# Artificial intelligence-assisted laser ultrasound method for the estimation of porosity in hairpin weld seams

## TL;DR

This paper explores using laser ultrasound and artificial intelligence to detect air inclusions in hairpin welds of electric motor stators, showing promising results compared to computed tomography.

## Contribution

The novel use of artificial intelligence to estimate porosity in hairpin welds from laser ultrasonic measurements is introduced.

## Key findings

- A median correlation of 0.6 was found between AI estimates and computed tomography pore volume data.
- Highly informative regions in B-scans were identified, which could improve data acquisition efficiency.
- Standard methods for porosity estimation were found unsuitable due to the complex geometry of hairpin welds.

## Abstract

Hairpin technology is being used as a replacement for the traditional winding stator in electric motors. In hairpin stator manufacturing, copper rods are used to achieve a higher slot fill factor. These rods are joined together in pairs through laser welding, forming a closed circuit. However, this welding process is prone to air inclusions in the welds, which can negatively impact the efficiency and durability of the motor. The present study aims to estimate the total volume of these air inclusions using laser ultrasonic measurements. Laser ultrasound is a fast, non-contact, non-destructive method that can cope with the limited sample accessibility, making it ideal for inline testing of these weld seams. To evaluate the effectiveness of laser ultrasound, a stator was intentionally manipulated prior to laser welding to favor the formation of air inclusions. The porosity of the weld seams was determined through computed tomography images. It was demonstrated that due to the complex geometry of the hairpin welds, leading to a complex ultrasound wave field, standard methods to estimate the porosity from laser ultrasound B-scans are difficult to apply. As an alternative approach, an algorithm that is based on artificial intelligence was utilized for the purpose of estimating the air inclusion volume in the welds from laser ultrasonic measurements. The outcomes demonstrated a median correlation of 0.6 between this estimate and the pore volume obtained from the computed tomography data, despite the utilization of only 48 samples. Moreover, these results were evaluated against a model where the labels were randomly mixed, and highly informative regions regarding pore volume were identified in the B-scans, which have the potential to accelerate the process of acquiring data.

•Laser ultrasound applicability for defect detection in hairpin welds demonstrated.•Artificial intelligence methods utilized to estimate porosity in hairpin welds.•Comparison of computed tomography with laser ultrasound results.•Regions in B-scans highlighted that are most informative for neural networks.

Laser ultrasound applicability for defect detection in hairpin welds demonstrated.

Artificial intelligence methods utilized to estimate porosity in hairpin welds.

Comparison of computed tomography with laser ultrasound results.

Regions in B-scans highlighted that are most informative for neural networks.