Quality Control and Fault Classification of Laser Welded Hairpins in Electrical Motors
Johannes Vater, Matthias Pollach, Claus Lenz, Daniel Winkle, Alois Knoll
Abstract
We present the development, evaluation, and comparison of different neural network architectures using different input data to detect and classify quality deviations in the welding of hairpins. Hairpins are copper rods that are located in the stator of electric motors in electric cars. We use both 3D data and grayscale images as input. The primary challenges are that only a small dataset is available and that high network accuracy is essential to prevent defects in the usage of an electrical engine and to enable a focused rework process. We were able to achieve a 99% accuracy using either 3D data or grayscale images.
Topics & Concepts
Computer scienceGrayscaleStatorArtificial intelligenceProcess (computing)WeldingArtificial neural networkReworkPattern recognition (psychology)Convolutional neural networkComputer visionPixelEngineeringMechanical engineeringEmbedded systemOperating systemWelding Techniques and Residual StressesIndustrial Vision Systems and Defect DetectionNon-Destructive Testing Techniques