Mr Jiawei Xie

Predicting rock fractures in unexcavated areas is a critical yet challenging aspect of geotechnical projects. This task involves forecasting the fracture mapping sequences for unexcavated rock faces using the sequences from excavated ones, which is well-suited for spatial¿temporal deep learning techniques. Fracture mapping sequences for deep learning model training can be achieved based on field photography. However, the main obstacle lies in the insufficient availability of high-quality photos. Existing data augmentation techniques rely on slices taken from Discrete Fracture Network (DFN) models. However, slices differ significantly from actual photos taken from the field. To overcome this limitation, this study introduces a new framework that uses Virtual Camera Technology (VCT) to generate "virtual photos" from DFN models. The external (e.g., camera location, direction) and internal parameters (e.g., focal length, resolution, sensor size) of cameras can be considered in this method. The "virtual photos" generated from the VCT and conventional slicing method have been extensively compared. The framework is designed to adapt to any distribution of field fractures and camera settings, serving as a universal tool for practical applications. The whole framework has been packaged as an open-source tool for rock "photos" generation. An open-source benchmark database has also been established based on this tool. To validate the framework's feasibility, the Predictive Recurrent Neural Network (PredRNN) method is applied to the generated database. A high degree of similarity is observed between the predicted mapping sequences and the ground truth. The model successfully captured the dynamic changes in fracture patterns across different sections, thereby confirming the framework's practical utility. The source code and dataset can be freely downloaded from GitHub repository (https://github.com/GEO-ATLAS/Rock-Camera).

In large-scale hydropower project construction, comprehensive internal and external deformation monitoring (periodic observation) of high-steep rock slopes is crucial for revealing slope deformation and rock mass deterioration. However, few studies have examined the impact of using both surface and subsurface monitoring data on the performance of probabilistic back-analysis (PBA). This study aims to fill this gap by developing an improved Bayesian PBA method. Using the left bank excavation slope of the Baihetan (BHT) hydropower station in China as a case study, we probabilistically calibrated the geotechnical parameters of the slope using multi-type surface (robotic total stations) and subsurface (multi-point displacement meters and inclinometers) monitoring data. The results indicate that compared to using a single type of monitoring data, using multiple types of monitoring data can further reduce the uncertainty of geotechnical parameters. Specifically, integrating surface and subsurface monitoring data for back-analysis can achieve an optimal three-dimensional model prediction and yield a reasonable parameter set. When there are differences in monitoring data stability, incorporating relatively stable monitoring (minor deformation) data into the Bayesian back-analysis can help improve the convergence speed of Bayesian sequential inversion. However, appropriate methods are required to evaluate the contribution of these data to the back-analysis model.

Slope failure mechanism and reliability assessment under rainfall usually not only ignores the nonstationary characteristics of soil hydraulic and shear strength parameters, but also does not make use of the freely available field observation that the slope remains stable under the natural condition. In this paper, the nonstationary characteristics and spatial variabilities of soil hydraulic and shear strength parameters, along with model bias, are explicitly accounted for. Firstly, Bayesian inverse analysis is conducted to infer the spatially varying shear strength parameters and reduce their uncertainties by incorporating the field observation. Following this, an infinite slope model is taken as an example to perform slope seepage, stability and reliability analyses subjected to a rainfall event based on the posterior statistics of soil shear strength parameters. The probabilities of slope failure and distributions of critical slip surface for various rainfall durations are then evaluated within a Monte-Carlo simulation framework. Based on these, the slope failure mechanism induced solely by the rainfall is investigated. The results indicate that the probability of failure of the infinite slope, when evaluated using the posterior statistics of soil shear strength parameters, is close to zero (7.24 × 10-2), which aligns with the field observation wherein the slope remains stable under the natural condition. The triggering factors for slope failure vary across different stages of rainfall infiltration are identified and elucidated in this paper. Ignoring the field observation and the nonstationary characteristics of soil properties can lead to inaccurate assessments of both the failure mechanisms and probabilities of slopes induced by the rainfall. The research can provide a new perspective for understanding the slope failure mechanism caused by the rainfall.

Geotechnical subsurface stratification based on sparse measurements presents a significant challenge. Learning from prior geological information, such as learning soil layer distribution patterns from stratification results (2D images) of adjacent data-rich projects, is an emerging approach to reducing uncertainties caused by data scarcity. Existing methods rely on pixel-based techniques that require training images to have identical soil types as the testing image. Additionally, pixel-based methods are prone to error accumulation. To address these issues, this study introduces a new framework that focuses solely on learning boundary information from training image rather than soil types. This eliminates the need for matching soil types between training and testing images. Tree-based model is proposed to learn boundary information from training images. Contrary to conventional tree-based models that use coordinates as input, this study employs a set of designed distance fields (boundary dictionary) to represent complex boundary patterns. A selection process is introduced to identify the most important distance fields from the training images. Using these selected distance fields for model input results in soil stratification that aligns well with both borehole data and training image boundaries. The proposed method's efficacy is validated through multiple simulated and real-world cases. The proposed method outperforms pixel-based methods in multiple cases, achieving up to a 25% improvement in accuracy. This method is robust and it yields consistent results for a variety of training images considered. Additionally, the proposed method also provides quantification of interpolation uncertainty through the Gini impurity method.

