. Wang, and H. Xu, “Time series data augmentation for deep learning: A survey,” in Proc. 30th Int. Joint Conf. Artificial Intelligence (IJCAI), Montreal, QC, Canada, Aug. 2021, pp. 4653–4660, doi: 10.24963/ijcai.2021/631.
[10] I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial nets,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), Montreal, QC, Canada, 2014, pp. 2672–2680.
[11] J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), Vancouver, BC, Canada, 2020, pp. 6840–6851.
[12] L. Ma, Y. Ding, Z. Wang, C. Wang, J. Ma, and C. Lu, “An interpretable data augmentation scheme for machine fault diagnosis based on a sparsity-constrained generative adversarial network,” Expert Syst. Appl., vol. 182, Art. no. 115234, Nov. 2021, doi: 10.1016/j.eswa.2021.115234.
[13] H. Chen, J. Wei, H. Huang, Y. Yuan, and J. Wang, “Review of imbalanced fault diagnosis technology based on generative adversarial networks,” J. Comput. Des. Eng., vol. 11, no. 5, pp. 99–124, Oct. 2024, doi: 10.1093/jcde/qwae075.
[14] C. Little, M. Elliot, R. Allmendinger, and S. S. Samani, “Generative adversarial networks for synthetic data generation: A comparative study,” arXiv preprint arXiv:2112.01925, 2021, doi: 10.48550/arXiv.2112.01925.
[15] P. Zhao, W. Zhang, X. Cao, and X. Li, “Denoising diffusion probabilistic model-enabled data augmentation method for intelligent machine fault diagnosis,” Eng. Appl. Artif. Intell., vol. 139, Art. no. 109520, Jan. 2025, doi: 10.1016/j.engappai.2024.109520.
[16] C. Fan, Y. Zhang, H. Ma, K. Yu, and Z. Ma, “A novel lightweight DDPM-based data augmentation method for rotating machinery fault diagnosis with small sample,” Mech. Syst. Signal Process., vol. 232, Art. no. 112741, 2025, doi: 10.1016/j.ymssp.2025.112741.
[17] G. Rubino, G. Tomassi, L. Ciprini, S. Ali, and F. Marignetti, “Speed sensorless control based on Luenberger observer for DC motors,” in Proc. 2nd Int. Conf. Sustainable Mobility Applications, Renewables and Technology (SMART), 2022, pp. 1–6, doi: 10.1109/SMART55236.2022.9990558.
[18] X. Lin, C. Wu, W. Yao, Z. Liu, X. Shen, R. Xu, G. Sun, and J. Liu, “Observer-based fixed-time control for permanent-magnet synchronous motors with parameter uncertainties,” IEEE Trans. Power Electron., vol. 38, no. 4, pp. 4335–4344, Apr. 2023, doi: 10.1109/TPEL.2022.3226933.
[19] "Starter Motor Teardown and Feature Study". Retrieved March 5, 2025, from https://advancedstructures.in/starter-motor-teardown-and-feature-study
[20] 蔡忠翰，「深度學習應用於啟動機動力傳動模組品質檢測」，碩士論文，國立成功大學機械工程學系，2025。
[21] "How an Engine Starter Motor Works". Retrieved March 5, 2025, from https://youtu.be/kFsl5r34lCI?si=03om5ei4yUAANhRR
[22] R. X. Gao and R. Yan, “Non-stationary signal processing for bearing health monitoring,” Int. J. Manuf. Res., vol. 1, no. 1, pp. 18–40, 2006, doi: 10.1504/IJMR.2006.010701.
[23] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), Long Beach, CA, USA, 2017, pp. 5998–6008.
[24] Z. Zhang, G. Qiu, Y. Cheng, and M. Wang, “Densely-connected decoder transformer for unsupervised anomaly detection of power electronic systems,” J. Autom. Intell., vol. 4, no. 3, pp. 217–226, 2025, doi: 10.1016/j.jai.2025.05.002.
[25] C. Fu, M. Quintana, Z. Nagy, and C. Miller, “Filling time-series gaps using image techniques: Multidimensional context autoencoder approach for building energy data imputation,” Appl. Therm. Eng., vol. 236, Art. no. 121429, Jan. 2024, doi: 10.1016/j.applthermaleng.2023.121429.
[26] M. Fawakherji, S. Alaoui, and S. M. Turjya, “Learning to see with less: A survey on computer vision with limited and imperfect data,” Arch. Comput. Methods Eng., 2026, doi: 10.1007/s11831-026-10558-y.
[27] Y. Zhao, T. Sheng, and D. Li, “Data augmentation fault diagnosis of rolling machinery using condition denoising diffusion probabilistic model and improved CNN,” IEEE Trans. Instrum. Meas., vol. 74, Art. no. 3517712, pp. 1–12, 2025, doi: 10.1109/TIM.2025.3545721.
[28] C. T. Alexakos, Y. L. Karnavas, M. Drakaki, and I. A. Tziafettas, “A combined short-time Fourier transform and image classification transformer model for rolling element bearings fault diagnosis in electric motors,” Mach. Learn. Knowl. Extr., vol. 3, no. 1, pp. 228–242, 2021, doi: 10.3390/make3010011.
[29] Auto 8, “Starter Motor for Toyota Camry Celica RAV4 Tarago MR2 Spacia Avalon Crown Lexus LS400 ES400 2.0L 2.2L 2.4L 2.5L 3.0L Engines,” Auto 8, Retrieved Jul. 28, 2026, from https://auto8.com.au/p/Electrics/Starter-System/Starter/b/Auto-8/S02-034.html
[30] S. Zhou, L. Zhang, X. Yang, R. Luo, B. Du, et al., “Remaining useful life prediction method of centrifugal pump rolling bearings based on digital twins,” Scientific Reports, vol. 15, Art. no. 19513, 2025, doi: 10.1038/s41598-025-03952-2.