Use of artificial neural networks in the prediction of horizontal and friction pressures in a slender silo

Abstract

Structural reliability studies in silo pressure predictions are a subject of great interest because the number of variables in silo pressure calculations is large, which can generate numerous failures and occurrences of collapse. The objective of this study was to propose a pressure prediction model using artificial neural networks (ANNs) in comparison with experimental data. For data collection, a pilot silo model, proposed by Pieper and Schutz (1980), adopted by the German standard DIN 1055 (1987), was used. The pressure measurements were performed during loading and unloading of the silo. Smooth walls with flat bottom were used in the silo, varying the height/diameter ratio between 4, 6 and 8. The pressures observed on the silo walls were the horizontal pressures and the friction pressure of the product with the wall. The results of the pressures obtained experimentally were fed into an algorithm using multilayer Perceptron ANNs. The pressures measured on the silo walls were compared with the values predicted by ANNs, linear and polynomial regressions, and also with the values calculated by means of the European Standard EN 1991-4: 2006 and Australian Standard AS 3774 (1996). Among the models used, the best performance was obtained by ANNs, which were able to predict silo wall pressures with smaller errors.