Abstract
Benchmark models are indispensable for gas turbine engine monitoring and diagnostics. In most cases, benchmark models are used to predict engine data. When the measured data from a real engine does not match up the predicted data, it means that something undesirable is going on and the engine may need servicing.
Generally, the benchmark models are also thermodynamic models. They are either offered by engine manufactures or obtained using the technique called “performance adaptation”. In both cases, those models have a major limitation: they cannot reflect the development of engine deterioration over time, i.e., deterioration parameters of these models are fixed unless manually adjusted. In practice, however, most engines deteriorate continuously and gradually over time. Due to the increasing mismatch of deterioration, the accuracy of the benchmark model will get worse gradually.
To address the mismatch, this paper presents a self-tuning model framework that intends to improve the existing benchmark models. In the framework, a set of deterioration models are introduced and attached to the original benchmark model. The deterioration models represent the deterioration conditions of all major components. They are machine learning models and can be updated automatically according to the engine measurements. With the aid of the k-nearest neighbours algorithm and gas path analysis techniques, the deterioration models can track the real deterioration conditions through self-tuning and are robust under noise. The proposed model framework is applied to a model turbofan engine. Simulation results show that the accuracy of the benchmark model improves significantly after using the self-tuning framework.
