Modeling Health Status Identification in a Gas Turbine System: Three-Class Classification Approaches

Abstract

In this work, the sensor data from a gas turbine system was analyzed with the objective of modeling the health status of the system. A variety of classification models of the system were developed in a three-class problem to differentiate healthy, deteriorated and failed system states to explore the ability of machine learning models to provide early warning of upcoming incidents. However, there are limited examples of failure incidents and the available examples do not cover the scope of possible failures. Therefore the challenge is to create a model that detects failures in new unseen incidents, while also successfully applying that model to different vehicles of the same type. Three approaches to selecting training data were used. The first followed a traditional method of randomly selecting data points from all data according to a target ratio between training and testing data for each data class. The second data selection strategy was to consider data related to failure incidents as a whole and select certain incidents to include in training, and the remaining ones to be unseen in testing. The third approach was a cross-validationinspired approach, separating data into folds but training and testing models based on failure incidents. In addition to investigating training and data selection strategies, the effect of hyperparameter optimization was explored as well as the effect of varying the time period of the deteriorated class. The classification methods included support vector machines, Gaussian Na¨ıve Bayes, Random Forest, Adaboost, multilayer perceptron, k-nearest neighbor, and extreme gradient boosting. Ensemble models were also created to leverage all the individual classification models that were developed. This paper describes the comprehensive results that were obtained using the various approaches and combinations, highlighting the respective benefits and limitations.