Joint maintenance and spare part ordering from multiple suppliers for multicomponent systems using a deep reinforcement learning algorithm

Reliability Engineering & System Safety, Journal Year: 2023, Volume and Issue: 241, P. 109628 - 109628

Published: Sept. 4, 2023

Language: Английский

---

A deep reinforcement learning approach for repair-based maintenance of multi-unit systems using proportional hazards model

Reliability Engineering & System Safety, Journal Year: 2023, Volume and Issue: 234, P. 109179 - 109179

Published: Feb. 20, 2023

Language: Английский

---

A two-stage RNN-based deep reinforcement learning approach for solving the parallel machine scheduling problem with due dates and family setups

Journal of Intelligent Manufacturing, Journal Year: 2023, Volume and Issue: 35(3), P. 1107 - 1140

Published: March 9, 2023

Abstract As an essential scheduling problem with several practical applications, the parallel machine (PMSP) family setups constraints is difficult to solve and proven be NP-hard. To this end, we present a deep reinforcement learning (DRL) approach PMSP considering setups, aiming at minimizing total tardiness. The first modeled as Markov decision process, where design novel variable-length representation of states actions, so that DRL agent can calculate comprehensive priority for each job time point then select next directly according these priorities. Meanwhile, state matrix action vector enable trained instances any scales. handle sequence simultaneously ensure calculated global among all jobs, employ recurrent neural network, particular gated unit, approximate policy agent. based on Proximal Policy Optimization algorithm. Moreover, develop two-stage training strategy enhance efficiency. In numerical experiments, train given instance it much larger experimental results demonstrate strong generalization capability comparison three dispatching rules two metaheuristics further validates superiority

Language: Английский

---

Deep reinforcement learning for maintenance optimization of a scrap-based steel production line

Reliability Engineering & System Safety, Journal Year: 2024, Volume and Issue: 249, P. 110199 - 110199

Published: May 18, 2024

This paper presents a Deep Reinforcement Learning (DRL)-based optimization approach for determining the optimal inspection and maintenance planning of scrap-based steel production line. The DRL-based recommends adequate time inspections activities based on monitoring conditions line, such as machine productivity, buffer level, demand. Some practical aspects system, uncertainty duration variable rate machines, were considered. A line was modeled multi-component system considering components dependencies. simulation model developed to simulate dynamics assist with development DRL approach. proposed is compared traditional policies, reactive maintenance, time-based condition-based maintenance. In addition, different algorithms PPO (Proximal Policy Optimization), TRPO (Trust Region DQN (Deep Q-Network) are investigated in case-based scenario. findings indicated potential significant financial savings. Therefore, demonstrates adaptability has be powerful tool industrial competitiveness.

Language: Английский

---

Dynamic robot routing optimization: State–space decomposition for operations research-informed reinforcement learning

Robotics and Computer-Integrated Manufacturing, Journal Year: 2024, Volume and Issue: 90, P. 102812 - 102812

Published: June 25, 2024

There is a growing interest in implementing artificial intelligence for operations research the industrial environment. While numerous classic solvers ensure optimal solutions, they often struggle with real-time dynamic objectives and environments, such as routing problems, which require periodic algorithmic recalibration. To deal deep reinforcement learning has shown great potential its capability self-learning optimizing mechanism. However, real-world applications of are relatively limited due to lengthy training time inefficiency high-dimensional state spaces. In this study, we introduce two methods enhance optimization. The first method involves transferring knowledge from during training, accelerates exploration reduces time. second uses state–space decomposer transform space into low-dimensional latent space, allows agent learn efficiently space. Lastly, demonstrate applicability our approach an application automated welding process, where identifies shortest pathway robotic arm weld set dynamically changing target nodes, poses sizes. suggested cuts computation by 25% 50% compared algorithms.

Language: Английский

---

Counterfactual-attention multi-agent reinforcement learning for joint condition-based maintenance and production scheduling

Journal of Manufacturing Systems, Journal Year: 2023, Volume and Issue: 71, P. 70 - 81

Published: Sept. 13, 2023

Language: Английский

---

A data-driven methodology with a nonparametric reliability method for optimal condition-based maintenance strategies

Reliability Engineering & System Safety, Journal Year: 2023, Volume and Issue: 241, P. 109668 - 109668

Published: Sept. 20, 2023

Language: Английский

---

A transformer-based deep reinforcement learning approach for dynamic parallel machine scheduling problem with family setups

Journal of Intelligent Manufacturing, Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 8, 2024

Abstract The parallel machine scheduling problem (PMSP) involves the optimized assignment of a set jobs to collection machines, which is proper formulation for modern manufacturing environment. Deep reinforcement learning (DRL) has been widely employed solve PMSP. However, majority existing DRL-based frameworks still suffer from generalizability and scalability. More specifically, state action design heavily rely on human efforts. To bridge these gaps, we propose practical learning-based framework tackle PMSP with new job arrivals family setup constraints. We variable-length matrix containing full information. This enables DRL agent autonomously extract features raw data make decisions global perspective. efficiently process this novel matrix, elaborately modify Transformer model represent agent. By integrating modified agent, representation can be effectively leveraged. innovative offers high-quality robust solution that significantly reduces reliance manual effort traditionally required in tasks. In numerical experiment, stability proposed during training first demonstrated. Then compare trained 192 instances several approaches, namely approach, metaheuristic algorithm, dispatching rule. extensive experimental results demonstrate scalability our approach its effectiveness across variety scenarios. Conclusively, thus problems high efficiency flexibility, paving way application solving complex dynamic problems.

Language: Английский

---

Condition-based maintenance with reinforcement learning for refrigeration systems with selected monitored features

Engineering Applications of Artificial Intelligence, Journal Year: 2023, Volume and Issue: 122, P. 106067 - 106067

Published: March 9, 2023

Language: Английский

---

Hierarchical framework for interpretable and specialized deep reinforcement learning-based predictive maintenance

Data & Knowledge Engineering, Journal Year: 2023, Volume and Issue: 149, P. 102240 - 102240

Published: Nov. 1, 2023

Language: Английский

---