Research Programs and Interests


My current theoretical research activities involve the development of adaptive machine learning algorithms for use in various nonlinear estimation and control problems encountered in the study of dynamic systems. Most of the algorithms we develop are intended for use with recurrent and feedforward neural networks. Recently, my applied research interests have concentrated on the application of these intelligent signal processing and control algorithm developments in two areas of system operations. The first area of application is the remote assessment of mechanical and electromechanical system life-cycle health, and the related area of life-extending controls. More recently, we have been focusing on the application of the fundamental algorithmic developments of the past, to solve (a) estimation and control problems related to inelastic (or real-time) traffic over best-effort computer networks and (b) estimation and control problems found in distributed software systems deployed over best-effort networks.

Research in Adaptive Estimation for Stochastic Systems: We have been working on the development of adaptive machine learning algorithms that could be used to solve nonlinear stochastic prediction and filtering problems encountered in many areas of engineering. We are primarily interested in developing algorithms that are applicable to adaptive problems not amenable to solutions using traditional techniques, such as the Kalman Filter and the Extended Kalman Filter. A major focus of this research has been the development of learning algorithms for accurate and accelerated training of recurrent neural networks. An additional thrust of this research has been to develop fast learning algorithms for training feedforward neural networks with large data sets. Results from both aspects of our research have been applied towards the solution of real-world adaptive state filtering and prediction problems with relative success. We are currently pursuing improvements in the accuracy and convergence speed of these algorithms with applications drawn from various problems in electromechanical systems, process systems, and real-time computer network traffic.

Research primarily suited for Ph.D. students. Requires background in stochastic processes, estimation and filtering theory, digital signal processing, discrete-time systems, neural networks and optimization.

Research in Intelligent Systems for Life-Cycle Health Assessment: One of the major thrusts of my applied research interests is the development of methods and algorithms that can be used to assess the health of various engineering systems throughout their life-cycle, as well as to obtain estimates of their expected end-of-life. We have been using adaptive filtering and prediction techniques based on recurrent neural networks, and multiresolution signal processing techniques to enable early detection and diagnosis of electrical and mechanical faults in electromechanical systems. Our thrust has been to develop systems that are non-intrusive to the operations and require the use of only standard sensors. Research on the equipment health assessment over computer networks is also part of this thrust. This research has both algorithmic and experimental components the latter used to test the various methods being developed. We have tested our developments in laboratory scale set-ups and in industrial scale electromechanical systems subjected to staged failures. Some field trial efforts are underway and some are being planned.

Research suited both for Ph.D. and M.S. students. Requires background in mechatronics, instrumentation, electromechanics, digital signal processing, neural networks.

Research in Intelligent Estimation and Control of Computer Networks and of Distributed Software Systems: The widespread deployment and use of best-effort networks, such as the Internet, and the use of distributed software systems over best-effort networks has brought forward numerous complex estimation and control problems with no known solutions. These problems become of particular interest as the fraction of real-time traffic transported over best-effort networks increase, and as more distributed software systems that are sensitive to network latency are deployed over best-effort networks. Obtaining solutions to some of these problems could significantly improve the operational performance of best-effort networks that form a large fraction of the existing Internet infrastructure, and could perhaps help enhance the performance of the planned next-generation infrastructure, e.g. Internet2. Furthermore, effective solution of some of these control problems could enable software engineers to enhance the performance of next generation distributed software systems. Adaptive estimation and control approaches appear to offer some possible solutions. Currently, we are using techniques from neuro-identificaiton, neuro-filtering, neuro-control and related control engineering techniques to solve various estimation and control problems that are characterized by significant time-varying time delays.

Research suited both for Ph.D. and M.S. students. Requires background in digital signal processing, discrete-time systems, dynamic systems and control system design, computer networks, neural networks, optimization.

NOTE: All of the above areas of research require extensive programming skills in C, C++, Matlab and some Java. Research involving system interfacing also requires extensive knowledge of LabVIEW and DSP programming.

 


Last Modified: September 19, 2001