Health Prognosis Research Of Mechanical Equipment Under Uncertain Data

With the advancement of modern industrial technology,many mechanical devices are becoming more and more complicated.In the event of failure of these devices,not only will it bring serious economic losses,but it will also pose a threat to personal safety.Diagnosing the current health state of mechanical equipment and predicting the remaining useful life can provide a theoretical basis for the development of corporate maintenance strategies.In production,due to human factors and environmental factors such as noise and disturbance,the sample data of health prediction will be uncertain.Uncertain sample data will reduce the ability of data mining,leading to deviations in prediction results,and reliable prediction results as the key to the maintenance of equipment status,to ensure the safety performance of equipment,to develop mechanical equipment maintenance plans and reduce maintenance Costs and other aspects play an important role.Based on the analysis of the current development of mechanical equipment health prediction technology and considering the uncertainty of sample data,this paper mainly does the following three research work:(1)Aiming at the missing data in the sample data,the segmented hidden semiMarkov model(SHSMM)architecture is established.The model expression of SHSMM is described by parameter ?,and the parameters of SHSMM are estimated by EM algorithm.Based on the WGM(1,1)model,a gray heuristic algorithm is proposed to fill the missing data in the monitoring samples.The complete data samples filled with gray heuristic algorithm are input into the SHSMM for health prediction of mechanical equipment.Finally,the effectiveness of the proposed method is verified by case study.(2)Aiming at the abnormal data in the sample data,based on the SHSMM model proposed in(1),in order to maximize the use of sample data information,the abnormal data in the sample data is treated as missing values,and a dynamic forward-backward gray filling method is designed.In the case study,the abnormal data eliminating and non-processing methods were used as comparison objects,and the dynamic forwardbackward gray filling method was proved to have better performance in the health prediction results of mechanical equipment.(3)Aiming at the inaccurate data in the sample data,based on Dempster-Shafer(DS)evidence theory and Markov chain,the DS-MM theoretical framework is established,and the model is reasoned and learned.Establish a state recognition framework and use interval numbers to represent inaccurate data,use distance and similarity between interval numbers as evidence to generate basic probability assignment(BPA),use Pignistic probability transformation to convert BPA into the underlying state probability distribution and perform equipment health predictions.Finally,an application case is used to verify the effectiveness of the proposed method.The above three research contents are closely related to each other.The missing data,abnormal data and inaccurate data in the sample data will affect the results of mechanical equipment health prediction.Therefore,this paper considers these three problems separately and proposes effective solutions accordingly.method.