Research On Convolution Neural Network Identification Technology For Wear State Of BTA Deep Drilling Bit

In the context of intelligent and informative manufacturing, through the collection of manufacturinog and related information in the manufacturing process, the use of these information to achieve the monitoring of the production process, and in these data to the production process problems The production process has an important role in making adjustments. Mechanical machining, the tool as of the direct contact with the machined surface, cutting tool is the weakest link in the whole machining system, to make the automation process, efficient and stable manner, tool condition monitoring technology in the manufacturing process of research and development is particularly important. And the existing tool state monitoring technology is mostly aimed at torning and and there are few methods for monitoring the state of the tools for deep hole drilling.This paper presents the characteristics of BTA deep hole, the whole process in a closed environment, the tool wear state can not be directly observed. By the inner relationship between the spindle motor current and bit wear, collected directly from the motor drive spindle motor current signal in the drilling process as monitoring objects. This paper introduces a method of tool state monitoing: for the fusion of signal analysis method and the pattern recognition method of convolution neural network.The spindle motor current signal acquisition system based on the NC module communication module of deep deep hole drilling CNC machine tool is established. The spindle motor current signal is obtained by drilling experiment, the main spindle current signal is analyzed.Recorded the drilling process drill bit wear and get the drill bit wear law information.Combined with the characteristicstics of the spindle current signal in different wear stages and the lack of signal time domain analysis and frequency domain analysis, in order to describe the signal frequency, the characteristics of the wear law of the drill are changed with time, using continuous wavelet transform get wavelet scalogram from different wear stages. The best wavelet basis function is determined by the characteristics of different wavelet bases, the optimal wavelet decomposition layer is determined by the method of wavelet signal entropy. The results show that the wavelet scalogram of the drill is different in different wear stages, wavelet scalogram is a good reflection of the signal frequency with the law of time. In wavelet scalogram, the high frequency components of the signal are gradually decreasing with time and the intermediate frequency are gradually increasing, wavelet scalogram can map the drill wear law very good.In view of the wavelet scale figure are obviously different in different wear stages, directly to the wavelet scalogram as state characteristics. Combine convolutional neural network can well identify images, feature extraction and pattern recognition process is completed in the convolutional neural network, eliminating the pretreatment before the pattern recognition process.The collected signal wavelet scale figure, part as the training set, a part as the training set, the network structure is determined by training and test results, including the number of network convolution layers, the size of the convolution kernel and the number of convolution cores.Through the visualization analysis of output map of each layer is completed, it is found that the image pixels are reduced after the layers are convolved and sampled, but different convolution extracts different image features. Multiple convolution cores guarantee the integrity of the image information. The convolution neural network can be good to extract image features. Finally, a well-trained network is used to identify the different wear and tear in the full drilling life cycle of a new drill, and the results are well recognized.