Remaining useful life estimation of rolling bearings based on sparse representation

This paper proposes a novel methodology of rolling bearings remaining useful life (RUL) estimation based on sparse representation theory. By analyzing the inner relationships among monitored data, three particular properties of remaining useful life estimation tasks are concluded as the key prior knowledge: monotonicity and continuity of RUL evaluation show the relationships of monitored data and the evaluated RUL; similarity of adjacent monitored data lays the foundation of data driven RUL evaluation methodologies. These properties are then integrated into dictionary learning and sparse coding of sparse representation progress to build a remaining useful life estimation model. In the dictionary learning step, monotonicity and similarity are integrated through a matrix operator to learn more effective and concise dictionary; in the sparse coding step, new sparse model is proposed involving continuity and similarity. In the last step, the life ratio of training set to testing set is calculated through a Hough voting with sparse codes acquired from sparse coding step. With this ratio, the useful life of testing set can be calculated. So, the RUL of the testing set can be evaluated through useful life subtracting the running duration. Experiments are conducted on the ball bearing data provided by FEMTO-ST Institute and the results show the efficiency of our methodologies.