A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Highlights Consider burst capacity models for pipelines containing dent-gouges. Collect a large database of burst tests of pipe specimens with dent-gouges. Improve the predictive accuracy of dent-gouge model using Gaussian process regression. Provide a new perspective on using full-scale test data and GPR to develop fitness-for-service assessment models.
Abstract The present study proposes an improvement of the well-known European Pipeline Research Group (EPRG) burst capacity model for pipelines containing dent-gouges by adding a correction term to the EPRG model. The Gaussian process regression is employed to quantify the correction term as a function of six non-dimensional input variables based on 190 full-scale dent-gouge burst tests collected from the open literature. The linear prior mean function and squared exponential kernel are considered in the Gaussian process regression, with the corresponding hyper-parameters evaluated using the maximum likelihood method from the training set of the collected full-scale test data. The correction term is shown to result in a marked improvement of the predictive accuracy of the EPRG model based on a comparison of the observed and predicted burst capacities for the test data. By applying the improved EPRG model to the regression set of the test data, the mean and coefficient of variation of the test-to-predicted ratios equal 1.01 and 9.2%, respectively. The analysis also sheds light on the relative importance of the input variables for the correction term.
Highlights Consider burst capacity models for pipelines containing dent-gouges. Collect a large database of burst tests of pipe specimens with dent-gouges. Improve the predictive accuracy of dent-gouge model using Gaussian process regression. Provide a new perspective on using full-scale test data and GPR to develop fitness-for-service assessment models.
Abstract The present study proposes an improvement of the well-known European Pipeline Research Group (EPRG) burst capacity model for pipelines containing dent-gouges by adding a correction term to the EPRG model. The Gaussian process regression is employed to quantify the correction term as a function of six non-dimensional input variables based on 190 full-scale dent-gouge burst tests collected from the open literature. The linear prior mean function and squared exponential kernel are considered in the Gaussian process regression, with the corresponding hyper-parameters evaluated using the maximum likelihood method from the training set of the collected full-scale test data. The correction term is shown to result in a marked improvement of the predictive accuracy of the EPRG model based on a comparison of the observed and predicted burst capacities for the test data. By applying the improved EPRG model to the regression set of the test data, the mean and coefficient of variation of the test-to-predicted ratios equal 1.01 and 9.2%, respectively. The analysis also sheds light on the relative importance of the input variables for the correction term.
Improvement of burst capacity model for pipelines containing dent-gouges using Gaussian process regression
Engineering Structures ; 272
2022-09-25
Article (Journal)
Electronic Resource
English
British Library Online Contents | 2000
Gouges for the Lathe: Selecting and sharpening spindle, bowl and roughing-out gouges
| British Library Online Contents | 1997
Modelling pile capacity using Gaussian process regression
| Elsevier | 2010
Modelling pile capacity using Gaussian process regression
| Online Contents | 2010
Modelling pile capacity using Gaussian process regression
| Elsevier | 2010