Home // International Journal On Advances in Intelligent Systems, volume 4, numbers 3 and 4, 2011 // View article

A Universal Model for Hidden State Observation in Adaptive Process Controls

Authors:
Melanie Senn
Norbert Link

Keywords: universal statistical process model; state prediction; regression analysis; dimension reduction; deep drawing

Abstract:
In many manufacturing processes it is not possible to measure on-line the state variable values that describe the system state and are essential for process control. Instead, only quantities related to the state variables can be observed. Machine learning approaches are applied to model the relation between observed quantities and state variables. The characterization of a process by its state variables at any point in time can then be used to adequately adjust the process parameters to obtain a desired final state. This paper proposes a general method to extract state variables from observable quantities by modeling their relations from experimental data with data mining methods. After transforming the data to a space of de-correlated variables, the relation is estimated via regression methods. Using Principal Component Analysis and Artificial Neural Networks we obtain a system capable of estimating the process state in real time. The general method features a high flexibility in adjusting the complexity of the regression relation by an adaptive history and by a variable determinacy in terms of degrees of freedom in the model parameters. The universal model is applied to data from numerical deep drawing simulations to show the feasibility of our approach. The application to the two sample processes, which are of different complexity confirms the generalizability of the model.

Pages: 245 to 255

Copyright: Copyright (c) to authors, 2011. Used with permission.

Publication date: April 30, 2012

Published in: journal

ISSN: 1942-2679