Pattern classification by an incremental learning fuzzy neural network

Abstract

PREFACETo detect and identify defects in machine condition health monitoring, classical neural classifiers, such as Multilayer Perceptron (MLP) neural networks, are proposed to supervise the monitored system. A drawback of classical neural classifiers, off-line and iterative learning algorithms, is a long training time. In addition, they are often stuck at local minima, unable to achieve the optimum solution. Furthennore, in an operating mode, it is possible that new faults are developing while a monitored system is running.These new classes of defects need to be instantly detected and distinguished from those Training by various torque levels, the network achieved 100% correct prediction for the same torque level of testing data. Furthermore, the classification performance of the network has been tested using other benchmark data, such as the Fisher's Iris data, the two-spiral problem, and a vowel data set. Comparison studies among other well-known classifiers were preformed. The ILFN was found competitive with or even superior to many classifiers.tV ACKNOWLEDGMENTS

An incremental neural learning framework and its application to vehicle diagnostics

“…The incremental neural learning algorithm was developed based on the theory of boosting. The boosting theory states that a system can be adaptively adjusted to the errors of the weak hypotheses returned by a weak learning algorithm [53] and then by summing the probabilistic predictions of the weak hypotheses [5,6,[19][20][21]44]. Schapire and Freund showed that any weak learning algorithm can be efficiently transformed or "boosted" into a strong algorithm [11,19,44] by altering the distribution of samples so as to increase the probability of the "harder" parts of the data space being recognized correctly.…”

“…Neural learning algorithms such as backpropagation are not incremental, since old instances need to be reexamined to determine whether weights need to be updated or the structure of a neural network system needs to be changed [22]. As Mandziuk and Shastri pointed out, one of the greatest impediments in building large, scalable learning systems based on neural networks is that when a network trained to solve task A is subsequently trained to solve task B, it often "forgets" the solution to task A [31,53]. A comprehensive discussion of incremental neural learning research is presented in Sect.…”

Health monitoring of vibration signatures

“…However, clusters that belong to the same class are grouped together via the pruning module. The details of learning algorithm can be found in [37], [38]. A trained ILFN does not exhibit a clear meaning of knowledge embedded inside its structure.…”

“…The comparison study was performed by Robinson [37]. In Robinson's stud}', the best results with the correct prediction of 56% were obtained using the nearest neighbor classifier.…”

“…Gaussian radial basis functions are used to form the distribution of the pattern space. The ILFN can learn in a supervised or an unsupervised fashion [37], [38]. The ILFN has four layers: one input layer, one hidden layer, one output layer, and one decision layer, as shown in Figure 2.…”