Numerical simulations play a crucial role in modern product development, requiring thorough analysis and comparison of results. This process can be time-consuming, highlighting the need for effective methods for Comparative Analysis that identify differences in outcomes and their interrelations. Dimensionality Reduction Methods (DRMs) have been employed for several years to address this need, with recent emphasis on nonlinear reduction approaches. One prevalent method, Difference Principal Component Analysis (DPCA), utilizes a linear reduction strategy to compute correlations between various simulation components. This dissertation aims to enhance DPCA by incorporating nonlinear Dimensionality Reduction (DR). The approach involves modifying the DPCA workflow in two steps. The first step extends several generative DRMs, while the second introduces a new generalized concept of Difference Dimensionality Reduction, demonstrated through two specific implementations. The effectiveness of these new methods was evaluated on artificial data to isolate individual steps and on realistic simulation results. The findings reveal that, in cases of nonlinear relationships, the new methods outperform linear approaches, while confirming other linear dependencies. These modifications render the DPCA workflow applicable to datasets with nonlinear dependencies, suggesting a wide range of potential applications, particularly in areas like topolog
