Time to Failure Prognosis of a Gas Turbine Engine Using Predictive Analytics

Abstract

Maintenance costs and machine availability are the most important concerns of any company that owns large machinery, especially gas turbine engines. With the advent of the 4th wave of the industrial revolution also known as Industrial Internet of Things (IIoT), the focus has shifted onto optimal utilization of the equipment. Reduction in the installation costs of sensors helped companies to install them on their key equipment. The live data from the sensors can now be utilized to monitor the health of the machines. This thesis proposes a prognostic technique to predict time-to-failure of gas turbine engines using standard machine learning and deep learning techniques that puts the data from sensors to good use. Our proposed approach provides accurate feedback on the health of the machine to the concerned personnel. Our approach includes developing multiple Recurrent Neural Network (RNN) models to predict the sensor readings of the engine and then using a Support Vector Machine (SVM) to classify these readings as safe or failure. The objective of this thesis is to establish supporting evidence for the proof of concept of the created time-to-failure prognostic technique. We have demonstrated the performance of our approach on the data-sets made available by NASA with different failure modes and then compared the performance of our approach with the current industry standard for land-based gas turbine engines.