ICIEA 2007

Robotics Applications Based on Computational Intelligence Technology

The authors group has been studying robotics based on computational intelligence technology, e.g., fuzzy control, neural networks, and chaos prediction. Developed robotics systems by the authors group in the last 25 years listed below have been surveyed by using DVD video images;

1. Conveyor Belt Robot
2. D Ping Pong Robot
3. Flower Arrangement Robot
4. Grasping 2D Irregular Moving Object
5. Yo-Yo Robot
6. Model Helicopter Hovelling
7. Prototype of Biped Robot
8. Irregular Moving Basket Shooting
9. Darts & Pin Ball Shooting
10. Moving Robots in Schools

Then the on going robotics project by the authors group, called - Development Project for a Common Basis of Next-Generation Robots - sponsored by NEDO (New Energy and industrial technology Development Organization, Japan), is presented. The outline of the project is presented by showing DVD video images.

Hirota, Kaoru
Professor
Tokyo Institute of Technology

Computational Intelligence As Modern Heuristics

The term computational intelligence has become increasingly fuzzy, as the words "intelligent" and "smart" are used for everything from clever design of cell phones, cars, to missiles. This talk will try to illustrate three different applications of computational intelligence heuristics. Training of neural networks by design is a very effective approach with applications where a good knowledge of a problem exists. Architectures such as counter propagation networks, genetic algorithm neural networks, or functional and polynomial networks are just some of them. Clearly the use of heuristic is one time honored form of an information based strategy to circumvent the learning process. However, an universal approach to determining network parameters is not known. A novel information base algorithmic design heuristic of a neural network classifier will be elaborated in support of an universal approach idea. Fuzzy arithmetic is one of the sub-domains of fuzzy logic concept. The use of fuzzy arithmetic has been widely exploited in various applications to enhance applicability of quantitative modeling heuristics. A novel heuristic to predictive analysis based on fuzzy arithmetic will be elaborated on a few applications of the computer and communication systems. Neuro-fuzzy controllers have been recognized by their inherent ability to manage two conflicting tasks - comprehensive learning and efficient control based on acquired knowledge. Various approaches exist in literature, often characterized by uncertain and extensive algorithms. A novel and elegant approach will be elaborated on a specific application of fuzzy-neural heuristic for bluetooth mobile devices.

Milos Manic
Professor
University of Idaho

Fractional Order Calculus and Its Applications in Mechatronics and Power Electronics - An Introduction

In the past, for our own convenience, the behavior of materials and systems were modeled using integer order derivatives. However, the real world is of fractional order in nature. The concept of fractional calculus has tremendous potential to change the way we see, model, and control the nature around us. The concept of a real line was not complete until the concept of zero, rational, and irrational numbers was introduced. Similarly, the concept of derivatives will not be complete until the concepts of fractional derivatives (or derivatives of arbitrary order) are introduced. In this talk, we focused on fractional order dynamic systems and controls for mechatronic systems and power electronic systems. Using a DC motor with an elastic shaft example, we demonstrate that FOC outperforms the best integer order counterpart. We also show a FOC application example in DC-DC buck converter. We will also illustrate how to solve some typical fractional order filtering and control problems in Matlab.

Dingyu Xue (Presenter) / YangQuan Chen
Professor
Northeastern University / Utah State University

High Performance Direct-drive Motion Systems

Most advanced industrial manufacturing processes require high-speed and precise motion for material transfer, packaging, assembly, and wiring. Examples are surface mounting of electronic components, wire bonding of semiconductor chips, and assembly of watches and hard-disks.

To achieve precise motion control, most of these manufacturing machines use standard rotary d.c. or a.c. motors as the prime motion actuator, and then couple their output shafts to mechanical motion translators (e.g. reduction gear, shaft coupler, belt, ball screw, etc.). Though this is the most widely used method, it has disadvantages of reduced accuracy, complex mechanical structure, difficult adjustments and alignments, high production cost, and low reliability.

This seminar discusses the worldwide trend of "simplifying the mechanics through specialized direct-drive actuators and advanced control methodologies" in high performance motion systems. It also exploits the possibilities of using variable-reluctance technology to custom design various motion actuators for high performance motion systems. Starting with the basic structure of electromagnetic actuators, the talk examines the possibilities and difficulties of designing innovative motion actuation systems. It then provides a few specialized direct-drive motion system examples (limited stroke actuators, embedded artificial limb, linear motion systems, planar motion systems, and magnetic levitated devices, etc.). The talk will highlight their features and advantages, and the challenges of controlling these devices.

