An Artificial Intelligence/Machine Learning Based Regression-Clustering-Sensor-Fusion Framework for Predicting Unmeasurable Time-Series Strain From Other Sensor Measurements

Abstract

Abstract Typically, in a reactor-coolant fatigue test loop, an autoclave is used for housing the test specimen. For this purpose, a small tubular autoclave is often used. This is for reducing the cost of building the test loop and for avoiding rigorous ASME pressure vessel qualification criterion as required for the qualification of a larger high-pressure-temperature vessel. However, the use of a small autoclave along with high-pressure flow inside the autoclave does not allow to put an extensometer (inside the autoclave) for measuring the strain. The measurement of strain during the fatigue test of a specimen is required to assess the accuracy of the test and for downstream activities such as for stress–strain based material model developments and stress analysis validation. In this paper, we discuss an artificial intelligence and machine learning framework for predicting time-series strain from other sensor measurements such as from load cell, frame pull-rod displacement, and actuator displacement sensors. First the framework was trained and validated against in-air condition fatigue test data (for which the strain was measurable). Then the validated model was used for predicting strain in a reactor-coolant fatigue test loop, in which direct strain was not measurable. The original aim of the research was to improve the mechanical testing and measurements capability for conducting fatigue tests in a high-temperature-pressure reactor-coolant-flow environment. However, a similar approach can be used for predicting strain in an actual reactor component.