在本論文中，將深度學習模型載入嵌入式系統中，以嵌入式系統實現一種基於卷積神經網絡的深度學習模型進行螺絲鎖附之異常偵測系統。
本論文所提出的方法，利用鎖螺絲機所產生的三種馬達參數(轉矩、角度、速度)資料重疊繪製成時間域波形圖，搭配在嵌入式系統中的卷積神經網絡模型進行圖像辨識分類，以實現更高且穩定的判定準確性。此方法無須專家進行參數調整，避免了外在因素干擾。在卷積神經網絡學習過程中，考慮到原始資料變化及時間的相對應關係，從原始資料中學習出代表性特徵，相較於傳統依賴檢測員經驗的手動特徵設定具有更高的準確性。
本論文通過鎖附母板螺絲取得鎖附成功與失敗兩種狀態各1,000筆的訓練資料集；以此資料集訓練模型1(VGG_19BN-GAP)以及模型2(ALEXNET-GAP)，最後以連續鎖附母板2,601次的整體測試來驗證模型的準確度。經過整體測試後，準確度模型1為99.3%、模型2為98.2%，相較於傳統手動特徵設定的準確度95.9%而言，判定正確率從95.9%上升至99.3%，結果表明該方法具有優異的判定能力。另外，由於該方法未來將實現於工廠的生產線上，因此針對嵌入式系統的效能與穩定度再做測試。測試結果顯示出對於推論時間較為嚴苛者，以網絡較複雜和低解析度的圖像為輸入的模型1(推論時間3.7秒，準確度98.2%，精確率99.8%)為最佳；若對於推論時間較為寬裕者，則以網絡較複雜和高解析度的圖像為輸入的模型1(推論時間7.03秒，準確度99.3%，精確率99.9%)為最佳。

In this research, the feasibility of introducing deep learning models based on convolutional neural network into embedded systems is studied for screwdriving fault detection. The proposed method uses the measured motor parameters, i.e. torque, angle and speed, to generate a time-domain waveform diagram of each screwdriving process and tries to use these waveform diagrams to train a convolutional neural network model. It is expected that such a trained model could be implemented in an embedded system to classify images for higher and stable performance, as compared to the current empirical method.
In the convolutional neural network learning process, considering the relationship between the raw data and time, the representative features are learned from the raw data, which are more accurate than traditional handcraft feature setting that rely on the experience of the inspector.
The results of the final states are defined as success and failure. Each result has a sample set of 1,000 waveforms(treated as images), which is used as the training dataset. The proposed Model 1(based on VGG_19BN-GAP) and Model 2(based on ALEXNET-GAP) were experimented, and then the accuracy of the models was verified by fastening(screwdriving) the motherboard 2,601 times. The experimental shows the accuracy of Model 1 is 99.3%, and Model 2 is 98.2%. The pre-recorded accuracy of traditional handcraft feature setting is 95.9%. The results show that the proposed two models are excellent in fault detection. Considering the proposed method will be realized in the assembly line of the factory using an embedded system, so the performance of the proposed methods in an embedded system is investigated. The test results show that with limited inference time, a refined Model 1 implementation using a more complex network and low-resolution images is the best (inference time:3.7 seconds, accuracy:98.2%, precision:99.8%). If the inference time is less constrained, another refined Model 1 implementation using a more complex network and high-resolution image is the best (inference time:7.03 seconds, accuracy:99.3%, precision:99.9%).

[1]S. Lu, Q. He, H. Zhang, and F. Kong, "Rotating machine fault diagnosis through enhanced stochastic resonance by full-wave signal construction," Mechanical Systems and Signal Processing, vol. 85, pp. 82-97, 2017.
[2]S. Lu, Q. He, and J. Wang, "A review of stochastic resonance in rotating machine fault detection," Mechanical Systems and Signal Processing, vol. 116, pp. 230-260, 2019.
[3]H. Oh, J. H. Jung, B. C. Jeon, and B. D. Youn, "Scalable and unsupervised feature engineering using vibration-imaging and deep learning for rotor system diagnosis," IEEE Transactions on Industrial Electronics, vol. 65, no. 4, pp. 3539-3549, 2017.
