A Systematic Workflow for Oscillation Diagnosis Using Transfer Entropy

Abstract

Causality analysis techniques, such as transfer entropy, have been proven to be useful tools for oscillation diagnosis in a variety of industrial processing applications. However, these tools have not gained widespread acceptance by industry. Causality analysis application procedures are complicated and time-consuming, requiring a large number of parameters to be selected. In this paper, a workflow for the application of transfer entropy for oscillation diagnosis is developed, emphasizing guidelines for the selection of parameters. Current parameter selection techniques require the user to search iteratively over a range of parameter values to find the optimum. With a computationally expensive technique, such as transfer entropy, this can be time-consuming. Alternatively, recommended default values are used, which may not be optimal. The process dynamics, characterized by time delay, time constants, and oscillation period, can have a strong influence on the optimal parameter selection. The relationship between process dynamics and the optimal transfer entropy parameters is investigated. These relationships are used to suggest guidelines for parameter selection. A strong relationship was found between the oscillation period and time delay and the optimal prediction horizon and time-interval parameters. The developed workflow for oscillation diagnosis was tested on an industrial case study. The workflow provided a robust systematic procedure for accurately identifying the propagation path of the oscillation.