Explainable Artificial Intelligence–Based Search Space Reduction for Optimal Sensor Placement in the Pipeline Systems of Naval Ships

Pipeline damage in mission-critical systems, such as pipelines within naval ships, can result in substantial consequences. Compared to manual inspection of pipeline damage by crew members onboard, structural health monitoring of pipeline systems offers prompt identification of damage sites, enabling efficient damage mitigation. However, one challenge of this approach is deriving an optimal sensor placement (OSP) strategy, given the large and complex pipelines found in real-scale naval vessels. To address this issue, a search space reduction method is proposed for OSP suitable for the large and complex pipeline systems found in naval ships. In the proposed method, the original search space for sensor placement is reduced to a manageable scale using an explainable artificial intelligence (XAI) technique, namely, a gradient-weighted class activation map (Grad-CAM). Grad-CAM enables quantification and visualization of the contribution of individual pipeline nodes to classify damage scenarios. Noncritical sensor locations can be excluded from the candidate search space. Furthermore, a peak-finding algorithm is devised to select only a limited number of nodes with the highest Grad-CAM values; in this research, the algorithm is proven effective in reconstructing the search space. As a result, the original OSP problem—which has an extremely large search space—is reconstructed into a new OSP problem with a computationally manageable search space. The new OSP problem can be solved using either meta-heuristic methods or exhaustive search methods. The effectiveness of the proposed method is validated through a case study on a real-scale naval combat vessel, measuring 102 m in length and carrying a full load of 2300 tons. The results show that the proposed XAI-based search space reduction approach efficiently designs an optimal pipeline sensor network in real-scale naval combat vessels.