ADMOW Pattern Recognition Method And Its Application In Fault Diagnosis Of Rolling Bearings

Rolling bearings are important components of rotating machinery.They play an important role in supporting shafts and on-shafts parts,maintaining the normal working position and rotation accuracy of the shafts in mechanical equipment,and are widely used in various fields of industry.Statistics indicate that rolling bearings are one of the most vulnerable parts of rotating machinery.Their working conditions directly affect the running of the entire equipment and even the production line.Therefore,it is of great significance,for the stable operation of mechanical equipment,to monitor and diagnose the working states of rolling bearings by using effective methods.However,in the fault diagnosis of rolling bearings,the real world collected vibration signals often can not reach the ideal stable state due to the noise of the field environment,the measurement error and other factors and usually express non-linearity and non-stationary.Meanwhile,some of the extracted characteristic data will be far from the group,which means outliers.These anomalous data have different degrees of impact on processing modeling,resulting in deviations between the established model and the actual situation,and then affect the accuracy of fault identification.In order to overcome the above deficiencies,an Agent Discriminate Model Based Optimization Weighted(ADMOW)is proposed.In the method,certain weights,determined by the contribution degree of each characteristic parameter,will be assigned to the sample eigenvalue,so as to weaken the model deviation,caused by the abnormal data.And then the Kriging agent discriminantion model is established according to the mutual intrinsic relationship between the eigenvalues.What's more,the parameters of the model will be optimized by the optimization algorithm to further weaken the influence of anomalous data.The emulational and experimental data analysis indicate that the proposed method has a good effect on the fault classification of rolling bearings and also provides a reference for other pattern recognition problem of outliers in eigenvalues.The main research contents of this paper are as follows:(1)Based on the Kriging model,a agent discriminant model is established.In the model,that the eigenvalues of the same category of samples have the same intrinsic relationship and the eigenvalues of different categories of samples have distinct internal relationships,are taken into consideration.Based on the relationship between the eigenvalues,a linear or nonlinear mathematical model is established,in which the prediction result of the eigenvalue is used as the classification basis for pattern recognition.(2)According to the outliers in the characteristic data set,the characteristic evaluation method and the entropy weight method are used to evaluate the contribution degree of each characteristic parameter,by which the characteristic value is weighted.The weighted model isapplied to the experimental data analysis of rolling bearings.The results suggest that both the characteristic evaluation method and the entropy weight method can improve the stability of the characteristic data,weaken the influence of outliers on the modeling,and improve the fault recognition rate of rolling bearings.In particular,the weakening effect of the former method on outliers is more obvious.(3)The particle swarm optimization algorithm is introduced to the agent discriminant model so as to optimize the parameters of established model.The rolling bearing experimental data analysis show that the parameter-optimized model can further weaken the influence of outliers and obtain a classification model that is more closer to the real world situation,and improve the classification accuracy and fault recognition rate.