Research On Tool Wear Detection Method Based On Machine Vision

The continuous development of manufacturing industry has brought profound technological change,and the demand for manufacturing industry from all industry is also increasing,among which precision and efficiency are important indicators to measure in the manufacturing production process.The precision and efficiency of manufactured products are mainly reflected in manufacturing tools.As one of the most important factors in processing,cutting tool has high research significance and value.This paper studied the wear state of the tool during high-speed milling,and proposed a tool wear method based on machine vision.The details are as follows:Based on the classification and existing problems of tool wear detection technology,and clarifying tool wear pattern and process,a tool wear detection method based on machine vision of tool wear surface image was proposed.For the obtained tool wear image,through image processing methods such as grayscale,median filter and Canny operator,the tool wear edges were effectively detected and the pre-processing of tool wear image was realized.After image pre-processing,the wide application of deep convolution neural network in image recognition and classification can realize easier and more convenient recognition methods and avoid complex feature extraction and calculation in traditional detection methods.Especially in the case of a small number of data sets,the network model with high processing efficiency and high accuracy can be trained by using the existing excellent network model through transfer learning.In this paper,the Alex Net and Res Net residual network in conventional neural network were optimized and trained,through modifying parameters and reconstruction the connection layers,using the tool wear images for transfer learning.The prediction effects of the two models were analyzed and compared.By comparing a small number of samples with full samples,pre-processed images and untreated images,the overall conclusion is that in the training of a small number of samples,Alex Net is slightly better than Res Net18 which has deeper multi-layer parameters.In the training of full sized samples,training results show that the accuracy of Res Net18 is slightly better than Alexnet.Based on trained Res Net18 network,the paper designed and developed the tool wear detection functional module software program framework based on machine vision,according to the functional objectives,combined with Matlab and App Designer,the hybrid programming of transfer learning network module,wear image and display module,image pre-processing module and wear type intelligent recognition module were designed.Finally,through the recognition status results,recognition rate and recognition time,the suggestions on whether the tool needs to be replaced during production and processing would be suggested.