<p>Many process control applications are best solved by heuristic or rule-based control. Unfortunately, conventional expert systems are generally large and slow, centralized on workstation computers, and incapable of continuous operation. Futhermore, few expert systems are able to process time-varying data or to reason about temporal relationships. Thus they are ill-suited to process-control which is inherently a continuous and temporal problem and which increasingly relies upon distributed networks of small embedded processors. A new expert system has been developed to overcome these limitations. This system achieves extremely high inferencing performance on "low end" microcontrollers, and requires very little memory. A new "cooperative/advisory" model of distributed problem solving allows networks of processors to cooperate on a problem while remaining able to work independently on distinct subproblems. Knowledge, in the form of facts or rules, may freely migrate around the network. the system incorporates a new algebra for time-valued data and a formal temporal logic for reasoning about this data. The capabilities of this system were demonstrated by automating for the first time the terminal charging subsystem of a model FN Tandem particle accelerator: a problem which is resistant to an analytic solution. Using the expert system, cooperating 68HC16 microprocessors have successfully operated the accelerator, performing as well, or better than an experienced human operator. During the course of these experiments new techniques for technology insertion were devised and a new local-area network for microcontrollers was invented.</p> / Doctor of Philosophy (PhD)

<p>Approaches to the investigation of myogenic and hormonal controls in the mediation of gut motility are well understood, however methods to analyze neural control remain to be developed. I have developed a cannula system with nerve cuff electrodes, subserosal bipolar electrodes, and extraluminal strain gauges to simultaneously monitor the vagal nerve, myoelectric (ECA), and contractile activities in chronic dogs. The cuff electrodes were used to both stimulate and record nerve signals. Five healthy dogs were implanted with such cannula on the gastric area, with the cuff electrodes placed on the branches of the anterior nerve of Latarjet. The condition of the cuff electrodes were monitored by impedance measurements, while that of the nerves under the cuff electrode were studied by electron microscopy (EM). Three Time Delay Estimation (TDE) algorithms: General Cross Correlation (GCC), Smooth Coherence Of Transform (SCOT), and Maximum Likelihood (ML) methods were simulated with three types of signals as inputs: Band limited Gaussian White Noise (GWN), Sine wave (SINE), and Impulses of random intervals (IMP). Results of the analysis of the recorded neural signals show that the three algorithms can be used to study the sensory and motor patterns of the compound nerve signals with the SCOT and ML methods being superior than the GCC method. The results from the EM studies suggested that the cuff electrode caused loss of the myelinated axons and the larger diameter unmyelinated axons. Based on the results of this study, relevant physiological interpretations were also discussed.</p> / Doctor of Philosophy (PhD)

<p>The problem of estimating evoked potentials and its pattern recognition and classification is addressed in this thesis. After providing the relevant physiological background and reviewing the various methods of processing the evoked potential, we propose the method of adaptive noise cancellation for estimating the evoked potential without stimulus repetition. A new weighted exact least squares lattice algorithm is derived for this purpose. The variable weighting factor can be used to make the algorithm robust. Its performance is compared to that of unnormalized and normalized exact least squares lattice algorithms and is shown to be superior. One example of using adaptive noise cancellation to estimate evoked potential without stimulus repetition is presented. Pattern recognition of evoked potentials is achieved by syntactic methods. We derive a finite-state grammar to represent the normal evoked potential. Suitable preprocessing using a zero-phase bandpass filter, parsing and attribute checking are the steps in this classification procedure. A database of normal evoked potentials and optimized acceptance criterion are used for checking the attributes. Detailed training and test runs are performed to demonstrate the performance of this classifier.</p> / Doctor of Philosophy (PhD)

<p>Understanding and minimizing loss are the main objectives of seeking solutions to the problems of stray loss in induction machines. This thesis contributes towards this objective by addressing the various problems of stray load loss. These problems include the questions of definition, origin, components, and effects; theoretical and experimental means of evaluation; and loss reduction.</p> <p>Insights into these problems are achieved through a comprehensive review of the state of the art of the subject. We have established that some commonly used terminologies in the subject area formed major obstacles to progress in definition.</p> <p>A conceptually simple and general theory of squirrel cage induction machines is presented. The theory results in a set of linear periodic differential equations, which has an infinite number of possible solutions. A suitable solution procedure is developed.</p> <p>Means for accounting for slot openings and saturation are developed. These have enabled various field waveforms in a practical machine to be generated and analyzed. The analysis produced insights into the interactions of harmonics and how harmonics contribute to stray load losses. Through this, an approximate means of separating a saturated non-sinusoidal waveform of an air gap flux density waveform into its fundamental, saturation, and other space harmonics is developed. An expression for determining the machine torque is derived. This torque expression, the ideas for manipulating and analyzing the field waveforms, and the presented theory are structured into an algorithm, modeling the behaviours of squirrel cage induction machines. The algorithm enabled the torque-speed characteristic of a practical machine to be predicted.</p> <p>The predicted characteristic is compared with that measured by means of accelerometer. The fact that it compares very well validates the theory and the developed model. The new steady state model has several advantages, including its easy application to the study and the evaluation of stray load losses. A brief study of how harmonics influence the developed torque is also conducted using the new model.</p> <p>Two theoretical methods are developed for predicting stray load loss at the design, manufacturing, or utilization stage of a machine. These methods, which employ the developed model, are applied to a practical machine.</p> <p>The predicted stray losses are compared with the measured, and that predicted using the nominal assignation technique. This draws attention to the need for experimental investigation of the subject. Consequently, two experiments, calorimetric method and novel experiment to study inter-bar current problem, are developed. Due to technical and economic problems, however, their implementation are not yet completed.</p> / Doctor of Philosophy (PhD)

