Research On Tool Condition Monitoring Method Based On Multi-source Signal Feature Fusion

During the machining process,the interaction of the tool,the workpiece,and the chips will transfer the material particles to the chips or the workpiece,causing the tool to wear,changing the tool state and affecting the machining accuracy of the workpiece.Research on intelligent monitoring and diagnosis methods that can accurately grasp the service status of tools in a timely and accurate manner is of great significance for improving the machining accuracy and efficiency of parts and extending the service life of tools.Taking the milling cutter as the research object,the signal acquisition,analysis and processing of the milling process are carried out,and the research on intelligent online monitoring of tool wear status is carried out.The main research contents are as follows:Online real-time collection of sound and vibration signals when the tool is in different stages of wear under variable working conditions;orthogonally arranged sensors are used to achieve a comprehensive and accurate description of the vibration signals during the cutting process,which solves the impact of the installation direction of a single sensor on information integrity.The Fourier transform method is used to map the time domain signal to the frequency domain,and the signal component distribution characteristics of the tool in the cutting state and the noncutting state are analyzed,and the low-frequency interference frequency band in the vibration signal is clarified.Based on the analog Butterworth low-pass filter model,the mathematical expression of the digital Butterworth high-pass filter is derived by using the denormalization method combined with the bilinear Z transform.Select appropriate filter parameters for the low-frequency strong interference signal in the vibration signal,and use the maximum flatness in the passband of the digital Butterworth high-pass filter to reduce the noise of the vibration signal;use the period translation autocorrelation coefficient calculation method proposed in this dissertation,The optimal filter decomposition layer number of the signal for nonlinear wavelet threshold denoising is determined.The sym5 wavelet basis function is used to eliminate the high frequency and random interference components that overlap the signal-to-noise frequency bands in the sound and vibration signals.By analyzing the component distribution in the signal of the tool under different wear conditions,using statistical methods,the first six-order principal components of the sound and vibration signals are obtained.The eigenvalues of the principal component amplitude ratio,the time domain,the frequency domain,and the energy ratio eigenvalues of different frequency bands decomposed by the wavelet packet in the sound and vibration signals are extracted as the original feature set for feature optimization.Optimizing sensitive features based on GA-RBF method to achieve feature dimensionality reduction.The chromosome genes are encoded in binary code,the RBF network model is used to calculate the chromosome fitness value based on the mapping relationship between the characteristic value and the tool wear level,and the recognition accuracy is used as the optimization goal.The population size and genetic algebra are set,and the results are found by iterative calculation It is suitable for the optimal feature combination under different cutting parameters.The feature value is preferably able to find sensitive features with high recognition degree and strong characterization ability,eliminate redundant features,reduce network input parameters,and improve its recognition accuracy and recognition speed.The two-dimensional convolutional neural network is used as a feature fusion neural network and a pattern recognition classifier,and the extracted feature values are arranged at equal intervals to form a feature matrix sample as the network input,and the network output is the degree of tool wear.Using the self-adaptive layer-by-layer feature recognition and extraction capabilities of the convolution operation,the feature-level fusion of multi-source signals and the recognition of the state of different wear levels of the tool are realized.The experimental results show the superiority of the multi-source signal eigenvalue fusion method and the validity of the established tool wear status identification theoretical method.In the stable wear stage,by monitoring the wear state of the tool to optimize the processing parameters and correct the tool to ensure the quality of the workpiece,prolong the service life of the tool,and reduce the cost of the tool;when the tool enters the severe wear stage,it will be detected and alarmed in time to avoid tool breakage and solve The problem of determining the best replacement time of the tool in the intelligent manufacturing process reduces the scrap rate of the workpiece,thereby reducing the production and manufacturing costs.