AttenMfg: An Attention Network Based Optimization Framework for Sensor-Driven Operations & Maintenance in Manufacturing Systems

Abstract

Operations and maintenance (O&M) scheduling is a critical problem in leased manufacturing systems, with significant implications for operational efficiency, cost optimization, and machine reliability. Solving this problem involves navigating complex trade-offs between machine-level degradation risks, production throughput, and maintenance team logistics across multi-site manufacturing networks. Conventional approaches rely on large-scale Mixed Integer Programming (MIP) models, which, while capable of yielding optimal solutions, suffer from prolonged computational times and scalability limitations. To overcome these challenges, we propose AttenMfg, a novel decision-making framework that leverages multi-head attention (MHA), tailored for complex optimization problems. The proposed framework incorporates several key innovations, including constraint-aware masking procedures and novel reward functions that explicitly embed mathematical programming formulations into the MHA structure. The resulting attention-based model (i) reduces solution times from hours to seconds, (ii) ensures feasibility of the generated schedules under operational and logistical constraints, (iii) achieves solution quality on par with exact MIP formulations, and (iv) demonstrates strong generalizability across diverse problem settings. These results highlight the potential of attention-based learning to revolutionize O&M scheduling in leased manufacturing systems.

Figures (5)

• Figure 1: Comparison of the mixed integer optimization model approach and the MHA-based method for operation and maintenance scheduling in leased manufacturing system. The benchmark model struggles with adaptability and scalability, while the MHA approach leverages real-time data for robust, scalable, and efficient maintenance scheduling across diverse scenarios.
• Figure 2: Framework for utilizing the proposed AttenMfg approach for operation and maintenance scheduling in leased manufacturing system. Input data, including dynamic maintenance costs, demands, penalties, and machine locations, is processed through multi-step embedding and encoded with multi-head spatial and temporal attention. The AttenMfg framework employs reinforcement learning for training and masking procedures to handle constraints, producing adaptive and efficient maintenance schedules.
• Figure 3: Summary of Solution Gaps for 20 Problems in Each Model.
• Figure 4: Generalizability analysis: solution gaps (log scale) when models trained on different datasets (x-axis) are evaluated on fixed target datasets (a–f). The results highlight that models perform best when the number of sites in the training and testing datasets match, while transitioning to larger configurations with different site numbers results in a significant performance drop, with a maximum observed gap of 53.09%. Additionally, models trained on larger and more complex datasets demonstrate strong backward generalizability, improving performance when evaluated on smaller datasets.
• Figure 5: Heatmap of gap percentages: models trained on random-site datasets evaluated across all configurations. The results highlight the strong generalization performance of random-site models ($LRP15M40$ and $LRP20M50$), achieving low solution gaps across diverse configurations. Notably, the larger random model ($LRP20M50$) performs comparably to testing on the same model and constructing new datasets for specific configurations.