Automated 3-D Piping Routing Synthesis Considering Seismic Resilience: Part II—Algorithmic Characterization, Statistical Assessment, and Proof-of-Concept Validation

Abstract

Stochastic optimization algorithms, such as the Multi-Objective Genetic Algorithm (MOGA) integrated into the MOGA-A* framework established in Part I, enable automated 3-D pathfinding in plant engineering. However, their inherent stochasticity introduces trial-to-trial geometric variability, raising reliability concerns for safety-critical applications. To evaluate the operational capabilities and limitations of this lightweight surrogate-driven solver, this paper (Part II) presents a three-part empirical assessment. First, to quantify search-space reproducibility, a computational sweep of 1,000,000 candidate layouts across distinct spatial topologies was conducted. Maximum Likelihood Estimation confirmed that compliant total evaluation indices follow predictable log-normal or gamma distributions, parameterized by a topological Designable Space Ratio metric incorporating space occupancy and obstacle clustering. Second, to assess whether the response-spectrum modal surrogate captures relative physical dynamic trends rather than absolute stress magnitudes, frequency-sweep shaking-table experiments were performed on 1:50 scaled models. The experiments reproduced the predicted downward shift in the dominant modal frequency and confirmed a lower peak dynamic response for the synthesized flexible-detour layout, which exhibited a 45.7% reduction in the experimentally derived response metric (11.98 MPa) relative to the unoptimized baseline (22.08 MPa). Third, to assess real-world decision-support utility, an algorithmically synthesized layout was benchmarked against a baseline manually drafted by an experienced engineer under identical spatial constraints from the Onagawa Nuclear Power Plant. Although the human designer achieved a lower peak dynamic stress (18.63 MPa vs. 145.46 MPa) by incorporating higher intermediate support density, the automated solver generated a code-compliant, collision-free candidate in 46 minutes. These results demonstrate that while support allocation heuristics benefit from post-processing refinement, the MOGA-A* framework functions as an effective decision-support engine that automates initial spatial pathfinding and reduces manual trial and error in early-stage plant design.