Diagnosis of an EPS module

Dissertação apresentada na Faculdade de Ciências e Tecnologia da Universidade Nova de Lisboa para obtenção do grau de Mestre em Engenharia Electrotécnica e Computadores

This thesis addresses and contextualizes the problem of diagnostic of an Evolvable Production System (EPS). An EPS is a complex and lively entity composed of intelligent modules that interact through bio-inspired mechanisms, to ensure high system availability and seamless reconfiguration. The actual economic situation together with the increasing demand of high quality and low priced customized products imposed a shift in the production policies of enterprises. Shop floors have to become more agile and flexible to accommodate the new production paradigms. Rather than selling products enterprises are establishing a trend of offering services to explore business opportunities. The new production paradigms, potentiated by the advances in Information Technologies (IT), especially in web related standards and technologies as well as the progressive acceptance of the multi-agent systems (MAS) concept and related technologies, envision collections of modules whose individual and collective function adapts and evolves ensuring the fitness and adequacy of the shop floor in tackling profitable but volatile business opportunities. Despite the richness of the interactions and the effort set in modelling them, their potential to favour fault propagation and interference, in these complex environments, has been ignored from a diagnostic point of view. With the increase of distributed and autonomous components that interact in the execution of processes current diagnostic approaches will soon be insufficient. While current system dynamics are complex and to a certain extent unpredictable the adoption of the next generation of approaches and technologies comes at the cost of a yet increased complexity.Whereas most of the research in such distributed industrial systems is focused in the study and establishment of control structures, the problem of diagnosis has been left relatively unattended. There are however significant open challenges in the diagnosis of such modular systems including: understanding fault propagation and ensuring scalability and co-evolution. This work provides an implementation of a state-of-the-art agent-based interaction-oriented architecture compliant with the EPS paradigm that supports the introduction of a new developed diagnostic algorithm that has the ability to cope with the modern manufacturing paradigm challenges and to provide diagnostic analysis that explores the network dimension of multi-agent systems.