機器學習應用於煞車來令片振動訊號缺陷檢測

Abstract

探討工業 4.0 的智慧製造在煞車來令片的振動訊號缺陷檢測方面的應用。相較於傳統用人耳聽取異音辨識缺陷的方法，本論文使用了機器學習對振動訊號的缺陷檢測，其中包含以有限元素的工具軟體 (COMSOL) 模擬來令片的損壞樣式作為訓練的樣本，以及使用實驗敲擊訊號作為測試樣本。 為了提高模型的準確性，機器學習使用了 Random Forest、XGBoost 和CatBoost 的分類方法，其中在架構調整上也使用了特徵重要性 (Feature Importance)判斷各個特徵對模型預測的貢獻程度和超參數 (Hyperparameters) 的調整，例如學習率、樹的深度等等。超參數調整的目的是為了找到最優的參數組合，再使用交叉驗證 (Cross Validation) 以避免過度擬合 (Overfitting)，以提高模型的準確度。 這種應用不僅可以提高檢測的精度，還可以節省人力資源和時間成本，並幫助製造商實現更高效的生產，並且有助於實現智慧製造的目標。

This paper discusses the application of Industry 4.0’s smart manufacturing in vibration signal defects detection in brake pads. In contrast to traditional methods of identifying defects by listening to abnormal sounds, using machine learning to detect vibration signal defects, which includes using finite element simulations (COMSOL) to train the model with brake pads’ damage patterns and use experimental tapping signals as test samples. To improve the model accuracy, Random Forest, XGBoost, and CatBoost classification methods were used in machine learning. Parameters adjusted include feature importance determines the contribution of each feature to the model’s prediction and hyperparameters such as learning rate and tree depth to find the optimal parameter combination, use cross validation to avoid overfitting, and improve the accuracy is 94% of the model. This application not only improves the accuracy of detection but also saves human resources and time costs. It also helps manufacturers achieve more efficient production and contributes to the goal of smart manufacturing.