Methodology to Use Failed Optimizer Cases to Achieve Low-Cost Computational Results

Trajectory optimization plays a crucial role in aerospace applications, but high computational costs and convergence issues pose significant challenges. To address this, we propose a data-driven approach on failed cases that utilizes feature extraction, clustering, and profile prediction to generate an efficient initial guess for the optimizer. First, data extraction is performed on data at selected time steps, capturing key statistical characteristics such as mean, variance, and absolute sum of changes. This allows for a structured representation of failed and successful cases. Next, a K-Means clustering algorithm is applied to categorize different trajectory profiles, grouping similar cases based on their deviation from the optimized altitude. The clustering process identifies the closest-matching trajectories to the best-case scenario, ensuring that only relevant data informs optimization. Finally, profile prediction leverages these clustered cases to construct a low-computation approximation of an optimal trajectory. This framework enhances Multi-Disciplinary Analysis (MDA) by streamlining trajectory simulations, making them more robust and computationally efficient. The proposed approach significantly accelerates the optimization process, paving the way for more effective aerospace trajectory planning and decision-making.