Recognition method of equipment state with the FLDA based Mahalanobis-Taguchi system

Mahalanobis-Taguchi system (MTS) is a kind of big data classification and reduction method which can be used in the fault diagnosis and maintenance modeling. Especially in the context of big data, it can get better results in application. And MTS uses Mahalanobis distance (MD) as the measurement scale to identify the system state with multidimensional characteristics. But when the benchmark and abnormal space which are constructed by the traditional MTS have a serious overlap, the model will perform imbalanced classification ability to identify the sample. In this paper, against the problem, a modified MTS amended by Fischer linear discriminant analysis (FLDA) is proposed, and to be used to recognize the running state of equipment. Firstly, the paper discussed the limitation to using MD as the measurement scale in the traditional model, and then to use the balance accuracy while balanced classification as the evaluation index for the balance ability of the model classification. And then the threshold optimization model was discussed with different weight coefficient considering the actual cost and loss of the missed-alarm and the false-alarm. Furthermore, FLDA was used to calculate the projection matrix and the best projection vector was selected to amend the tradition measurement scale. Finally, the modified model amended by FLDA was compared with the traditional MTS and FLDA model form two aspects of accuracy index and the size of abnormal samples by using the bearing running data. The result proved the effectiveness and superiority of the modified model.

Automated summary

The paper proposes a modified Mahalanobis-Taguchi system (MTS) method amended by Fischer linear discriminant analysis (FLDA) for recognizing the running state of equipment. By discussing the limitations of using Mahalanobis distance (MD) as the measurement scale and introducing balanced accuracy as the evaluation index, the paper improves the imbalanced classification ability of the traditional MTS. The results demonstrate the effectiveness and superiority of the modified model in terms of accuracy index and the size of abnormal samples.