Development of Method to Detect Steel Fractures in Concrete Members Using Magnetic Flux Leakage Method with Machine Learning

Abstract

One of the techniques to detect steel fractures in reinforced concrete and prestressed concrete bridges is the magnetic flux leakage method. Judgment using machine learning requires a large dataset of many types to determine the presence or absence of steel fractures. To increase the amount of training data from actual structures with steel fractures, a mockup was created in the laboratory to confirm the parameters that affect measurements. The experimental parameters included concrete cover thickness, horizontal spacing of steel bars, fracture spacing of steel bars, number of horizontally arranged steel bars, stirrup spacing, stirrup cover thickness, and vertical spacing of steel bars. Machine learning was applied using the 754 data points obtained from both actual structures and laboratory test specimens.