基于卷积神经网络和Huffman编码的电火花加工穿透检测技术研究

Study on EDM Based on Convolutional Neural Network and Huffman Coding for Penetrating Testing Technology

为了解决电火花加工薄壁中空复杂零件穿透检测难题,提出一种通过采集电极工件间电流、电压数据并利用卷积神经网络进行分类的算法。针对在不同放电条件、不同工件材料、不同电极材料条件下,构造神经网络时需要选择不同权值参数造成的实时运算数据量过大问题,采用Huffman编码方式对权值参数进行压缩处理,并在运算时写入高速缓存中,便可针对不同加工条件灵活选择最优的权值参数,提高CPU利用率。通过累计实验表明,该方法可对不同加工条件的穿透深度进行有效控制,减小硬件开销,降低算法复杂度。

In order to solve the penetrating testing problem of thin-walled hollow complex parts produced by EDM,this paper presents an algorithm for classifying the current and voltage data between electrode workpieces by convolution neural network. Meanwhile different weighting parameters are needed to construct neural networks by different discharge conditions,different workpiece materials and different electrode materials,resulting in excessive amount of real-time data. In this paper,Huffman encoding method is used to compress the weight parameters and write them into the cache during operation. This method can flexibly select the optimal weight parameters according to different processing conditions and improve the utilization of CPU. The accumulative experiments show that this method can effectively control the penetrating depth under different processing conditions,reduce hardware resource and algorithm complexity.