This paper presents a continuation-based computational framework designed to automate the exploration of standing-wave thermoacoustic engines (SWTE) modeled in DeltaEC. While shooting-based solvers are commonly used, they are often unstable when exploring wide design spaces. This framework provides a solution to these convergence issues by using a "staged seeding" process to systematically map variations in stack position and length. Using this automated workflow, a robust dataset of 526 converged operating points was generated for a quarter-wavelength engine. A Random Forest regression model was trained on this data achieving a predictive performance (R 2 ) of 0.97 and was used to assess efficiency trends across the full design space. This model enabled the identification of global design drivers and the examination of local behavior through normalized, one-at-a-time ±5% perturbations. The results indicate that efficient operation is restricted to a narrow phase-feasible region, where the framework successfully identified a path to increase efficiency from a 2.8% baseline to 6.9%. The strong alignment between global and local behavior validates that the workflow accurately captures the underlying physics. Ultimately, this approach offers a dependable automation solution for shooting-based software, enabling high-quality data generation for successive optimization.
Arnaoud et al. (Thu,) studied this question.