Skip to main content
Nuclear Science and Engineering Division | Artificial Intelligence and Machine Learning

Feedwater Pump-Motor Set

The goal of cost reduction through improved maintenance procedures requires the ability to ascertain and predict component health

Applications targeted to enable staffing and equipment cost reductions at nuclear power plants include the following:

  • Safety system sensors require periodic calibration. Being able to unobtrusively determine the calibration status could obviate the need for performing unnecessary instrument calibration.
  • Safety system equipment requires periodic testing. Being able to unobtrusively determine performance status of equipment could provide additional safety credit.
  • Lifetime extension applications are to include plans for the digital upgrade of balance-of-plant systems. Understanding how the selection of an upgraded sensor set contributes to improved monitoring and control can provide for more informed staffing reductions and reliable operation.

Recent work with a nuclear utility is addressing monitoring of feedwater pump performance. Economic incentives are driving a need to rework maintenance procedures so that sensor calibration and pump overhaul are scheduled on an as-need basis rather than a periodic basis. Prior experience with pump maintenance activities indicated that overhaul procedures were being performed on otherwise serviceable equipment that was operating within specifications.


The feedwater pumps found in commercial boiling water and pressurized water reactors are typically multi-stage centrifugal pumps while the pump motor is typically a synchronous induction motor. In this application the pump-motor set was treated as a single system for fault diagnosis under the PRO-AID code. The accompanying table shows the row-based sensor residual patterns given the available sensor set in this application and the corresponding level of fault discrimination as appears column-wise in the table. One notes that the sensor set supports discrimination between sensor faults and pump-motor faults and further allows the unique diagnosis of the fault appearing in any one column of the table.

The performance of the PRO-AID algorithms for diagnosing performance degradation was evaluated using archived plant data provided by the utility. The faults cases included sensor calibration errors and equipment degradation. The sensor fault scenarios included two separate cases, the first adding a bias of 2% to the measured values of the flowrate sensor and second a similar bias added to the motor power sensor. The equipment degradation scenario involved a blind increase in shaft bearing friction losses causing a decrease on pump-motor system efficiency. All degradations were successfully diagnosed. The fault signature giving rise to the diagnosis of increased bearing friction loss appears in the first figure while the estimated efficiency loss of 1.6% is shown in the second figure.


Presently preventative maintenance (PM) procedures schedule maintenance at best using a generic likelihood of failure for a component without regard for the component current condition. There is the opportunity to improve on this by first incorporating a mechanistic treatment of the evolution of a degradation process and second by initializing the process using the diagnosis of current condition as described here.