Interpretable failure risk assessment for continuous production processes based on association rule mining

Abstract

Continuous production processes are often highly complex and involve machine failures as well as unscheduled process downtimes. Failures result in the production of waste and in high opportunity costs, but their causes are not always apparent to machine operators. As a result, identifying failure root causes and avoiding risky process states is of high interest for producers. This work presents an approach for a data-driven failure risk assessment that is validated on real-world process data of a nonwovens production line. In this approach, association rule mining is adapted to continuous processes for producing highly interpretable results in the form of association rules that represent the main causes for failures. The methodology includes data preparation, modelling of production states and the evaluation of root causes using an associative classification algorithm. The result of this paper is a method for an interpretable risk assessment in continuous production processes. By using the method in live production, causes of failures can be detected and interpreted. The universal structure of the developed method supports applications in many other continuous production processes.

Keywords

• Nonwovens production
• Root cause analysis
• Risk management
• Machine learning
• Association rule mining