F. Declercq

Using Levenberg-Marquardt Minimization in Neural Model Based Predictive Control

Abstract Simulation is a commonly used technique to learn the behaviour of a system. In a large number of situations the model of a system is not known and can only be characterized by the relationship between the input and output data. Fitting the model to the data is better known as identification of the system. For linear systems this technique is generally well known. Identification of non-linear systems is much more difficult. When no a priory knowledge is available, the structure of the model depends only on the choice of the control engineer. Neural nets can be a good choice for the system model. Instead of using the backpropagation training algorithm, the parameters are calculated using the Levenberg-Marquardt algorithm. The neural model is used in a model based predictive control algorithm.

A Real-Time Vision Controlled Robot Tracking a Randomly Moving Object

Abstract The environment of a robot is not fixed anymore like in the early days of robotics. The production engineers demand robots to work in a flexible environment. To fulfill this requirement we investigated a case where a CCD camera was connected to a robot (SCARA-type). The robot has the task to track and find an object which is moving in and out the workspace of the robot. The positions of the servo motors of the robot are controlled by a transputer network using 4 T800 transputers.

Object Recognition and Selective Handling by a Robot

A small flexible production cell has been built around a selectively compliant articulated robot arm. Moving on a conveyor belt, boxes marked with different labels are presented to the robot in a random order. Using a camera and a vision card, the labels on the boxes are recognized. Each one of the labels can be rotated, translated or scaled. Three different invariant feature extraction methods (signature, invariant moments of Hu and Zernike) are compared. A neural net is used to classify the labels. The task of the SCARA robot is to pick up the moving boxes and to sort them according to their labels.

A neural model based predictive control toolbox for MATLAB

Using the simulation package MATLAB, a toolbox with a graphical users interface has been built. The main use of this toolbox is to control non-linear systems using the predictive control strategy. This control technique needs a model of the system which can predict the output of the system. The non-linear identification can be done using a set of different neural models such as feedforward, radial based and Elman neural nets.

A flexible production cell

Abstract A small flexible real-time production cell was built around a SCARA robot. On a PLC-controlled conveyor, boxes marked with different labels are presented to the robot in a random order. Using a CCD-camera and a vision card, the labels (squares, triangles,…) on the boxes are recognized. Each one of the labels can be rotated, translated or scaled. Three different invariant feature extraction methods (signature, invariant moments of Hu and Zernike) are compared. A neural net is used to classify the labels on the boxes. The task of the SCARA robot is to pick up the moving boxes and to sort them.