ABSTRACT
This case study describes an ontogenic CID3 algorithm and its application to recognition of defects in a floating glass ribbon. The structure of this case study is as follows. First, the CID3 algorithm is described in sufficient detail to give the reader the feeling of how ontogenic algorithms generate their architectures. Second, a step-by-step application of the algorithm to a problem of distinguishing true defects (bubbles, stones and tin drops) from surface anomalies (water droplets and water spots) is provided. The second step also includes a description of the preprocessing steps crucial for achieving high accuracy of recognition. Finally, the ontogenic CID3 algorithm results are compared with those obtained by RBF and backpropagation algorithms on the same data.
