Classification of defect types in flow drill screw fastening using a machine learning approach

Abstract

Defect types in flow drill screw fastening are diverse and occur under a variety of influences. Some of these cannot be predicted on the basis of the set screw parameters. Depending on the materials to be joined, the parameters must be set precisely and individually. In this paper, the correlation between defect types and influencing parameters is analyzed and used to develop a defect classification tool. To this end, an extensive experimental program was conducted using a variety of steel and aluminum alloys, sheet thicknesses and screw types with different surface coatings. A broad range of parameter settings was deliberately applied to intentionally produce defective joints for training purposes. The generated dataset enabled the training of a supervised machine learning model. Therefore, a feedforward neural network was trained to be capable of classifying defect types based on process data.