Research On Tool Wear Monitoring Based On Deep Learning

With the proposal of the Made in China 2025 plan,the manufacturing industry is gradually upgrading in the direction of intelligent manufacturing,leading to the gradual development trend of Computer Numerical Control(CNC)machine tools toward intelligence,automation and efficiency.The tools as a direct execution part of the workpiece forming can affect machining accuracy,and have an indirect impact on processing efficiency and production costs.Therefore,the real-time monitoring of tool wear state is of great significance for improving product quality,avoiding machine tool damage,and improving manufacturing technology and service levels.In this paper,the research object is the wear state of the milling cutter in the CNC milling process,the vibration signal generated during the cutting process is the monitoring signal,and the deep learning network model and data processing algorithm are combined to identify the wear state of the tools.The main research contents of are as follows:(1)The formation and characteristics of tool wear were discussed.The vibration signal collected during the cutting process of the tool was used.The signal-to-noise ratio was used to determine the selection of wavelet base.Based on the analysis of the signals in the time and frequency domains,wavelet packet technology was used to extract the frequency band energy characteristics of the signals,and it's determined that the frequency band energy characteristic value distributions of different tool wear stages were also different,which was a deep learning network model.(2)By labeling the energy feature samples,a one-dimensional convolutional neural network(1D CNN)model was established to learn features,and the structure and parameter tuning of the 1D CNN model were deduced and analyzed in detail.The feasibility of the model was analyzed.Finally,different deep learning models were used to predict the test results,and verified the effectiveness of the 1D CNN model and the high accuracy characteristic.(3)Aiming at the inefficiency of the traditional signal feature extraction process,a tool wear monitoring model based on a one-dimensional residual convolutional neural network was proposed.Using its advantages of adaptively extracting spatial features,the frequency domain data containing a mapping relationship between high-dimensional features and tool wear status was established.The one-dimensional residual convolutional neural network model was used to recognize tool wear state,and the network structure of the model and the parameter optimization process were described.First,the validity and time-consuming of the model are verified by comparative experiments.The results show that the tool wear status can be identified effectively and with low time-consuming.At the same time,the adaptability of the model is verified to obtain better results.(4)In order to realize the recognition of the tool wear status,a tool wear state monitoring system was developed based on the Lab VIEW and the MATLAB.This system was tested by software then.