IMS 10-Validation of a co-evolving diagnostic algorithm for evolvable production systems

Research output: Contribution to journalArticlepeer-review


With the systematic implantation and acceptance of IT in the shop-floor a wide range of production paradigms, relying in open interoperable architectures, have been developed. Exploring these technological novelties, they promise to revolutionize the way current plant floors operate and react to emerging opportunities and disturbances. There is a high interest of module providers in the adoption of these open mechatronic architectures as they may provide a new business model where the automation solution can be easily tailored for each customer in due time and ships with a significant part of the control solution (high added value). Final customers on their side can contract operating hours rather than buying modules. Moreover, the automation solution can be swiftly modified to meet changing requirements. The necessary increase in the number of distributed and autonomous components that interact in the execution of processes implies that new diagnostic approaches should be developed to tackle the network layer of these highly dynamic systems. In fact fault propagation events can be harder to understand and can affect the system in unpredictable and pervasive ways. Following this rationale the paper presents a potential diagnostic solution that targets multiagent-based mechatronic systems where their components are highly decoupled from a control point of view. The diagnostic architecture presented tackles the problem of fault propagation while preserving the decoupled nature of the Mechatronic Agent concept. In this context the diagnostic system explores selforganization to enact an emergent response that denotes macro-level coherence. The system's response is the result of an individual probabilistic diagnostic inference based on Hidden Markov Models that capture the propagating nature of a failure. The validation results of the proposed diagnostic approach are detailed for the system's response in simulation (highlighting the main variables that affect the performance of the system) and compared to the system applied to a pilot assembly cell. The simulation model and the performance metrics considered are detailed and discussed along with the main implementation details. (c) 2012 Elsevier Ltd. All rights reserved.
Original languageUnknown
Pages (from-to)1142-1160
JournalEngineering Applications Of Artificial Intelligence
Issue number6
Publication statusPublished - 1 Jan 2012

Cite this