A defect detection system uses a CNN to detect cracks in images, then organises the detected objects and spatial relationships into a knowledge graph to infer construction status. This is best described as a hybrid approach, combining connectionism and symbolic AI.

A system uses an ontology that says “excavator is a type of heavy equipment” and “heavy equipment near a worker is a safety risk”. It then reasons that Worker A standing 1.5 metres from Excavator 01 is in a risk situation. This is mainly symbolic AI.

A computer vision model learns from thousands of labelled site images to recognise cracks, spalling and corrosion staining under different lighting conditions. This is mainly connectionism because it learns visual patterns from examples.

A site safety system applies a fixed rule: “If a worker is within 3 metres of heavy equipment, generate a safety-risk alert.” This is mainly symbolic AI because the reasoning is based on an explicit rule.

A rule-based model says: “All cracks are black and vertical.” Because the rule is explicit, it will usually be more accurate than a CNN trained on diverse real-site crack images.

A knowledge graph is closer to connectionism than symbolic AI because it stores data in a computer.

A neural network is always fully explainable because it represents knowledge as human-readable rules and ontologies.