Pile load tests have been utilized to reduce the uncertainty of pile resistance, thus leading to a higher resistance factor used in the Load and Resistance Factor Design (LRFD). Previous studies have primarily focused on calibrating resistance factors for single piles based on load tests. This calibration hinges upon the resistance bias factor of single piles, defined as the ratio of measured resistance to predicted resistance. Due to the redundancy in the pile group system, it is conventionally assumed that if the individual piles within the group achieve a lower reliability index (e.g., 2.0¿2.5), the pile group as a whole attains the target reliability index of 3. However, the approach is empirical as it does not consider system redundancy directly. Moreover, this empirical approach disregards the correlation between resistance bias factors of individual piles, which is inherently influenced by the spatial variability of soils. In this study, the random finite difference method (RFDM) is employed to evaluate the correlation between resistance bias factors of individual piles in spatially variable soils. The resultant correlation matrix is subsequentially employed in Bayes' theorem to update resistance bias factors using individual pile load test results and their corresponding test locations. The updated resistance bias factors are then used for the direct calibration of resistance factors for pile groups within the framework of LRFD. A pile group subject to vertical loading in undrained clays is adopted for illustration. Comparative analyses between the proposed approach and the empirical approach demonstrate that the latter tends to overestimate the resistance factor. Furthermore, the proposed approach enables the determination of optimal locations for conducting subsequent load tests based on previous test results.

Automatic and timely identification of mud pumping is important for the reliability and safety of railroads. The current mud pumping prediction model is based on monitoring the dynamic response of railway tracks. The essential geotechnical trigger factors such as the hydrological conditions are not well-considered in these prediction models, as that information is hard to be quantified, which unavoidably reduces the accuracy of the prediction. This paper proposes to utilize the Geographic Information System (GIS) to quantify the hydrological information along railway tracks. Through GIS analysis, the hydrological variables including elevation, near-river distance, rainfall, sink depth, and soil types are estimated and combined with in-service train monitoring data for model development. To deal with multi-attribute data, a dual-channel neural networks model is proposed to separately mine the characteristics in different attributes data for prediction. To further understand the prediction model, Shapely addictive explanations (SHAP) method is applied to estimate the importance of hydrological variables and reveal the possible relationships between the variables and the probability of mud pumping. The proposed approach is applied to a real-life case from the railway tracks in Australia to validate its effectiveness. The prediction results show that the proposed approach can predict mud pumping with balanced accuracy of 90.84%. The results confirm that integrating of GIS information and monitoring data can generate more accurate prediction and reduce the false prediction rate. Based on the explainable results, it is observed that rainfall is the most important hydrological variable that influences mud pumping occurrence. Apart from rainfall, groundwater-related variables show a greater impact on mud pumping occurrence than surface water-related variables. The explainable results also can help infrastructure managers to identify the most vulnerable sections in railway tracks, which facilitates targeted maintenance planning and track substructure design.

Numerical simulation is a powerful technique for slope stability assessment and landslide hazard investigation. However, the physicomechanical parameters of the simulation results are susceptible to uncertainty. Displacement back-analysis is considered an effective method for the prediction of the geomechanical parameters of numerical models; therefore, it can be used to deal with the parameter uncertainty problem. In this study, to improve the interpretability of the back-analysis model, an analytical function relationship between slope displacements and physicomechanical parameters was established using geographically weighted regression. By combining the least-squares and linear-algebra algorithms, a displacement back-analysis method based on geographically weighted regression (DBA-GWR) was developed; in particular, the multi-objective displacement back-analysis was represented as an analytical problem. The developed method was subsequently used for a slope of the Guiwu Expressway in Guangxi, China. Simulation experiments and GNSS real-data experiments demonstrated that the GWR could achieve high-precision deformation modelling in the spatial domain with model-fitting precision in the order of mm. Compared with state-of-the-art methods, the precision of the simulated displacement with the proposed method was significantly improved, and equivalent physicomechanical parameters with higher accuracy were obtained. Based on the corrected numerical model, the most severely deformed profiles were forward-analysed, and the simulated deformation and distribution patterns were found to be in good agreement with the field investigation results. This approach is significant for the determination of geomechanical parameters and the accurate assessment of slope safety using monitoring data.

