abstract
This report presents the design, implementation, and evaluation of a Multi-Agent System (MAS) for autonomous vehicle predictive maintenance that unifies three graduate courses of the University of Bologna’s Master’s programme in Computer Science and Engineering: Machine Learning and Data Mining (Codice 95631), Distributed Systems (Codice 87474), and Embedded Systems and IoT (Codice 77780).
The system demonstrates how three seemingly independent technical domains combine to solve a real automotive problem: keeping a distributed fleet healthy while minimising unplanned downtime and emergency repair costs. At the edge, ESP32 telematics devices collect sensor data, derive ECDSA signatures from each vehicle’s VIN, and publish authenticated telemetry over MQTT, with a dedicated Arduino controller acting as a hardware authentication gateway.
A permissioned Hyperledger Fabric network of five peer nodes provides Byzantine fault-tolerant, immutable storage for telemetry and service records, sequencing transactions through a Raft ordering service. A Random Forest classifier trained on 50,000 vehicle records predicts maintenance needs with high accuracy and exposes feature-importance reasoning for explainability.
Above these three layers, a Jason AgentSpeak multi-agent system coordinates vehicles, service centres, and a fleet coordinator through stigmergy-based signalling and consensus-validated agreements. The integrated system targets an 80% reduction in unplanned downtime and an estimated saving of €165,800 per 10,000 vehicles annually (€16.58 per vehicle).
This document also details the Service Center Agent contribution, Mary Anne Selirio, framed as a capacity-focused resource manager that tracks technician availability and parts inventory, allocates service slots in response to fleet booking pressure, and logs immutable service records to the blockchain.
keywords
Multi-Agent Systems, Predictive Maintenance, Stigmergy, Byzantine Fault Tolerance, Hyperledger Fabric, Raft Consensus, Random Forest, ESP32, ECDSA, MQTT, Jason, AgentSpeak, Intelligent Embedded Systems.
outcomes
