Research On Intelligent Monitoring And Pattern Recognition Of Product Quality And Tool Condition

In manufacturing industry,the processing quality of products is not only related to the benefit and survival of an enterprise,but also closely related to the stable development of other industries and even the country because manufacturing is the foundation of many industries.This topic focuses on the intelligent monitoring of product quality and machining process,and studies some key issues of statistical process control and tool condition monitoring.A control chart and histogram pattern recognition method based on deep learning is proposed.A tool wear prediction method based on deep learning and semi supervised learning is proposed.It provides support for enterprises to improve the automation and intelligence level of quality management,maintain stable processing state and processing quality.Some research results in this paper have been rapidly concerned by scholars in related fields.The main contents of this paper are as follows:(1)In order to solve the problem of control chart pattern recognition in the field of statistical process control,it is necessary to extract all kinds of complex features manually,and the level of quality control automation and intelligence is not high.Based on the classical two-dimensional convolutional neural network,a control chart pattern recognition method based on one-dimensional convolutional neural network(1D-CNN)is proposed in this paper.This method does not need to extract complex features manually,but uses 1D-CNN to learn the optimal features from the original control chart data.Compared with the traditional recognition method based on expert features and multi-layer perceptron,it not only solves the shortcomings of the traditional method,but also improves the accuracy of control chart pattern recognition.(2)In order to solve the shortcomings of the existing histogram pattern recognition research,and to further improve the accuracy of control chart pattern recognition,based on the above research,a control chart and histogram pattern recognition method based on bi-directional long short term memory network(Bi-LSTM)is proposed.This method also belongs to the deep learning method,which has the ability of feature learning and does not need to extract features manually.Because of its recursive algorithm mechanism,it has unique advantages in processing time series data.Compared with other deep learning methods including 1D-CNN algorithm,Bi-LSTM has the best feature quality and pattern recognition accuracy.In addition,the method can effectively identify the abnormal patterns in the actual production data.(3)In the task of tool condition monitoring,it is expensive and time-consuming to obtain a large number of labeled training data.And in the case of limited data,the traditional supervised learning algorithm is difficult to obtain satisfactory results.In order to overcome these shortcomings,this paper proposes a tool wear prediction method based on semi-supervised learning and deep learning,and establishes the regression relationship between multi-sensor signal and real-time wear.In this method,1d-cnn is trained to extract the multi-sensor signal features,then k-Nearest Neighbor(KNN)algorithm is used to mark the unlabeled samples,and finally the pseudo marker samples with high confidence are selected to be added to the training set.A large number of unlabeled data that is easy to obtain is fully learned by the network,so that its potential value can be effectively used.Using the tool wear detection system based on machine vision and PHM2010(Prognostic and Health Management)tool wear data set,the performance of the proposed tool wear prediction method is verified by experiments.The experimental results show that the prediction accuracy of this method is significantly improved compared with that of the small sample supervised learning method.