ABSTRACT: The accurate prediction of cement failure is significant to maintain wellbore integrity. In the downhole situations, the failure criteria widely used for rock are not guaranteed to be suitable for cement. Till now, seldom researchers have paid attention to the right choice of failure criterion for cement and most of them simply used the Mohr-Coulomb criterion. In this work, comprehensive research is performed to collect published conventional and true triaxial compression strength data of cement. All the accessible data are used to evaluate the accuracy of six different failure criteria. A machine learning algorithm is developed to generate cement parameters, such as cohesion and friction angle. The statistic value, Misfits, is used for the quantitative comparison of the accuracy of different failure criteria. Results show that using only the conventional triaxial strength data, the influence of intermediate stress cannot be reflected and Mohr-Coulomb could provide similar accuracy compared to the three stress dependent criteria. However, when the true triaxial compression strength data is applied, the Mohr-Coulomb could generate significant errors. Three stress dependent criteria should be used for predicting cement failure. This study fills the gap of finding the right failure criterion for cement and provides guidance for wellbore integrity.

ABSTRACT: Parameter calibration is an indirect problem in which responses of a system are known, but the properties of the said system are not. Often, trial and error strategies are used for parameter calibration. However, they can be labor intensive and time consuming. Machine learning methods have been employed to automatize the process. The present work proposes a combination of Genetic Programming (GP) with an optimization method to estimate the elastic parameters of a finite element model of an oil and gas field. The estimated parameters were the Young's modulus and the Poisson's ratio of each rock layer and the target responses were the field measurements of horizontal and vertical total stress profiles and the maximum surface subsidence. GP provided functions relating each response to the elastic properties. An optimization procedure using the interior-point algorithm was applied in order to estimate the values for the properties that would lead to the target outputs. Results show that the calibration procedure provided properties that, in most cases, differed in less than 10% from the expected values. Thus, the proposed method has potential to be a relatively straightforward method to estimate parameters of a geomechanical model.

ABSTRACT: The one-step acid systems, mainly based on the chelating agents, can greatly simplify process of sandstone acidizing operation, and have a great application prospect. But for now, there are still no standard one-step acids optimization methods. Existing studies mostly conduct at room temperature conditions, and cannot reflect the field situation under the reservoir conditions. So current optimized acids still produce secondary precipitation and difficultly apply in sandstone acidizing. Herein, based on the complexometric titration, an effective optimization method of one-step sandstone acid systems at formation conditions is proposed. The method can quantitatively measure the chelating ability of chelating agents with respect to each type of metal ions under different formation conditions, especially temperature. According to this method, five chelating acids for four metal ions are tested. The results show chelating acid 4# can better chelate the four metal ions under different temperature. The core permeability after treatment of the preferred chelating acid 4# is increased by 1.68 times, 3.14 times, and 9.21 times at 25 C°, 60 C°, and 90 C°. In addition, mineral composition tests indicate the core after the treatment with chelating acid 4# has no secondary precipitation.

ABSTRACT: Harnessing the power of predictive statistical modeling and advanced machine learning techniques to evaluate and assess the impact of various well completion and stimulation parameters on the wells’ production performance have become one of the recent growing interest in the oil and gas industry and especially in the development of the unconventional resources. The main objective of this study is to utilize the partial least square (PLS), a machine learning technique, to create predictive models to evaluate the impact of several well completion parameters such as lateral length and gross perforated interval coupled with different stimulation parameters such as the total pumped water and sand volumes in addition to the concentrations of different stimulation additives. The completion and stimulation parameters will be used as a predictor variable to the short- and long-term gas production in the Marcellus shale play. Another outcome from this study is to gain insights about the relationship between those parameters and the wells production in order to optimize the future designs of those wells. Data of more than 2700 wells were utilized from two sources, the stimulation parameters were gathered from FracFocus website and the well completion and production data were gathered from DrillingInfo database. Three production predictive models were presented in the form of mathematical functions that were created based on the wells’ data which can be considered as the training dataset. Unlike many other prediction modeling approaches, this study applies cross validation procedures to assure the reliability of the constructed prediction models.

