Our main goal in creating our development environment for inverse control is to bring up-to-date control solutions to the industry. We carefully scrutinized the practical considerations in designing our environment. We believe that the resulting development environment using adaptive inverse control is very appealing from a practical perspective. Our implementation of adaptive inverse control is shown in Figure 2, where we have split the controller and the adaptation into two blocks to map it to the reality of the industrial environment. We have the microcontroller in series with the plant which provides on-line control, i.e. the plant is under control at all times. We implemented this controller in an industry standard, off-the-shelf, low-cost microcontroller board. This board runs an algorithm that implements the feedforward, non-adaptive, part of the adaptive inverse control architecture.

Figure 2 - Basic Architecture and Implementation

The off-line portion of the controller, runs on a PC using NeuroSolutions. The PC is interfaced with the controller through a serial port (although a LAN or other high-speed/long-distance connection is also possible). It receives the plant output and the command input from the microcontroller and through adaptation finds the best parameters for the controller. The updated parameters are then copied to the on-line microcontroller. The advantages of this arrangement are the following:

• The plant is always under control with a set of weights which are optimal or near optimal.
• Since the training is done separately from the control of the plant, the stability of the control can be maintained using appropriate checks before updating the weights of the controller.
• This scheme can be easily integrated with the existing hardware (most plants already have simple microcontrollers) because the task of the on-line controller is normally rather simple to implement.
• The computationally demanding tasks are executed on the host PC. Here we can choose more costly hardware and run sophisticated learning algorithms, under the control of an operator. A single PC can be multiplexed to update the parameters of several microcontrollers in a network.
• There is no need to develop code for the microcontroller since the code of NeuroSolutions can be recompiled with great savings.
• The PC can run a flexible and user friendly development environment, which is helpful to validate solutions and answer what-if scenarios.

Ventilator Home | Next Page

Products | Support | Order | Download | Search | Contact

Product questions? Contact info@nd.com Website questions? Contact webmaster@nd.com
Web Site Design and Implementation Copyright © 2002 NeuroDimension, Inc.