Developing novel models using neural networks and fuzzy systems for the prediction of strength of rocks from key geomechanical properties. (May 2017)

• Record Type: Journal Article
• Authors: Sharma, L.K.; Vishal, Vikram; Singh, T.N.
• Is Part Of: Measurement. Volume 102(2017)
• Page Start: 158
• Page End: 169
• Publication Date: 2017-05
• Subjects: Unconfined compressive strength -- Multiple regression -- ANN -- ANFIS; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2017.01.043
• Languages: English
• ISSNs: 0263-2241

Abstract

Graphical abstract: Highlights: Key rock index properties were measured along with the UCS. Prediction models for UCS were developed using statistical and intelligence methods. Prediction potential of SR, MR, ANN, and ANFIS was evaluated. All the proposed models stand statistically meaningful. The best UCS prediction accuracy was achieved using the ANFIS model. Abstract: This research study was conducted to predict the unconfined compressive strength (UCS) of the rocks by applying the adaptive neuro-fuzzy inference system (ANFIS), and the outcomes were compared with the traditional statistical model of multiple regression (MR) analysis and artificial neural network (ANN). 13 types of rock samples collected from 5 geological horizons in India were tested in the laboratory as per the International Society for Rock Mechanics (ISRM) standards. In developing the predictive models, ultrasonic P-wave velocity, density and slake durability index were considered as model inputs, whereas UCS was the output parameter. The prediction performance of ANFIS model was checked against the MR and the ANN predictive models. It was found that the constructed ANFIS model exhibited relatively high prediction performance of UCS than the MR and the ANN models. The performance capacity of the predictive models were evaluated based on the coefficient of determination (R 2 ), the mean absolute percentage error (MAPE), the root mean square error (RMSE) and the variance account for (VAF). The ANFIS predictive model had R 2, MAPE, RMSE and VAF equal to 0.978, 10.15%, 6.29 and 97.66%, respectively, superseding the performance of the MR and the ANN models. The performance comparison revealed that soft computing is a good approach for minimizing the uncertainties and inconsistency of correlations in geotechnical engineering.