Applications of pattern recognition to the diagnosis of equipment failures

Abstract Some contributions of the pattern recognition methodology to maintenance control, design review and automated testing, are stated and illustrated. A feature extraction method, inspired from principal component analysis, is applied to the information in a reliability data bank once transformed; the failure patterns and time-observations are displayed simultaneously for maintenance control and design review. Sequential classifications by the generalized nearest neighbour rule applied to different sets of learning data, achieve real time diagnosis. A 92 per cent true classification rate for acceptable items is reported in automated testing with 21 classes and 82 dimensional equipment description vectors.