Quantum Annealing Boosts Reinforcement Learning for RUL Prediction

Predicting the Remaining Useful Lifetime (RUL) of assets is a critical component of predictive maintenance, where unexpected failures can incur substantial costs. Traditional statistical degradation models often struggle with the strong nonlinearities inherent in real-world systems, while data-driven models can converge to suboptimal solutions in complex, high-dimensional search spaces. To address these limitations, a new framework called Quantum Annealing enhanced Q-Learning (QAQL) has been developed. This approach seamlessly couples the unique sampling capabilities of quantum annealing with the sequential decision-making process of Q-learning. Instead of acting as a deterministic optimizer, the quantum annealer returns a distribution over near-optimal actions, providing the necessary exploration to avoid premature convergence along nonlinear degradation trajectories. Each Q-value update is formulated as a small quadratic unconstrained binary optimization (QUBO) problem, solved on a D-Wave Advantage system. Validated on NASA C-MAPSS turbofan engine datasets and a device-fleet predictive maintenance dataset, QAQL consistently outperformed both classical and other quantum baselines across multiple error metrics. This demonstrates that quantum annealing can be a practical and effective optimizer within reinforcement learning loops for industrial applications.