State Recognition And Residual Life Prediction Of Tool Wear Based On Data-driven Model

The proposed "Made in China 2025" plan has brought about the vigorous development of the manufacturing industry.The CNC machine tools,which occupy an important position in the manufacturing industry,are also developing rapidly toward the target of intelligent and high-quality processing.As the key parts of NC machine tool,the wear state of the tool directly affects the quality of the workpiece and the efficiency of the machine tool.Therefore,the study of tool wear is of great significance to guarantee the machining quality and improve the efficiency of equipment.With the rapid development of deep learning and signal monitoring technology,data-driven tool wear state recognition and residual life prediction based on tool milling signal data have been widely explored.This paper is based on PHM 2010 high speed milling tool dataset to study this aspect.Firstly,the raw data of the tool is analyzed by a series of steps including preprocessing,feature extraction and selection.The invalid values and outliers in the original data were removed by truncation and filtering,and the time-frequency domain features were extracted by combining EMD and SVD.Then,the 154 extracted multi-domain features were calculated by taking the intersection of MI and 32 highly sensitive features were selected to form the feature space.Secondly,the method of tool wear state identification is discussed.Considering the complex feature engineering and low recognition accuracy of machine learning method,this paper innovatively applies deep learning 1D-CNN network to tool state recognition.This model is capable of adaptive feature extraction and final state recognition.By comparing it with other commonly used machine learning recognition models on experimental data,the results show that the 1D-CNN recognition model has higher reliability and accuracy.Then,the method of predicting residual life of tool wear is explored.In order to ensure the overall reliability of residual life prediction,this paper establishes a model from two aspects of wear value and the number of remaining tool runs.In this paper,a method combining CNN network and BILSTM network with the Attention mechanism is proposed to automatically extract features and build a deep model to complete the prediction.Finally,a variety of commonly used machine learning prediction models and deep learning prediction models were selected to compare,and the stability and high precision of the proposed method in predicting the residual life of the tool were proved.Finally,the tool wear monitoring system is designed.In order to better promote the application of the research content in this paper,PyCharm programming interface is used to visualize all the research.Based on the tool milling data of PHM 2010,four modules of the system,including user login,data analysis and processing,wear state recognition and residual life prediction,were realized and verified.