雷射切雕機的精度調校自動化

Abstract

由於快速、消耗功率小，雷射加工機已被廣泛應用在光罩製作、IC打印（Marking）等方面。當應用在打印時，往往需要將一個圖案的陣列打在IC上，即使每個圖樣的誤差很小，由於數量龐大而產生相當大的誤差累積，所以誤差的校正非常重要。 誤差參數的調整必須先觀察加工的結果，找出誤差的形式，再調整相關的參數。但由於誤差的現象是由數個參數所造成，而且參數間會交互影響，所以非常難調整，而且當機台有所變動時，如：更換鏡頭、改變焦距時，均必須重新調整，所以需要相當長的時間和豐富的經驗，因此參數調整程序的自動化非常重要。 首先將以田口法的理論，建立直交配置表，分析各可控因子對各種加工誤差影響的程度，接著利用田口法所得的結果，選定數組實驗參數作為實驗數據，接著利用類神經網路，配合最佳化理論，以前述的實驗參數及所得的數據來訓練類神經網路，找出誤差參數及加工結果的數學模式，再求出此數學模式的反函數，就可利用加工結果及初始的參數設定找出加工誤差最小的參數值。

Because of high speed and low power consumption, laser machine has been applied to manufacturing of mask and IC marking. We always need to mark an pattern array on a workpiece, even if the error of each pattern is small, the total error will be large because of large amount of patterns, so correction of error is very important. In order to adjust parameters, we should observe results of machining first, find the trend of error. But errors are related to several parameters, and the correlations between each errors are nonlinear, so it’s hard to adjust, further more, when alterations happen to the machine, we should readjust parameters, it takes a lot of time and depends on experiences of engineers, so it’s important to automate computation of parameters. First, we build orthogonal array on the basis of Taguchi Method, analyze the degree of influence of controllable factors to machining errors. Then use previous result to determine several parameters of experiment, and use these data as training sets of neural network, find the mathematical model between error parameters and machining results, and inverse function of this model, then we can use Parameters and results to find optimal parameters.