[4]J. Pan, Y. Zi, J. Chen, Z. Zhou, and B. Wang, "LiftingNet: A novel deep learning network with layerwise feature learning from noisy mechanical data for fault classification," IEEE Transactions on Industrial Electronics, vol. 65, no. 6, pp. 4973-4982, 2017.
[5]H. Shao, H. Jiang, H. Zhao, and F. Wang, "A novel deep autoencoder feature learning method for rotating machinery fault diagnosis," Mechanical Systems and Signal Processing, vol. 95, pp. 187-204, 2017.
[6]W. Sun, R. Zhao, R. Yan, S. Shao, and X. Chen, "Convolutional discriminative feature learning for induction motor fault diagnosis," IEEE Transactions on Industrial Informatics, vol. 13, no. 3, pp. 1350-1359, 2017.
[7]M. Xia, T. Li, L. Xu, L. Liu, and C. W. De Silva, "Fault diagnosis for rotating machinery using multiple sensors and convolutional neural networks," IEEE/ASME Transactions on Mechatronics, vol. 23, no. 1, pp. 101-110, 2017.
[8]C. Sun, M. Ma, Z. Zhao, and X. Chen, "Sparse deep stacking network for fault diagnosis of motor," IEEE Transactions on Industrial Informatics, vol. 14, no. 7, pp. 3261-3270, 2018.
[9]S. Lu, G. Qian, Q. He, F. Liu, Y. Liu, and Q. Wang, "Insitu Motor Fault Diagnosis Using Enhanced Convolutional Neural Network in an Embedded System," IEEE Sensors Journal, 2019.
[10]Y. Chen, Z. Lin, X. Zhao, G. Wang, and Y. Gu, "Deep learning-based classification of hyperspectral data," IEEE Journal of Selected topics in applied earth observations and remote sensing, vol. 7, no. 6, pp. 2094-2107, 2014.
[11]J. D. Rennie, L. Shih, J. Teevan, and D. R. Karger, "Tackling the poor assumptions of naive bayes text classifiers," in Proceedings of the 20th international conference on machine learning (ICML-03), 2003, pp. 616-623.
[12]A. Liaw and M. Wiener, "Classification and regression by randomForest," R news, vol. 2, no. 3, pp. 18-22, 2002.
[13]K. Makantasis, K. Karantzalos, A. Doulamis, and N. Doulamis, "Deep supervised learning for hyperspectral data classification through convolutional neural networks," in 2015 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2015, pp. 4959-4962: IEEE.
[14]U.-P. Chong, "Signal model-based fault detection and diagnosis for induction motors using features of vibration signal in two-dimension domain," Strojniški vestnik, vol. 57, no. 9, pp. 655-666, 2011.
[15]Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998.
[16]V. Nair and G. E. Hinton, "Rectified linear units improve restricted boltzmann machines," in Proceedings of the 27th international conference on machine learning (ICML-10), 2010, pp. 807-814.
[17]L. Wen, X. Li, L. Gao, and Y. Zhang, "A new convolutional neural network-based data-driven fault diagnosis method," IEEE Transactions on Industrial Electronics, vol. 65, no. 7, pp. 5990-5998, 2017.
[18]H. Wang, S. Li, L. Song, and L. Cui, "A novel convolutional neural network based fault recognition method via image fusion of multi-vibration-signals," Computers in Industry, vol. 105, pp. 182-190, 2019.
[19]A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," in Advances in neural information processing systems, 2012, pp. 1097-1105.
[20]K. Simonyan and A. Zisserman, "Very deep convolutional networks for large-scale image recognition," arXiv preprint arXiv:1409.1556, 2014.
[21]M. Lin, Q. Chen, and S. Yan, "Network in network," arXiv preprint arXiv:1312.4400, 2013.
[22]K. He and J. Sun, "Convolutional neural networks at constrained time cost," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 5353-5360.
[23]A. Krizhevsky and G. Hinton, "Learning multiple layers of features from tiny images,"http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.222.9220&rep=rep1&type=pdf, last retrieve 11 November 2019.