Design And Implementation Of Tool Wear Detection System Based On Deep Learning

As machining becomes more intelligent,automated,and efficient,mechanical fault detection becomes an indispensable part,and the tool is the most important part in machining,but it is difficult to detect the wear state of the tool during high-speed milling,besides,when the tool wear is severe,it will affect the machining accuracy and production cost,and will indirectly affect the machining efficiency and product quality.Many scholars are eager to try to realize the online detection of tool wear more efficiently,more accurately and more steadily,so the research on the detection technology of tool wear state has very high application value.This paper takes online tool wear detection as the research object.By collecting data and analyzing and learning the data,a tool wear detection method based on 3-KMMBS is proposed,and a tool wear detection system based on deep learning is built.The main research contents of the paper are as follows:1.First,collect the vibration and acoustic emission signals in different axes during high-speed tool milling,and apply wavelet analysis to preprocess the data,including wavelet packet transformation of the vibration signal,then make use of the extracted approximate coefficients and detail coefficients to determine whether it has singularity;Cepstrum analysis is performed on the original signal to determine that the different wear levels of the tool have different AE values,and the improved 3-K-Means clustering algorithm is used to cluster the three wear state intervals of the tool.2.Secondly,based on labeled samples and unlabeled samples,a multi-select multi-hidden layer neural network structure is established to perform feature learning,and then make use of the Softmax classifier to classify;among them,the multi-select multi-hidden layer neural network contains a variety of hidden Neural network with a layer structure,which provides support and learning for different data sources,and the Softmax classifier is used to classify the output of the neural network to obtain the label value of tool wear;the label value in the sample will be used to compare the accuracy of each branch in multiple choices and multiple hidden layers neural network,and the deepest network with the highest accuracy will be selected.3.Meanwhile,according to the existing part of the labeled samples,apply random gradient descent to fine-tune the parameters of the selected deep network to establish a tool wear detection model;4.Finally,the model is integrated into the system to establish a tool wear detection system based on 3-KMMBS.The final experimental results show that the method proposed in this paper not only has stronger learning ability and less human intervention than the traditional shallow model,but also has higher and more stable detection accuracy.At the same time,the system design complies with the system design principles,and is more humane,smarter,and more efficient than other tool wear detection systems.