| As the core component of the centrifugal pump,the milling quality of the impeller runner directly determines the operating performance of the centrifugal pump.If the tool wear is serious,it will not only reduce the machining quality of the impeller,but also affect the production efficiency and increase the production cost.At present,tool wear condition monitoring(TCM)mainly relies on manual identification,which not only disrupts the processing schedule and causes tool waste,but also has errors and instability.Therefore,how to achieve the online TCM of impeller milling process needs further research.In view of this,the main research contents of this paper are as follows:Firstly,by analyzing the research status of online TCM methods at home and abroad,the machine tool spindle current signal was chosen as the monitoring signal of tool wear condition.According to the basic principle of tool wear,the tool wear condition was divided into three stages: initial wear,normal wear and rapid wear.Based on the three stages of tool wear,tools were prepared in advance,a data acquisition experimental platform was set up,a data acquisition program was written based on Lab VIEW,and experimental procedures were designed.Secondly,the effective current signals and the thermal deformation values of the machine tool spindle were preprocessed to form 156 signal samples.After extracting time-domain,frequency-domain and time-frequency domain features from the signal samples,32 features with strong correlation to tool wear condition were obtained,an original feature set with a feature space of 32*156 was formed.To improve the characterization ability of the tool wear condition,sample sensitivity analysis and deep auto-encoder were used to reduce the original feature set into a low-dimensional feature set with a feature space of 12*116,which achieved the evolution of the feature set from high-dimensional to low-dimensional.Thirdly,a TCM model was built by using back propogation(BP)neural network,the low-dimensional feature set was mapped to the three tool wear states to achieve the purpose of TCM,and recognition error was 11.13%.Considering the limitations of the BP neural network,a method based on the genetic algoritm(GA)to optimize the initial weights and bias of the BP neural network was proposed to imporve the network recognition performance and reduce the recognition error.The results showed that,compared with the traditional BP neural network,the recognition accuracy of tool wear condition of the optimized BP neural network based on genetic algorithm was improved by5.42%.On this basis,a comparative analysis of different feature set construction methods was carried out based on the GA-BP neural network recognition model.The comparison results showed that,compared with traditional feature extraction,the feature set evolution method proposed in this paper can obtain a low-dimensional feature set with strong representation ability for tool wear condition changes,and the recognition accuracy can reach 94.59%.Finally,an online TCM system based on Lab VIEW platform was developed.The system combining a well-trained GA-BP neural network can realize real-time visualization of current signal and tool wear status during the milling process.The application of the system demonstrated the value of the research method proposed in this paper in engineering applications and attained the transformation of scientific research results into practical engineering applications. |
