Data-driven Prognostics And Health Management For Rolling Bearing

As an indispensable part of the production and operation of one enterprise,the maintenance and management of mechanical equipment has a huge impact on the economic benefits of enterprises,which is an important part of modern enterprise management.To overcome the difficulties of traditional method of equipment maintenance management that cannot identify the fault timely and accurately as well as cannot predict the remaining useful life of equipment precisely,the maintenance and management of equipment based on PHM technology has become a research hotspot in recent years.This paper proposes a solution that based on vibration signal for the fault identification and remaining useful life prediction of rolling bearing.The proposed solution is consisted of vibration data collection,signal analysis,fault identification and remaining useful life prediction.Firstly,the vibration signal of rolling bearing is collected from experiment platforms,and then,the classical time domain,frequency domain and time-frequency domain signal analysis techniques are used to preprocess the collected vibration signal,finally and most importantly,the intelligent models of fault identification and remaining useful life prediction are constructed to equipment predictive maintenance management.(1)The construction of fault identification model.The features of time domain,frequency domain and time-frequency domain of vibration signal are fused to a feature vector as the input parameters of deep learning models(Stack auto-encode,SAE;deep Boltzmann machines,DBM;deep belief networks,DBN),respectively,and then,the parameters related with deep learning models are deeply analyzed,such as number of hidden layers,the number of hidden units and the times of iteration.The experimental results show that 3 kinds of deep learning models can all get more than 99% fault identification accuracy.Furthermore,compared with DBM and DBN,the SAE can get a highest fault classification accuracy of rolling bearing under the same optimal parameters.(2)The construction of remaining useful life prediction model.the energy values of 3-layer of wavelet packet decomposition are selected to describe the degradation of state-of-health of rolling bearing by comparing with shape factor,mean,impulse factor,and kurtosis.To overcome the influence of SIR(Sampling importance resampling)particle filter on prediction accuracy resulting from ignoring the latest observation,this paper use energy values to train the state-space model and utilize auxiliary particle filter(APF)to prediction remaining useful life of rolling bearing.Finally,the performance of predictors is evaluated by mean absolute error.The experimental results show that the predictor based on auxiliary particle filter algorithm has lower mean absolute error than SIR particle filter predictor,so the APF-based predictor has stronger prediction performance.Overall,the proposed solution can identify fault effectively and predict the remaining useful life accurately for the key component of equipment.Therefore,it can provide equipment management decision support for manager.