Engineering Applications of Artificial Intelligence
IMS 10—Validation of a co-evolving diagnostic algorithm for evolvable production systems☆,☆☆
Introduction
Modern control approaches already heavily rely in networking for the synchronization of production processes. With the pervasiveness and availability of affordable tiny computing devices it is anticipated that the overall complexity of industrial networks will increase quite substantially in incoming years.
Companies are not obviously introducing ad-hoc and purposeless complexity in their automation solutions. With a significant pressure towards achieving a more sustainable production environment it is fundamental to tight the system's regulatory mechanisms in order to avoid premature disposal of materials and other sources of energy waste.
The disposal of equipment, mostly motivated by the introduction of changes at shop floor level, is a significant threat to sustainability. The adoption of open automation architectures, as envisioned by a series of emerging production paradigms: Bionic Manufacturing Systems (BMS) (Ueda, 1992), Holonic Manufacturing Systems (HMS) (Babiceanu and Chen, 2006), Reconfigurable Manufacturing Systems (RMS) (Koren et al., 1999), Evolvable Assembly Systems (EAS) (Onori, 2002) and Evolvable Production Systems (EPS) (Barata et al., 2007a); while not relaxing the complexity issue provides an opportunity to tackle pressing complexity related events (failures, fault propagation, tracking, etc.) in an integrated and holistic approach.
This integrated approach is also interesting from a business point of view for module providers that could potentially add significant value to their modular systems by embedding part of the control solution.
Emergent control approaches have developed towards the Lego metaphor where self-contained and function-specific autonomous blocks are combined to implement a specific process. The metaphor itself has evolved starting with the notion of Flexibility whereby one of these constructs is able to perform many functions and can simultaneously take part in many processes mostly associated with the production of a specific product line. Becoming agile was the next paradigmatic step in competitiveness. Agility is different from Flexibility. The latter often refers to the ability of producing a range of products (typically predetermined). It is also different from being Lean (producing without waste). Agility implies understanding change as a normal process and incorporating the ability to adapt and profit from it. Agility covers different areas, from management to shop floor control and regulation aspects. It is a top down enterprise wide effort. The agile company needs to integrate design, engineering, and manufacturing with marketing and sales, which can only be achieved with the proper IT infrastructure (Kidd, 1994, Goldman and Nagel, 1995, Goranson, 1999, Christopher, 2000, Christopher and Towill, 2001, Barata, 2003, Ribeiro and Barata, 2009). Sustainability is therefore the next step for agile companies which can be attained as these increasingly modular systems become adopted by industry.
The notion of Evolvable Production System (EPS) is consolidating in this context and can be envisioned as a broader umbrella for a wide range of design, architectural and technical considerations firstly explored under the framework of Evolvable Assembly Systems (EAS). The initial EAS concept dating from 2002 (Onori, 2002) was further developed under the FP6 EUPASS project. Current research includes the developments in the scope of the FP7 IDEAS project where the application of EPS concepts in networks of embedded tiny controllers is being pursuit. The essence of EAS/EPS resides not only in the ability of the system's components to adapt to the changing conditions of operation, but also to assist in its overall evolution in time so that processes become more robust.
In this context a system is a highly dynamic entity whose structure and processes evolve. From a diagnostic point of view it is somehow difficult to characterize and model such a system using conventional approaches as these have either been applied within the scope of specific devices or to entire installations with constraining degrees of freedom regarding evolution or adaptation. To a certain extent, diagnostic systems are one of a kind (tailored accordingly to the installation's peculiarities (Thybo and Izadi-Zamanabadi, 2004)).
Preserving the decoupled nature of modules is crucial for the robustness and prompt response of an EPS-based system. It is also the key to support system's evolution and adaptive response on the imminence of change. Hence, it is required a diagnostic system that captures the evolving nature of faults (exploring the network dimension of the system and complementing local/specific diagnostic methods) while ensuring that the system's components remain autonomous and decoupled units.
The self-organizing nature of EPS implies that the convergence of the system to macro-states and the events at module level are of probabilistic nature. In the proposed diagnostic approach self-organization and a local and probabilistic diagnostic model based on a Hidden Markov Model (HMM), which was designed to capture fault propagation events, are combined so that a global and consistent diagnostic consensus emerges. It is important to stress that the considered diagnostic approach envisions the subject of diagnosis from a systemic perspective essentially and evaluates how the system components, which may be other systems themselves, affect each other. The purpose of this paper is therefore twofold:
- •
To show that it is significant to consider this network/systemic dimension of the diagnostic problem detailing a fault simulation model that considers agent-based mechatronic networks of different complexity and whose components are more or less prone to fault propagation events while clarifying how small changes in parts of the system may influence the propagation of a fault.
- •
To demonstrate one solution to perform diagnosis at this level that consumes local network connectivity information as well as harmonized agent's sensorial data.
The decoupled, distributed and self-organizing nature of the diagnostic response is analysed for different operating conditions to shed light of the potential and limitations of this approach. Hence, the proposed approach does not perform diagnosis at component level, it concentrates on explaining how one of these component specific events may be abstracted and interpreted from a network perspective and used to diagnose pervasive failure scenarios.
This rationale reflects in the subsequent details that are organized as follows: Section 2 briefly presents the related literature, Section 3 depicts the system architecture, Section 4 presents the testing set-up and briefly addresses some implementation details, Section 5 details the main results, Section 6 discusses the main conclusions and points future research directions.
Section snippets
Traditional diagnostic approaches
Fault diagnosis activities are fundamental to ensure the sustainability of systems and have taken many different forms and approaches throughout the years while accompanying the main technological and socio-economic challenges and opportunities.
