Research On Tool Data Anomaly Detection Method Oriented Processing

Abnormal detection of tool status in the machining process can ensure the safety and stability of tool processing.The traditional tool data detection method is relatively simple,and the traditional 3sigmod method in the industry is used for calculation,and abnormal data is obtained,and then processed.Such a processing method may prolong the detection time,thereby shortening the tool life and more serious wear.How to detect abnormal tool data using artificial intelligence methods has become the key to tool state detection in the machining process.Aiming at the characteristics of tool data,this paper proposes an LSTM-based abnormal detection of tool single-sequence data and the prediction of multi-dimensional tool data.The association analysis method is used to analyze and predict multi-sequence data.The first is for the cutting force data of a single sequence,by introducing the LSTM neural network algorithm,and setting the number of neurons and the number of neural network layers,using the existing data to set the parameters first,and adjust the number of iterations,sliding window parameters,etc.After reaching the best,use the LSTM model for training,and then calculate the difference between the predicted result and the true value,obtain the difference probability density function value,compare it with the set normal threshold,and get the abnormal value.The experimental results show that,This method obtains the abnormal value of the cutting force,and the result is more accurate.The next method is to calculate the correlation of the tool data.Since the collected tool data is not only one-dimensional data,but also other aspects of data,in order to better predict the tool data,first use the Pearson coefficient calculation method to calculate the correlation between the sequences to obtain a strong correlation Matrix,put the data in the strong association matrix into the trained LSTM model to train the prediction,and compare the result with the single-sequence prediction result.The result shows that the prediction accuracy of multiple sequences is stronger than the prediction accuracy of single-sequence,and The loss function is smaller and the accuracy is higher.After combining the neural network algorithm with the traditional 3sigmod and time correlation analysis,the abnormality that may occur in the future can be effectively predicted,which can effectively avoid the problems in the processing process and improve the processing efficiency.