Content area
The Department of Defense (DoD) and other maintenance stakeholders are beginning to implement predictive maintenance, and 75% of its agencies attempting at least one-use case will likely see a 35% increase in mission readiness and cost reduction (Logan et al., 2023). However, because of the massive annual cost of maintenance that the DoD still expends, it continues to seek optimized methodologies, tools, and technologies to cut maintenance costs, especially the cost of unscheduled maintenance. To enable DoD to continue embracing predictive maintenance, many researchers have evolved different approaches, such as analytical or physics-based and data-driven. One of the data driven methodologies for predictive maintenance is to be able to train and predict remaining useful life, end of life, and current condition of a component, set of components of systems using computing intelligence (Machine Learning Algorithms). To be able to train the model, a dataset that represents a true run-to-failure of the system with key features is required. The researcher used datasets from the National Aeronautics and Space Administration (NASA) Integrated Vehicle Management Program (IVMP) (NASA, 2024).
The focus of this research was to analyze the run-to-failure of a Turbofan Engine dataset curated by NASA IVMP, to determine the key features that influence the remaining useful life of the engines and to use machine learning as a data-driven approach to predict the remaining useful life (RUL) of turbofan engines and monitor the health of turbofan aircraft engines. Machine learning models can learn the patterns or behavior of a system from historical datasets. Multiple models such as Convolution Neural Network (CNN) model, Simple Recurrent Neural Network (RNN) model, and Long-Short Term Memory (LSTM) RNN model were trained and their output compared to determine which model performed better. The research showed that LSTM RNN performed better in predicting RUL as a time series problem. Finally, aside from predicting the RUL of the engine, the researcher also used a random forest classifier model to assess the engine's health or condition. Thus, we are able to predict the RUL and also assess the health of aircraft engines.