ABSTRACT: Rockfalls constitute an abrupt type of landslide process with manifold impact at local level, on economy and mostly on a human modified environment. Rapid automated and accurate mapping of rockfalls’ characteristics are crucial in order to identify areas of potential risk and propose mitigation measures. Simultaneously, urban expansion in hazardous regions in combination with climate change variations had played an important role on the increase of rockfall events. The current study presents a novel semi-automated approach to detect rockfall hazard from detailed 3D point clouds that were collected with an Unmanned Aerial Vehicle (UAV), right after a recent rockfall event, triggered by heavy rainfall that destroyed several houses in the base of a steep terrain. Block geometry measurements of sliding slope have been provided in relation with the geometrical and mechanical characteristics of the main discontinuity surfaces from the photogrammetric 3D point cloud. In addition, this study was contributed by field measurements and laboratory tests of the rock joint strength for the analysis of the slope stability of the rock blocks. A set of corresponding discontinuity factors have been detected and characterized in terms of point cloud analysis and provide a qualitative hazard system to assess rockfall probability by extensive characterization and parametrization of the joint systems. Results have demonstrated the capabilities of combining UAV platforms with computer-based methods for rapid and accurate identification of discontinuity parameters subjectively and even from inaccessible areas of the rockfall body.

ABSTRACT: High-fidelity flow and transport simulation on large discrete fracture networks (DFN) are computationally expensive. This makes uncertainty quantification studies of quantities of interest such as travel time through the network computationally intractable, since multiple runs of the DFN model are required to get good bounds on the uncertainty of the predictions. In this context, we present a system-reduction technique for DFNs using supervised machine-learning via a Random Forest Classifier. The in-sample errors (in terms of precision and recall scores) of the trained classifier are found to be very accurate indicators of the out-of-sample errors, thus exhibiting that the classifier generalizes well to test data. Moreover, this system-reduction technique yields sub-networks as small as 12% of the full DFN that still recover transport characteristics of the full network such as the peak dosage and behaviour of the breakthrough curve for late times. Most importantly, the sub-networks do not get disconnected, and their size can be controlled by a single dimensionless parameter. The computational efficiency gained by this technique depends on the size of the sub-network, but large reductions in computational time can be expected for small sub-networks, yielding as much as 90% computational savings for sub-networks that are as small as 10-12% of the full network. The method utilizes the inherent trade-off in all classification methods between precision and recall to use it to advantage and obtain connected networks in all instances considered. This happens because for small backbones, as one looks for high recall, the precision is low, and thus it adds more fractures to the network than an ideal classifier, and these end up providing the desired connectivity.

ABSTRACT: Flying automated Unmanned Aerial Vehicle (UAV) missions has become commonplace in aboveground mining applications. In global positioning system- (GPS-) and radio frequency- (RF-) denied environments, such as underground mines, using UAVs has been limited to areas where an experienced pilot could maintain line of sight and successfully navigate hazards. RESPEC and AutoModality are working together to leverage technology advancements in mobile computing and remote sensing to create a fully autonomous UAV platform that can travel more than 1 mile through an underground mine and perform complex tasks, such recognizing defects in equipment or delivering a breathing apparatus to a miner in an emergency. Powered by artificial intelligence running on an advanced onboard mobile processor, the UAV collects data from light and detection ranging (LiDAR), infrared (IR), and visible light sensors and controls the UAV's movement based on a map of the mine and mission specifications that direct the UAV. The technology developed through this project, which is partially funded by the National Institute for Occupational Safety and Health (NIOSH), will reduce personnel exposure to unsafe conditions and eliminate the need for expensive ground-support rehabilitation when attempting to identify and access problems in unmaintained, underground areas.

ABSTRACT: Obtaining permeability data has proved to be a difficult and expensive process. A permeability profile could determine which sections in a well are most profitable to stimulate and help characterize a reservoir very well. Utilizing well log data to find permeability could potentially reduce the need for laboratory testing which could also minimize the cost and lower the risk of the associated process. Permeability estimation in carbonate reservoirs is a challenging task to be handled accurately. Many researches tried to relate permeability and reservoir properties using complex mathematical equations which resulted in inaccurate estimation of the formation permeability values. The objective of this study is to develop a Functional Networks (FN) model that can be used to predict the permeability of heterogeneous reservoir based on three logs only, namely, resistivity, bulk density, and neutron porosity. In addition to the FN model, in this paper an empirical correlation of the FN model to predict permeability is also presented that can be used for carbonate rocks as far as given inputs lies within the range defined in the paper.

