Development for a Prediction System of Chatter Vibration by using Fuzzy Neural Network Model

本文研究了基于模糊神经网络模型的颤振预报系统。机械加工过程中产生的颤振会导致被加工零件的表面光洁度和尺寸精度差，降低刀具寿命，甚至损坏机床。在过去的几十年里，人们对其预测和避免进行了各种研究。首先，利用小波变换和模糊神经网络模型开发了高速立铣刀颤振预测专家系统。其次，将小波包变换用于模糊神经网络模型。通过使用小波包，扩展了输入层神经元数目的选择灵活性，而不是通常的小波变换。此外，隐藏层被分为两个块，第一个块与条件部分相关，第二个块由切割声音部分组成。对学习的收敛和决策的准确性进行了研究。将该方法应用于一台夹具磨床上，验证了颤振预测方法的有效性。但是，在运行过程中，切割声音的测量是很困难的。测量的声音包括环境噪声。进行了从测量数据中分离出切割声的研究。由于该问题被看作是一个盲源分离问题，因此利用ICA算法和子空间辨识方法实现了切割声与测量数据的分离。该程序需要在未来进一步改进。

This research is concerned with Chatter prediction system by using a fuzzy neural network model. The chatter vibration occurring in mechanical machining gives rise to poor surface finish and dimensional accuracy in machined part, reduction of tool life, and even damages machine tools. Various kinds of researches concerning its prediction and avoidance have been carried out over the last several decades. First, this study is to develop an expert system for predicting chatter vibrations in high-speed end milling using wavelet transform and a fuzzy neural network model. Second, wavelet packet transform is used for the fuzzy neural network model. By using wavelet packets, flexibility for selection of the number of neurons in input layer extend rather than the usual wavelet transform. Additionally, the hidden layer is divided into two blocks, the first block is related to the condition part and the second one consists of the cutting sound part. The convergence of learning and the accuracy of decision are investigated in this paper. The proposed method is applied to a jig grinding machine, and the results demonstrate the effectiveness of the chatter prediction procedure. But, it is difficult to measure cutting sound in operation. The measured sounds include environmental noise. The research has been carried out to separate the cutting sounds from the measured data. As the problem is regarded as a BSS problem, the separation of cutting sounds from the measured data is performed by ICA algorithm and the subspace identification method. The procedure needs further improvement in the future.