基于协同神经网络的系统状态识别

System State Recognition Based on Cooperative Neural Network

现代机械加工对安全性和效率的要求越来越高,声发射以高灵敏度、高可靠性、高安全性等优点,成为机械加工动态监测中最具前景的监测方法之一。声发射现象在机械加工中广泛存在,通过研究声发射技术的基本原理可以明确声发射信号的变化规律。基于协同神经网络的机械加工工作状态识别的实现过程包括学习训练阶段和迭代识别阶段。网络实现过程首先将输入待识别模式的特征向量进行去均值归一化等预处理,求出序参量的初始值,然后根据序参量的动力学方程进行演化,只有一个序参量演化的值达到稳定,最终模式类在竞争中胜出,网络识别结果是把待识别模式归类为相应的初始训练模式类。

Modern machining has higher and higher requirements for safety and efficiency.Acoustic emission(AE)has become one of the most promising monitoring methods in machining dynamic monitoring with its advantages of high sensitivity,high reliability and high safety.Acoustic emission phenomenon is widely existed in machining.By studying the basic principle of acoustic emission technology,the variation law of acoustic emission signal is clarified.The realization process of machining working state recognition based on cooperative neural network includes learning training stage and iterative recognition stage.The process of network realization is to preprocess the input feature vector of the pattern to be recognized,such as de-mean normalization,to find out the initial values of the order parameters,and then evolve according to the dynamics equation of the order parameters.Only one of the order parameter evolution values reaches stability,and the final pattern class wins the competition.The result of network recognition is to classify the pattern to be recognized as the corresponding initial training pattern class.