Quantitative Evaluation of Fault Diagnosability Based on Maximum Mean Discrepancy

Quantitative evaluation of fault diagnosability is a crucial part of system design. Therefore, we propose a data-driven method for the quantitative evaluation of fault diagnosability. The evaluation results reflect the characteristics of the fault and are independent of any diagnosis algorithm. Assuming the measurement data follow the unknown multivariate distribution. Then, the multivariate distribution is mapped to the reproducing kernel Hilbert space (RKHS) through the characteristic kernel, and the distance of the multivariate distribution is represented by maximum mean discrepancy (MMD), so the quantitative evaluation of fault diagnosability is transformed into calculating the distance of the multivariate distribution in RKHS. Finally, an example is given to verify the effectiveness of the proposed method. Experimental results show that the proposed method can quantitatively evaluate the difficulty of fault diagnosis without relying on any diagnosis algorithm.