Norbert Cheung
Professor
Hong Kong Polytechnic University

Advance in Signal and Image Processing for Hyperspectral Remote Sensing

With the rapid development and deployment of sensor technology for remote sensing, remote sensing data with diverse spatial, spectral even temporal resolutions that cover different scopes in electromagnetic spectrum are abundant. With such high dimensional data it will be possible to obtain deeper and more accurate information on objectives and their environment, which are of great potential values in scientific remote sensing, lunar observation, and deep-space exploration etc. In this talk, advance in signal and image processing for hyperspectral remote sensing data with large quantity and complicated characteristics thoroughly, will be presented, including, 1) feature extraction by static analysis and intelligent processing, 2) band selection by neural weighting, 3) unmixing method for mixed pixel detection and classification, 4) compression by wavelet transform, 5) fusion of hyperspectral images for classification.

Mingyi He
Professor
Northwestern Polytechnical University

Two Fundamental Challenges in Perceptual Picture Coding and Image Restoration

This paper examines two fundamental challenges in two areas, respectively, which have been intensively researched in the field of image processing and communications, i.e., digital picture coding/compression and digital picture (including both video and still images) restoration (or de-noising). It reflects on historical developments and reviews the state-of-the-art in the area of digital picture coding. Quantitative perceptual distortion measure based on the human visual system is identified as the weakest link in the current picture coding framework and remains a fundamental challenge in devising the next generation picture coding systems. In the area of picture restoration, the paper takes a special interest in surveillance and security video/image de-noising task for forensic investigations. Highlighting most recent advances and the inadequacies in existing image de-noising techniques, it focuses on a fundamental challenge in designing a unified picture de-noising framework, including noise modeling, for removal of analogue and digital surveillance video distortions.

Henry Wu
Professor
RMIT of Australia

Study on the Distributed hybrid control system for continuous industry production line

with the development of modern industry, the scale of production line is larger with more equipments and technics flow is more complex. Therefore, there exist the space distribution and coupling among system state parameters and time -space interrelated relation ,it is a kind of classic distributed hybrid control system. In the paper the continuous hot mill production line is taken as example to establish the hybrid analysis and control model. a hot continuous rolling production line is about 1Km long, the running parameters of the all rolling machines on the rolling line should be computed and set according to rolling technology, which are correlated with a technology constraint. the paper makes a full nonlinear analysis by means of neural network information fusion with application of the real-time running state signals, which include rolling force, steel width, density, steel velocity etc based on a distributed measurement system, then the hybrid control model is given to optimize the rolling parameters progressing. The modelling and control algorithm study of distributed hybrid system are a hot and difficult topic at present, a new innovative concept- State parameter field is presented in the paper, which is used to describe the time-space interrelated relation among the variables of distributed hybrid system The main contents of the paper include : 1) introduction and summary of hybrid system modelling; 2) hot rolling production technology and the mathematical description of State parameter field 3) The analysis of the rolling states with application of data fusion method based on neural network rolling technology parameter interrelated constraint equations 4) optimizing model of the rolling parameters. 5) simulation and experiment results analysis.

Ge Lusheng
Professor
AnHui University of Technology

Assessing the Value of Information in IS-Intensive Automation Industry

What is the value of the information existing in a given automation equipment? The answer is relevant for most decision makers in Industry and Academia when making investments in intellectual property (IP) and information systems (IS). Without a clearly defined price for the information, it is difficult to sustain huge investments in information systems. A related question is about the orthogonality of IP relative to IS: can intellectual property of a company be developed in any kind of information system? Can a global information system such as a digital ecosystem assists intellectual property development? Can the value of vertical integration be measured in Automation?

Clearly, these questions are hard to answer precisely in a field that is so very dynamic and large. We need to consider some restrictions and identify a useful form for the type of answer.

The automation industry has an extensive set of types of information. Control programs, process specifications (ISO 18629), software models (UML), mathematical models, alarms, signals, data on enterprise resource planning - ERP, are just a few. At a more abstract level, the information available for some automation equipment - both information that is stored and information acquired from different processes and internets - can be seen as a chain of symbols si, where 'i' is the generation index of the software. The number of symbols in si, say ni=|si|, does not give alone the size of the problem as the information contained in the chain is defined not only in a Shannon logarithmic fashion, but also by the meta-description of the information type, which is often also included or is implicit in si. Such information is for example the ontology of the domain, semantics, prototype examples or translation rules. Let the set of types in si be pi. Normally technology has an incremental, additive development, with the exception of rare cases when a disruptive technology is introduced. That means, development is a sequence of generations of software, , where typically both the size and the typology of the software increases continuously: n0 > n1 > … > nm and p0>p1>…>pm. If the set of all chains si is S (i=0,…,m), then we can define a metrics r in S as r:S x S e R that quantifies the relative distance between two chains of symbols such that a larger index indicates a higher value for the information. One would expect that the distance r(si, si+1) decreases with each generation, else it would be hard to claim that any purposeful development is taken place. In other words, an additive normal technology development is expected to have a Cauchy fundamental sequence property, where decreased distance with increased i, means the existence of an accumulation point s8. That means, the value of an automation program being at generation 'j' could be expressed as r(sj, s8). Two programs of the same age, 'i' and 'j' but with different properties, could be compared as r(si, sj).

