Data-driven approach to predict the sequence of component failures: a framework and a case study on a process industry

Abstract

PurposeThe research approach is based on the concept that a failure event is rarely random and is often generated by a chain of previous events connected by a sort of domino effect. Thus, the purpose of this study is the optimal selection of the components to predictively maintain on the basis of their failure probability, under budget and time constraints.Design/methodology/approachAssets maintenance is a major challenge for any process industry. Thanks to the development of Big Data Analytics techniques and tools, data produced by such systems can be analyzed in order to predict their behavior. Considering the asset as a social system composed of several interacting components, in this work, a framework is developed to identify the relationships between component failures and to avoid them through the predictive replacement of critical ones: such relationships are identified through the Association Rule Mining (ARM), while their interaction is studied through the Social Network Analysis (SNA).FindingsA case example of a process industry is presented to explain and test the proposed model and to discuss its applicability. The proposed framework provides an approach to expand upon previous work in the areas of prediction of fault events and monitoring strategy of critical components.Originality/valueThe novel combined adoption of ARM and SNA is proposed to identify the hidden interaction among events and to define the nature of such interactions and communities of nodes in order to analyze local and global paths and define the most influential entities.