A Life Clustering Framework for Prognostics of Gas Turbine Engines under Limited Data Situations

The reliability of data driven prognostics algorithms severely depends on the volume of data. Therefore in case of limited data availability, life estimations usually are not acceptable; because the quantity of run to failure data is not sufficient to train prognostics model efficiently. To board this problem, a life clustering prognostics (LCP) framework is proposed. LCP regenerates the train data at different ages and outcomes to increment of the training data volume. So, the method is useful for limited data conditions. In this research, initially LCP performance is studied in normal situation is; successively robustness of the framework under limited data conditions is considered. For this purpose, a case study on turbofan engines is performed. The accuracy for the proposed LCP approach is 71% and better than other approaches. The prognostics accuracy is compared in various situations of data deficiency for the case study. The prognostic measures remain almost unchanged when the training data is even one third. Successively, prognostics accuracy decreases with a slight slope; so that when the training data drops from 100 to 5%, the accuracy of the results drops 26%. The results indicates the robustness of the proposed algorithm in limited data situation. The main contribution of this paper include: (1) The effectiveness of life clustering idea for use in prognostics algorithms is proven; (2) A step-by-step framework for LCP is provided; (3) A robustness analysis is performed for the proposed prognostics algorithm