The talk gives a number of definitions for selecting appropriate metrics r, and show by simulations and examples from top automation companies how such assessments could be used in industrial and academic work.

George A. Fodor
ABB Process Automation

Adaptive Control of a Class of Nonlinear Systems with Dominant Hysteresis

Hysteresis nonlinearities are very common in industrial control systems. For decades, the existence of such nonlinearities have provided one of the most difficult challenges to control design engineers since the entire Laplace domain and most state space control design techniques were developed exclusively for differentiable linear or nonlinear systems. Hence, the existence of hysteresis nonlinearities in actuators and systems were neglected and the controllers were designed based on the nominal smooth systems. When the systems are considered with non-differentiable nonlinearities, these methods encountered substantial difficulties in the analysis, model fitting and control design stages. It was extremely difficult, if not impossible, to design or prove stability of such systems. The development of techniques for the identification of such nonlinearities in realistic industrial plants has emerged as a significant problem in itself.

This talk is intended to raise awareness of modeling and control techniques and to provide an opportunity to discuss state-of-the-art solutions for the problems. The presentation and discussion i will range from modeling of hysteresis, to the design of corresponding control schemes, especially in the absence of complete information concerning the system model and state. The talk is designed to appeal to an audience from different backgrounds. People working in the area of control will have a chance to interchange ideas and to view problems from different perspectives. People working in other areas will also benefit by understanding the new methods and technologies developed for control's point of view.

Chun-Yi Su
Professor
Concordia University

Fault-Tolerant H Control of Linear Systems: An Indirect Adaptive Approach

More and more advanced technological systems rely on sophisticated control systems to increase their safety and performances. In the event of system component failures, the conventional feedback control designs may result in unsatisfactory performances or even instability, especially for complex safety critical systems such as aircrafts, space crafts, nuclear power plants, etc. This has ignited enormous research activities in search for new design methodologies for accommodating the component failures and maintaining the acceptable system stability and performances, so that abrupt degradation and total system failures can be averted. This type of control is often known as fault-tolerant control (FTC). Faut-tolerant control design approaches can be broadly classified into two types: Passive approach such as robust fault accommodation approach, and active approach such as linear quadratic regulator(LQR); eigenstructure assignment(EA); multiple model (MM); adaptive approach; pseudo-inverse; and neural networks. In the passive approach, the same controller is used throughout normal and faulty cases, and such that this passive fault-tolerant controller is easily implemented. Moreover, several performance indexes such as H1, H2 and cost functions mainly based on algebraic Riccati equation (ARE) or linear matrix inequality (LMI) methods can be used to describe the performances of closed-loop systems with ¯fixed controller gains. However, as the number of possible failures and the degree of system redundancy increase, a controller designed based on the passive approach becomes more conservative and attainable control performances may not be satisfactory. On the other hand, a fault-tolerant control system based on active approach can compensate for faults either by selecting a pre-computed control law or by synthesizing a new control strategy on-line. Adaptive approach is very important in active FTC, which relies on the potential of the adjustments of parameters to assure reliability of closed-loop systems in the presence of a wide range of unknown faults. Most of the results in adaptive fault-tolerant control are based on model reference adaptive control (MRAC), where the outputs of closed-loop systems can track the prescribed referent outputs. But the disturbance attenuation performances of systems have not been addressed yet within the MRAC framework. Another typical approach for fault compensations is based on fault detection and isolation (FDI) subsystems. However, it should be noted that the fault detection diagnosis (FDI) mechanism might not always give the exact fault information. In this paper, the fault-tolerant H1 control problem for linear time-invariant systems against actuator faults is studied. A general actuator fault model is considered, which covers the outage cases and the possibility of partial faults. A notion of adaptive H1 performance index is proposed to describe the disturbance attenuation performances of systems, and the adaptive approach and LMI approach to robust control are combined successfully to give adaptive fault-tolerant H1 controller design methods for both state feedback case and dynamic output feedback case. Furthermore, the proposed design methods can deal with different fault modes. With the on-line estimations of fault values using adaptive laws instead of an FDI mechanism, a re-configurable control law can be designed to maintain satisfactory adaptive H1 performances. Sufficient conditions for the existence of the above-mentioned adaptive fault-tolerant H1 controllers are given, and it is shown that these conditions are more relaxed than those for the passive fault-tolerant controller designs with fixed controller gains. Comparing with the FTC approach with the need for an FDI mechanism to provide the exact fault information, the new proposed design methods rely on the on-line estimations of fault values using adaptive laws, where it is not necessary for the estimations to give the exact fault information.

Yang Guang-Hong
Professor
Northeastern University