Community self-organization based multi-agent collaboration for industrial sensor fault detection

Industrial sensor networks in monitoring and control systems often suffer from missing data and abnormal faults during long-term operation, threatening production safety and system reliability. Traditional monitoring methods based on isolated single-variable analysis lack robustness under complex and varying operating conditions. To address these challenges, we propose (community-collaborative multi-agent abnormal detection), a community-constrained multi-agent framework for robust sensor fault detection in industrial monitoring systems. First, a unified single-agent module integrating temporal deep learning and hybrid probabilistic modeling is developed to perform probabilistic modeling of prediction errors for anomaly detection while serving as a basic unit for collaborative data reconstruction. Second, a community self-organization mechanism is proposed to dynamically group agents based on spatio-temporal similarity, enabling community-level information sharing through shared error distributions and fault-context templates. Finally, a reliability-driven collaborative imputation strategy is proposed to select trustworthy peers for missing-data reconstruction, with instance-level post-processing to prevent masking underlying fault patterns. Experiments on real-world sensor datasets demonstrate that improves data availability and enhances the robustness and sensitivity of sensor fault detection, supporting reliable industrial monitoring under complex conditions.