Solution: The companies provided us a large amount of measurements from their control valves, along with the maintenance reports that could be used to roughly label the valve conditions. According to the data, there were 4 possible fault cases, including the medium leak, pressure leak, positioner failure, and stiction. We proposed a technique, called hybrid FDD, which applies a nonlinear dynamic modelling technique in fault detection process. When a valve is identified as faulty, a set of carefully selected features are extracted and used in a supervised classifier to assign the most possible fault type of the valve.

Benefit: The solution developed was able to predict the current health of valve with satisfactory accuracy. This allows predictive maintenance and automated root cause analysis to be done which can result in more accurate input for making diagnostic decisions as well as for asset management program.