臺灣博碩士論文加值系統

本研究主要在探討電動車電源供應器生產線後段的成品瑕疵檢測，目前生產現場的作法會由專業聽音員將電源供應器進行撞擊測試，讓沒有焊接好的元件脫落，再手動將電源供應器拿起並進行搖晃，若有元件脫落則會導致異音的產生，從而判斷出是否為瑕疵品。但即使專業聽音員經過長時間的訓練，也無法辨識出脫落元件的種類，若可分辨出脫落元件的種類，即可讓維修人員快速針對問題而進行維修。有鑑於此，本研究提出一種自動異音檢測之方法，使用電腦硬體及演算法進行脫落元件之辨識，本研究方法分為四個步驟：數據收集、數據降噪、特徵提取和元件辨識。首先數據收集將分為兩種做實驗，實驗一為使用人手進行搖晃，實驗二為使用伺服馬達進行搖晃，並分別利用麥克風將搖晃電源供應器裡脫落之元件所發出的異音進行收集。再使用帶通濾波器進行降噪，把背景雜音去除來獲得乾淨之聲音數據。第三步，將所有聲音數據轉化成對數梅爾譜圖後進行元件辨識。第四步元件辨識，將使用卷積神經網路，學習這些異常聲音數據，並經過最大池化層的特徵篩選，訓練出預測模型，最後將額外準備的測試資料進行測試，手動搖晃之數據準確率達到91.67%，伺服馬達搖晃之數據準確率達到90%，可以取代專業聽音員去分辨電源供應器是否瑕疵，並且能分辨出脫落元件的種類。

The purpose of the study was to discuss the defect detection of products in the electric vehicle power supply units after the production line. Currently, the production site method, professional listeners the power supply unit conduct a crash test, allow the falling off improperly soldered components. Then, the power supply unit is manually shaken. If any audible anomalies resulting from falling off components are identified to determine they are defective. However, even with extensive training, professional listeners are unable to recognize the specific type of dislodged component. If the type of dislodged component can be identified, maintenance personnel can quickly address the problem. In this paper proposes an automatic abnormal sound detection method, computer hardware and algorithms will be utilized to identify dislodged components.
The research methodology consists of four steps: data collection, data denoising, feature extraction, and component recognition. Firstly, data collection involves two experiments. Experiment one involves manual shaking by hand, while experiment two utilizes a servo motor for shaking. In both cases, the abnormal sounds generated by the loose components in the shaken power supplies are collected using microphones. Next, a bandpass filter is applied to denoise the data and remove background noise, resulting in clean audio data. In the third step, all audio data is transformed into Log-Mel spectrograms for component recognition. The fourth step, component recognition, employs a Convolutional Neural Network to learn from the abnormal sound data, followed by feature selection through a max pooling layer to train the predictive model.
Finally, additional test data is used for evaluation. The accuracy of manually shaken data reaches 91.67%, while the accuracy of data shaken by the servo motor reaches 90%. This approach can replace professional listeners in discerning defects in power suppliers and identify the type of dislodged components.

中文摘要 …………………………………………………………… i
Abstract …………………………………………………………… ii
誌謝 …………………………………………………………… iv
目錄 …………………………………………………………… v
表目錄 …………………………………………………………… vii
圖目錄 …………………………………………………………… viii
第一章 緒論……………………………………………………… 1
1.1 前言……………………………………………………… 1
1.2 研究動機與目的………………………………………… 1
1.3 文獻回顧………………………………………………… 2
1.4 論文架構………………………………………………… 4
第二章 特徵提取與辨識模型原理……………………………… 5
2.1 訊號處理………………………………………………… 5
2.1.1 數據分幀………………………………………………… 5
2.1.2 加窗函數………………………………………………… 7
2.1.3 快速傅立葉轉換………………………………………… 8
2.2 梅爾濾波器……………………………………………… 9
2.3 CNN卷積神經網路……………...…………...…………. 10
第三章 實驗數據收集與異音辨識模型建立方法……………… 12
3.1 數據收集………………………………………………… 12
3.1.1 實驗一：手動搖晃……………………………….……... 13
3.1.2 實驗二：馬達搖晃………………………………….…... 13
3.2 數據降噪………………………………………………… 15
3.3 特徵提取………………………………………………… 16
3.4 異音辨識………..……………………………………….. 18
3.5 實驗設備………………………………………………… 19
第四章 實驗結果………………………………………………… 20
4.1 不同參數比較…………………………………………… 20
4.1.1 不同幀長與分幀之比較………………………………… 20
4.1.2 不同快速傅立葉點數之比較…………………………… 21
4.1.3 有無濾波器之比較……………………………………… 23
4.2 各元件之對數梅爾譜圖………………………………… 25
4.2.1 實驗一：手動搖晃之對數梅爾譜圖…………….……... 25
4.2.2 實驗二：馬達搖晃之對數梅爾譜圖…………….……... 29
4.3 模型訓練………………………………………………… 34
4.3.1 實驗一：手動搖晃之模型訓練……………….……...… 34
4.3.2 實驗二：馬達搖晃之模型訓練………………………… 35
4.4 模型測試………………………………………………… 36
第五章 結論與未來展望………………………………………… 37
5.1 結論……………………………………………………… 37
5.2 未來展望………………………………………………… 37
參考文獻 …………………………………………………………… 38
Extended Abstract …………………………………………………………… 40

