Enhancing the effectiveness of lean clustering in establishing benchmarks for automatic classification systems

The classification of workpieces according to their shape can enhance the efficiency of workpiece design and production. The lean clustering method employs only a small number of workpieces to perform the similarity comparison and the data obtained will be used to infer the similarity data of other workpieces in order to form a workpiece similarity matrix. With the aid of such a similarity matrix, one can perform the workpiece classification. To enhance the effectiveness of the lean clustering method and to extend its applicability, the study investigated this issue with regards to the following three aspects: 1. add the hypothesis of skew-to-left, centralised, and skew-to-right workpiece distribution patterns in addition to the uniform distribution of workpiece similarity; 2. beside the most commonly used max-min method, apply the Hamming method, interval average method, and weighting method to the workpiece similarity inference; 3. compare the effectiveness of rough classification and fine classification in lean clustering method. The results revealed that there is no significant difference among the four similarity distribution patterns and among the four similarity inference methods. However, the degree of consistency can reach up to 89% when the rough classification is used for workpiece classification. The authors, therefore, suggest that there is no need to test the similarity distribution in advance, in application of lean clustering method when the rough classification will be used for the classification task of a large number of workpieces so as to reduce the cost.