Summary of all THREE submitted Cluster Proposals (including Research Title, Partners, Industrial Collaborators, Project Objectives & Summary)

 

 

1. FIRST (‘small-scale’) Cluster Proposal:

 

1.1 Title: Modular Learning and Co-ordination in Complex Systems

 

1.2 Project Partners: Universities of Strathclyde & Stirling

 

1.3 Industrial Partners: INCITE (Stirling), ACT Club (Strathclyde), Axeon Ltd., National Instruments Corporation,  Emerson Process Management Ltd. (Total committed value: £60,356)

 

Total grant value requested: £556,110

 

1.4 Objectives:

The aim of the project is to develop a new computational framework for coping with significant complexity in dynamical systems.  This will involve developing new co-operative multi-agent based learning control algorithms.  Objectives are:

1.         Introduction of a new theoretical framework and algorithms for autonomous modular learning, hybrid multiple-model, and switching control of complex systems.

2.         Inclusion of novel event-driven adaptation techniques for complex dynamical systems.

3.         Production of a solution with integrated monitoring and diagnostic techniques for complex systems based on a learning multiple model and switching framework.

4.         Development of new computational intelligence based techniques for on-line controller design and re-configuration.

5.         Construction of algorithms for implementation of robust adaptive intelligent controllers for non-linear complex systems and production of software facilities to enable the algorithms to be assessed and validated.

6.         Testing and validation the robust adaptive intelligent control strategy on two major industrial simulations and a real robotic system.

 

1.5 Summary:

The aim is to develop intelligent control numerical algorithms that have the capacity to adjust and optimise complex systems behaviour when responses deteriorate, and to enhance on line incremental system performance.  The new approach will stem from employing new techniques in computing science and non-linear mathematics integrating ideas of robust intelligent control, multi-agent architectures, hybrid multiple-models, switching techniques, data fusion, on-line condition monitoring and soft computing tools.

The proposed computational methods for the control of very complex processes are ambitious and will include learning and intelligence to simplify the design process and provide enhanced overall performance. They will be adaptive so that the algorithms are: able to cope with unpredictable deterioration; autonomous, so no intervention of operators will be required to preserve reliability and performance; robust, so tolerating uncertainties and disturbances; and, modular, using multiple agents, contributing to the overall capabilities.  The resulting novel computational framework will be evaluated on a diverse set of complex real-world applications in order to demonstrate its generic intelligent modelling and control capability, using a real-time robot application and two validated industrial simulations, working closely with companies throughout.

 

 

2. SECOND (‘medium-scale’) Cluster Proposal:

 

2.1 Title:  Towards Multi-agent based Learning Modelling & Estimation in Complex Systems

 

2.2 Project Partners: Universities of Strathclyde, Stirling, Hull & UMIST

 

2.3 Industrial Partners: INCITE (Stirling), ACT Club (Strathclyde), Mondi Paper (UK), alstom Power, Rolls-Royce Naval Machine (Total committed value: £76,000)

 

Total grant value requested: £698,966

 

2.4 Objectives:

The aim of the project is to develop a new modelling framework to cope with complexity in dynamic systems.  The research will develop a new computationally-efficient framework based on multi-agent learning, modelling and estimation framework for coping with this complexity. This will involve new robust modelling techniques based on co-operative learning in a multi–agent based framework including integrated condition monitoring, estimation and diagnosis.  The methodology to be addressed will use learning of hybrid models, receding horizon techniques, evolutionary and biologically-inspired computing, along with game-type optimisation.  Thus the objectives are:

1.         Introduction of a novel theoretical framework and algorithms for distributed learning based modelling.

2.         Development of new agent based estimation for complex inter-connected dynamic systems.

3.         Production of new distributed condition monitoring and fault diagnosis algorithms.

4.         Aggregation algorithms for the implementation of diverse multiple-agents and the associated   development of software facilities.

5.         Assessment, testing and validation of the proposed design framework on two major complex industrial (paper machine & gas turbine) systems.

 

2.5 Summary:

This project is concerned with a new multi-disciplinary approach to intelligent modelling, estimation of variables, and of parameters in complex systems. Modern systems, for example a network of embedded systems in automobiles, intelligent homes, intelligent manufacturing and production systems and multi-robotics are complex in terms of both behaviour and the large-scale nature of their connected and embedded structure.  These systems are difficult to model, manage and control using classical techniques.  Hence, an alternative methodology and as a result the learning and identification issues are a vital challenge for further understanding.

