Toward Quantum Reinforcement Learning for Automated Neural Architecture Optimization in Structural Health Monitoring

Source raw items (1)

• Semantic Scholar2026-07-22 07:15:10
  Toward Quantum Reinforcement Learning for Automated Neural Architecture Optimization in Structural Health Monitoring

  Structural Health Monitoring (SHM) increasingly benefits from AI models for automated damage detection and classification. Reinforcement Learning (RL) has already shown strong performance in this domain by enabling adaptive and accurate image-based classification. However, the next challenge lies not in classifying damage but in automating the design of the networks themselves - a task addressed through Neural Architecture Search (NAS) but still computationally demanding due to the exponentially growing architecture and hyperparameter search space. The present work builds upon the Q-learning Non-Dominated Sorting Algorithm (QNSA), which integrates RL with multi-objective optimization for automated neural architecture design within a NAS framework. In QNSA, an RL agent adaptively selects metaheuristic crossover operators through a three-dimensional Q-tensor that encodes action–state–variable interactions, enabling efficient exploration of mixed binary, categorical, and numerical parameters. In this study, the QNSA framework is reproduced and validated on the NEU steel surface defect dataset to establish a baseline for future Quantum Reinforcement Learning (QRL) extensions. The ongoing phase of this research investigates a QRL approach to automated model optimization in SHM. As illustrated in Figure 1, the proposed framework is designed for deployment within a structural health monitoring system, where quantum-inspired decision policies enhance both model adaptation and efficiency during real-time data acquisition and analysis. In this next phase, quantum computing principles such as superposition, interference, and probabilistic policy encoding are leveraged to expand the exploration capabilities of the RL agent. Instead of operating on single deterministic states, the QRL agent represents multiple potential policies simultaneously, enabling a richer and more efficient search across the vast architecture space. The overarching objective is to investigate how quantum-enhanced decision policies can improve the efficiency, adaptability, and scalability of NAS frameworks applied to vision-based SHM. By merging quantum computation with reinforcement learning, this research aims to accelerate the discovery of optimized deep architectures while reducing computational cost. Ultimately, this study represents an initial step toward quantum-enhanced intelligent SHM systems, where hybrid quantum–classical learning can support autonomous, adaptive, and resource-aware model optimization for next-generation sensing and diagnostic platforms.