Unknown bridge foundations pose a significant safety risk due to stream scour and erosion. Records from older structures may be non‐existent, incomplete or incorrect. We evaluate 2D and 3D electrical resistivity imaging (ERI) as a means to reliably identify the depth of unknown bridge foundations. A survey procedure is described for mixed terrain/water environments in the presence of rough terrain. Some electrodes are installed on the stream banks while others are adapted for underwater use. Tests were conducted at five field sites, including three roadway bridges, a geotechnical test site and a railway bridge, containing drilled shafts and spread footings of both known and unknown depth extent. The 2D data acquisition was carried out in the dipole‐dipole configuration. The 2D ERI method resolved the shape and depth extent of the larger bridge foundations but, with less accuracy, the shape and depth extent of the smaller foundations. The 3D ERI method is time‐consuming and does not add sufficient additional value over 2D ERI to become a practical tool for unknown bridge foundation investigations. The 2D ERI method is a cost‐effective geophysical method that is relatively easy to use by bridge engineers.

Bridge scour is a challenge throughout the U.S.A. and other countries. Despite the scale of the issue, there is still a substantial lack of robust methods for scour prediction to support reliable, risk-based management and decision making. Throughout the past decade, the use of real-time scour monitoring systems has gained increasing interest among state departments of transportation across the U.S.A. This paper introduces three distinct methodologies for scour prediction using advanced artificial intelligence (AI)/machine learning (ML) techniques based on real-time scour monitoring data. Scour monitoring data included the riverbed and river stage elevation time series at bridge piers gathered from various sources. Deep learning algorithms showed promising in prediction of bed elevation and water level variations as early as a week in advance. Ensemble neural networks proved successful in the predicting the maximum upcoming scour depth, using the observed sensor data at the onset of a scour episode, and based on bridge pier, flow and riverbed characteristics. In addition, two of the common empirical scour models were calibrated based on the observed sensor data using the Bayesian inference method, showing significant improvement in prediction accuracy. Overall, this paper introduces a novel approach for scour risk management by integrating emerging AI/ML algorithms with real-time monitoring systems for early scour forecast.

Missing substructure information has impeded the safety assessment of bridges with unknown foundations, especially for scour-prone bridges. An approach based on artificial neural networks (ANNs) was developed to identify the inherent patterns in the substructure design of bridges with commonly available evidence (e.g., geometric characteristics of superstructures, loading conditions, soil properties, year built, and location) and then to generalize them further to bridges with unknown foundations. The proposed ANN models were trained with information collected for an inventory of bridges with available foundation records located in the Bryan District of the Texas Department of Transportation. Results showed that the proposed ANN models were able to make successful predictions about the foundation type and the embedment depth for deep foundations. In addition, the degree of uncertainty in the models’ predictions was evaluated by performing the random subsampling method. Graphs of the probability of exceedance were generated that allowed for factoring the predicted pile depth on the basis of a reasonable probability of failure caused by scour. As a consequence, departments of transportation can adopt a similar methodology to reclassify the bridges with unknown foundations included in the National Bridge Inventory.

This paper presents the experimental database and corresponding statistical analysis (Part I), which serves as a basis to perform the corresponding parametric analysis and machine learning modelling (Part II) of a comprehensive study on organic soil strength and stiffness, stabilized via the wet soil mixing method. The experimental database includes unconfined compression tests performed under laboratory-controlled conditions to investigate the impact of soil type, the soil’s organic content, the soil’s initial natural water content, binder type, binder quantity, grout to soil ratio, water to binder ratio, curing time, temperature, curing relative humidity and carbon dioxide content on the stabilized organic specimens’ stiffness and strength. A descriptive statistical analysis complements the description of the experimental database, along with a qualitative study on the stabilization hydration process via scanning electron microscopy images. Results confirmed findings on the use of Portland cement alone and a mix of Portland cement with ground granulated blast furnace slag as suitable binders for soil stabilization. Findings on mixes including lime and magnesium oxide cements demonstrated minimal stabilization. Specimen size affected stiffness, but not the strength for mixes of peat and Portland cement. The experimental database, along with all produced data analyses, are available at the Texas Data Repository as indicated in the Data Availability Statement below, to allow for data reproducibility and promote the use of artificial intelligence and machine learning competing modelling techniques as the ones presented in Part II of this paper.

Acoustic emission (AE) reading is among the most common methods for monitoring the behavior of brittle materials such as rock and concrete. This study uses discrete element method (DEM) simulations to explore the correlations between the pre-failure AE readings with the post-failure behavior and residual strength of rock masses. The deep learning (DL) method based on long short-term memory (LSTM) algorithms has been applied to generate predictive models based on the data from DEM simulations of biaxial compression. The dataset has been populated by varying interparticle friction while keeping bond cohesion constant. Various configurations of the LSTM algorithm were evaluated considering different scenarios for input features (strain, stress, and AE energy records) and a range of values for the key hyperparameters. The prime AI models show promising accuracy in predicting residual strength decay with strain based on pre-failure patterns in AE readings. The results indicate that the pre-failure AE indeed encapsulates information about the developing failure mechanisms and the post-failure response in rocks, which can be captured through artificial intelligence.