Stock Status: In Stock
In the recent past, neural networks have been used to create intelligent (or soft) sensors to predict key process variables that cannot be measured on-line, to pre-empt lab analysis delays and to validate physical sensors. However, they have had limited application due to two significant drawbacks: first,traditional neural net development is a fairly complicated task requiring extensive third party “expert” intervention; and second, due to their inherent nature, neural nets do not handle changes in process operation over time very well, often requiring retraining. In this paper we present an approach that addresses both problems. Neural Nets are represented as a function block in a structured system, and by using state-of-the-art techniques and intuitive GUI applications, the commissioning process becomes fast and easy. Automation of the following: data collection and pre-processing, identification of relevant process inputs and their delays, network design, and training, enable the process engineer to develop neural net strategies without the need for rigorous techniques. In online mode, the actual process variables, for example those obtained as a result of lab analysis, are used as inputs to the neural net for automatic adaptation of its prediction in response to changes in process. The paper details these and other features of convenience.
- Downloadable files require Adobe Acrobat Reader.
All contents copyright of ISA © 1995-2013 All rights reserved.