Prediction of the Remaining Useful Life of Engines for Remanufacturing Using a Semi-supervised Deep Learning Model Trained by the Bees Algorithm

Abstract

The estimation of the remaining useful lifeRemaining Useful Life (RUL) of a component is one of the most important tasks for predictive maintenance systems (PMSs). This research aims to predict the remaining useful lifeRemaining Useful Life (RUL) of aircraft engines by using the Bees AlgorithmBees Algorithm, THE (BA)-optimised semi-supervised deep learningDeep learning model. To this end, the LSTMLong Short Term Memory architecture has been implemented as a successful deep learningDeep learning architecture to process time-series datasets. The predictionsPrediction have been made by using the NASA Turbofan Engine Corruption Simulation (C-MAPSS) dataset. The proposed predictionPrediction model has been formulated as a binary classificationClassification task based on the semi-supervised labelling process. Unlike the conventional predictive maintenance models, the data-driven LSTMLong Short Term Memory architecture automatically learns features of the multivariate time series. The model has been verified based on sixteen different time cycles. The proposed semi-supervised deep learning model has been trained using the BA to improve the accuracy value for the safe and unsafe conditions of aircraft engines. The experimental results reached 98% accuracy for the test dataset, and the proposed model performed better in terms of the F1 measure for both the training and test datasets. Experiments proved that the proposed deep learningDeep learning model can be used as a promising RULRemaining Useful Life predictionPrediction model.