Abstract

In this paper, an approach for data-based process monitoring and fault diagnosis in dynamical systems is presented. Given data of the nominal operation of a system, the approach determines a set of stochastically dependent variables for each sensor using mutual information. The sets of dependent variables provide information about the system dynamics and the resulting factorization of the probability density function, which is represented as a factor graph. The probability density function is estimated by non-parametric kernel density estimation using Gaussian kernels. Based on the nominal system description provided by the factor graph, the probability of abnormal events and faults is calculated and their root cause is identified. The approach allows an iterative integration of detected faults such that the diagnosis of repeated faults is possible. The effectiveness of the approach and the iterative integration of faults is illustrated for the heat exchanger circuit of an air supply unit using real world measurement data.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call