A Web-Based Predictive Maintenance Framework for Turbofan Engines Using Support Vector Regression on NASA CMAPSS Dataset

Abstract

This trend has witnessed modern aircraft systems to become more complex thus making predictive maintenance a primary requirement in ensuring safety operations and achieving reduced downtime at the same time increased-reduced costs of maintenance. Conventional maintenance plans, including time based and reactive plans are usually associated with either time based or unanticipated failures, which is not an efficient or economical way of maintenance. This paper proposes a web based predictive maintenance system to address these problems, but it will predict Remaining Useful Life (RUL) for turbofan engines using machine learning methods. The proposed system is put into action on the NASA CMAPSS data set, that supposes simulating the degradation of a real world turbofan engine, using multi-sense sequential data. The non-linear degradation features of the engine behavior are systematized by the Support Vector Regression (SVR) model based on Radial Basis Function (RBF). It uses sophisticated data preparation, such as the loss of redundant sensors, MinMax scaling normalization and feature engineering with lag-based temporal features. In addition to prediction, the framework uses a Flask-based web application that allows real-time interaction with a user, tracking of the fleet, and schedule of maintenance. The engine health is also categorized into stepable levels, i.e., Healthy, Warning, and Critical with a fixed level and a percentile level to make a better decision. The experimental findings indicate that the given model provides the consistent prediction accuracy in a variety of CMAPSS sub-datasets and maintains the high computational efficiency. The most important impact of this work is that integrating machine learning-based prognostics with a web-based interface that can be configured to scale up makes it possible to practically implement predictive maintenance in a real-life aviation environment.