Markov Nonlinear System Estimation for Engine Performance Tracking

Abstract

This paper presents a joint state and parameter estimation method for aircraft engine performance degradation tracking. Contrast to previously reported techniques on state estimation that view parameters in the state evolution model as constants, the method presented in this paper treats parameters as time-varying variables to account for varying degradation rates at different stages of engine operation. Transition of degradation stages and estimation of parameters are performed by particle filtering (PF) under the Bayesian inference framework. To address the sample impoverishment problem due to discrete resampling, which is inherent to PF, a continuous resampling strategy has been proposed, with the goal to improve estimation accuracy of PF. The algorithm has shown to be able to detect abrupt fault inception based on the residuals between the estimated results from the state evolution model and actual measurements. The developed technique is evaluated using data generated from a turbofan engine model. Simulation of engine output parameters over a series of flights with both nominal degradation and abrupt fault types has been conducted, and error within 1% for performance tracking and degradation prediction has been shown. This demonstrates the effectiveness of the developed technique in fault detection and degradation tracking in aircraft engines.

Artificial Intelligence in Advanced Manufacturing: Current Status and Future Outlook

“…Recently, LSTMs have also been investigated in conjunction with model-based techniques, such as particle filters (PF), in order to alleviate the limitation of insufficient observations for degradation model parameter estimation. PFs have been widely studied for the prognosis of complex engineering systems such as HVAC [125] and aircraft engines [126] in which the degradation model can be continuously fine-tuned by the incoming sensing observations based on Bayesian inference. In Ref.…”

A hybrid model of LSTM neural networks with a thermodynamic model for condition-based maintenance of compressor fouling

“…The GPA method, one of the common performance-based condition monitoring methods, is a cost-efficient approach to representing the degradation as the deviation of measurements [21][22][23]. There are three key issues that need to be investigated in the GPA method [24]: (i) nonlinear tracking: studies have shown that the degradation curve of gas turbine performance is nonlinear, including the polynomial term [25], the logarithmic term [26], and the exponential term [27]. (ii) Health parameters: the health parameters should represent the health status of the gas turbine and are independent of the environmental factors or control factors that are not related to the degradation degree [28].…”

Degradation Modeling and Remaining Useful Life Prediction of Aircraft Engines Using Ensemble Learning

“…The existing literature pertaining to RUL prediction for aircraft engines can be classified into two categories: model-based and data-driven prognostics [3][4][5]. Model-based prognostic methods describe system behavior and system degradation using physics-based models typically in combination with state estimators such as the Kalman filter, the particle filter, and the hidden Markov model [6]. While model-based prognostic methods provide closed-form solutions, certain assumptions must be made.…”