傾斜線切割放電加工粗割輪廓與其類神經網路模型探討

Abstract

線切割放電在實際加工中，影響工件好壞的因素除了加工參數外，加工中線振位移、極間排渣等皆會造成工件表面形貌的變化，當嚴重的線振位移發生時、極間排渣不良造成頻繁的短路、電弧放電，進而造成極間間隙距離無法保持一穩定狀態，皆會使得加工表面品質欠佳，故上述的監控在放電加工中有其必要性。然欲量測上述三者有其困難性，本實驗嘗試藉由擷取出放電加工過程中的放電波形用以描述極間狀況，探討參數、訊號與加工結果三者間的關係，從中分析關聯。 實驗透過田口法探討參數與物理量對響應的關係，響應包含工件角度誤差、材料移除率以及垂直形狀精度。其中垂直形狀精度是指放電加工後，工件垂直方向的真直度情況。本實驗透過表面粗糙度輪廓儀量測垂直形狀精度，從加工中的原始波形擷取物理量，建立同參數下不同時刻垂直形狀精度與物理量之關係。研究主要探討相對穩定製程下粗割一刀參數，並對垂直形狀精度深入探討其參數及物理量對形狀與量值的影響。 深入分析垂直形狀精度在形狀與量值上的影響，分別在真平度以及真直度方面皆有著較為新穎的發現。透過區分不同加工段的方式，分別統計該加工段下訊號之特徵，以此與加工參數共同預測加工響應，並比較不同模型對問題處理所具有之物理意義，以類神經網路建立放電預測模型，實驗結果顯示模型的誤差率約落在9.35%左右，在數值與趨勢上一定程度實現加工響應垂直形狀精度的預測，又透過長短期記憶模型與倒轉傳遞模型比較收斂性的驗證，基於參數與訊號對垂直形狀誤差有著時序關係的基礎下，架設實時數據傳輸系統。

While WEDM is processing, in addition to the processing parameters, the factors that affect the quality of the workpiece, the linear vibration displacement during processing, the slag discharge between the poles, etc. will all cause changes in the surface profile of the workpiece. When severe linear vibration displacement and poor inter-electrode slag discharge cause frequent short circuits and arc discharges, and thus cause the inter-electrode gap distance to be unable to maintain a stable state, the quality of the surface profile will be poor. Therefore, the above monitoring is necessary in WEDM. However, it is difficult to measure. In this experiment, the discharge waveform during the WEDM process is captured to describe the inter-electrode condition, and the relationship between the parameters, the signal and the machining result is discussed and analyzed. The experiment uses the Taguchi method to explore the relationship between parameters and physical quantities to the response. The response includes the workpiece angle error, material removal rate and vertical shape accuracy. The vertical shape accuracy refers to the true straightness of the workpiece in the vertical direction after machining. In this experiment, the vertical shape accuracy was measured by a surface roughness profiler, and the physical quantities were extracted from the original waveform during processing to establish the relationship between the vertical shape accuracy and physical quantities at different times under the same parameters. The research mainly discusses the parameters of rough cutting in a relatively stable process, and deeply discusses the influence of its parameters and physical quantities on the shape and value of the vertical shape accuracy. Analysis of the influence of vertical shape accuracy on shape and value, there are relatively novel findings in terms of true flatness and true straightness, respectively. By distinguishing different processing sections, the characteristics of the signals under the processing section are counted, and the processing response is predicted together with the processing parameters, and the physical significance of different models for problem processing is compared, and a neural network is used to establish a discharge prediction model. The experimental results show that the error rate of the model is about 9.35%, and the prediction of the vertical shape accuracy of the machining response can be achieved to a certain extent in terms of numerical value and trend. The real-time data transmission system is set up on the basis of the time-series relationship between the signal and the vertical shape error.