Abstract
The application of Artificial Neural Networks (ANNs) as a basis for new generation of rock failure criteria has been extended to polyaxial state of stresses. A comprehensive strength database including the results of 254 polyaxial tests on six different rock types is collected, and for each rock type the data are randomly divided into two subsets of training and test sets. Following the optimization of architecture and configuration of ANNs, the training data are used for training the networks to predict the value of major principal stress at failure from the values of minor and intermediate principal stresses. The same training data are also used to obtain the parameters of two selected conventional failure criteria, namely the Modified Wiebols–Cook and Rafiai criteria. Then, the test data are used to compare the accuracy of the different failure criteria. It is observed that in all cases, the ANN-based failure criterion leads to more accurate results, and the average value of root mean squared error for this criterion is about 50 and 20% less than those of the Modified Wiebols–Cook and Rafiai criteria, respectively. Subsequently, the explicit formulation of the ANN-based criterion as an equivalent Mohr–Coulomb criterion with stress-dependent parameters is presented. Finally, the formulation is incorporated into a numerical code which is used to simulate the polyaxial tests included in the strength database. Excellent agreement between the expected and simulated values of strength shows that the ANN-based failure criterion can be implemented in numerical simulations.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have