Automated Inspection Utilizing Deep Learning for Polished Skin

This technical paper reports the development of an automatic defect detector utilizing deep learning for “polished skins”. Materials with a “polished skin” are used in the fabrication of the external plates of commercial airplanes. The polished skin is obtained by polishing the surface of an aluminum clad material, and they are visually inspected, which places a significant burden on inspectors to find minute defects on relatively large pieces of material. Automated inspection of these skins is made more difficult because the material has a mirror finished surface. Defects are broadly classified into three categories: dents, bumps, and discolorations. Therefore, a defect detector must be able to detect these types of defects and measure the defects’ surface profile. This technical paper presents details related to the design and manufacture of an inexpensive automated defect detector that demonstrates a sufficiently high level of performance. The system employs multiple line sensor cameras and image processing including deep learning to find defects. By developing an effective automated inspection method combined with a relatively large apparatus and an appropriate processing algorithm, defect detection performance of the system was found to be equal to or higher than existing visual inspection methods. Furthermore, although the measurement time depends on the size on the piece of material, most inspections take short time to generate a report, which is significantly less than the time needed to conduct a visual inspection. This technology could be used in the automated inspection of large pieces of material with a mirror-surface that are currently inspected visually.