Surrogate-Based Thermal Risk Tool of Aircraft System Placement Optimization

Placement optimization of aircraft systems is a multidisciplinary problem that includes thermal, maintainability, and safety considerations. This work focuses on an existing thermal tool that yields risk scores based on component positioning, orientation, and heat loads. These scores are characterized by first-order discontinuities, which poses difficulties to gradient-based optimization methods. To address this issue, surrogate modelling is employed to fulfill two objectives that appear to be at trade-off: (i) provide accurate predictions of the thermal score and (ii) provide a differentiable response surface. Two surrogate modeling techniques—Kriging-based methods and tree-based classifiers—are tested on the most placement-sensitive score category, namely TR3. The surrogate models performance is evaluated based on prediction accuracy and numerical stability improvements. Preliminary results indicated the elimination of areas with zero gradients using both models, with prediction accuracy shown to be higher in the Kriging model.