Engine Parameter Estimation in Test Cells Using Hybrid Physics/Empirical Models

Abstract

Estimation of engine parameters such as thrust in test cells is a difficult process due to the highly nonlinear nature of the engine dynamics, the complex interdependency of thrust and the engine’s health condition, and factors that corrupt thrust measurements due to test stand construction. Because the frequency content of the corrupting dynamics is close to the engine’s dynamics, filtering the thrust signal is not sufficient for extraction of the true dynamic content. A configurable thrust estimation system is developed for accurate data reduction which provides “virtual” measurements of thrust and other necessary parameters at steady state and during aggressive engine transients. The thrust estimation framework consists of a representative nonlinear engine model coupled with an adaptive structural dynamics model. To account for discrepancies between the physics-based model and the true engine, a hybrid model using a novel neural network (NN) enhancement to a physics-based engine model is presented that reduces certain modeling errors between the engine model and the physical plant. This includes engine-to-engine variation, engine degradation and any essential neglected dynamics. To fuse the model and sensor measurements, this hybrid model is used within a constant-gain extended Kalman filter batch estimator which is able to reconstruct the true dynamic performance of the engine using noisy or corrupted sensor measurements and control inputs. The Kalman filter estimates measured and unmeasured parameters and state variables such as engine component deterioration parameters and effective flow areas.