A Bayesian network-based approach for fault analysis

For high-value assets such as certain types of plant equipment, the total amount of resources devoted to Operation and Maintenance may substantially exceed the resources expended in acquisition and installation of the asset, because high-value assets have long useful lifetimes. Any asset failure during this useful lifetime risks large losses in income and goodwill, and decreased safety. With the continual development of information, communication, and sensor technologies, Condition-Based Maintenance (CBM) policies have gained popularity in industries. A successfully implemented CBM reduces the losses due to equipment failure by intelligently maintaining the equipment before catastrophic failures occur. However, effective CBM requires an effective fault analysis method based on gathered sensor data. In this vein, this paper proposes a Bayesian network-based fault analysis method, from which novel fault identification, inference, and sensitivity analysis methods are developed. As a case study, the fault analysis method was analyzed in a centrifugal compressor utilized in a plant.