Madhukar Pandit

Optimizing iterative learning control of cyclic production processes with application to extruders

Conventional and new methods for the control of cyclic processes are described and compared on the basis of their performance results achieved in an aluminum extruder plant. The thrust of the work lies in the area of iterative learning control systems. After a brief description of (linear) iterative learning control, the optimizing iterative learning control of cyclic processes is presented. In this method the control input is adjusted from cycle to cycle such that a prescribed quantitative performance index is made to take on an extremum. The results which the presented methods of cyclic control yield when applied to a simulation model of an aluminum extruder are compared with one another. Finally, results obtained in an actual industrial extruder plant are given. The new method yields an increase of production by 10% as compared to methods in current use.

An iterative learning controller with reduced sampling rate for plants with variations of initial states

In this paper a discrete-time iterative learning controller for single input single output systems is presented. The iterative learning controller works with a reduced sampling rate that ensures the reduction of an appropriate norm of the error trajectory from cycle to cycle and can cope with initial state error. Initial state error occurs when the initial state of the system is different from the initial state that is implicitly given by the reference trajectory. If the initial state changes for every learning iteration, then the controller cannot achieve ideal tracking but the error trajectory is bounded. Using two different sample times together with a potentially time variant learning gain improves the controller performance for dealing with initial state error. Simulation examples are presented to show the results of the proposed iterative learning controller with reduced sampling rate.

Isotropic two-dimensional Fractional Brownian Motion and its application in Ultrasonic analysis

In this paper, we first present a new algorithm to generate an isotropic 2 dimensional discrete Factional Brownian Motion, using the method of Cholesky decomposition. Secondly, we derive a two dimensional Maximum Likelihood Estimator to extract the fractal parameters from a given stochastic fractal image. Finally, this algorithm is applied to scale independently parameterize the textures displayed in Ultrasonic medical images. The application is to detect immune reactions in transplanted kidneys.

Speckle reduction in ultrasonic imaging for medical applications

The application of SAR speckle reduction filters for the purpose of ultrasonic imaging is introduced. A new adaptive filter is proposed which yields a significant improvement of sharpness in edge areas and better smoothing in homogeneous regions. The new filter incorporates high statistical moments which are weighted by their normalized local entropy determined by a moment window around the central pixel. Furthermore, a recursive version, which is algorithmically more efficient, is suggested. The results are illustrated by applying the filter to noise-corrupted artificial edge areas and ultrasonic agar phantoms. As a practical application the diagnosis of immune rejections in transplanted kidneys is demonstrated.<<ETX>>

A discrete-time iterative learning control law with exponential rate of convergence

When dealing with the convergence properties of iterative learning controllers, an exponential rate of convergence is desirable. That means a suitable norm of the error trajectory should be reduced from cycle to cycle. In this paper a discrete-time iterative learning controller for single input single output systems is presented. It works with a reduced sampling rate in order to guarantee an exponential rate of convergence. The controller is robust with respect to model uncertainties and excites the system well for performing a system identification. A simulation example shows that the ILC with reduced sampling rate can even cope with initial state error.

A Novel Spread-Spectrum Receiver Synchronization Scheme Using a SAW-Tapped Delay Line

In spread-spectrum communication systems, a basic task which has to be performed is the synchronization of the receiver pseudonoise (PN) signal with the pseudonoise signal contained in the input. In this paper, a synchronization system employing a surface acoustic wave tapped delay line (SAW TDL) matched filter for both initial synchronization (acquisition) and tracking is presented. The periodic correlation peaks at the SAW TDL output repeatedly correct the epoch of the local PN signal clock phase and the code generator initial condition to their correct values. As the correlation impulses are distorted and attenuated due to the effects of message modulation, Doppler frequency shifts and unavoidable interferences, a modulation canceller and a differentiating device are employed to improve synchronization performance. Formulas for estimating the performance of a system incorporating these ideas are given. A hardware implementation of the suggested system has been built and tested. Experimental results obtained with the system are presented.

Analysis of pilot-in-the-loop oscillations due to position and rate saturations

In this paper we deal with the analysis of category II (nonlinear) pilot-in-the-loop oscillations (PIO). PIO phenomena are originated by a misadaptation between the pilot and the aircraft that causes sustained or uncontrollable oscillations, which especially occur during some tasks where tight closed loop control of the aircraft is required from the pilot. Category II PIO are those oscillations that can strictly be correlated with the activation of rate and position limiter elements in the closed loop pilot-vehicle system. This kind of nonlinearity is unavoidably present in every aircraft, because of physical constraints of elements such as stick/column deflections, actuator position and rate limiters, limiters in the controller software and so on. In this paper we propose an approach, based on the describing function technique, to evaluate the nonlinear effects of the simultaneous presence of position and rate saturations in the control loop. The X-15 landing flare PIO is used as test case to demonstrate the effectiveness of the method.

Texture analysis with a texture matched M-channel wavelet approach

A novel method of texture analysis based on the wavelet transform is described. An M-channel extension of the existing two-channel biorthogonal wavelets is proposed. The extension offers a compact and efficient decomposition, a higher degree of freedom in the design of the filter coefficients, and the facility of an iterative linear solution. In contra-distinction to the classical, purely mathematical design procedures of wavelet filters, the proposed design takes into account texture-relevant features. Texture-matched, asymmetric separable 2-D FIR (finite impulse response) filters are obtained which permit the decomposition of the image into texture-feature-dependent pyramid structures downsampled by a factor of M for each direction. The performance of the new filters is tested with the Brodatz textures and compared with the results of other wavelet approaches.<<ETX>>

Automatic generation of morphological opening-closing sequences for texture segmentation

In this paper we present two algorithms to derive sequences of morphological gray-scale opening/closing operators from gray-scale (anti-) granulometries which transform images containing textures of two previously known classes into images within which segmentation can be done simply by using a threshold.

Algorithms for iterative learning control of nonlinear plants employing time-variant system descriptions

Deals with the design of iterative learning controllers (ILC) based on extended state space models for nonlinear cyclic process control. In order to design a suitable learning operator, knowledge about the plants dynamical behaviour is needed which implies that a system model has to be set up. It is expedient to acquire a state space model of the plant using identification methods. Here we deal especially with the case, that a linear model represents system dynamics inadequately. We start with a nonlinear model and linearize the system along the current trajectory, thus obtaining a linear time variant model. Using this as the basis, we develop methods for identification and control of the nonlinear process. Experimental results show that a good system model is also useful to perform a pre-training for the ILC; this is especially interesting in case large deviations from a desired system output trajectory must be avoided. The presented algorithms have been implemented and tested experimentally with a real-life nonlinear processing plant.