Fuzzy-Pattern-Classifier Based Sensor Fusion for Machine Conditioning
Volker Lohweg, Uwe Mnks · Sciyo eBooks · 2010
In this chapter we have reviewed fuzzy set theory based multi-sensor fusion built on FuzzyPattern-Classification. In particular we emphasized the fact that many traps can occur in multi-sensor fusion. Furthermore, a new inspection and conditioning approach for securities and banknote printing was presented, based on modified versions of the FPC, which results in a robust and reliable detection of flaws. In particular, it was shown that this approach leads to reliable fusion results. The system model “observes” the various machine parameters and decides, using a classifier model with manually tuned or learned parameters, whether the information is as expected or not. A machine condition monitoring system based on an adaptive learning was presented, where the PCA is used for estimating significance weights for each sensor signal. An advantage of the concept is that not only data sets can be classified, but also the influence of input signals can be traced back. This classification model was applied to different tests and some results were presented. In the future we will mainly focus on classifier training with a low amount of samples, which is essential for many industrial applications. Furthermore, the classification results should be improved by the application of classifier nets.