應用於顫振辨識之特徵選擇與分類方法之研究

Abstract

維持高產率對於銑削加工的效率而言十分重要。顫振是加工時發生的一種自激式振動，在實務中限制了產率。過去的研究提出了許多顫振偵測的方法，利用各種訊號處理的方法如快速傅立葉轉換(FFT)，小波包轉換(WPT)，及希爾伯特-黃轉換(HHT)。許多資料分類演算法也被應用於顫振偵測。雖然顫振偵測的領域已有許多文獻，我們仍不清楚何種方法可以達到較佳的正確率與偵測速率。 在本研究中，我們將測試多種訊號處理方法以及資料分類的演算法，使用的資料集中包含各種主軸轉速及切深。我們結合多種訊號處理方法及分類演算法，開發了一個顫振辨識平台以建立分類模型並評估其性能。資料分類方法包含了固定的閾值，最近鄰居法(k-NN)，單純貝氏分類器，支持向量機(SVM)，局部密度因子(LOF)，以及類神經網路。以分類精準度而言，結果顯示最近鄰居法搭配小波包轉換及希爾伯特-黃轉換是最佳的方法，誤判率僅2.2%。

Maintaining high production yield is important for efficiency in the milling process. Chatter is a type of self-excited vibration that can occur during machining, and limits the production yield in practice. In the past, many chatter detection methods were proposed using different signal processing methods such as Fast Fourier transform (FFT), wavelet packet transform (WPT), and Hilbert-Huang transform (HHT). Several classification methods were also applied in chatter detection. Despite the large amount of researches regarding chatter detection, it is unclear which of these proposed methods are better in terms of accuracy and detection speed. In this research, we test the signal processing methods and classification algorithms against the entire dataset, with a wide range of spindle speeds and depth of cuts. A chatter identification platform is developed to train models and evaluate their performance, using combinations of signal processing methods and classification algorithms. The classification methods include numerical threshold, k-nearest neighbors (K-NN), Naïve Bayes, support vector machine (SVM), local outlier factor (LOF), and artificial neural network. K-NN proves to be the optimal method when using WPT and HHT for signal processing, with an error rate of 2.2%.