Feature Extraction And Fault Diagnosis Of Rolling Bearing Acoustic And Vibration Signal Based On Differential Evolution

With the rapid development of industrial Internet of Things,big data and artificial intelligence technology,mechanical equipment fault diagnosis has also entered the intelligent era.As a key component of mechanical equipment,rolling bearing condition monitoring and fault diagnosis can ensure the safe and stable operation of the equipment.The traditional condition monitoring is mainly based on vibration signal.However,acoustic signal also contains important information about the running state of rolling bearing.Moreover,the acoustic and vibration signals have complementary characteristics and have their own advantages in signal acquisition methods.Therefore,the health state of bearing can be understood by acoustic and vibration signal integrating analysis.In recent years,signal processing technology,intelligent evolutionary algorithm and machine learning theory are also developing rapidly,their provide theoretical basis for the research of rolling bearing fault diagnosis method based on vibration and acoustic signals.This dissertation deeply studies the acoustic and vibration signal feature extraction and intelligent fault diagnosis of rolling bearing under complex working conditions based on differential evolution algorithm,signal adaptive decomposition,blind signal processing and machine learning.The main research contents can be summarized as follows:1.The standard differential evolution algorithm has some problems,such as lack of diversity,easy to fall into local optimization,slow convergence speed and difficult to determine the control parameters.In this paper,five improved algorithms are proposed from the aspects of adaptive setting of control parameters,optimization of population structure and mixed use with other intelligent algorithms,which contain multiple population differential evolution(MPDE),simulated annealing differential evolution(SADE),grid search differential evolution(GSDE),multiple population differential evolution based on parameter adaptation(PA-MPDE)and simulated annealing differential evolution based on parameter adaptation(PA-SADE).Then,11 typical standard test functions are used to analyze these improved algorithms.From the results,PA-SADE has obvious advantages compared with other algorithms.Thus,it will be used to solve various optimization problems in the process of rolling bearing acoustic and vibration signal fault diagnosis.2.An improved ensemble empirical mode decomposition(EEMD)algorithm is proposed to detect the impact components caused by early bearing faults and extract fault features.At the same time,an improved variational mode decomposition(VMD)algorithm is used to process the complex and interfering components in the bearing compound fault signal.Firstly,PA-SADE algorithm is introduced to improve EEMD and VMD,which realized the adaptive and optimal setting of decomposition algorithm parameters.Secondly,PA-SADE-EEMD is combined with similarity measure criterion to preprocess early fault vibration signal and PA-SADE-VMD is combined with the average kurtosis criterion to preprocess the compound fault vibration signal.Thirdly,adaptive resonance demodulation method and synchronous compressed wavelet transform are used to extract frequency domain and time-frequency domain feature information.Finally,the simulation signal and measured signal of bearing early fault and compound fault are experimentally analyzed.The results show that this method can effectively extract fault characteristics.3.Blind signal processing theory is applied to bearing noise signal processing and feature extraction.In order to solve the problem of uncertain separation results caused by different time delay parameters,PA-SADE algorithm is used to search the optimal delay in blind deconvolution.Then,Fuzzy c-means clustering(FCM)is also optimized by PA-SADE algorithm to realize the adaptive determination of the number of clusters,which can not only balance the problem of global optimization and local fast convergence,but also improve the performance of FCM.Finally,an improved time-domain blind deconvolution algorithm is proposed based on the above research.The experimental results show that this method can effectively extract and separate the fault features of bearing noise signal.4.In order to realize intelligent fault diagnosis of rolling bearing,deep learning,unsupervised learning and transfer learning are applied to identify fault patterns under different conditions.Firstly,the fault diagnosis method based on improved HHT algorithm and PA-SADE-CNN,which is established by optimizing the AlexNet convolutional neural network,is proposed for the identification of bearing compound fault.Secondly,the fault diagnosis method,which is combining the deep convolution generative adversarial network in unsupervised learning and VGGNet network,is proposed aiming at small sample fault identification for imbalance between fault classes.Finally,in order to share the fault characteristic information of vibration and noise signals,the fault diagnosis method combining deep transfer learning and deep residual network is proposed.It can unify the bearing vibration and noise fault diagnosis methods in the same framework.Through experimental analysis,the results show that these methods can effectively identify fault types in different situations.