Toward imprecision-aware RUL forecasting: ANFIS-ensemble approach

Abstract Engine health monitoring in the aeronautical domain is crucial. Aircraft engines operate under crucial conditions, and their failure can have major safety and operational consequences. The prognostic and health management (PHM) of engines plays an important role in keeping the operation of engines steady and secure. The main purpose of PHM is to predict future machinery faults by estimating their remaining useful life (RUL). To do so, PHM methods could apply machine learning methods using monitoring data to predict industrial machinery RUL. However, these predictions could differ from one source (predictor) to another, which is not totally reliable. In addition, this imperfection is due to data imprecision that needs to be managed. In this investigation, the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset is used, as engine health monitoring is highly critical and has a great impact on the maintenance of aeronautical systems. To tackle this challenge, we combine the process of studying sliding windows on time series data and taking advantage of the integration of learning algorithms. Specifically, we merge the convolutional neural network (CNN) with multi-layer perceptron, and CNN with long short-term memory. To further enhance the performance, we employ an ensemble learning approach to manage imprecision by applying fuzzy concepts embedded within the adaptive neuro-fuzzy inference system model and a decision tree as a meta-learner. Experimental errors are reduced compared to predictions offered by hard prediction models.