Advanced Approach for Optimizing Predictive Maintenance and Reducing Industrial Failures

Abstract

Optimizing intervention strategies and reducing unplanned downtime are major challenges in predictive maintenance, particularly in the aeronautics industry, where engine reliability is crucial. This work aims to develop a long-short-term memory (LSTM) neural network model to predict the remaining useful life (RUL) of aircraft engines from simulated sensor data. The main objective is to enable proactive maintenance planning by anticipating potential failures, thus minimizing operational risks and associated costs, the study aims to estimate the number of additional cycles an engine can withstand before failure, while assessing the probability of a failure occurring within a defined time interval. The project uses regression methods to estimate RUL continuously and uses binary classification techniques to predict the probability of failure over a specific cycle. The method used involves an organized approach that includes literature review, preliminary processing and normalization of data, development and training of LSTM models, and assessment of their predictive performance based on relevant indicators.