ANN-based prediction of fractured rock mass hydraulic conductivity for the Frieda River Copper-Gold Mine in Papua New Guinea | |
| Author | Yervang Wang |
| Call Number | AIT Thesis no.GE-15-07 |
| Subject(s) | Rock mechanics--Papua New Guinea Soil permeability--Papua New Guinea |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Geotechnical and Earth Resources Engineering, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Thesis ; no. GE-15-07 |
| Abstract | A complex nonlinear relationship exists between indicative rock parameters (lithology, weathering, fracturing, unconfined compressive strength (UCS), defect angle (alpha), Rock Quality Designation (RQD)) and hydraulic conductivity. Literature review provides various empirical models for the prediction of hydraulic conductivity, which commonly suffer from insufficient representation of the nonlinear relationship between parameters though. In this study, an artificial neural network analysis was conducted for the prediction of hydraulic conductivity of fractured rock masses at the Frieda River Copper-Gold Mine in Papua New Guinea. In this study, the backpropagation neural network (BPNN) was trained to successfully map the relationship between indicative rock parameters and hydraulic conductivity using a variety of rock data sets provided by Geotech International Pte Ltd that collected in the site investigation of a feasibility study for the study mine. The data include five full sets of geotechnical core logging and packer testing from five boreholes (HTBG001 to HTBG005) with depths ranging from 360 m to 500 m. Input data selected for ANN analysis include lithology, weathering, fracturing, UCS, defect angle and RQD. Those parameters that are qualitative were coded into digital values. Three sets of input data were employed for study on the input layer effects, and namely 4-input data set including lithology, weathering, fracturing and UCS; 5-input data set including lithology, weathering, fracturing, UCS and defect angle; and 6-input data set including lithology, weathering, fracturing, UCS, defect angle and RQD. Borehole HTBG005 was selected for trial analyses to obtain appropriate inputs and models for prediction of hydraulic conductivity. Six ANN models were employed for study on the effects of transfer and training functions where the ANN structure applied is a multilayer model with transfer functions of log-sigmoid, tan-sigmoid and purelin, and training functions of Gradient descent with momentum backpropagation (traingdm) and Levenberg-Marquardt backpropagation (trainlm). The training process was generally conducted with 1000 iterations, 6 validation checks, learning rate of 0.005. At the same time, other training parameters for trainlm function were set with their default values except performance goal including initial mu of 0.001, mu decrease factor of 0.1, mu increase factor of 10 and epochs between displays of 25. As a result of study on the effects of inputs, transfer and training functions, the optimal ANN model for prediction of hydraulic conductivity was found to be a network with 3 hidden layers, 20 neurons per each, transfer function of sigmoid for all hidden layers and training function of Levenberg-Marquardt backpropagation (trainlm) using five inputs including lithology, weathering, fracturing, UCS and alpha. Prediction of hydraulic conductivity for other boreholes of HTBG001, HTBG002, HTBG003 and HTBG004 showed a good regression of 0.88, 0.83, 0.74 and 0.84 while their average errors were estimated to be 17.78%, 18.34%, 20.42% and 23.33%, respectively. The range of predicted hydraulic conductivity of borehole HTBG001 showed 0.205 to 1.871 μm/s while the measured values range from 0.176 to 2.348 μm/s, borehole HTBG002 showed 0.004 to 2.286 μm/s while the measured values range from 0 to 2.379 μm/s, borehole HTBG003 showed 0.005 to 0.398 μm/s while the measured values range from 0.004 to 0.295 μm/s and borehole HTBG004 showed 0.001 to 2.708 μm/s while the measured values range from 0 to 3.671 μm/s. |
| Year | 2016 |
| Corresponding Series Added Entry | Asian Institute of Technology. Thesis ; no. GE-15-07 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Geotechnical Engineering (GE) |
| Chairperson(s) | Pham Huy Giao; |
| Examination Committee(s) | Noppadol Phienwej;Silver, Marshall L.; |
| Scholarship Donor(s) | Asian Institute of Technology Fellowship; |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2016 |