Questions and Answers
Each answer opens with the result, gives the number as the paper reports it, and links the DOI so it can be checked. Where I am not the first or corresponding author, the answer says so. Last reviewed 16 August 2026 by Dr.-Ing. Daguang Han.
Automatically generated CAD models reached an Intersection over Union of up to 91% for critical structural outlines. The same testing reported dimensional inaccuracies of about 8% by F1-measure and an 83% success rate in identifying primary components. The case study was the Baijusi Bridge, a large-scale structure where manual modelling would have been prohibitively slow.
Geometric prediction deviation was less than 1/1800 of the experimental bridge span, and the mean stress predicted by the method was 90% of the measured mean stress. The framework couples a terrestrial laser scanning point cloud, a BIM model and finite element analysis, so that the fit-up check happens before components reach site. Point cloud registration accuracy is the limiting factor on the whole chain.
The corrosion rate constant and the zinc coating thickness together govern 65% of the total reliability variance. That result comes from a field demonstration on a 40-year-old galvanised lattice tower in an ISO 9223 C4 coastal environment. The practical consequence is that a risk-ranked selective maintenance strategy beats uniform treatment of a tower population, because two measurable quantities carry most of the uncertainty.
On a real power infrastructure project the framework was 95.8% accurate in translating written rules and 98.3% feasible in rule execution, and it cut human effort by 90% — 168 hours against 1620 hours. The method keeps a human in the loop rather than removing one: the machine proposes the formalised rule, a person accepts or corrects it, and the correction updates the rule base. That is why the accuracy holds on messy real documents.
For an arch bridge, alignment deviation did not exceed 2.5 mm horizontally and laterally, and did not exceed 1.4 cm vertically. The vertical figure is an order of magnitude larger than the horizontal one, which is the honest and expected result: vertical alignment accumulates error from temperature, creep and support settlement in a way that plan position does not.
Yes — multiple CFRP layers increased plateau force, reduced maximum deflection to 42%, and shortened impact duration. The effect is not linear in the number of layers, and the gain is in energy absorption behaviour rather than in raw static capacity, which is why impact testing rather than static testing is the right way to qualify it.
YOLOv8s-SNC improved precision by 2.6% and recall by 7.6% over the original YOLOv8 for helmet detection in industrial scenarios. Recall matters more than precision for this task: a missed unhelmeted worker is a safety failure, while a false alarm is only a nuisance, so the larger recall gain is the more useful of the two numbers.
A mechanical state inversion method using 3D laser scanning was verified on a completed suspension bridge with a span of 808 m. The approach infers the force state from measured geometry, which matters for bridges in service where the original construction records are incomplete or the structure has since been modified.