Detection of Overlapping and Simultaneous Two Faults Based on the Improved LDA

 In the two-dimension space, when there is a correlation between two critical-to-quality (CTQ) attributes, two predefined faults overlap each other. In this case, if fuzzy  c-means  (FCM) method is applied to detect these two faults that happen simultaneously, the detection rate is low. In order to solve this problem, a new algorithm called linear discriminant analysis (LDA) with principal component analysis (PCA) and discriminant intervals (PILDA) is proposed. By this algorithm, PCA-shaping is used to eliminate the effects of overlapping and to determine the discriminant intervals so as to overcome the difficulty in detecting two faults that are not predefined. To verify the proposed method, simulation is made for the detection of simultaneous faults and single faults by using 864 fault combinations in different variable relations and different variable shifts. By the proposed method, the average detection rate is 84.94% compared with 58.13% by FCM. This is a significant improvement.