Using AI-Based Tools to Quantify the Technical Condition of Bridge Structural Components

The main task of the operation of engineering structures is to ensure the stability of structures with aggressive external influences, which have a complex probabilistic nature. The reliable functioning of bridge structures requires the development and application of modern systems for inspection and assessment of the technical condition of the structure to take timely measures to ensure the safe operation of the structure in changing operating conditions. With the rapid development of AI, modern approaches are increasingly adopted, offering distinct advantages compared to classical methods. The article aims to develop an AI-based model for quantifying the technical condition of bridge structural components based on data obtained from the survey. To achieve this goal, the authors analyzed existing approaches to the inspection and assessment of bridges and studied the experience of using AI in bridge assessment. Based on the Polish Principles of Bridge Technical Condition Assessment, three datasets were formed to quantify the condition of the bridge components made from reinforced concrete: bridge deck, span structures, and piers and abutments. This study created and compared the performance of five AI-based models: XGBoost, Decision Trees, Random Forest, Support Vector Regression, and Artificial Neural Networks (ANNs). The initial comparison revealed relatively low performance across all models, with the ANN model showing a slight advantage. Subsequently, nine ANN models were optimized to achieve higher performance levels. The performance of models was conducted based on a comparison of mean absolute percentage error (MAPE) and R2 metrics. The ANN model with ReLU activation functions for hidden layers and the RMSprop optimizer achieves optimal performance at 100 epochs (MAPE = 3.5%; R2 = 0.994). The practical implementation of such a model can considerably reduce uncertainties stemming from subjective expert judgments and enhance the accuracy of assessments.