home blog research publications projects cv github

PhD Thesis

Abstract

“Improving the Decision Support in Shop Floor Operations by Using Agent-based Systems and Visibility Frameworks” Pablo García Ansola — Universidad de Castilla-La Mancha (UCLM), 2024

This doctoral dissertation addresses the design and implementation of distributed control architectures grounded in intelligent agent technology and Radio Frequency Identification (RFID) for the real-time management of physical and human resources in complex operational environments. The research focuses primarily on airport ground handling operations and industrial manufacturing plants, domains characterised by high operational uncertainty, time-critical constraints, and the need for continuous coordination among heterogeneous actors.

The thesis was awarded the International Doctor distinction (cum laude) by the Universidad de Castilla-La Mancha (UCLM) in June 2012, and a research residency at the Institute for Manufacturing (IfM), University of Cambridge, provided the international dimension required for that designation.

Research Problem and Motivation

The efficient allocation of resources — equipment, personnel, gates, and vehicles — in airport ground handling is an NP-hard combinatorial optimisation problem further complicated by real-time disruptions: flight delays, equipment failures, and demand fluctuations. Classical centralised planning systems lack the responsiveness and fault tolerance demanded in these settings. Similarly, manufacturing environments with RFID-instrumented production lines generate continuous streams of visibility data that existing ERP and MES architectures are unable to exploit for closed-loop adaptive control.

The central research hypothesis posits that a Multi-Agent System (MAS) architecture, in which autonomous agents perceive their environment through RFID and Wireless Sensor Network (WSN) infrastructure and reason via BDI cognitive models, can provide distributed, proactive, and resilient decision support superior to monolithic planning approaches.

Methodology

The research follows a constructive design science methodology structured around three iterative cycles:

  1. Architecture design — Formal specification of agent roles, interaction protocols, and ontologies for resource and event representation.
  2. Prototype development — Implementation of agent platforms using JADE (Java Agent DEvelopment Framework) and JADEX (BDI extension), integrated with RFID middleware (Fosstrak/EPCIS) and WSN gateways.
  3. Experimental validation — Simulation-based evaluation using historical operational data from real airport scenarios, complemented by controlled laboratory deployments on instrumented conveyor and sorting systems.

Technical Architecture

The proposed framework, denominated AUTOLOG-MAS, consists of four principal layers:

  • Perception layer — RFID readers (UHF Gen2, ISO 15693) and ZigBee-based WSN nodes provide real-time asset tracking and environmental sensing. Raw tag events are filtered and aggregated by a Complex Event Processing (CEP) engine before being published to the agent platform.
  • Agent layer — A society of BDI agents, including Resource Agents (representing physical assets), Task Agents (representing operational tasks such as aircraft turnaround subtasks), Broker Agents (mediating resource-task assignments via Contract Net Protocol), and Monitor Agents (detecting anomalies and triggering replanning).
  • Coordination layer — Multi-attribute negotiation protocols and constraint-satisfaction mechanisms enable decentralised optimisation without a single point of failure. Coalition formation algorithms address the combinatorial assignment problem under dynamic constraints.
  • Integration layer — A service-oriented interface (SOA/web services) connects the MAS to existing airport operational databases (AODB), flight information systems (FIDS), and ERP platforms, ensuring interoperability with legacy infrastructure.

Key Contributions

  1. BDI agent architecture for airport resource allocation — A reusable agent model in which each resource (vehicle, personnel, gate) is represented as an autonomous BDI agent with explicit beliefs about its state, desires for task completion, and intention hierarchies for execution. The architecture supports preemptive replanning triggered by RFID-detected disruption events, reducing average resource idle time by 18% in simulation experiments.

  2. RFID-MAS integration framework — A middleware architecture bridging the EPC Information Services (EPCIS) standard event repository with the JADE agent messaging layer. The framework introduces an ontology-based event translation mechanism that maps low-level RFID tag observations to high-level operational events interpretable by BDI agents without manual rule authoring.

  3. Distributed decision support for supply chain traceability — Extension of the framework to milk collection and food-grade cold chain logistics, demonstrating generalisability beyond airport settings. The system provides end-to-end product visibility and automated exception management using the same agent coordination primitives.

  4. WSN-enhanced visibility architecture — Integration of Wireless Sensor Networks with RFID for supplementary environmental monitoring (temperature, vibration) in industrial settings, feeding contextual data into agent deliberation cycles for condition-based maintenance triggering.

  5. Intelligent product concept with embedded agency — A theoretical contribution formalising the notion of products as active participants in manufacturing systems: RFID-tagged items carry processing requirements encoded as agent goals, enabling self-directed routing through reconfigurable production cells.

Results and Validation

The system was evaluated across multiple industrial and airport case studies in collaboration with Spanish logistics operators and European airport ground handlers. Key quantitative outcomes include:

  • Up to 23% reduction in ground handling resource conflicts during simulated peak-hour operations.
  • Average event-to-response latency of under 800 ms from RFID tag detection to agent replanning initiation.
  • Near-linear scalability demonstrated for fleets of up to 200 concurrent resource agents under a distributed JADE container architecture.
  • Successful real-world pilot deployment on an RFID-instrumented picking line, validating the theoretical models under operational conditions.

The research generated over 20 peer-reviewed publications in international journals and conference proceedings, including contributions to IFAC, IEEE, and Springer venues.

Reference Implementation

The complete open-source Python reference implementation of the thesis architecture is available at MAS-DUO. It reproduces the BDI–MDP–IS Platform three-layer architecture, the 4W/EPCIS visibility model, and the Ciudad Real Central Airport case study (Chapter 4.1) using the PettingZoo AEC framework. See the MAS-DUO implementation notes for a detailed mapping between thesis concepts and code.

Thesis Supervisor and Institution

  • Supervisor: Prof. Juan Pavón Mestras / Dr. Holger Billhardt (UCLM)
  • Institution: Universidad de Castilla-La Mancha, School of Computer Science, Albacete, Spain
  • International mention: Institute for Manufacturing (IfM), University of Cambridge, United Kingdom
  • Defence date: June 2012
  • Distinction: Cum Laude, International Doctor