Development of a tool wear online monitoring system for dry gear hobbing machine based on new experimental approach and DAE-BPNN-integrated mathematic structure

The hob wear and tear of hob adversely affect the stability of dry gear hobbing accuracy. Quick and accurate online recognition of current hob wear state is crucial to curb this type of machining error during the continuous gear hobbing process. For most of CNC machine tools, an in-depth analysis of the spindle power consumption signal is the common way to monitor the tool wear condition. However, the variation of the spindle current signal is not only related to tool wear state but also the thermal-induced error and the different machining allowances of workpieces to be cut. Therefore, it is barely possible to find one or two individual signal features to precisely characterize the change of hob wear state. In this study, to find the features with strong characterization abilities, we proposed a new tool wear experiment strategy to collect the power consumption data and the thermal-induced error in the entire thermal deformation developing period as well as the representative value of different machining allowances of workpieces. Based on collected spindle power consumption data, we indiscriminately extracted both the time-domain and the frequency-domain features and formed a high-dimensional 47*100 feature dataset. Subsequently, some extracted features were excluded by conducting a correlation analysis with the help of collected thermal error data and machining allowance data, and the rest was further compressed to nine new synthetic features with a dimension-reduced treatment by using the deep autoencoder algorithm. Finally, a mapping model between the synthetic features and different hob wear states was developed by adopting the back propagation neural network classification algorithm. To prove the practicability of this mapping model, another gear cutting experiment was conducted and the results showed the accuracy of this hob wear state online monitoring approach can reached to over 90%.