Rock tensile strength prediction using empirical and soft computing approaches
The tensile strength of rocks plays an important role in site investigation during the initial phase of many construction projects. However, laboratory tests as direct methods for tensile strength prediction are not considered efficient in terms of time and cost. The present study is focused on deve...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Published: |
Springer Verlag
2019
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/87409/ http://dx.doi.org/10.1007/s10064-018-1405-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The tensile strength of rocks plays an important role in site investigation during the initial phase of many construction projects. However, laboratory tests as direct methods for tensile strength prediction are not considered efficient in terms of time and cost. The present study is focused on developing new hybrid intelligent models for rock tensile strength prediction. Rock samples which were gathered from a tunnel in Malaysia were subjected to testing in the laboratory to prepare a database comprising 80 datasets for analysis. These rock index tests include Schmidt hammer test, dry density test, and point load test, together with the Brazilian tensile strength (BTS). Moreover, simple regression analysis was conducted and the results showed that a model with multiple inputs needs to be developed for tensile strength prediction. Subsequently, three hybrid intelligent systems, namely, an imperialist competitive algorithm (ICA)-artificial neural network (ANN), a genetic algorithm (GA)-ANN, and a particle swarm optimization (PSO)-ANN, were developed for estimating the tensile strength of rocks. In fact, the ICA, PSO, and GA were employed for weight and bias adjustments in ANN models. The values of the variance accounted for (VAF), coefficient of determination (R2), and root mean square error (RMSE) were obtained for the evaluation of the developed hybrid models. For the training datasets of the GA-ANN, PSO-ANN, and ICA-ANN models, R2 values of 0.911, 0.933, and 0.923, respectively, were obtained. On the other hand, R2 values of the testing datasets of the GA-ANN, PSO-ANN, and ICA-ANN models were 0.912, 0.929, and 0.920, respectively. The results showed that, although all the developed predictive models can predict the BTS within acceptable levels of accuracy, the PSO-ANN predictive model provided the best performance. Consequently, it was concluded that, due to the higher capability of the PSO-ANN model, it could be introduced as a new model for the prediction of the tensile strength of rocks. |
|---|