Efficient maintenance has always been essential for the successful application of engineering systems. However, the challenges to be overcome in the implementation of Industry 4.0 necessitate new paradigms of maintenance optimization. Machine learning techniques are becoming increasingly used in engineering and maintenance, with reinforcement learning being one of the most promising. In this paper, we propose a gamma degradation process together with a novel maintenance model in which repairs are increasingly imperfect, i.e., the beneficial effect of system repairs decreases as more repairs are performed, reflecting the degradational behavior of real-world systems. To generate maintenance policies for this system, we developed a reinforcement-learning-based agent using a Double Deep Q-Network architecture. This agent presents two important advantages: it works without a predefined preventive threshold, and it can operate in a continuous degradation state space. Our agent learns to behave in different scenarios, showing great flexibility. In addition, we performed an analysis of how changes in the main parameters of the environment affect the maintenance policy proposed by the agent. The proposed approach is demonstrated to be appropriate and to significatively improve long-run cost as compared with other common maintenance strategies.

Main control room simulators are an important information source for improving the empirical basis of human reliability analysis (HRA). Both quantitative and qualitative observations provide valuable insight on the operators’ ability to successfully perform their tasks. However, relevant portions of this evidence are not used in typical approaches for estimation of human error probabilities (HEP), which are based on failure counts. With small sample sizes and the operator performance levels representative for nuclear power plants, very little information enters the probability estimates. This paper presents a novel approach to formalize the collection and use of reliability-relevant evidence from simulators. The methodology treats task reliability with the plant-centred orientation characteristic of the systems or Probabilistic Safety Assessment (PSA) perspective as well as a human-centred view of task performance based in human factors engineering. These are combined to form a Task reliability index (TRI) able to distinguish finer performance variations and consider indications of potential failures. Bayesian updating of HEP distributions is suggested by translating the TRI data into pseudo-failures and trials, preserving the convenient coupling of Beta-Binomial distributions in the Bayesian analysis. The approach is demonstrated using data from the International HRA Empirical Study and is shown to produce results that are comparable to those obtained in the empirical study, which were based on comprehensive expert analyses.