To develop a model for the predictive analysis of the condition of a turbine-driven feed pump and showcase the adaptive capabilities of the Clover PMM software within the power-generating equipment maintenance and overhaul control system.
The application of Clover PMM as a system for the diagnosis and forecast of the technical condition of equipment.
Input data: telemetry data over a three-year period of operation and data regarding 10 defects with detection dates.
Sequence of actions
- sampling was classified according to the operating conditions of the equipment and broken down into five-minute intervals to optimize the quantity of observations
- a decision tree algorithm was selected for data processing
- using the established dependence on the values of all variables, the algorithm calculated the value of the base variable and compared it to the actual feed water flow in the turbo pump delivery head
- the difference between the calculated value of the feed water flow at the head of the feed turbine pump and the actual value was an error
- an acceptable range of error was calculated for each five-minute interval. The principle of calculating the acceptable value was similar to the principle of calculating the root-mean-square deviation
- if the estimated error fell outside the range, it was an anomaly (any „abnormal“ behavior of the equipment)
20 groups of anomalies with high error values were detected.