An improved Kalman filter based on neural network for turbofan engine gas-path health estimation

The Kalman filter is the most commonly-used state estimation method for turbofan engine health monitoring. It achieves the state estimation on condition that the number of the available measurement sensors is more than the number of the health parameters to he estimate. However, it is hard to hold this assumption in the turbofan engine gas-path health monitoring application. Thus, in this paper, an improved Kalman filter based on neural network is proposed to improve the filter estimation accuracy. The improved Kalman filter consist of a master filter and a neural network based estimator. During each sampling period, the estimation result of the neural network based estimator is integated to the master filter as a penalty term to update posterior state, which completes a better trade-off between the estimation accuracy and computational efforts. Moreover, a mind evolutionary algorithm is adopted to optimize both the weights and thresholds of the estimator. The simulation results of a turbofan engine health monitoring application in the flight envelope show that the proposed method yields a significant improvement of the estimation accuracy and robustness, it achieves better trade-off between the estimation accuracy and computational efforts.