The primary focus of the research is to integrate state-of-the art sensor modalities and advanced data analytics to enhance the condition-based maintenance of secondary piping structures in nuclear power plants. The periodic piping inspections are not reliable and increases operation and maintenance costs. The developed and validated sensor system in conjunction with data analytics method should be capable of detecting corrosion in piping structures and make a pathway to move from inspection-based maintenance to condition-based maintenance. A testbed was developed with pipe sections fabricated out of 2-inch inner diameter schedule 80 carbon steel pipe and utilized a standard 4-bolt flange connections. The test bed also included a centrifugal pump (130 gpm at 0 head) for fluid flow, a 30-gallon tank with a heating element, a pump, an elbow, a tee, and a network of pipes that fed back into the tank intake. Six accelerometers were placed on an elbow and five accelerometer were placed on a Tee structure. The pump was run at 60Hz on a variable frequency drive controller and temperature was held at 80 deg. C. With the initial setup baseline sensor data was collected for 3.2 seconds with a sampling rate of 52100 samples/sec which indicates the healthy pipe condition. To determine the response of accelerometers, masses of 2.9g, 6.4g, 11.4g, 13.8g, and 23.5g are added at different levels to pipe elbow and data was collected separately. Then, to emulate the corrosion process the masses of 19g, 32g, 48g, 63g, 79g, and 95g were removed sequentially at pipe elbow and data was collected. The accelerometer data is processed using the Hilbert-Huang Transformation(HHT), which works well for non-linear and non-stationary signals. The HHT using empirical mode decomposition generates multiple intrinsic mode functions (IMF) of different amplitude and frequency levels from raw vibration data at each time point in X, Y and Z directions. From each IMF, features including phase, energy, entropy, mean, standard deviation, skewness, and kurtosis were extracted. To deal with the curse of dimensionality, 1. Principle Component Analysis was used for dimensionality reduction and, 2. Permutation Importance was used for feature selection. Random Forest, Support Vector Machine, and Logistic Regression methods were used to distinguish mass addition and mass removal from the Baseline. Logistic regression using only phase information from Y-direction classified mass additions from baseline data with testing accuracy of 94%. Prediction accuracy varied substantially among trials of the same run, however, performed well when averaging over trials at each time point. Random Forest resulted in \>95% testing accuracy for classifying both mass addition and mass removal data from baseline data. Whereas SVM with PI based feature selection showed \~ 94.7% testing accuracy for mass removal and \~ 92.53% testing accuracy for mass addition. In the case of SVM with PCA, testing accuracy was \~ 88% for mass addition and \~96% for mass removal.


Diagnosis of a Corrosion Process in Secondary Piping Structures in Nuclear Power Plants

Division / Sponsors: 
Pressure Vessels & Piping (PVP)


Paper Proceedings: https://asmedigitalcollection.asme.org/PVP/PVP2020/volume/83815

PVP2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.