Adaptive Online Fault Mitigation using Hierarchical Engine Control

This paper proposes a hierarchical control architecture to mitigate faults and modeling errors in an engine system. A hybrid approach is used to model the complete engine system with the main cylinder combustion process represented using a neural network model and the rest of the system is modeled using well-studied physics-based analytical equations. A control calibration map that consists of the optimal engine control parameters to maintain the desired engine torque using the minimum fuel consumption rate is generated offline by performing Bayesian optimization on this high-fidelity hybrid engine model. We then use proportional-integral (PI) and extremum seeking (ES) controllers on top of the offline map to compensate for any engine faults and modeling errors for online calibration. The work is motivated by optimally reconfiguring autonomous systems. We test three different scenarios through numerical simulations which require online calibration of the engine control parameters. It is shown that the PI+ES controller can overcome the fault by maintaining the desired torque and also lowers the fuel consumption rate compared to a PI controller.