Research On Condition Monitoring And Life Prediction Of Ball Screw Pair System Based On Deep Learing Hybrid Model

Ball screw system has been widely used in high-precision CNC machine tools,aerospace equipment and deep-sea equipment because of its high transmission efficiency,low environmental load influence and strong adaptability to working conditions.It has been used as a key core component for accurate transmission of motion and power in high-end precision mechanical equipment in manufacturing industry.The fault and performance degradation of the ball screw system directly affect the running accuracy and life of high-end precision equipment and the processing quality of the workpiece.However,the previous research mainly focuses on the structural design and manufacturing,dynamic analysis,performance testing and other aspects of the ball screw system.In order to ensure the safe,stable,efficient and precise operation of the ball screw system,grasp its health status in real time,and then extend its service life,it is imperative to conduct in-depth and systematic research on the condition monitoring,fault diagnosis,performance degradation and life prediction of the ball screw system during operation.This paper proposes a method of condition monitoring and life prediction for ball screw system based on deep learning hybrid model.It focuses on the fault mechanism of ball screw system,the design of multi-sensor condition monitoring test bench,the monitoring and processing of multi-source heterogeneous information,the condition monitoring and life prediction of key components of the system,and the design of life cycle intelligent monitoring and maintenance system.Based on the fault mechanism,dynamic performance analysis and simulation results of the ball screw system,a multi-sensor condition monitoring test bench for the ball screw system was developed.The multi-source heterogeneous information monitoring and processing system of vibration,noise and temperature of ball screw pair system is designed,which provides experimental verification for the theoretical research of condition monitoring and life prediction of ball screw system based on deep learning hybrid model.Aiming at the problem that the traditional degradation index cannot effectively judge the abnormal points of the early degradation of the ball screw system,a signal reconstruction noise reduction method based on ICEEMDAN-WS is proposed.This method preprocesses the collected original vibration signals,and uses the ICEEMDANWS noise reduction method to extract the health indicators that reflect the early degradation state of the ball screw pair.The extracted health indicators are input into the CNN-GRU hybrid model to realize the prediction of the degradation trend of the ball screw pair and the monitoring and diagnosis of early fault points.In order to solve the limitation of single signal reflecting degradation information,a monitoring method of ball screw system condition based on multi-source heterogeneous information and multi-domain feature fusion is proposed.This method extracts multidomain features from multi-source heterogeneous information of the running state of the ball screw system collected by multi-sensors after preprocessing such as filtering and denoising.The multi-domain characteristic parameters are used to comprehensively characterize the running state of the ball screw system.The DCNN-SVM network model is established,and the multi-domain characteristic parameters are input to realize the condition monitoring of the ball screw system.In view of the fact that there are many types of ball screw faults and there are obvious uncertainties,the data collected by a single sensor has certain limitations.Combining the advantages of various intelligent identification methods,a fault diagnosis method based on DCNN-IDST multi-model decision fusion is proposed.The DCNN model is used to adaptively extract the multi-source heterogeneous information of multiple sensors.After normalization,it is used as the basic probability distribution function of the D-S evidence theory based on the improved fuzzy consistent matrix.The decision-level fusion of multiple models realizes the quantitative evaluation of uncertainty in the ball screw system and effectively improves the accuracy of fault diagnosis.Aiming at the problem that the single network model has poor generalization ability under variable working conditions and is prone to multiple performance degradation caused by multiple failure modes,a hybrid prediction model of ball screw life based on DITCN-ABi GRU under variable working conditions is established.Improve the TCN network to realize multi-domain feature convolution extraction;the global attention mechanism is introduced to obtain the sensitive feature quantity.Based on the Bi GRU network,the dependence information of the completely training process on the variable working conditions of the ball screw pair is captured.The health index HI is constructed by polynomial fitting to predict the failure point and life trend of the ball screw pair.The modular system of life cycle intelligent monitoring and maintenance of ball screw system is designed,which provides a practical method and means for the health monitoring and predictive maintenance of key components of ball screw system.