Machine learning-driven ontological knowledge base for bridge corrosion evaluation

In bridge maintenance, assessing structural performance requires adherence to rules outlined in safety and regulatory standards which can be effectively and formally represented in both human and machine-readable formats using ontologies. However, ontology-based semantic inference alone falls short when faced with the complicated mathematical operations required for structural analysis. The increasing digitization of bridge engineering has opened doors to data-driven prediction methods. Machine learning (ML)-based models, in particular, have the capacity to learn from historical data and forecast future structural performance with remarkable accuracy. This paper introduces an innovative approach that integrates ML models with an ontological knowledge base for evaluating bridge corrosion. Web Ontology Language and Semantic Web Rule Language are combined to develop the knowledge base. Random forest algorithm is used to train the ML model with a good agreement (coefficient of determination of 0.989 and root mean square error of 1.200). A Python-based module is designed to seamlessly integrate ML predictions with ontology-based semantic inference. The proposed approach not only infers the corrosion ratings based on the rules defined in the Network Rail standard, but also infers the structural safety performance based on predicted structural response under the action of corrosion. To demonstrate the effectiveness of the developed method in enabling accurate and rational evaluations, a real bridge in the UK is showcased as a practical application