Bearing Remaining Useful Life Prediction by combining CNN with PSO_LSSVM

The remaining useful life (RUL) for bearings is of crucial importance to ensure system availability and reduce maintenance costs. In this article, a novel approach combining Convolution Neural Nets (CNN) with Particle Swarm Optimization Least_Squares Support Vector Machine (PSO_LSSVM) is adopted to predict the RUL of bearings. To be specific, firstly, the Relative Root Mean Square (RRMS
norm
) not affected by individual differences is calculated as training label to depress noise in raw vibration signals. Then, the CNN is trained by the raw data and its training label, which makes it possible to extract a new degradation feature. From the new degradation features, the prediction model based on the PSO_LSSVM is constructed to predict the RUL of the bearings. Note that Particle Swarm Optimization (PSO) is introduced to automatically optimize the important parameters of Least_Squares Support Vector Machine (LSSVM), which is a contribution of the proposed methodology. Finally, the performance of the proposed method is verified by actual vibration data from the experiment platform.