An unsupervised learning method for comparing the quality of the soft computing algorithms in analog systems diagnostics

The paper presents the method of assessing the difficulty of the analog systems´ fault detection and location procedure, using the soft computing algorithm. The analysis of learning and testing of data sets (generated for the selected analog system) allows determining the difficulty of the system for the diagnostic procedure. This helps to compare different artificial intelligence and machine learning approaches used to detect, locate and identify faults. The versatile method of the data sets´ difficulty based on the graph clustering algorithm is proposed and explained. It is then applied to test two advanced methods: fuzzy logic and rough sets against the sixth order Butterworth lowpass filter. The paper is concluded with future prospects of the proposed method.