Tool Wear Detection Method for Lightweight Convolutional Neural Network Combined with Time Series Fusion
Aiming at the problems of the large size and complex structure of traditional convolutional neural network tool wear rec-ognition method,and the difficulty of obtaining tool wear image data online,a research method of applying lightweight convolutional neu-ral network to tool wear recognition was proposed.The force signal and vibration signal generated during milling cutter machining were converted into an image data set through Gram angle field processing.Then the image data were entered into the lightweight convolution-al neural networks MobilenetV2,MobilenetV3 and ShuffleNet models for comparative analysis.The results show that MobilenetV2 can be used in the initial wear and normal wear stage,and MobilenetV3 can achieve the best effect on tool wear recognition in the rapid wear stage.Comparison with general convolutional neural networks shows that lightweight convolutional neural networks are more effective in tool wear recognition than general convolutional neural networks.Using this method,the problem of online acquisition of tool image data is solved,the fault tolerance of information processing is increased,the volume and calculation amount of the model are effectively re-duced,and the implantation into embedded system and integration into the machine tool system are facilitated.
tool weardeep learningconvolutional neural networkGram angle field
NETL
NSTL
万方数据
