Multiobjective Optimization for Turbofan Engine Using Gradient-Free Method

Turbofan engine performance optimization is usually formulated as a single objective, closed-form optimization problem by employing a prior mechanism model with an additive, user-preference weight. However, in practical scenarios, the conventional single objective performance optimization may not satisfy the high-performance requirements. For instance, pursuing high-effective thrust will lead to high-turbine inlet temperature due to generating extra heat. Moreover, the system model may be inaccurate or even unavailable, mainly due to the degradation factor, manufacturing tolerance, or time-intensive experiments. Traditionally, the multiobjective optimization methods may require a certain amount of function evaluations, or the convergence properties may not be guaranteed explicitly. To tackle the above-mentioned issues, we formulate the performance optimization of turbofan engines as a multiobjective optimization problem and construct a gradient-free framework to deal with the issue of an inaccurate/unavailable turbofan engine model. Then, to ensure the safety requirement of the turbofan engine operating processes, a multiobjective optimization algorithm is proposed utilizing a gradient-free method, termed Hessian aware gradient estimation-based randomized search (HAGE-RS), and we analyze the corresponding convergence properties of the solved candidates. Finally, we illustrate the proposed algorithm on benchmarks and the performance optimization problem using real-world turbofan engine data under different operating conditions to show superior performance.