<p>This thesis deals with the analysis of emergency locator transmitter (ELT) signals used for locating crash sites of aircraft or ships through the Search and Rescue Satellite Aided Tracking (SARSAT) System. This system uses the doppler shift phenomenon to achieve its goal.</p> <p>Experiments conducted by the Communications Research Centre (CRC) in Ottawa, discovered problems associated with a family of ELT units produced by certain manufacturers. One of the major problems is the random-phase characteristic of these ELT signals, producing bad spectra, where the doppler shift cannot be easily estimated.</p> <p>In this thesis, the performance of different processors, including the Periodogram (FFT) and the autoregressive moving average (ARMA) method, is investigated as applied to estimating the spectrum of ELT signals. The spectrum is used to obtain the doppler shift. Results of processing computer-generated signals and real test ELT signals indicate the inefficiency of the FIT for random phase signals. However, low order ARMA models are seen to be more effective.</p> <p>A novel processor is suggested here for handling the random-phase signals. This processor is based on a new approach (proposed here) for carrier frequency estimation of signals, using low order MEM or ARMA. Not only is random phase problem solved, but also simplified processing and more accurate results can be achieved.</p> / Doctor of Philosophy (PhD)

<p>Biomagnetism is a two-sided discipline encompassing the magnetic stimulation of excitable tissue and the recording of magnetic fields generated by the action currents in those tissues. The present work explores new avenues in developing potential research and clinical tools for magnetically stimulating and recording muscle activity. This dissertation is presented within a cohesive experimental framework of stimulation, recording and processing. We sought to improve the magnetic stimulation of mammalian nerve, with a view towards niche applications where a minimally invasive procedure may be justified. Increased stimulus repetition rates and excellent spatial focus were achieved using small implantable "magnodes" in dog and rat hindlimbs, respectively. Generally, a three-fold increase in stimulus response was obtained, thus enabling a greater sustained stimulus rate at a given power level. An insensitivity to coil positioning was also observed in the rat. We believe two distinct mechanisms are responsible for the observed improvements in response. A toroidal coil magnetometer was developed, intended as a non-invasive tool for rapidly assessing overall activity in intact muscle. Design of the complete recording system included a low noice amplifier and a novel low frequncy magnetic shield composed of amorphous alloy. Employing non-physiological test sources, the unshielded system could resolve a multipole configuration as an equivalent dipole source. Unfortunately, the shielded could not be completed and the magnetometer's sensitivity remains inadequate for physiological work at this time. A signal processing method was developed for the purpose of filtering individual toroid recordings, and hence preclude the need for multiple stimuli. Due to the shielding problems the method was applied to data from another magnetometer. The filtering strategy offers the speed of fixed filtering with the quality of signal estimates typically found with adaptive techniques. This research has demonstrated the efficient stimulation of mammalian nerve using magnetic fields; the detection of magnetic fields from multipole current sources using non-superconducting technology; the design of lightweight and robust low frequency magnetic shields; and the efficient filtering of individual muscle response records. Continuation of the present work requires the design of a high speed magnetic stimulator and the completion of the magnetic shield.</p> / Doctor of Philosophy (PhD)

<p>This thesis discusses a design methodology that can be employed as a framework for the design of computer-centered performance enhancement systems for technology insertion into non-computerized human-machine systems. This methodology seeks to hybridize and explout the benefits of several diverse areas of electrical and computer engineering, specifically knowledge-based reasoning, fuzzy logic-based control, real-time system design, human-machine interaction, computer-human interface, and software engineering. The methodology for combining these disciplines is used in the realization of a specific application: The Particle Accelerator Control Expert System (PACES), an artificial intelligence-based performance enhancement and control system for a KN-3000 Van de Graaff particle accelerator. PACES combines knowledge-based reasoning and human factors concept to furnish accelerator operators with a computer-centered operations support facility. Knowledge-based reasoning is coupled with real-time systems concepts to provide the capability for autonomous accelerator control. Technology insertion, human-machine interaction and human-computer interface concepts are used to design a simple, safe, flexible and low-cost operations support system that is accepted by both veteran and novice operators alike.</p> / Doctor of Philosophy (PhD)

