Monitoring Method Of Tool Wear Condition Based On Vibration And Power Signals Fusion

With the rise of personalized and quality demands of industrial products,the demand for intelligent and efficient machining of NC machine tool is increasing.The tool is responsible for the cutting task of machine tool,and its wear state has a decisive influence on machining efficiency and parts quality.Hence,accurate and reliable monitoring of tool wear is necessary for the machining quality and efficiency,and also the inevitable requirement to realize the intelligent manufacturing.At present,information technology and manufacturing industry deeply integrated development,nc machine tool processing produces industrial big data flow,with the help of modern data analysis methods,such as machine learning and its branch deep learning,industrial big data flow can be used for tool wear condition monitoring.After discussing the research status of tool wear monitoring methods,this paper studied tool wear monitoring through signal acquisition and processing,feature extraction and selection,machine learning and deep learning network modeling.The main details as follows:Firstly,the power instrument and dynamic signal acquisition analysis system were used to collect the vibration and power signals in the process of processing,the processing work such as noise reduction was excuted,and then the features of signal are extracted.The vibration and power signals in the designed experiment were collected;the vibration signals were processed by wavelet threshold denoising and other methods,and the power signals was processed according to the power behavior of the machine tool;37 features were extracted from the unidirectional vibration signal analyzed by time domain,frequency domain and wavelet,10 time domain features were extracted from the power signals;121 features collected from vibration in 3 directions and power signals constituted the feature list.Secondly,the signal features for tool wear condition monitoring of single cutting parameters were selected by constracted feature selection system,the tool wear condition monitoring model was constructed by the differential evolution gray wolf optimization support vector regression(DE-GWO-SVR),bidirectional long-short term memory network(Bi-LSTM)and depth residual network(DRN),and then the monitoring effect were compared.33 vibration and 3 power signal features were selected by using the feature selection system based on Fscore and Pearson correlation coefficient;the tool wear condition monitoring models were established by using DRN network,the improved DE-GWO-SVR based on iterative mapping and exponential convergence factor,the Bi-LSTM network base on bayesian optimization(BO)and signal features training set;the monitoring effects of different model were compared based on the same evalution index.The results show that the Bi-LSTM and DE-GWO-SVR model have high accuracy,DE-GWO-SVR model consumes shortest prediction time,DRN model has low accureay,but it is less affected by network depth,and the training speed of model is fast.Finally,the signal features for multi cutting parameter tool wear condition monitoring were selected,the effects of DE-GWO-SVR model,DRN transfer model and Bi-LSTM transfer model in multi cutting parameter tool wear condition monitoring were further researched by introducing Transfer learning(TL),and a tool wear condition monitoring system was designed.25 vibration and 3 power signal features were selected by using the feature selection system under the condition of multiple cutting parameters;the monitoring effect of multi-cutting parameters of DE-GWO-SVR model and the influence of cutting parameter features on monitoring effect were reasearched;the Bi-LSTM and DRN model were applied to multi cutting parameter tool wear condition monitoring by TL.The analysis show that the TL combined with DRN model has the best effect in monitoring tool wear with multiple cutting parameters.After completing the model construction,a tool wear condition monitoring system including system operation and data display area was designed based on the Appdesigner module of Matlab software,the operability and visualization of the research content were realized.The tool wear prediction errors of the proposed model are basically within± 0.02mm,which proves that the proposed method has high accuracy and stability.The method provides technical support for accurate monitoring of tool wear during automatic machining process,thus improving machining quality and efficiency and realizing intelligent manufacturing.