The prediction of the diameter of columns is a fundamental step for the design of jet grouting applications, as harmful consequences may derive from an inadequate selection of the treatment setup. Starting from different perspectives, this goal has been pursued by several authors who provided empirical or theoretical correlations between the mean diameter of columns, the treatment parameters and the mechanical properties of native soils. However, the margin of uncertainty with these relations is still relatively large, mostly because of arbitrary assumptions made in their formulation. In order to reduce as much as possible the role of preliminary choices, a method based on Artificial Neural Networks (ANN) is proposed. It consists in training a computer code with a set of experimental observations and in using the established correlations between input and output variables to predict future occurrences. After a brief introduction of the principles and limitations of ANN’s, the paper describes the logical procedure followed for the selection of the variables which better describe the mechanism of columns formation. A database of more than 130 case studies, where jet grouting parameters, properties of soil and diameters are simultaneously reported, has been collected from the literature to train the network. Systematic analyses have been then performed, parametrically varying the structure of the network and the use of data, in order to improve the accuracy of prediction. The comparison with other methods recently published in the literature confirms the good predictive capability of the proposed method. For its practical application, a set of design charts has been produced where the mean diameters of columns are expressed, for all injection systems and soil types, as functions of the soil penetration index NSPT and the specific energy of treatment. Safety factors have been finally computed to take into account the inaccuracy of prediction.
Przewidywanie ṡrednic kolumn iniekcyjnych przy zastosowamiu sztucznych sieci neuronowych
MODONI, Giuseppe
2015-01-01
Abstract
The prediction of the diameter of columns is a fundamental step for the design of jet grouting applications, as harmful consequences may derive from an inadequate selection of the treatment setup. Starting from different perspectives, this goal has been pursued by several authors who provided empirical or theoretical correlations between the mean diameter of columns, the treatment parameters and the mechanical properties of native soils. However, the margin of uncertainty with these relations is still relatively large, mostly because of arbitrary assumptions made in their formulation. In order to reduce as much as possible the role of preliminary choices, a method based on Artificial Neural Networks (ANN) is proposed. It consists in training a computer code with a set of experimental observations and in using the established correlations between input and output variables to predict future occurrences. After a brief introduction of the principles and limitations of ANN’s, the paper describes the logical procedure followed for the selection of the variables which better describe the mechanism of columns formation. A database of more than 130 case studies, where jet grouting parameters, properties of soil and diameters are simultaneously reported, has been collected from the literature to train the network. Systematic analyses have been then performed, parametrically varying the structure of the network and the use of data, in order to improve the accuracy of prediction. The comparison with other methods recently published in the literature confirms the good predictive capability of the proposed method. For its practical application, a set of design charts has been produced where the mean diameters of columns are expressed, for all injection systems and soil types, as functions of the soil penetration index NSPT and the specific energy of treatment. Safety factors have been finally computed to take into account the inaccuracy of prediction.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
