A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Surrogate Models in Rock and Soil Mechanics: Integrating Numerical Modeling and Machine Learning
https://orcid.org/0000-0001-6289-3666
Abstract A surrogate model is an engineering method to predict the outcome of some process rather than model the real process itself. Here, we use surrogate models based on the machine learning that have been trained using a synthetic data set generated by a numerical model. The concept of a surrogate model is a powerful link between the traditional paradigm of numerical modeling and the new paradigm of machine learning. Surrogate models can provide results in seconds and predict a set of outputs of a complex process for a range of input parameters with an accuracy greater than 99%. This has an advantage over numerical models, that may need hours to attain the same accuracy. Surrogate models can be used for probabilistic analysis, numerical pre-conditioning, training and education, and for application by field practitioners. Machine learning has seen rapid growth in the last 10 years with the advent of large data sets and powerful new techniques. There is currently a unique opportunity to apply data science and machine learning methodologies as a complement to advanced numerical modeling of the subsurface. This paper describes a practical methodology for training surrogate models and gives three examples. The examples include predicting the crushed zone size in rock blasting, the effective properties of a discrete fracture network (DFN), and the bearing capacity of a layered soil. We show that surrogate models perform well when feature engineering (the application of domain knowledge to machine learning) is applied and when the number of input parameters is limited.
Highlights A methodology for creating rock and soil mechanics surrogate models is presented along with practical examples.Synthetic data sets describing the rock blasting crushed zone, the bearing capacity of layered soils, and the effective properties of discrete fracture networks are generated using numerical models.Artificial neural networks are able to rapidly and accurately predict some aspects of the numerical model results.The number of synthetic data points needed is problem dependent and a methodology to ensure accurate results over the entire target range is presented.
Surrogate Models in Rock and Soil Mechanics: Integrating Numerical Modeling and Machine Learning
https://orcid.org/0000-0001-6289-3666
Abstract A surrogate model is an engineering method to predict the outcome of some process rather than model the real process itself. Here, we use surrogate models based on the machine learning that have been trained using a synthetic data set generated by a numerical model. The concept of a surrogate model is a powerful link between the traditional paradigm of numerical modeling and the new paradigm of machine learning. Surrogate models can provide results in seconds and predict a set of outputs of a complex process for a range of input parameters with an accuracy greater than 99%. This has an advantage over numerical models, that may need hours to attain the same accuracy. Surrogate models can be used for probabilistic analysis, numerical pre-conditioning, training and education, and for application by field practitioners. Machine learning has seen rapid growth in the last 10 years with the advent of large data sets and powerful new techniques. There is currently a unique opportunity to apply data science and machine learning methodologies as a complement to advanced numerical modeling of the subsurface. This paper describes a practical methodology for training surrogate models and gives three examples. The examples include predicting the crushed zone size in rock blasting, the effective properties of a discrete fracture network (DFN), and the bearing capacity of a layered soil. We show that surrogate models perform well when feature engineering (the application of domain knowledge to machine learning) is applied and when the number of input parameters is limited.
Highlights A methodology for creating rock and soil mechanics surrogate models is presented along with practical examples.Synthetic data sets describing the rock blasting crushed zone, the bearing capacity of layered soils, and the effective properties of discrete fracture networks are generated using numerical models.Artificial neural networks are able to rapidly and accurately predict some aspects of the numerical model results.The number of synthetic data points needed is problem dependent and a methodology to ensure accurate results over the entire target range is presented.
Surrogate Models in Rock and Soil Mechanics: Integrating Numerical Modeling and Machine Learning
https://orcid.org/0000-0001-6289-3666
Abstract A surrogate model is an engineering method to predict the outcome of some process rather than model the real process itself. Here, we use surrogate models based on the machine learning that have been trained using a synthetic data set generated by a numerical model. The concept of a surrogate model is a powerful link between the traditional paradigm of numerical modeling and the new paradigm of machine learning. Surrogate models can provide results in seconds and predict a set of outputs of a complex process for a range of input parameters with an accuracy greater than 99%. This has an advantage over numerical models, that may need hours to attain the same accuracy. Surrogate models can be used for probabilistic analysis, numerical pre-conditioning, training and education, and for application by field practitioners. Machine learning has seen rapid growth in the last 10 years with the advent of large data sets and powerful new techniques. There is currently a unique opportunity to apply data science and machine learning methodologies as a complement to advanced numerical modeling of the subsurface. This paper describes a practical methodology for training surrogate models and gives three examples. The examples include predicting the crushed zone size in rock blasting, the effective properties of a discrete fracture network (DFN), and the bearing capacity of a layered soil. We show that surrogate models perform well when feature engineering (the application of domain knowledge to machine learning) is applied and when the number of input parameters is limited.
Highlights A methodology for creating rock and soil mechanics surrogate models is presented along with practical examples.Synthetic data sets describing the rock blasting crushed zone, the bearing capacity of layered soils, and the effective properties of discrete fracture networks are generated using numerical models.Artificial neural networks are able to rapidly and accurately predict some aspects of the numerical model results.The number of synthetic data points needed is problem dependent and a methodology to ensure accurate results over the entire target range is presented.
Surrogate Models in Rock and Soil Mechanics: Integrating Numerical Modeling and Machine Learning
Furtney, J. K. (author) / Thielsen, C. (author) / Fu, W. (author) / Le Goc, R. (author)
2022
Article (Journal)
Electronic Resource
English
BKL:
38.58
Geomechanik
/
56.20
Ingenieurgeologie, Bodenmechanik
/
38.58$jGeomechanik
/
56.20$jIngenieurgeologie$jBodenmechanik
RVK:
ELIB41
Soil mechanics - rock mechanics
| UB Braunschweig | 1962
International symposium on numerical methods in soil mechanics and rock mechanics
Springer Verlag | 1974
Soil mechanics, rock mechanics and mining
| Engineering Index Backfile | 1968
| UB Braunschweig | 1989
| TIBKAT | 1989