Data processing: measuring – calibrating – or testing – Measurement system – Performance or efficiency evaluation
Reexamination Certificate
2002-10-02
2004-09-14
Barlow, John E. (Department: 2863)
Data processing: measuring, calibrating, or testing
Measurement system
Performance or efficiency evaluation
C162S198000, C703S007000, C702S183000
Reexamination Certificate
active
06792388
ABSTRACT:
DEFINITION OF THE NATURE OF THE INVENTION
The invention concerns a method and system for monitoring and analyzing a paper manufacturing process, in which method a large number of quantities are measured from the process and entered as an input vector into a neural network which produces an output vector as a continuous quantity, and in which method at least one fingerprint representing a good process situation, i.e. an optimal output vector is defined, and fingerprints or output vectors obtained in a normal process situation are compared to the said optimal finger-print(s) or output vector(s) substantially in real time and, based on this comparison, the difference is determined and presented to the user in a graphic form.
BACKGROUND OF THE INVENTION
Using teachable neural networks, large amounts of data can be effectively classified and links or groupings present in measurements and large masses of data can be revealed that are very difficult to detect via statistical calculation, mathematical models and logical rules. The learning ability makes it possible to perform various functions with a reasonable accuracy by the aid of examples without detailed programming of all different situations and exceptions.
A neural network consists of simple computing elements, neurons, having a plurality of inputs and one response. A neural network is created by linking neurons to each other. The overall operation of the network is a combination of several weighted values which is difficult to understand via individual weights. The weights in the neural network are generally determined by teaching using examples. There are a fairly large number of different network structures and teaching algorithms.
Diversified descriptions of the principles and applications of neural networks are to be found in the publications “Neurolaskennan mahdollisuudet”; Pasi Koikkalainen, Tekes 1994, and “Neural Computing Theory and Practise”; Philip D. Wassermann, New York 1989, ISBN 0-442-20743-2.
In a paper manufacturing process, the operator should be able to obtain concentrated information about how well the process is performing as compared with previously identified good situations. So far, the paper manufacturing process has been analyzed e.g. by the SOM (Self Oriented Map) technique. Quite often, history data for a large number of variables is available in the form of various curves, of which the operator can select desired ones to be presented on his display. Generally, detecting a significant change in the set of history values is impossible because such changes are lost in normal random variation. Therefore, a method is needed that allows the entire process situation to be analyzed using a neural network and divergent situations to be classified as early as possible to enable the operator to start searching for the cause that led to the situation.
Neural network solutions so far applied in the paper manufacturing industry have not produced satisfactory results. Analyzing the results is difficult for the user; especially the results obtained by the above-mentioned SOM method are unclear.
Finnish patent application 941058 (Taipale) presents a method for the treatment of especially a paper manufacturing process. According to this application, a set of measurement results from the process is fed into a neural network, e.g. a perceptron network, and, using a special algorithm, adjustable variables are corrected in order to correct the operating point of the process. The method aims at directly optimizing the quality of the final product, which is determined via 100-300 measurements for each machine roll.
European patent 815320 (Furumoto) discloses a method for process control of a paper machine by utilizing a neural network. A set of measurements comprises spectral values of the substances used, and statements regarding the quality of the product are transmitted via the neural network and corresponding control signals are passed further to a so-called stock preparation stage. In this case, too, product quality is the primary control variable. Spectral (optical) measurements are ill suited for the determination of chemical changes.
U.S. Pat. No. 5,347,446 (Iino) presents a regulation system resembling the process and based on creating a model of the process. Computations produce a scalar cost function as a result. In practice, creating a model is probably only possible in the case of limited processes. It is not possible to create a model for the entire paper machine by this method.
In general, the experiences about control processes designed to control the entire paper machine are not encouraging. Among other things, incidental factors have a considerable effect on the process, which means that a rationally calculated correction in itself produces more changes and instability.
BRIEF DESCRIPTION OF THE INVENTION
The object of the present invention is to achieve a new type of method utilizing a neural network in a paper manufacturing process whereby the process can be more easily and accurately analyzed and monitored than before. The basic idea of the invention is only to monitor the process in a reliable manner especially in regard of its runnability. The task of monitoring the quality of the final product is entrusted to other measurement processes because a high quality of a good process situation is generally always achieved when the process situation itself is equally stable. The primary aim of the method of the invention is to describe the runnability of a paper machine. The instant of occurrence of a change in the fingerprints can be seen from history data, so it is possible to find out what changes have taken place in the output variables at this instant. Correcting a bad situation is outside the scope of the present invention, because correcting a given output variable is not a straightforward task as a given difference may be the result of many factors. Local know-how is preferably utilized. The personnel of each plant know their own plant, and this special knowledge is important when the process is to be corrected after it has got into an unstable condition.
The features characteristic of the method of the invention are presented in the claims below.
The output vector of the neural network is processed so as to produce a scalar or otherwise unambiguous quantity. The processing consists of applying a mathematical algorithm, and this can be done using a general-purpose computer.
According to a preferred embodiment, a polar conversion of a continuous difference quantity, expressing the state and history of the process in an extremely concentrated form, is presented to the user. According to a third embodiment, both the aforesaid polar conversion and sets of history values are presented side by side to the user, so that, when the process gets into an abnormal state, it will be easier to establish the reason for this. Even incidental causes of process instability can often be established by utilizing local know-how because the precise instant of occurrence of a change can be determined from history data. Preferably a multi-layer perceptron neural network is used.
The other advantages and embodiments of the invention will be described below in conjunction with examples of embodiments.
REFERENCES:
patent: 5347446 (1994-09-01), Iino et al.
patent: 5671335 (1997-09-01), Davis et al.
patent: 5960381 (1999-09-01), Singers et al.
patent: 6187145 (2001-02-01), Furumoto et al.
patent: 6208953 (2001-03-01), Milek et al.
patent: 0895197 (1999-02-01), None
patent: 9629468 (1996-09-01), None
Barlow John E.
Fildes & Outland, P.C.
Le John
Liqum Oy
LandOfFree
Method and system for monitoring and analyzing a paper... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and system for monitoring and analyzing a paper..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system for monitoring and analyzing a paper... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3217314