Geotechnical and geophysical testing data are conventionally considered as separated information or combined based on deterministic methods in site investigation programs, which causes loss of information and introduces additional uncertainties. This study aims to reduce the uncertainties and costs in inhomogeneous soil profile characterization and geotechnical analysis by quantitatively integrating these data. The intrinsic collocated co-kriging method (ICCK) is utilized to integrate Multi-channel analysis surface wave (MASW) and CPT data. This research shows the potential of using the MASW data to explore the horizontal scale of fluctuation (SOF) of the cone tip resistance (qc) field which is very difficult to get using the conventional methods due to the limited CPTs. The Markov model is used to avoid the tedious modeling of the cross-covariance relationship in the ICCK method. A series of synthetic case studies show that the combined soil profile is in good agreement with the "true" qc field with significantly reduced uncertainties. Based on estimating the uncertainties, the optimal distance between CPTs is suggested to be 1¿2 horizontal SOF. This framework is also applied to a real case in the Christchurch area, which involves integration of CPTs and MASW tests. Cross validation and edge detection methods are used to quantitatively compare the integration results, confirming that the proposed framework is fully applicable to field data and can provide realistic and reliable estimations of qc field.

Timely detection and identification of rail breaks are crucial for safety and reliability of railway networks. This article proposes a new deep learning-based approach using the daily monitoring data from in-service trains. A time-series generative adversarial network (TimeGAN) is employed to mitigate the problem of data imbalance and preserve the temporal dynamics for generating synthetic rail breaks. A feature-level attention-based bidirectional recurrent neural network (AM-BRNN) is proposed to enhance feature extraction and capture two-direction dependencies in sequential data for accurate prediction. The proposed approach is implemented on a three-year dataset collected from a section of railroads (up to 350 km) in Australia. A real-life validation is carried out to evaluate the prediction performance of the proposed model, where historical data are used to train the model and future 'unseen' rail breaks along the whole track section are used for testing. The results show that the model can successfully predict nine out of 11 rail breaks three months ahead of time with a false prediction of nonbreak of 8.2%. Predicting rail breaks three months ahead of time will provide railroads enough time for maintenance planning. Given the prediction results, a Shapley additive explanations (SHAP) method is employed to perform a cause analysis for individual rail break. The results of cause analysis can assist railroads to plan appropriate maintenance to prevent rail breaks.

Timely detection and identification of substructure defects in railway track are crucial for the safety and reliability of railway networks. Instrumented in-service trains can provide daily data for assessing the track conditions. This study tries to develop a data-driven model for the prediction of mud pumping defects using daily in-service train data. The data-driven model is based on long short-term memory (LSTM) networks. Bayesian optimization method is used to select the optimal hyper-parameters in LSTM. Genetic algorithm (GA) method is used for feature selection. A four-year real-world dataset from a section of railway network in Australia is used to train and test the data-driven model. The t-distributed Stochastic Neighbor Embedding (t-SNE) method is used to examine the resulting LSTM networks. The results show that the proposed approach can be used to predict the mud pumping defects in advance leaving enough time for maintenance.

Conventional planning of maintenance and renewal work for railway track is based on heuristics and simple scheduling. The railway industry is now collecting a large amount of data with the fast-paced development of sensor technologies. These data sets carry information about the conditions of various components in railway track. Since just before the beginning of the 21st century, data-driven models have been used in the predictive maintenance of railway track. This study presents a systematic literature review of data-driven models applied in the predictive maintenance of railway track. A taxonomy to classify the existing literature based on types of models and types of applications is provided. It is found that applying the deep learning methods, unsupervised methods, and ensemble methods are the new trends for predictive maintenance of railway track. Rail geometry irregularity, rail head defect, and missing rail components detection were the top three most commonly considered issues within the application of data-driven models. Prediction of rail breaks has received increasing attention in the last four years. Among these data-driven model applications, the collected data types are the most critical factors which affect selecting suitable models. Finally, this study discusses upcoming challenges in the predictive maintenance of railway track.

The discreteness of fatigue test results of inorganic bonded stabilized materials is very voluminous. In this study, the fatigue crack checking process of cement stabilized macadam is analyzed by using measurement uncertainty and interval analysis theory in order to improve the efficiency of fatigue test data. Firstly, the measurement uncertainty of the bending strength and fatigue life of cement stabilized macadam is evaluated based on the measurement uncertainty theory. Secondly, the regression analysis of the fatigue life interval of different stress ratio conditions is realized, following which the regression model of the fatigue life interval of cement stabilized macadam is derived. Finally, the calculation model of fatigue cracking life interval is deduced, and the fatigue cracking life interval of cement stabilized macadam is then obtained by interval algorithm. Meanwhile an evaluation method of interval comparison results is proposed in order to undertake a comparative analysis of the fatigue cracking life interval of cement stabilized macadam. According to the findings, the non-uniformity of the specimen is determined as the root cause of the large discreteness of fatigue test results. A fatigue interval model with 95% guarantee rate can be obtained based on the regression analysis method of fatigue life interval, which is proved to be effective in improving the efficiency of fatigue test data. There is a case of interval expansion in calculating the equation of fatigue cracking life interval. It is verified that the extended interval result has engineering applicability. Therefore, it can be surmised that the interval result comparison method proposed in this paper can achieve both quantitative and qualitative evaluation, which is particularly suitable for the engineering field of interval analysis.