Πλοήγηση ανά Συγγραφέας "Martini, Stavros"
Τώρα δείχνει 1 - 1 of 1
Αποτελέσματα ανά σελίδα
Επιλογές ταξινόμησης
Τεκμήριο Model-based synthesis and validation of high-performance supervisory controllers for event-driven industrial automation(ΕΛΜΕΠΑ, Σχολή Μηχανικών (ΣΜΗΧ), Τμήμα Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών, 2026-10-27) Martini, Stavros; Μαρτίνι, Σταύρος; Kornaros, Georgios; Κορνάρος, ΓεώργιοςCoordinating logical decision-making with real-time physical control is one of the toughest challenges in embedded systems. Typically, the discrete logic and continuous control are developed separately using different tools, then manually stitched together. Most logic errors happen during this manual integration step. This thesis proposes a two-layer architecture based on formal supervisory control synthesis to solve this problem. The upper layer uses the Eclipse ESCET toolkit and CIF specification language. Each subsystem is modeled as an Extended Finite Automaton (EFA). Safety and coordination requirements are written as formal constraints, and ESCET automatically synthesizes a supervisor. The resulting supervisor has mathematical guarantees: it is controllable, non-blocking, and maximally permissive. Before code generation, we verify additional properties like bounded response time and confluence. The lower layer handles real-time control. Each motor node runs field-oriented control (FOC) using the SimpleFOC library at 20 kHz, completely independent of the network supervisor. Communication between layers happens through a minimal, formally-defined interface: CIF input variables receive sensor data that trigger uncontrollable events, while controllable events generate callbacks that send actuator commands to local nodes. Importantly, local nodes handle their own safety reactions without waiting for supervisor approval. We tested the architecture on a physical two-node BLDC motor network connected via CAN 2.0B, implementing a Heat Recovery Ventilation (HRV) system. The application required 46 formal requirements covering coordinated startup, sequential calibration, fault recovery with bounded retries, and strict “both or nothing” fan operation. We injected electrical faults, thermal faults, and power loss on live hardware. In all cases, the supervisor behaved exactly as the formal model predicted—no deadlocks, no requirement violations. The proposed approach drastically reduces logical integration errors and provides a commercially viable solution from the formal specifications to the embedded code.