An Improved Similarity-based Prognostics Method for Remaining Useful Life Estimation of Aero-Engine

Remaining Useful Life (RUL) estimation is the most common task in the research field of prognostics and health management (PHM). Accurate RUL estimation can avoid accidents, maximize equipment utilization, and minimize maintenance costs. RUL estimation based on performance degradation data is a hot spot in current research. The data-driven method can find out the relationship between the sensor data and the system degradation level with run-to-failure data and do not require any domain knowledge. RUL estimation can be carried out even when it is difficult to obtain the mathematical model of system degradation process. Sensors are used to collect data and monitor performance index. The actual system will experience multiple working conditions from the initial state to the performance failure process, and different working conditions have different impact on system degradation. In order to solve the problem that the degradation trend of sensor data is not declining obviously and the prediction of residual life is not accurate, a similar residual remaining useful life prediction method based on operating conditions clustering analysis and information fusion is proposed. Similarity-based methods are suitable for RUL estimation when complex systems cannot use data learning to build a global model. The core idea of RUL estimation based on similarity method is that if the test samples have similar degradation performance as the reference samples, then they may have similar RUL. In this paper, considering the influence of system operating conditions and sensor sensitivity on aero-engine life prediction, a remaining life estimation method based on multi-information fusion residual similarity model is proposed. Firstly, different working conditions were analyzed by clustering, and the data of various sensors were normalized. Then, the data of multiple sensors with different sensitivity were fused into a health index related to system degradation by the information fusion method. The distance between the degradation curve of the test sample and the degradation trajectory of the similar model was taken as the scoring basis, and the closest degradation curves were selected according to the scoring level. Finally, the closest similar degradation curves were selected according to the scores, and the Remaining Useful Life was predicted based on the residual life of these curves. The validity of the proposed method is verified by the failure data test of aero turbofan engine. The experimental results show that the proposed method has high accuracy and versatility when a large number of historical data are available. By comparing the estimated life of different breakpoints, it is found that the Remaining Useful Life estimation becomes more accurate with the increase of the proportion of verified data. Compared with other related methods, this method has achieved better results in predicting accuracy.