ABSTRACT: The failure of slopes and dams has major consequences worldwide, including extensive damage to infrastructure and related economic losses and, in the worst case, loss of life. The ability to monitor and forecast failure is a major concern for risk management and it is often hindered by lack of data. Recent technological advances enable the use of multiple sources of information, such as Earth Observation (EO), imagery analysis, real time monitoring, which can be combined with numerical analysis and machine learning techniques for monitoring, as well as forecasting the safety of civil works. This paper presents three applications concerning slope stability and dam safety: i) a practical research study for forecasting rainfall-induced landslides that combines remote sensing, geo-engineering and hydrology with machine learning; ii) image-based analysis for slope monitoring applied to the Brumadinho Dam (Minas Gerais, Brazil) pre-failure video footage; and iii) a pilot project for monitoring the safety of the Bhadra Dam, in India, combining the use of high-frequency in-situ measurements and low-frequency satellite-based measurements with numerical modeling. The results show how these enabling technologies can contribute with information to safer infrastructure in a lean, remote approach, which is particularly suited for data-scarce environments and the detection of potential anomalies.

ABSTRACT: Operators and service companies are always interested to have a clear insight about the rate of penetration (ROP) since it will provide a good estimate for the time and the cost of the drilling operations. The aim of this work is to use recurrent neural networks (RNNs) to accurately predict ROP for any formation prior to drilling. A recurrent neural network was created to estimate the ROP based on some key drilling parameters and unconfined compressive strength (UCS). The data were divided into three sets; training, validation, and testing datasets. 70% of the data used for training, 15% for validation, and the rest for testing. The optimum number of hidden layers and the number of neurons in the hidden layer were obtained using trial and error by calculating the mean square of error (MSE) for each trail and then the lowest MSE was chosen. The final results showed that the supervised RNN has the ability to predict ROP prior to drilling for any formation assuming the key drilling parameters that are desired to be used to drill the formation were known as well as UCS of the formation. The network predicted ROP with an overall R 2 of 0.94. This estimation is tangibly valuable for providing a robust image in regard the cost and time of the drilling operations.

ABSTRACT: An evaluation of the blasting performance at Erdenet Mine was conducted. Rock fragment size distributions were measured and analyzed with the digital image processing method employing the SPLIT_Desktop system. Regression prediction models were developed based on the field measurements and blasting parameters. Comparisons were made among the Kuz-Ram model predictions, the digital image measurements and the regression prediction models.

ABSTRACT: In tunneling with TBMs, a reliable assessment of the TBM rate of penetration (RoP) is needed for planning, cost control or other decision making purposes. An incorrect estimate of the RoP could delay the project and induce unplanned costs, making the project less profitable. Thus, extensive studies with the aim of estimating TBM's performance have been conducted in the last three decades. As a result, several empirical equations and models have been developed for different tunneling situations. The purpose of this study is to propose empirical models capable of assessing the RoP of TBM and to critically compare them. Two approaches were investigated on the basis of the rock mass parameters including the uniaxial compressive strength (UCS), intact rock brittleness (BI) alpha angle (a) the distance between planes of weakness (DPW). Datasets from the Queens Water Tunnel No. 3 project, New York City, were compiled and used to establish the models. On one hand, four (4) candidate models were proposed and the Bayesian inference approach was implemented to identify the most suitable among the candidates. The models were computed in WinBUGS software which employs the Bayesian analysis of complex statistical models. Next the models were ranked on the basis of the deviance information criterion (DIC) values taking into account both their fit and complexity. On the other hand, Multi-Layer Perceptron network a well-known approach used in rock engineering to map complex correlation between dependent and independent variables, was employed to predict the RoP. Both methods were validated and compared. The overall results were in good agreement with the field observations. In addition, the Bayesian-based models showed more flexibility in terms of probabilistic interpretation of the results.

