Temperature-Effect Compensation for Leak Detectors by Using Machine Learning Techniques
Egileak: Andoni Bilbao Eneko Fernández Zelmar Etxegoien
Data: 23.09.2021
Abstract
Although Differential Pressure Decay Testing (DPDT) is less influenced by external environment than pressure decay testing, the different temperatures involved in the process still affect the leak measurements, particularly in quick changing conditions. This paper investigates the impact of air, injected air and part temperature on leak measurements and develops a compensation model based on Machine Learning (ML) algorithms that uses these temperatures as predictors, as well as other such as maximum and minimum pressure during stabilization stage. An automated machine for data capture has been developed to simulate varying conditions. The results show that under the conditions investigated, the part temperature has the greatest impact on leak measurements. For the regressive model used in the compensation model, several ML algorithms are investigated, and the best results are obtained by using multilayer perceptron, reducing the mean absolute error measured by a commercial leak detector by 91%.
BIB_text
title = {Temperature-Effect Compensation for Leak Detectors by Using Machine Learning Techniques},
pages = {536-545},
keywds = {
Leak detection, Machine learning, regressive model, compensation model, temperature compensation
}
abstract = {
Although Differential Pressure Decay Testing (DPDT) is less influenced by external environment than pressure decay testing, the different temperatures involved in the process still affect the leak measurements, particularly in quick changing conditions. This paper investigates the impact of air, injected air and part temperature on leak measurements and develops a compensation model based on Machine Learning (ML) algorithms that uses these temperatures as predictors, as well as other such as maximum and minimum pressure during stabilization stage. An automated machine for data capture has been developed to simulate varying conditions. The results show that under the conditions investigated, the part temperature has the greatest impact on leak measurements. For the regressive model used in the compensation model, several ML algorithms are investigated, and the best results are obtained by using multilayer perceptron, reducing the mean absolute error measured by a commercial leak detector by 91%.
}
isbn = {978-3-030-87869-6},
date = {2021-09-23},
}