Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Predicting Wall Thickness Loss in Water Pipes Using Machine Learning Techniques
Wall thickness loss in water pipes has been found to be positively correlated with water pipe failure. The reliability of water pipes reduces as their wall thickness loss increases. Although previous studies have investigated pipe failure modeling using historical failure data, however, indirect failure modeling via wall thickness loss is yet to be explored. Hence, this study develops machine learning (ML) models to predict wall thickness loss in water pipes. Random Forest (RF) and Gradient Boosting Machine (GBM) are chosen as the base models and are integrated with Bayesian Optimization (BO) algorithm for hyperparameters selection. The predictive models are evaluated using root mean square error (RMSE), mean absolute error (MEA), mean absolute percentage error (MAPE), and coefficient of determination (R2). Based on the evaluation metrics, the hybrid models (i.e., RF+ BO and GBM+BO) outperformed the base models (RF and GBM), showing the importance of the systematic selection of hyperparameters. The best model (RF + BO) achieved an RMSE, MAE, MAPE, and R2 value of 3.212, 2.494, 11.506, and 0.910, respectively. These metrics show the high predictive capability of the model, which can be used by water infrastructure management to predict wall thickness loss in water pipes.
Predicting Wall Thickness Loss in Water Pipes Using Machine Learning Techniques
Wall thickness loss in water pipes has been found to be positively correlated with water pipe failure. The reliability of water pipes reduces as their wall thickness loss increases. Although previous studies have investigated pipe failure modeling using historical failure data, however, indirect failure modeling via wall thickness loss is yet to be explored. Hence, this study develops machine learning (ML) models to predict wall thickness loss in water pipes. Random Forest (RF) and Gradient Boosting Machine (GBM) are chosen as the base models and are integrated with Bayesian Optimization (BO) algorithm for hyperparameters selection. The predictive models are evaluated using root mean square error (RMSE), mean absolute error (MEA), mean absolute percentage error (MAPE), and coefficient of determination (R2). Based on the evaluation metrics, the hybrid models (i.e., RF+ BO and GBM+BO) outperformed the base models (RF and GBM), showing the importance of the systematic selection of hyperparameters. The best model (RF + BO) achieved an RMSE, MAE, MAPE, and R2 value of 3.212, 2.494, 11.506, and 0.910, respectively. These metrics show the high predictive capability of the model, which can be used by water infrastructure management to predict wall thickness loss in water pipes.
Predicting Wall Thickness Loss in Water Pipes Using Machine Learning Techniques
Taiwo, Ridwan (Autor:in) / Seghier, Mohamed El Amine Ben (Autor:in) / Zayed, Tarek (Autor:in)
ce/papers ; 6 ; 1087-1092
01.09.2023
6 pages
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Predicting IRI Using Machine Learning Techniques
| Springer Verlag | 2023
High strength and heavy wall thickness steel pipes for building structures
| Tema Archiv | 2008
High strength and heavy wall thickness steel pipes for building structures
| Tema Archiv | 2009
Predicting disc cutter wear using two optimized machine learning techniques
| Springer Verlag | 2024