Quality improvement of individual cores of distribution transformers usingdecision trees

The estimation of individual core losses of wound core distribution transfo rmers (T/Fs) is very important because these constitute one of the main par ameters of their quality and moreover core costs represent about 30% of the total T/F is very important because these constitute one of the main param eters of their quality and moreover core costs represent about 30% of the t otal T/F material cost. In addition, accurate calculations of individual co re actual losses is extremely difficult, since actual losses show a diverge nce up to +/-20%, in relation with the theoretical individual core losses. In this paper a decision tree (DT) method is presented, suitable for the pr ediction of individual core losses in wound core distribution transformers (T/Fs) at the early stages of the production process. For the creation of t he learning and test set actual industrial measurements are used. Data comp rise grain oriented steel electrical characteristics, and quality control m easurements of cores production line. The resultant DT presents a success r ate of 94 %. Based on this DT, rules comprising the most important paramete rs and their threshold values can be derived. These are used to improve the quality of individual cores by reducing their actual losses.