ABSTRACT: Seismic hazards have become one of the common risks in underground coal mining and their assessment is an important component of the safety management. In this study, a methodology, involving nine fractal dimension-based indices and a fuzzy comprehensive evaluation model, has been developed based on the processed real time microseismic data from an underground coal mine, which allows for a better and quantitative evaluation of the likelihood for the seismic hazards. In the fuzzy model, the membership function was built using a Gaussian shape and the weight of each index was determined using the performance metric F score derived from the confusion matrix. The assessment results were initially characterised as a probability belonging to each of four risk levels (none, weak, moderate and strong). The comprehensive result was then evaluated by integrating the maximum membership degree principle (MMDP) and the variable fuzzy pattern recognition (VFPR). The model parameters of this methodology were first calibrated using historical microseismic data over a period of seven months at Coal Mine Velenje in Slovenia, and then applied to analyse more recent microseismic monitoring data. The results indicate that the calibrated model was able to assess seismic hazards in the mine.

ABSTRACT: The mechanical properties (i.e. elastic constants, Poisson's ratio, hardness, and modulus) of rock have been mostly obtained from rock core mechanical testing. Retrieving core samples from unconventional wells (i.e. gas/oil shales and tight gas sands) are known to be costly and with high risk. Previously considered having no value for rock mechanical evaluation, tons of drill cuttings are inevitably produced while drilling wells. In this project, we aim to extract the mechanical properties of shale rocks using just retrieved drill cuttings, thus making the characterization technique economical and efficient. Shale is classified as transversely isotropic material and its mechanical properties depend on the direction of loading with respect to the bedding orientation. Unlike core samples, the bedding orientation is difficult to be determined on small samples such as drill cuttings. To address this challenge, we performed microindentation testing on artificially made shale drill cuttings with varying orientations. The contact mechanic solutions for anisotropic and transversely-isotropic material are then utilized to correlate the indentation modulus acquired from randomly oriented drill cuttings to obtain components of the stiffness tensor for transversely isotropic material. An inverse problem is formulated to identify the mean values for the quantities of interest that best fit the data. This is done by performing a constrained optimization. The constraints represent physical information about the bounds on elastic properties as well as the mathematical constraint on the structure of elasticity tensor ensuring the accuracy and robustness of solutions to this optimization problem. Results are validated by performing microindentation on artificially made cuttings prepared from shale samples with known mechanical properties.

ABSTRACT: Lithological classification of geological formations usually depends on the geologist’ expertise through a laborious and time consuming process. Automating the rock classification procedure is a step towards fast and accurate characterization of geological formations where large data is available. In this study, we assessed the use of a clustering algorithm, namely self-organizing maps, for classification of different lithological classes of limestone. The input parameters are geo-mechanical and geological properties of limestones such as density, uniaxial compressive strength, p-wave velocity and geological origin. The results indicate that the self-organizing maps have a promising potential in lithological clustering. The algorithm accurately clusters limestone classes that have consistent high or low values for at least one of the input parameters. However, the accuracy decreases if the limestone class have inherent high variance. The results provided in this paper provides a preliminary confirmation of the applicability of the self-organizing maps. Future work will focus on the replication of the results with a larger dataset.

ABSTRACT: Experimental studies on fracture flow processes are vital to understand the dependence of fracture permeability on various complex processes. A key parameter in such experiments is the fracture aperture, which can be determined using light transmission techniques when using transparent fracture models. For this, scientific-grade monochrome cameras are used to record light intensity under different conditions. The output data of scientific cameras is linearly related to the radiance scene, which allows one to do quantitative analysis using such data. On the contrary, consumer-grade digital (CGD) cameras usually output nonlinear color images (e.g. JPG or TIFF files), which are not suitable for most scientific uses. In this paper, we outline an image analysis procedure to use CGD cameras for light intensity measurements. We demonstrate that the direct use of raw data from such cameras is adequate for scientific quantification and prevents the introduction of artefacts from further image processing that usually occurs within the built-in software of most cameras.