[1]S. Dol, and R. Bhinge, “SMART Motor for Industry 4.0,” 2018 IEEMA Engineer Infinite Conference, New Delhi, India, pp. 1-6, 2018.
[2]B. Mrazovac, V. Ilian, and M. Hulea, “Beyond Industry 4.0: Leveraging AI-Powered Anomalous Sound Detection for Smart Maintenance,” 2021 Zooming Innovation in Consumer Technologies Conference, Novi Sad, Serbia, pp. 43-47, 2021.
[3]W. Baek and D. Y. Kim, “In-Process Noise Inspection System for Product Fault Detection in A Loud Shop-Floor Environment,” in IEEE Transactions on Instrumentation and Measurement, Vol. 70, pp. 1-11, 2021.
[4]E. Cakir, T. Heittola, H. Huttunen, and T. Virtanen, “Polyphonic Sound Event Detection Using Multi Label Deep Neural Networks,” 2015 International Joint Conference on Neural Networks, Killarney, Ireland, pp. 1-7, 2015.
[5]K. In Tan, S. Yean, and B. Sung Lee, “Sound Classification Using Sound Spectrum Features and Convolutional Neural Networks,” 2022 3rd International Conference on Human-Centric Smart Environments for Health and Well-being, Levi’s, QC, Canada, pp. 94-99, 2022.
[6]Q. Han, H. Zhang, J. Sun, D. Yuan, and J. Zhang, “ARResNet: A Convolutional Neural Network Based on Human Ear Features to Construct Abnormal Sound Detection System for Air-Conditioning,” 2021 International Conference on High Performance Big Data and Intelligent Systems, Macau, China, pp. 162-166, 2021.
[7]A. Shaukat, A. Younis, U. Akram, M. Mohsin, and Z. Mustansar, “Towards Automatic Recognition of Sounds Observed in Daily Living Activity,” 2019 IEEE 18th International Conference on Cognitive Informatics & Cognitive Computing, Milan, Italy, pp. 66-74, 2019.
[8]S. Hatanaka, and H. Nishi, “Efficient GAN - Based Unsupervised Anomaly Sound Detection for Refrigeration Units,” 2021 IEEE 30th International Symposium on Industrial Electronics, Kyoto, Japan, pp. 1-7, 2021.
[9]Z. Feng, Z. Zhao, H. Chen, B. Dou, and L. Hu, “Power Plant Production Equipment Sound Recognition Method Combined with Attention Mechanism, ” 2022 IEEE 4th International Conference on Power, Intelligent Computing and Systems, Shenyang, China, pp. 185-188, 2022.
[10]T. Tran, and J. Lundgren, “Drill Fault Diagnosis Based on The Scalogram and Mel Spectrogram of Sound Signals Using Artificial Intelligence,” in IEEE Access, Vol. 8, pp. 203655-203666, 2020.
[11]A. Meghanani, A. C. S., and A. G. Ramakrishnan, “An Exploration of Log-Mel Spectrogram and MFCC Features for Alzheimer’s Dementia Recognition from Spontaneous Speech,” 2021 IEEE Spoken Language Technology Workshop, Shenzhen, China, pp. 670-677, 2021.
[12]T. Yan, H. Meng, S. Liu, E. Parada-Cabaleiro, Z. Ren, and B. W. Schuller, “Convoluational Transformer with Adaptive Position Embedding for Covid-19 Detection From Cough Sounds, ” ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing, Singapore, Singapore, pp. 9092-9096, 2022.
[13]M. Zhang, L. Yan, G. Luo, G. Li, W. Liu, and L. Zhang, “A Novel Insect Sound Recognition Algorithm Based on MFCC and CNN,” 2021 6th International Conference on Communication, Image and Signal Processing, Chengdu, China, pp. 289-294, 2021.
[14]L. Chen, R. Wang, F. Hu, Z. Li, L. Liang, and C. Lei, “Research on Voice Print Recognition of Electrical Faults Based on Attention-MFCC Algorithm, ” 2021 Power System and Green Energy Conference, Shanghai, China, pp. 748-751, 2021.
[15]P. Mahesha, and D. S. Vinod, “LP-Hillbert Transform Based MFCC for Effective Discrimination of Stuttering Dysfluencies,” 2017 International Conference on Wireless Communications, Signal Processing and Networking, Chennai, India, pp. 2561-2565, 2017.
[16]V. T. Tran and W. H. Tsai, “Acoustic-Based Emergency Vehicle Detection Using Convolutional Neural Networks,” in IEEE Access, Vol. 8, pp. 75702-75713, 2020.
[17]J. W. Cooley, and J. W. Tukey, “An Algorithm for The Machine Calculation of Complex Fourier Series,” Math. Comput., Vol. 19, pp. 297-301, Apr. 1965.
[18]S. S. Stevens, J. Volkmann, and E. B. Newman, “A Scale for The Measurement of the Psychological Magnitude Pitch,” Journal of the Acoustical Society of America, Vol. 8, pp. 185-190, 1937.
[19]J. Makhoul and L. Cosell, “LPCW: An LPC Vocoder with Linear Predictive Spectral Warping, ” ICASSP '76. IEEE International Conference on Acoustics, Speech, and Signal Processing, Philadelphia, PA, USA, pp. 466-469, 1976.
[20]D. H. Hubel, and T. N. Wiesel, “Ferrier Lecture-Functional Architecture of Macaque Monkey Visual Cortex,” Proceedings of the Royal Society of London, Series B. Biological Sciences 198.1130, pp. 1-59, 1977.