Shahid Beheshti University International Journal of Research and Technology in Electrical Industry 2821-0190 4 2 2025 07 01 Data-Driven Distributed Fault Diagnosis in Multi-Agent Systems with Fractional-Order Sliding Mode Estimator 107101 10.48308/ijrtei.2026.242242.1106 EN Mohsen Shiri Department of Electrical Engineering, Faculty of Engineering, Bu-Ali Sina University, Hamedan, Iran. 0009-0000-2966-4239 Hadi Delavari Electrical Engineering Department, Hamedan University of Technology, Hamedan, Iran 0000-0002-5718-933X Younes Solgi Department of Electrical Engineering, Faculty of Engineering, Bu-Ali Sina University, Hamedan, Iran 0000-0003-1279-0153 Journal Article 2025 10 29 As large-scale systems and industrial processes become more complex and widespread, guaranteeing their safety and reliability is now a primary concern for control and intelligent monitoring. The performance of these Multi-Agent Systems (MAS) is highly vulnerable to faults, leading to significant disruptions. This paper introduces a new Model Free-Distributed Sliding Mode Control (MF-DSMC) method to ensure agents in a homogeneous nonlinear MAS accurately track a reference signal. This approach is combined with a Model Free-Distributed Fractional-Order Sliding Mode Estimator (MF-DFOSME) to provide fault diagnosis. Within the fault detection framework, we develop a Model-Free-Distributed Sliding Mode Observer (MF-DSMO) that utilizes distributed estimation errors to facilitate rapid fault detection. This is done by monitoring the residual signal; if its magnitude crosses a predefined threshold, it indicates a system fault. By integrating the MF-DFOSME, we enhance the performance of the fault diagnosis system (FDS). This combination provides key advantages, including greater robustness, long-term memory, and highly accurate and rapid fault estimation. Finally, we optimize the design parameters using a Reinforcement Learning (RL) algorithm. We use Lyapunov functions to provide a formal stability guarantee for the proposed algorithms. Simulation results confirm that our methods significantly boost MAS performance. The MF-DSMC method yielded a 17% performance improvement, while the proposed FDS method achieved a 26% enhancement over the baseline techniques.