<p>The thesis is devoted to investigating neurocontrol of nonlinear systems with uncertain and unknown dynamic models. The aim of my research in the neural network area is to search for fast learning algorithm with reduced computation burden. Novel theoretical synthesis and analysis of neurocontrol systems have been conducted and applied to control a robotic manipulator. Modified back propagation learning algorithm making use of the changeable shape of the nonlinear function of node is introduced. The resulting algorithm results in a better accuracy and faster convergence. Neural network based control scheme is used to control the motion of a manipulator. The neural network plays the role of an approximate inverse model of a robot and are then used in conjunction with a conventional proportional plus derivative (PD) controller. To demonstrate the feasibility of the proposed algorithm and neurocontrol scheme, intensive computer simulations were conducted. Different types of adaptive tracking problems and regulation problems are considered. The proposed scheme possesses robustness to systems model uncertainty and changing conditions of operations. Simulation results are quite promising. It is concluded that neural networks, by virtue of their natural ability to learn from data, are well suited for dynamic reconstruction, bringing the world of nonlinear dynamics closer to practical use. To demonstrate the practicability of the proposed scheme, experiments were conducted on an existing two-link manipulator and a single link manipulator. Results confirm the practicability of the proposed scheme. The thesis concludes that the neurocontrol approach is capable of learning the tasks of reasonable complexity and it should be possible to train a system for a variety of operations using a neural network of practical size.</p> / Doctor of Philosophy (PhD)

<p>The major effort in this thesis has been directed towards the performance modelling and reliability estimation of large-scale power systems subject to probable single or multiple contingencies. A scheme for performance data pooling is introduced, which overcomes the problems of the limited available performance history and manipulation of large numbers of data sets associated with various components of a power system. A simulation technique which adopts a Monte-Carlo scheme is developed-to estimate component reliability measures and to select the probable contingency states of the system. A combined optimization/reliability formulation is introduced which provides improved reliability figures of the power system. In this formulation the system control parameters are manipulated to simulate practical contingency situations in which suitable system controls are invoked to preserve as much as possible, the continuity of supply. An efficient network partitioning scheme based on Ward equivalence has been developed. Only those parts of the system which are mostly affected by a contingency are retained for detailed analysis while the rest of the system is modelled by network equivalents. It is demonstrated that the use of such a partitioning scheme yields significant reduction in computational time and storage for contingency analysis while maintaining sufficiently accurate solutions. The partitioning scheme is subsequently combined with the optimization/reliability formulation which adds to the efficiency of both approaches. A technique is developed to guide the partitioning scheme via predicting the change in a performance index due to a contingency. The predicted changes are efficiently calculated using either a perturbed matrix inversion scheme or a suitable nth order derivative formulation. Using these predicted changes, a generalized contingency ranking methodology is introduced, for generation and/or transmission contingencies. The contingency ranking scheme is based on an important class of line loading functions representing system security constraints. A novel and computationally efficient technique for fast linear contingency analysis and ranking is introduced. This technique utilizes a bilinear-based formulation to calculate efficiently, the changes of individual bus voltage angles after the occurrence of a transmission contingency. An alternative equivalent formulation of the technique is aIso presented. It adopts reduced gradients of system states and provides the post-contingency exact changes of these states with significantly reduced online computations.</p> / Doctor of Philosophy (PhD)

<p>This thesis presents a comprehensive approach for the steady state stability analysis of AC-DC power systems. A new method is presented for the evaluation of the system state matrix which is then used to determine system stability and develop new algorithms for the stability analysis and control of large power systems.</p> <p>The method exploits the powerful features of the Component Connection Method for power system modelling and overcomes the disadvantages of the earlier methods. The state matrix is formulated from two separate sets of equations. One set models the component subsystems whereas the other defines the interconnection between the subsystems. The main advantage of this is the great flexibility provided in the modelling of the power system components. As long as the input-output quantities are fixed the modelling complexity of the subsystems may be changed without affecting the interconnection equation. A compact interconnection equation has been derived relating machine voltages and currents in the presence of a multiterminal HVDC network. The subsystems retain their physical identity in this formulation and allow the derivatives of the system state matrix to be easily obtained. The power system operating point is determined by a new sequential AC-DC loadflow scheme. Any AC loadflow method can be used. The DC network is solved using the Gauss-Siedel method and any HVDC network configuration and terminals control scheme can he accommodated. The DC network solution need not be repeated and the method ensures that a feasible HVDC system operating point is selected.</p> <p>A new eigenvalue tracking algorithm has been developed based on the evaluation of the sensitivity of a matrix determinant. It iteratively updates the eigenvalues following any change in the system state matrix at one-third the cost of eigenvalue computation using the QR algorithm. Used together with the proposed state matrix formulation method, it is particularly useful for identifying the modes due to any particular subsystem.</p> <p>Two new methods for decentralized placement have been developed. The first method assigns the given poles among the various subsystems and the elements of the feedback gain matrix are varied to cancel the effects of the system interconnection. The second method is based on the sensitivity of a matrix determinant and solves the decentralized pole placement problem as an inverse eigenvalue problem. Both methods are easy to implement and computationally efficient.</p> <p>The methods presented in this thesis have all been verified by applying them to realistic power system models. These have included a single machine infinite bus system, a three-machine AC system with six buses and nine lines and a three-machine three-terminal AC-DC system. These applications include simulation, analysis and decentralized controller design.</p> / Doctor of Philosophy (PhD)