Modern construction and industrial automation environments generate large volumes of heterogeneous data from sensors, visual inspections, equipment logs, maintenance records, and human observations. Although artificial intelligence methods are increasingly used for anomaly detection, object recognition, and predictive analytics, many existing approaches remain isolated from domain knowledge and provide limited explainability for engineering decision-making. This paper proposes an AI-driven multimodal sensor intelligence architecture that combines embeddings, ontology-based knowledge representation, and autonomous agentic workflows for contextual interpretation of sensor and inspection data.
The proposed architecture introduces an ontology layer that represents physical assets, sensor observations, construction activities, defects, operational states, risks, and recommended actions. Image and text embeddings are used to connect visual evidence, sensor readings, and inspection descriptions in a shared semantic space. A knowledge graph organizes these relationships and enables explainable reasoning over detected events. On top of this semantic layer, agentic workflows coordinate retrieval, validation, summarization, rule checking, and notification tasks. This allows the system to support human experts by identifying relevant evidence, detecting inconsistencies, and recommending follow-up inspection or maintenance actions.
The approach is positioned for construction monitoring and industrial automation scenarios, where sensor signals, images, and technical documentation must be interpreted together. The expected contribution is a reusable framework for integrating artificial intelligence with sensor data, semantic models, and human-supervised autonomous workflows. Compared with isolated AI models or conventional sensor dashboards, the proposed approach improves contextual retrieval, explainability, and decision support for complex engineering environments.