Research On Rolling Bearing Fault Diagnosis Method Based On Compressive Sensing And Machine Learning

Rotating machinery equipment has been working in harsh environments for a long time.Complex structures and precision processes have caused the failure of parts and the reasons and types of equipment failure.Rolling bearings are extremely susceptible to damage,which indirectly leads to the collapse of the entire rotating machinery system or major dangerous accidents.Bearing fault diagnosis and online condition monitoring is crucial.Traditional mechanical fault diagnosis methods are faced with thorny problems such as massive test data,difficulty in feature extraction,and low classification accuracy.Based on the vibration analysis diagnosis method,this paper combines compressive sensing theory and machine learning model,and proposes three effective methods for rolling bearing fault diagnosis,which reduce the amount of calculation and storage resource overhead,greatly ease the mode classification workload,and effectively improve the fault diagnosis efficiency.The research works are shown as follows:(1)The research status of rotating machinery fault diagnosis and compressive sensing at home and abroad is introduced.The basic structure,failure modes and vibration mechanism of rolling bearings are summarized.The realization method of mechanical fault diagnosis is analyzed.The basic principles of sparse representation,compressive measurement and signal reconstruction involved in the compressive sensing technology are expounded.(2)Proposed a rolling bearing fault diagnosis method based on compressive sensing and heuristic neural network.Because the collected vibration data is often submerged in environmental noise,and taking into account the limitations of traditional wavelet and empirical mode decomposition,the signal pre-processing is implemented by combining the second-generation wavelet transform and the cubic Hermite interpolation-local mean decomposition algorithm.Compressive sensing technology is applied to vibration data dimension reduction and feature extraction.The constructed concise sample sets are sent to the heuristic neural network for fault classification.A large number of simulation experiment results show that the pre-processed vibration signal facilitates fault information extraction,the proposed feature extraction scheme effectively improves fault classification accuracy.(3)Proposed a rolling bearing fault diagnosis method using PSO-SVM classifier based on compressive sensing and neighborhood rough set.For the characteristics of the collected vibration data and the phenomena that are submerged in the noise,so the threshold-based adaptive redundant lifting scheme packet algorithm is used for vibration signal de-noising.Combining compressive sensing technology with neighborhood rough set model is applied to extract and select sensitive features.And then the constructed concise sample sets are input into the PSO-SVM hybrid classifier for fault classification.A large number of simulation experiment results show that this diagnosis method is dominant in the vibration signal de-noising,sensitive feature selection and classification accuracy.(4)Proposed a rolling bearing fault diagnosis method based on hybrid time-frequency analysis and random forest.Based on the fault diagnosis technology of modern rotating machinery and taking into account the characteristics of vibration data,a hybrid timefrequency analysis algorithm is proposed for signal analysis and fault feature extraction.The constructed high-dimensional sample sets are input into the random forest model for fault classification.A large number of simulation experiment results show that this method has good generalization ability and improves fault diagnosis efficiency.