AI based Aircraft Health Management - Early Detection of Engine Anomaly using Machine Learning and Deep Learning

Abstract

An aeroplane is a very complicated system that is comprised of a variety of different subsystems, assemblies, and individual parts and pieces. To the extent that aircraft maintenance is concerned, forty percent of the total gross operating costs of aviation systems can be attributed to these expenses. The proper health monitoring that is required for the operation of an aircraft will increase operational efficiency and reduce the need for maintenance on the aircraft. The traditional approach to determining the health of aviation engines entails coming up with a plan for and building a facility known as an Engine Test Bed (ETB), which can cost up to millions of US dollars. Building systems that make use of prediction algorithms in order to forecast the state of an engine is absolutely necessary. Predictive maintenance, also known as PdM, is a method that is both reliable and effective in predicting the condition of an engine. PdM helps to reduce the costs associated with engine overhaul, improves engine safety, and reduces the overall costs of an engine’s life cycle. An aviation engine prognostic system needs to be developed so that engine anomalies can be predicted and aircraft with engine flame out issues can receive assistance. Because of the rapid increase or decrease in engine temperature that can occur when ever an engine flame-out is reported. The engine oil temperature (EOT) and the cylinder head temperature (CHT), both of which can change depending on the flight mode that is being used and the flight operating parameters, are intimately connected to this problem. Additionally, the flying behaviours that change an aircraft’s engine EOT and CHT also have an effect on the characteristics of the aircraft’s subsystems. Using machine learning and deep learning models for aircraft engines, the purpose of this work is to attempt to predict engine EOT and CHT. The forecast is supported by the information that is extracted from the flight data recorder (FDR) that is installed within the aircraft and is part of aircraft avionics. In this particular investigation, the estimation of EOT and CHT was accomplished by utilising the MLR, DTR method of machine learning, as well as the ANN and RNN/LSTM deep learning algorithms. The MAE, RMSE, and r2 evaluation methods were utilised in the course of this work. It is necessary to independently estimate, and then compare, the errors throughout the entire flight and cruise phase. It was discovered that, out of all the algorithms, the LSTM algorithm performed the best.