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.

Digital Twins (DTs) have attracted significant interest across industries over the past decade, and they play a vital role in modern Structural Health Monitoring (SHM). A key component in DT development is the digital model of the monitored asset. In complex systems, these models can impose a considerable computational workload, necessitating the use of model surrogates to enable real-time simulations. While many effective methodologies for surrogate modelling have been developed, they still only serve as approximations of the original system. Recent advancements in Quantum Computing present the opportunity to create algorithms that can run full-order models significantly faster than traditional computers. The Quantum Finite Element Method (QFEM) is an innovative framework for solving Finite Element Method (FEM) problems using quantum algorithms, aiming to leverage the speed of quantum computers to handle large-scale models without the need for surrogates.

This study introduces a QFEM-based application for beam model updating, implemented within a Monte Carlo Markov Chain Metropolis-Hastings (MCMC-MH) algorithm for Bayesian inference. Specifically, this model estimates beam degradation by inferring changes in stiffness, which decrease as the structure wears. The methodology focuses on developing a QFEM model capable of rapidly simulating the structure while varying its parameters. The model's accuracy is evaluated by comparing it with a traditional FEM formulation. Furthermore, using QFEM in conjunction with MCMC-MH allows for the identification of the posterior distribution of the structural parameters. This demonstrates that QFEM's accuracy is sufficient to enable reliable model updating, providing a foundation for future Digital Twin applications.

Ultimately, the successful identification of structural parameters confirms that integrating QFEM into Bayesian frameworks is a promising approach for the realisation of next-generation Digital Twins, enabling increased computational speed while maintaining physical accuracy.

Recent advances in quantum photonics and single-photon detection are enabling new sensing capabilities for vibration-based diagnostics and non-destructive evaluation. This work introduces the Quantum Photonic Vibrometer (QPV), a quantum-enhanced, non-contact sensor designed for sub-nanometric displacement and vibration measurements within Structural Health Monitoring (SHM).

The QPV operates at a telecom wavelength of 1550 nm using a mode-locked laser emitting 20 MHz optical pulse trains. Each optical pulse is paired with a precisely timed detection gate on an avalanche photodiode (APD) operated in Geiger mode. Rather than measuring the time-of-flight of individual photons, the system records photon counts within a narrow 2.4 ns gate. These photon-count fluctuations arise from phase-induced speckle modulation in the light backscattered from the structure. Because surface vibration changes the optical phase of the coherent return field, the resulting speckle intensity varies pulse-to-pulse, producing a photon-count signal that directly encodes the structural vibration.

In contrast to a classical Laser Doppler Vibrometer (LDV), which relies on measuring Doppler frequency shifts of a continuous beam, the QPV extracts displacement from \textbf{phase-sensitive intensity variations of the reflected photons}. This single-beam, pulse-gated approach enables operation at extremely low optical powers - including the single-photon level - making the technique suitable for delicate, optically sensitive, or remotely located specimens.

Several experimental parameters were identified as critical for achieving high-contrast and reproducible measurements, including APD gate width (≈2.4 ns), laser repetition rate (20 MHz), internal optical crosstalk around 40 ns, and APD bias stability. Understanding and mitigating these factors will be essential for advancing the technique toward calibrated SHM applications.

This study represents the first demonstration of time-gated single-photon vibrometry for structural diagnostics, bridging quantum photonics with conventional SHM methodologies. By leveraging coherent scattering, speckle-induced phase sensitivity, and single-photon statistics, the QPV establishes a foundation for quantum-enhanced vibrometry offering improved sensitivity, noise robustness, and new opportunities for non-contact damage detection.