Main idea: Fusion of information acquired from various sources with a proper consideration of their uncertainty characteristics;
Use of this information in design and implementation of intelligent embedded systems for different applications
Main information sources: Real (comes from measurements, statistics, manuals, publications) data and expert (uncertain, opinion-based, often linguistically expressed) information
Main methods: Conventional techniques and intelligent heuristics
Main goals: Development of the practical methodology with software and hardware tools, user recommendations and manuals, etc.
1. Feasibility investigation of applying new emerging intelligent methods such as expert systems, fuzzy logic, neural networks and interval analysis for expression and evaluation of the measurement uncertainty. Fuzzy and interval theory models are examined for describing some uncertainty components. Research towards the creation of a general theory of measurement and perception.
Rationale for research
The evaluation of uncertainty of measurement is a central concern in all quantitative fields of measurements. The International Standards Organization (ISO) Guide to the Expression of Uncertainty in Measurement establishes a consistent and widely agreed upon methodology based on probability and statistical theories. Since the early 90s and in particular since publication of the ISO Guide, there has been a widening recognition that uncertainty of measurement is no less critical than the value of the measurement result itself. In parallel with this the development of more rigorous, systematic and sophisticated approaches to its evaluation, based on the ISO Guide, have led to an increase in the complexity of its application. The rationale for intelligent technologies applications in the modern metrology environment is driven by:
I believe that a neuro-fuzzy approach can produce models, which can be applied to describe measurement (and more generally perception) uncertainty for both qualitative and quantitative measurements with applications ranging from engineering to the social sciences.
Results achieved so far
2.Feasibility study of measurement methodology being developed above for computer and network security evaluation
Rationale for research
Information assurance and computer security has become one of the most important aspects of information technology and the hottest research field. The goal of this research and development is to improve system security. However, we still do not know how to measure and /or evaluate security and its attributes. Designers often apply information assurance or security technology to systems without the ability to evaluate the impact of those mechanisms to the overall system. And as usual our inability to measure definitely acts as a main constraint to our ability to improve.
Computer and network system security is recognized as a complex issue with no clear definition, with the research domain having no sharp boundaries. Over recent years and especially months, one could see substantial changes in the security problems pattern and their significance on a national and international scenes. While information assurance and security traditionally has been focused on confidentiality of information, the problems of greatest concern today relate to the availability of information and continuity of essential services, which emphasizes the domain of a new emerging discipline of system survivability.
This area will serve as a major application of the previous research with fuzzy and neural networks models being investigated for applications in measurement and evaluation of computer security properties.
1. Design methodology of embedded intelligent systems
Rationale for research
Embedded computer market and product constraints have resulted in new combinations of design constraints, such as very good performance at very low power and very low cost. Technically, this often translates to fast specialised computing using very small memories. There is a strong need in a design methodology providing tools for fast capturing designers expertise on one hand and quick low cost design implementation on embedded microprocessors on another hand. The proposed way is to express the design ideas with the help of fuzzy rules initially, optimise it with neural networks and evolutionary programming for specified implementation constraints. The proposed methodology allows combining the fuzzy system capability of capturing and expressing a knowledge frame with the neural network learning and optimisation abilities. It should develop an engineering approach with with comprehensive consideration of all problems involved in reaching the results from an initial design knowledge formulation to the implementation.
Results achieved so far:
2. Investigation of a neural network (NN) application for a fuzzy system (FS) implementation after its formulation by an expert with fuzzy rules.
Rationale for research
This research is to study NN capabilities of a practical, cost effective implementation of FS and to develop a comprehensive methodology, covering all aspects from an initial FS design through to its microprocessor implementation. This approach allows complementing a convenience of fuzzy rules design frame with an NN flexibility. In order to develop such a methodology different NN types and FS structures have to be analysed and compared. The main problems, which should be addressed, are:
Results achieved so far:
3. Development of a practical user-friendly methodology of a fuzzy controller (FC) design
Rationale for Theoretical Research
Attempts to develop systematic FC design procedures date back to the earliest years of the theory and different approaches have been proposed. Until now, the design process has been mainly conducted in an intuitive, ad-hoc manner. FC design is still more an art than a technology, the area where a designers experience plays the main role. There is a strong need in developing a unifying methodology covering all FC design aspects from its initial formulation with fuzzy rules by an expert through to an implementation on a low cost, general purpose microprocessor, which is very popular with the industry.
Results so far:
Fuzzy controller design for industrial applications.
A universal adaptive fuzzy controller structure allowing for on-line tuning of the scaling factors (both input and output) has been developed and tested by computer simulation and in real control of a synchronous power generator. The structure of this FC is system independent. Its on-line operation supports the overall control loop robustness under various operating conditions. The structure application in different projects is expected.
Fuzzy logic applications in mobile communications
1. The method of fuzzy logic application for mobile locating and positioning has been proposed. Different implementations have been simulated. Simulation results prove the advantage against the conventional methods. Further research and practical implementation is proposed.
2. Fuzzy handover method is proposed and researched. Further development based on an application of extra information available is anticipated.
Fuzzy logic application for electrical power transformer state monitoring and prediction.
In order to predict the additional load that may be placed on a transformer it is necessary to know the temperature regime of different winding parts. The research examines the possibility of neuro-fuzzy models application in an identification of the transformer temperature behaviour patterns. The derived models describe the relationship between the spotted temperatures and external factors such as load and ambient conditions. These models are derived as a neuro-fuzzy system, initially formulated as a fuzzy rules frame based on conventional knowledge. Later the models are tuned using measurement results applied to neural networks. The proposed algorithm has been tested by computer simulation and verified by practical experiments with real transformers.
AI techniques application in electrical energy contracting in competitive electricity market.
Neuro-fuzzy techniques are applied in the development of an effective tool allowing market participants to estimate spot prices and to reduce the risk of electricity contracting. Two methodologies are to be investigated. One is based on the application of fuzzy logic techniques to incorporate human attitudes to a market risk analysis and imprecision of linguistic modeling. Another one is based on combination of fuzzy systems and neural networks and an application of ANFIS systems for modelling the dependence of the spot price upon the predicted demand. The case is tested with the real data from the Victorian electricity market.
Development and implementation of the industrial fuzzy logic controller for power generator.
The fuzzy logic controller was designed and implemented in an excitation control of a synchronous generator connected to an infinite bus through a transmission line. The adaptive fuzzy excitation control system replaces both the automatic voltage regulator and the power system stabilizer. It has been achieved through an introduction of the pre-control stage where a single parameter representing the rotor angle and the terminal voltage was introduced. The control system is implemented on a DSP processor and applied to control a laboratory prototype generator. The reported test results demonstrate both efficiency and robustness of the structure developed.
Development and implementation of the industrial fuzzy logic controller for multi-zone domestic HVAC control system.
The control system has been designed to facilitate the commercial viability, while also achieving robustness in multi-zone control. The cost has been minimised by eliminating the need for airflow sensors, complex duct design and complicated installation and commissioning procedures. The system allows for a distributed control, whereby the fuzzy controllers operation can be disseminated to prevent a large number of zones to overload the master controller. Each fuzzy controller has similar implementation requirements, simplifying software and communication.