整合LSTM及去噪自編碼器於刀具磨耗之預測
|
| 摘要 銑銷加工是製造業常見的加工技術之一，業界對於銑削加工的平面經常有精度上的要求，銑刀的磨耗程度自然也成為了影響加工精度的主要因素。本研究援引了 PHM2010 大數據競賽提供的銑刀磨耗資料集，收集了銑刀的磨耗值與各個傳感器偵測的震動數據，透過快速傅立葉轉換(FFT)與濾波過濾較不重要之雜訊並接續後面的單/ 雙向網路LSTM 預測銑刀磨耗值，而實驗結果也表明，本研究提出的LSTM-DAE能有效提取特徵，供後續模型進行訓練，且雙向LSTM的表現要優於單向LSTM。
關鍵字：銑刀磨耗預測、快速傅立葉轉換、自動編碼器、LSTM。 ABSTRACT Milling is one of the common processing technologies in the manufacturing industry. The industry often has precision requirements for the plane of milling. The degree of wear of the milling cutter naturally becomes the main factor affecting the machining accuracy. This study cites the milling cutter wear data set provided by the PHM2010 big data competition. The wear value of the milling cutter and the vibration data detected by each sensor are collected, and the less important noise is filtered through fast Fourier transform (FFT) and filtering. The following unidirectional/bidirectional network LSTM predicts the wear value of the milling cutter, and the experimental results also show that the LSTM-DAE proposed in this study can effectively extract features for subsequent model training, and the performance of the bidirectional LSTM is better than that of the unidirectional LSTM. KEYWORDS: Milling Cutter Wear Prediction; Fast Fourier Transform; Autoencoder, LSTM. |