Intelligent Fault Diagnostic Methods For CNC Spindle Bearing Under Variable Conditions

Bearing is one of the core components of the machine tool spindle,and its health condition will directly affects the spindle's operating state.Once the spindle bearing fails unexpectedly,product quality loss or production rhythm chaos occurs.Therefore,real-time monitoring of the spindle bearing condition is a necessary step to achieve reliable production.Spindle bearing operating conditions are complex and changeable.It is not possible to obtain bearing fault data sets in full operating conditions during production.Intelligent diagnostic models established based on data of specific operating conditions(source domain)are not capable of achieving good diagnosis performance under new operating condition(target domain).It is of great practical significance for the realization of intelligent operation and maintenance and intelligent manufacturing of machine tools to deeply study the effective intelligent diagnosis scheme of the spindle bearing under variable working conditions.In order to solve the problem of spindle bearing fault diagnosis under variable operating conditions:A spindle bearing experiment was designed and conducted.This experiment uses a commonly used CNC machining center of a company as an experimental platform to obtain bearing fault data sets under variable load and speed.Through comprehensive analysis of the experimental data in time and frequency domains,such conclusion is obtained: The amplitude range and dimensional characteristics of the signal is sensitive to working conditions as well as bearing health condition.Dimensionless features are less sensitive to working conditions,at the same time,they are not strongly discriminatory to bearing health condition.Signal rotation frequency and its multiplication are strongly correlated with working conditions.A fault diagnosis model based on Convolutional Neural Network is established.Models based on Multi-layer Perceptron,Stack Autoencoder and Recurrent Neural Network are used as comparison methods to carry out multiple sets of cross-load and cross-speed diagnostic experiments to verify the superiority of the diagnostic method based on Convolutional Neural Network.It is also found from the experimental results that the intelligent diagnosis model has good load generalization performance,but the generalization performance of the cross-speed is greatly reduced as the training conditions are reduced.An intelligent diagnosis method based on Feature-alignment Convolutional Neural Network is proposed.In order to solve the problem of poor model generalization performance caused by the difference of local feature positions of different samples,this method optimizes the design of convolutional layer and pooling layer,and cooperates with the global average pooling layer to achieve feature alignment between samples.Experiments show that the proposed structure has higher diagnostic accuracy,higher diagnostic confidence,and better feature robustness than conventional models,thus improving the generalization performance of the operating conditions.A cross-condition diagnosis scheme based on style transfer strategy is proposed.This solution aims at the problem of poor generalization performance of a trained model under a single working condition source domain.The method of generating samples through style migration uses the source domain error samples and the target domain normal samples to generate joint domain fault samples to generate sample-assisted intelligence Diagnosis model training to achieve fault diagnosis in the target domain.Experimental samples,the proposed scheme can achieve effective diagnosis in the target domain under the condition of only a single working condition in the source domain.This paper proposes a variety of solutions for the fault diagnosis of spindle bearings under variable operating conditions,and achieves effective intelligent diagnosis under variable load and variable speed conditions.New solutions are proposed to tackle the problem of lack of fault data in the target domain.