Equipment Fault Diagnosing Algorithms Based On Vibration And Acoustic Signal

In the industrial production,the health condition of equipment is positively linked to production safety and quality.The researches on fault diagnosis have special meaning in theory and application as the technique is an important guarantee of equipment protection and accidents prevention.For traditional fault diagnosis methods,the equipment status is estimated based on current,temperature,vibration or other signal analysis and expirical parameters.The equipment fault diagnosis method based on vibration signal became the research hotspot as vibration signal is pointed and better relatedness to fault point.In this paper,in consideration of extreme detecting conditions such as high temperature and corrosion will restrict vibration signal acquisition,non-touching acoustic signal acquisition is presented for faut diagnosis.Aiming to the challenges in practical fault diagnosis based on vibration and acoustic signal,fault diagnosis algorithms have been studied in this work.For traditional fault diagnosis methods,the equipment status is estimated by manual observation or condition parameters feedback generally,which gives lot of room to improve real timing and precision.Although the fault diagnosing accuracy improved based on the combination of traditional diagnosis method,signal analyzing and machine learning methods,there are still challenges in practical application,such as environmental noise interference which will lead to blur fault characteristics and loss of fault classification accuracy.Meanwhile,under the condition of insufficient known fault data and fault patterns,anomaly conditions can not be detected accurately by the fault classification methods.Besides,vibration signal acquisition has some limits in practical application,such as high temperature and corrosion,and the signal-to-noise ratio is relatively low under some circumstance like vibration on outer shell large scale equipment,which will affect the detecting accuracy.To solve these problems,and taking the needs of equipment and key components health monitoring in practice into account,equipment fault classification and anomaly detection methods are studied in this paper,the main researching problems and research content of this paper are summarized as follows:1.An equipment fault classification method combinded fault characteristics enhancement based on spectral kurtosis(SKFE)with dual-stream 2-dimension convolutional neural networks(DS2DCNN)is proposed to address the fault diagnosis problem under noise interference.The fault classification algorithm SKFE-DS2DCNN can be summarized as:(1)Fault characteristics are enhanced by the presented SKFE method,the information unconnected with fault characteristics are suppressed.(2)The relevant information between frames are include in the feature map by the dual-stream 2D feature map extension methods,which leads to more accurate fault classification result.(3)Comparing to 1D fault feature and previous fault classification methods,DS2DCNN can improve the classification accuracy under Gaussian noise influence by feature fusing and deep feature extracting,and the presented method can classify faults accurately under the interference of complex industrial environmental noise.2.In response to the problem of fault classification accuracy affected by strong Gaussian noises,a lower order moment spectra(LOMS)based fault classification method and a spectrogram local fluctuation feature(SLFF)based fault classification method are proposed in this paper.The fault classification accuracy is improved by extracting more effective and robust feature of equipment signal to against the interference of Gaussian noise.The research content of this method can be summarized as:(1)In the LOMS,the differences of fault characteristic frequency are improved by lower order moment spectra,and the noise interference are reduced,the classification result is improved.(2)In the SLFF,the fluctuation indicators describe the basic shape of waveform and the variation tendency of spectral vectors in spectrogram,which improve the accuracy and robustness of feature expression under strong Gaussian interference,and further improve the fault classification result.3.Aiming at the problem of inaccurate fault classification result caused by insufficient known fault data and fault patterns,a multi-branch hierarchical gaussian model(MBHGM)and an antibody population optimized artificial immune system(APO-AIS)based anomaly detection method are proposed in this paper,the propose anomaly detection methods can recognize the known conditions of equipment and the unknown anomaly issues accurately.The research content of this method can be summarized as:(1)In MBHGM based anomaly detection method,samples are detected by the presented multi-branch feature Gaussian models and hierarchical score Gaussian model,which can separate high dimension feature vectors by Gaussian distribution model appropriately,the method is more simple,effective and practical than Gaussian mixture model and multivariate Gaussian model.(2)In APO-AIS,the antibody population and recognizing area are optimized by the proposed antibody selecting,screening method and recognizing area determining method,which improve the quality of antibody population and the freedom degree of recognizing area for accurate anomaly detection.Meanwhile,aiming at different objective conditions of practical diagnosis tasks,diagnosis methods proposed in this paper are appropriate for vibration and acoustic signal.Compare to vibration signal,fault diagnosis based on acoustic signal makes up for the shortage of vibration signal acquisition,widen the applicable scene of fault diagnosis methods.In this paper,a public bearing vibration dataset and equipment acoustic signal collected by high precision microphone are applied to evaluate the effectiveness and robustness of the proposed fault classification and anomaly detection algrithms.