Applicability of Map generated by Self-Organizing Map Algorithm in Hammering Sound Inspection

The hammering sound inspection for pieces plays the Digital Twin part in the Hybrid Twin approach. The map generated by Self-Organizing Map (SOM) algorithm makes it possible to cluster the pieces into defected and non-defected groups by checking the corresponding hammering sounds. However, the applicability of map generated by SOM has not yet been discussed. This paper suggests a procedure how to improve a map generated by SOM algorithm. It shows that if the hammering sound of a piece is mapped on the boundary between the areas corresponding to defected and non-defected products, the map cannot be applied to inspect this piece and another map must be generated by repeating the training including the hammering sounds of such pieces by SOM algorithm. By executing the procedure some number of times, it is likely that the improved map will be used for hammering sound inspection.