The aim of this project is to develop a new computational framework for modelling and estimation that can cope with levels of complexity and effectively deal with changes in the system produced by non-linearities, disturbances, failures and faults.   This will be achieved by drawing inspiration from computing, behavioural, engineering and mathematical sciences; integrating the ideas of estimation and fault detection (condition monitoring), risk-sensitivity, multi-agent architectures, hybrid models and soft  computation.

Modelling and estimation must be seen as part of a larger picture of managing and monitoring the performance of complex systems.  A multi-agent framework will be an invaluable tool for the development of an efficient method of handling complex systems.  Such a novel framework for modular learning has been proposed for further investigation as part of another proposal submitted by the Strathclyde and Stirling applicants, under the EPSRC Novel Computation Cluster initiative. The goal of this project is to develop new types of learning incorporating event-triggered multiple agents for the design of improved modelling and fault detection algorithms, which forms an integrated part of system condition monitoring and fault tolerance.  The classical fault detection approaches for dynamic systems are based on analytical modelling of continuous systems and cannot immediately cope with the hybrid (mixed continuous and discrete-event) nature of the fault-tolerance in complex systems.

 

 

3. THIRD (‘large-scale’) Cluster Proposal:

 

3.1 Title: Novel Computational Framework for Control of Complex Engineering Systems

 

3.2 Project Partners: Universities of Stirling, Strathclyde, Hull, UMIST, Edinburgh, Sheffield, Brunel, Glasgow

 

3.3 Industrial Partners: INCITE (Stirling), ACT club (Strathclyde), National Instruments Corporation, UK Paper Industry Technical Association, Caterpillar (USA)   (Total committed value: £48,500)

 

Total grant value requested: £1,342,642

 

3.4 Objectives:

 

The project aim is to develop novel computational techniques for analysing, modelling and controlling complex systems.  The research results will be a benchmark for the future development of this particular subject area, thereby enhancing its applicability and acceptance for the analysis and design of modern complex control systems.  To achieve this aim the objectives are to:

1     Establish a novel computational framework for the theory, analysis, design and operation of complex control systems.  This will include the following developments:

1.1        Architectures for the design of systems with learning, fault tolerance and self-repairing features.

1.2        Computational strategies for modelling complex systems

1.3        New control strategies for complex embedded systems

1.4        Optimal decision-support systems for enhanced autonomy and fault-tolerance

2     Demonstrate the generic nature of the framework by assessing, testing, and validating on exemplar systems. 

3     Establish a virtual centre to run major international workshops and conferences on complex control systems studies and provide a forum for stimulating further research in the UK and overseas.

 

3.5 Summary:

Recent advances in digital and computing technology have resulted in a need for a better understanding and development of complex engineering systems.  Advances, particularly in digital technology, are creating the potential for higher levels of integrated, multi-loop distributed and networked control in systems containing large numbers of dynamically interacting uncertain components.  So far these systems have been engineered in an ad hoc or piecemeal manner.  However, the lack of a systematic framework to model complex system has sometimes led to poor system performance, as a consequence of the inability to properly understand the system behaviour.   The result is that we are attempting to build systems where interactions among sub-systems and components cannot be thoroughly planned, understood, anticipated, or guarded against.  This deficiency has resulted in system malfunctions, failures and breakdown or accidents, such as the Northeast America and Swiss power grid failures of 2003.

To cope with this complexity there is a critical need to develop a formal computational framework along with reliable computational paradigms to maintain quality of service along with system safety and security. This project aims to develop the required framework and develop a thorough understanding of the analysis and design issues associated with the complexity of modern systems.  Clearly, such a project requires multi-disciplinary team effort and this research consortium plans to draw inspiration from the emerging areas of science, such as “systems biology” and “bio-complexity” including notions of hierarchical and distributed structures.  This implies the synthesis of computing, behavioural, engineering and mathematical sciences, synthesising control and multiple-agent architectures for distributed decision-making.  In order to tackle these multi-disciplinary issues, this consortium comprises leading experts and their teams from 8 UK universities in a Virtual Centre for Control of Complex Systems.