Research On Rolling Bearing Fault Diagnosis And Prediction Method Based On Support Vector Machine

As an important component of the machine,the state of the rolling bearing has an important impact on the operation of the entire mechanical system.If the bearing fails,the mechanical system will deviate from the normal working state,and even cause downtime,production stoppage,and mechanical damage.Therefore,it is particularly important to carry out the fault diagnosis and prediction research of rolling bearings,which plays an important role in ensuring the long-term and stable operation of the mechanical equipment.Through the acquisition of bearing signals,feature extraction and intelligent analysis we can detect or predict the faults occurring in the bearing as soon as possible.Measures such as repairing or replacing bearings can be taken to avoid a series of production accidents and unnecessary economic losses caused by bearing failures.In order to solve the problem of slight fault diagnosis of the bearing and improve the accuracy of fault diagnosis,this paper uses the time domain parameters and frequency domain parameters selected by distance evaluation technique to combine with EMD energy moment and features extracted by EMD-SVD method to form multi-domain feature.The multi-domain feature can discover the fault information hidden in dynamic signals.Based on the established multi-domain feature set,the Gaussian kernel function support vector machine is used to diagnose bearing faults.The analysis of the experimental data shows the diagnostic accuracy rate of the bearing fault diagnosis with the multi-domain feature is as high as 98.75%,which is higher than the fault diagnosis accuracy rate with any single feature.In addition,this paper also analyzes the sensitivity of single features to different faults.In the practical industrial applications,the number of normal bearing samples is far more than the number of fault bearing samples.The sample imbalance will cause the classification hyperplane of the support vector machine to tilt,resulting in inaccurate model classification.A similar problem also exists in other pattern recognition methods.In order to solve this problem,a two-step fault diagnosis model is proposed,which combines the anomaly diagnosis method based on feature extraction method with symbolization method of probability density space partition and the support vector machine(SVM).The experimental data analysis shows that the correct rate of fault diagnosis of the two-step diagnostic model is 3.34%higher than that of the single-step fault diagnosis based on support vector machine.In addition to the above imbalance between the normal bearing samples and the faulty bearing samples,there is also a sample imbalance problem in fault samples.Based on the two-step diagnostic model,the paper generate "fewer samples" fault samples by genetic operation and some of the generated samples are added to the original samples to achieve the balance between samples with different faults.Then the support vector machine is trained with the balanced samples.The weighted Euclidean distance is used to select the generated"fewer samples" fault samples.Since the feature parameters with larger contribution to the classification are endowed with larger weights,the samples slected by this method are more conducive to establishing accurate classification model.This paper refers to the proportion of bearing fault samples in industrial applications proposed by previous scholars,and selects the rolling element fault bearing samples and the inner ring fault bearing samples to form an unbalanced sample set.Experiments show that the fault diagnosis accuracy is 90%by directly training the support vector machine with unbalanced samples.The fault diagnosis accuracy is 100%by training the support vector machine with balanced samples,which is increased by 10%.This paper proposes a combined prediction model based on support vector machine with IMF component values as input features.The model firstly decomposes the vibration signal by EMD,and divides the IMF components into two groups according to the degree of fluctuation:the high frequency vibration sequence and the low frequency vibration sequence.Then the two groups are respectively synthesized into a high sub-frequency sequence signal and a low sub-frequency sequence signal.The high-frequency sub-sequence signal and the low-frequency sub-sequence signal are respectively predicted by the support vector machine with the IMF components as the input features.Finally,the two predicted values are superimposed to obtain the final prediction result.Since the high frequency subsequences and the low frequency subsequences have different trends,it is theoretically possible that the combined prediction model can improve the accuracy of the prediction.This paper uses bearing experimental data to verify the effectiveness of the combined prediction model.The results show that the root mean square error,the average absolute error and the average absolute percentage error of the combined prediction method are much smaller than the direct prediction method.The average absolute percentage error of the direct prediction method is 11.49%,and the average absolute percentage error of the combined prediction method is only 6.54%,which is much lower than the direct prediction method.Experimental data analysis results demonstrate the effectiveness of the combination prediction model.This paper also applies this combined prediction model to the situation prediction of the rolling bearing of the equipment in a coal washing plant,and the forecasting situation is consistent with the actual situation.