ABSTRACT: In underground mines, it is of paramount importance to ensure that local support design and installation is carried out in the best possible manner. A study was done to evaluate the adequacy of tendon support used at a Great Dyke platinum mine in anisotropic jointed rockmass. A comprehensive analysis of the geotechnical data collected was done using the Q rating, RMR and MRMR systems. The current tendon support system was evaluated using the roofbolt factor of safety approach, fallout thickness evaluation and numerical modelling for adequacy in geotechnically challenging ground conditions. Rockmass classification techniques used showed that the ground condition where the study was carried out is indeed weak, hence an adequate local support system is required to prevent falls of ground attributable to wedge failure. GPR scans and observations showed that the fallout height thickness is now 1.8m, instead of the 1.5m used in the preliminary stages of mining operations. Numerical modelling using JBlock software was used to deduce potentially unstable blocks and the probability of wedge failure under the current tendon regime. From the evaluation of the tendon systems used at the research area, the authors concluded that the current tendon system used is not adequate to meet the desired demand. A newly designed system that can improve safety and productivity was proposed following a comprehensive empirical and numerical analysis. The proposed tendon system is more economical with further benefits of improved safety. It is also recommended to mining companies to incorporate the use of risk evaluation softwares such as RiskEval to assist in the quantification and prediction of fall of ground related injuries in the design of support systems Local support systems also need to be updated with respect to geotechnical conditions in order to optimize the mine's key performance indicators (KPIs).

ABSTRACT: Tunnel Boring Machine (TBM) has been widely used in various types of tunnel construction. The disc cutters are key component of TBM which affecting the construction duration, cost and safety of TBM tunnel significantly. According to site investigation and field data, the cutter wear mechanism is analyzed. The influence factors of cutter wear mainly include geological parameters, tunneling parameters and cutterhead design parameters. An initial cutter wear prediction network is established based on Bayesian theorem. The structure of the net (a Directed Acylic Graph, DAG) is formulated according to the inter-relationship among cutter wear and various factors. Prior Conditional Probability Table (CPT) is obtained by training the model with field data by Netica. As an illustration, a simplified network is applied to YHJW project. The cases test indicates that the method is feasible and the predicted values of cutter wear rate are basically equal to the data collected from YHJW south tunnel under the same geology and tunneling parameters. The predicted results can also reflect the regulation of disc cutter wear under the separate influence of Equivalent Quartz Content (EQC), Uniaxial Compressive Strength (UCS) and thrust. Finally, the conception of an intelligent TBM construction management platform is proposed.

ABSTRACT: Predicting rockburst intensity is an important task in mining since rockburst occurs as a violent expulsion of rock in high geo-stress condition which causes considerable damages to underground structures, equipment and most importantly presents serious menaces to workers’ safety. It has been responsible for numerous deaths and injuries in underground mines across the world. Due to this importance, the current study aims at predicting the intensity of rockburst on the basis of 174 rockburst events that were compiled. Several existing criteria were considered to model the rockburst intensity. The inputs parameters included the maximum tangential stress, the uniaxial compressive strength, the uniaxial tensile strength of the surrounding rock and the elastic strain energy index. A Bayesian inference approach was implemented to identify the most appropriate models for estimating the rockburst intensity category among three rockburst criteria. The WinBUGS software was used to compute the posterior predictive distributions of the model parameters and the deviance information criterion (DIC) corresponding to the models. The DIC and the percentage of correctly predicted rockburst category were employed to assess the model performance. Overall, the results indicate that the Bayesian inference allows achieving satisfactory predictive performance in modelling the rockburst intensity. Also, the associated predictive uncertainty can be improved when new data are available. The results suggest that the implemented Bayesian models can be helpful in managing rockburst events in mines using site specific data and therefore, reducing the casualties induced by rockburst.

ABSTRACT: In recent years, efforts dubbed “i-Construction” have been intensified to improve the productivity of construction sites in Japan, and various studies have considered related topics, such as actively utilizing information and communications technology (ICT). In the study described, we utilized ICT technologies including artificial intelligence, construction-information modeling (CIM), and image processing at mountainous civil engineering work sites to automatically determine geological conditions on site. We then built a system for centralized management of the results and summarized details, which could be performed by a geological engineer in a remote location via the cloud. This paper describes (1) the overall outline of the system, and describes (2) the creation/storage system for construction records utilizing CIM and geological observation records, (3) the sophistication and 3D rendering of slope measurements, and (4) the quantification and automation of geological evaluations of the tunnel face or slope excavation as it exists at the construction site.