Abstract
The self-organizing map (SOM) is a kind of artificial neural networks that forms a topological map to cluster data as well as to reduce dimensionality of the data based on unsupervised learning. In this study, the SOM is introduced to the quality inspection process of beverage cans where the input vector with large dimension is composed of a series of frequency response data that is obtained by the magnetic hammering test. In response to the input vectors, the system predicts the internal pressure of cans in a way that a certain unit comprising the learned network responds to an input vector having minimum distance with the weight vector of the selected unit. Test results have shown that the SOM can roughly identify the difference of the internal pressure of cans, however, the estimate accuracy is not satisfactory in comparison to the prediction results obtained by the system which is based on the k-nearest neighbor algorithm. Toward the further improvement of the suggested classification system based on SOM, the investigation into the well-suited choice of input vectors, proper arrangement of the network parameters are necessary.
