An artificial neural network based on adaptive resonance theory for fault classification on an automated assembly machine
An unsupervised artificial neural network (ANN) based on the ART2-A algorithm is compared to a rule-based method for fault classification on an automated assembly machine. Machine data is collected using three greyscale sensors and two redundant limit switches for 11 different operating conditions. Descriptive features are extracted from the raw data and two data sets, each containing 180 feature vectors, are created for testing both methods. The first data set contains 'real' feature vectors obtained from the original sensor signals, and the second data set contains 'simulated' feature vectors obtained by scaling the 'real' feature vectors. The second data set is used to test the performance of each system when variations are present in the input space. During testing, the rule-based system correctly classified 98.3% of all feature vectors, but its classification thresholds needed to be manually adjusted to accommodate the 'simulated' data set. The ART2-A network perfectly classified the 'real' data set into 13 clusters, and then correctly classified the 'simulated' data into the same 13 clusters without any modification to the algorithm's tuning parameter, vigilance.
Includes bibliographical references