Control Chart Pattern Recognition Using Spiking Neural Networks
Author : D.T. Pham and S. Shahnorbanun
Statistical process control (SPC) is a method for improving the quality of products. Control charting plays the most important role in SPC. A control chart can be used to indicate whether a manufacturing process is under control. Unnatural patterns in control charts mean that there are some unnatural causes for variations. Control chart pattern recognition is therefore important in SPC. In recent years, neural network techniques have increasingly been applied to pattern recognition. Spiking Neural Networks (SNNs) are the third generation of artificial neural networks, with spiking neurons as processing elements. In SNNs, time is an important feature for information representation and processing. Latest research has shown SNNs to be computationally more powerful than other types of artificial neural networks. This paper proposes the application of SNN techniques to control chart pattern recognition. The paper focuses on the architecture and the learning procedure of the network. Experiments show that the proposed architecture and the learning procedure give high pattern recognition accuracies.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
NOTE: TO READ THIS AND OTHER IPROMS 2006 PAPERS, PLEASE REGISTER FOR THE CONFERENCE.
REGISTRATION IS FREE.
CLICK here TO REGISTER.
| Attachment | Size |
|---|---|
| FV_CCPRSNN.pdf | 238.53 KB |
| CCPR_Pr.wmv | 3.15 MB |
| iproms2006.pdf | 173.37 KB |
login or register to download the paper.
These are excellent questions. Thank you.
For details on spiking neurons, we would refer you to the book by Wulfram Gerstner and Werner Kistler, "Spiking Neuron Models: Single Neurons, Populations, Plasticity", published by Cambridge University Press in 2002.
We have not tried the spiking neural network on problems other than those involving single-parameter control charts and cannot give a definitive answer to your second question.
However, if the problem of automatically recognising patterns in multi-parameter control charts can be solved using BP networks, then we believe we should be able to apply the spiking neural network to it as well.
Dear Professor and,
Assalamualaikum Shahnorbanun,
Interesting work!
i) In section 5.3 (Training set), I am wonder whether fully developed unstable patterns or transitional patterns were used in the study?
ii) The performance measure used was % accuracy. Did the study also consider other measures such as ARL, type I and type II errors?
Cheers.
Adnan.
Hi,
This is another good book for further details about spiking neural networks, in an application point of view;
Pulsed Neural Networks,
Wolfgang Maass and Christopher M. Bishop (Editors),
The MIT Press, Cambridge, 2001.
Cheers,
Charles.
Hai Mei Choo. Thank you for your interesting question.
The differences between spiking neurons and conventional neuron that enable spiking neuron to outperform conventional neuron are:
1) the models of SNN are much more realistic as biological neural networks than the conventional networks.
2)their computational power: the models of SNN are much more nonlinear and more parameters are considered than the conventional networks.
Hence networks of spiking neurons are capable of exploiting time as a resource for coding and computation in a much more sophisticated manner than the conventional networks.
Wualaikumsalam Dr.Adnan, and
greetings ladies & gentlemens.
Thank you for the interesting question.
1)In this paper we are using fully developed unstable patterns. We might be interested to consider transitional patterns in our future study.
2)At this stage, we only consider the training and testing accuracy as the performance measurement.
Regards,
Shahnorbanun
Can you give further insights on how the differences between spiking neuron and conventional neuron that enable spiking neuron to outperform conventional neuron?
Also, your paper has shown a case study based on control chart of a single parameter, and there are relatively small amount of patterns (6 patterns) to be recognised. How do you think your method can handle multiple co-related parameters?
Thanks