One can relate the emergence of diagnosis practices with the beginning of machine instrumentation in the early industrial revolution whereby human operators were assigned the task of checking operational limits and performing minor maintenance tasks such
The interaction metaphor
One of the fundamental characteristics that render natural systems rich is their ability evolve and adapt encompassing small changes that gradually build up. As far as can be observed natural systems interconnect. This interconnection happens at an outstanding number of levels and interfaces not always perceived or understood.
In the context of this work industrial set-ups are perceived in this perspective: a pool of interactive entities (modules, systems and subsystems) in many-to-many
Preliminary notes on the Mechatronic Agent framework
From an IT perspective EAS/EPS are built upon the Mechatronic Agent construct. A Mechatronic Agent (MA) is an agent-based entity that encapsulates all the necessary mechatronic control and regulatory (monitoring, diagnosis, recovery and maintenance) aspects of a system or a process oriented module harmonizing the physical characteristics and functions of the module with the Multiagent System runtime environment.
One of the cornerstones of EPS is the integration of legacy equipment. It should be
Validation of the fault propagation model
The fault propagation model (working assumptions and conditions detailed in Section 3.1) was introduced as a mean to evaluate the performance of the proposed diagnostic approach and its response to significant network measurements such as complexity and node vulnerability. The behaviour of the proposed model is statistically characterized to justify the working points and networks later considered when assessing the diagnostic approach. The following are the control variables of the propagation
Conclusions and future research directions
The results attained suggest that it is in fact possible to explore local interactions and information to promote a self-organizing diagnostic response in an EPS system or any other system constituted by decoupled interacting entities for this matter.
There are however some limits that ought to be considered prior to the instantiation of such an approach. While the number of agents in the network does not seem to significantly affect the performance of the diagnostic algorithm, the complexity of
References (93)
- et al.
Fault diagnosis of the multi-stage flash desalination process based on signed digraph and dynamic partial least square
Desalination
(2008) - et al.
A multiagent-based control system applied to an educational shop floor
Robot. Comput. Integr. Manuf.
(2008) - et al.
Neural networks: applications in industry, business and science
Commun. ACM
(1994) Elephants don't play chess
Robot. Auton. Systems
(1990)- et al.
Intelligent mobile agents in elderly care
J. Robot. Auton. Systems
(1999) - et al.
Development of fault diagnosis strategies based on qualitative predictions of symptom evolution behaviors
J. Process Control
(2009) The agile supply chain: competing in volatile markets
Ind. Mark. Manage.
(2000)- et al.
Multi-agent framework based on smart sensors/actuators for machine tools control and monitoring
Eng. Appl. Artif. Intell.
(2006) - et al.
Diagnosing multiple faults
Artif. Intell.
(1987) - et al.
Sensor fault tolerant control using sliding mode observers
Control Eng. Pract.
(2006)
Internet based diagnosis of assembly systems
CIRP Ann.—Manuf. Technol.
Supervision of nonlinear adaptive controllers based on fuzzy models
Control Eng. Pract.
Intelligent automation systems for predictive maintenance: a case study
Robot. Comput. Integr. Manuf.
An extension of QSIM with qualitative curvature
Artif. Intell.
An algorithm for diagnosis of system failures in the chemical process
Comput. Chem. Eng.
Process fault detection based on modeling and estimation methods—a survey
Automatica
A review on machinery diagnostics and prognostics implementing condition-based maintenance
Mech. Systems Signal Process.
Reconfigurable manufacturing systems
CIRP Ann.—Manuf. Technol.
Qualitative simulation
Artif. Intell.
A model-based approach to robot fault diagnosis
Knowledge-Based Systems
Industrial applications of agent technologies
Control Eng. Pract.
Observer-based fault detection and isolation: robustness and applications
Control Eng. Pract.
Application of signed digraphs-based analysis for fault diagnosis of chemical process flowsheets
Eng. Appl. Artif. Intell.
Supporting agile supply chains using a service-oriented shop floor
Eng. Appl. Artif. Intell.
An integrated knowledge representation scheme and query processing mschanism for fault diagnosis in heterogeneous manufacturing environments
Robot. Comput. Integr. Manuf.
Robust state estimation and fault diagnosis for uncertain hybrid nonlinear systems
Nonlinear Anal.: Hybrid Systems
A survey of design methods for failure detection in dynamic systems
Automatica
Development and applications of holonic manufacturing systems: a survey
J. Intell. Manuf.
A probabilistic approach to fault diagnosis in industrial systems
IEEE Trans. Control Systems Technol.
Bibliography on the application of probability methods in power system reliability evaluation
IEEE Trans. PAS-91 Power Appar. Systems
Swarm Intelligence: From Natural to Artificial Systems
Fundamentals of expert systems
Annu. Rev. Comput. Sci.
An integrated model for the design of agile supply chains
Int. J. Phys. Distrib. Logistics Manage.
Cited by (2)
Artificial immune systems based multi-agent architecture to perform distributed diagnosis
2019, Journal of Intelligent ManufacturingBIOSOARM: a bio-inspired self-organising architecture for manufacturing cyber-physical shopfloors
2018, Journal of Intelligent Manufacturing
- ☆
The material in this paper was partially presented at the 2010 10th IFAC Workshop on Intelligent Manufacturing Systems (IMS 2010), July 1–2, 2010, Lisbon, Portugal.
- ☆☆
Note: This paper is an extended version of a communication presented in the IMS10 workshop.