本研究目標為利用工具機的振動數據判別車刀的狀況。振動數據來源是是安裝在在車床主軸上的振動感測器，每秒量測1,660筆，單位是〖10〗^(-3)g。本論文的數據有A、B兩組；分別來自不同工序。A數據共有14筆：11筆為正常車刀，2筆為車刀磨損，1筆為切斷刀磨損。B數據紀錄的是某工序持續重複作95次---量測車刀從全新到磨損的連續過程---的振動數據。本論文利用移動標準差作數據前置處理，找出正常車刀與磨損車刀之間的明顯不同特徵，並與其他的常用的訊號分析方式做比較。最後嘗試利用深度學習的模型建立預測車刀狀況的鑑別系統。 This article investigates the relationship between the vibration data and the status of lathe tools. The vibration data comes from the sensor mounted on the spindle of the lathe to measure the pressure in 〖10〗^(-3)g 1660 times per second. There are two sets of data, called data set A and B, respectively. The data set A consists of 14 time series, among which 11 data record the pressure of normal lathe tools and 3 data record the pressure of abnormal lathe tools. The data set B consists of 95 time series, which consecutively record the pressure for the same working process, as the lathe tool turns from normal status into abnormal status. We preprocess the time series by using moving standard deviation then find the characteristics which is capable of discriminating the status lathe tools. We also use other methods such as the short-time-Fourier-transformation, Hilbert-Huang transformation to preprocess the time series for comparison. Finally, we try to establish a LSTM (long short-term memory) deep learning model to predict the moving standard deviation of normal lathe tools according to the past vibration. Our future goal is to adopted a LSTM as a generator in a GAN (generating adversarial network) to automatically discriminating the status of lathe tools.
