US 6,212,438 B1Granted

Method and apparatus for generating a model of an industrial production

AssigneeSchenk Panel Production Systems GmbH

InventorFrank Reine

Life of this patent

Estimating the term…

Loading patent…

Abstract

A process model of an industrial process or system is generated. The model correlates a first number M of process parameters forming input values with a second number L of quality characteristics forming output values, which are processed to form feedback control signals for the process or system. A third number N of training data sets of the industrial process are first gathered and processed during a learning phase of the model with the help of a central processing unit, whereby a preliminary approximately model is used including a neural network with local approximation characteristics. The neural network is connected in parallel with a linear network. Both networks are connected to the same inputs. The neural network initially has a number N of neural cells corresponding to the number of training data sets. A weighted linear combination of the M process parameters is performed. The linear network and the neural network are connected with their outputs through weighting circuits to a common summing point. A stepwise regression is performed to reduce the number of neural cells from N to K and of linear paths from M to M-R. Closed loop feedback signals control the industrial process.

A method for generating a model of an industrial production to provide control signals for optimizing said industrial production, comprising the following steps:

(a) gathering a number N of training data sets for said industrial production,
(b) connecting a first neural network (5) in parallel to a second linear network (6), said neural network being formed initially by a number of neural cells corresponding to said number N of training data sets, said linear network being formed by M linear paths corresponding to a number M of linear inputs provided in common for said first and second networks (5, 6), to provide a preliminary model for simultaneously performing linear combinations of input values in the first and second networks to thereby train and optimize said first and second networks together,
(c) processing by said neural cells with the application of radial basis functions an input vector into individual first activating values,
(d) applying by said neural cells first weighting factors to said individual first activating values to provide first weighted values,
(e) linearly combining said first weighted values to provide first combined values,
(f) supplying said first weighted values and said first combined values to an output summing circuit,
(g) applying further weighting factors to said input values in said linear second network to provide second weighted values,
(h) simultaneously linearly combining said further weighted values to provide second combined values,
(i) supplying weighted and combined second values to said output summing circuit,
(j) performing an R number of regression steps and terminating said regression steps in accordance with a stop criterium which determines an over-adaptation when said initial number N of neural cells is reduced to a lower number K of neural cells in said neural network and when said number M of linear paths is reduced to a lower number M-R of linear paths in said linear network, to provide a reduced final model,
(k) ascertaining actual process or production parameters and supplying said actual process or production parameters to said reduced final model,
(l) determining by said reduced final model expected or rated quality characteristics of a product to be manufactured by said industrial production to provide respective rated quality output values,
(m) processing said rated quality output values with the help of an optimizing algorithm to provide production control values, and
(n) controlling said industrial production by said control values.

11 more elements

Abstract

Claims9