Automatic Detection of Surface Damages on Steel Structures using Near Infrared Hyperspectral Imaging

Publication Type:

Conference Abstract


9th Workshop on Hyperspectral Image and Signal Processing : Evolution in Remote Sensing (WHISPERS) (2018)


Surface damages such as corrosion and coating delamination are very common on steel structures and, if not prevented, can accelerate degradation and eventually lead to failure of the whole structure. Hence, field monitoring is of utmost importance to guarantee the structural health and reduce the maintenance cost. Visual inspection is very perilous (especially for offshore structures), time-consuming, and largely depends on the experience of the inspector (Price & Figueira 2017). Current non-destructive testing (NDT) techniques (e.g. based on ultrasonic or eddy current technology, acoustic emission ([Calabrese et al 2016; Wu et al 2016] or Electrical Resistance [ Li et al 2007; Mathiesen et al 2016]), require a direct contact with the surface, which causes difficulties when the structure or parts of it are not accessible.
Image processing techniques have been considered as a fast and easier alternative to contact-based techniques. These techniques are usually based on measuring the textural and morphological changes in digital images from the structures (Feliciano et al 2015; Li & Cheng 2016). However, in reality other materials on steel structures such as dirt, paint, bird droppings, oil leakage, and biofouling can significantly complicate the inspection. Moreover, some effects of corrosion not necessarily reveal themselves in the visible wavelength region. Hyperspectral imaging outside the visible range is an advanced imaging technique that can reveal chemical-mineralogical surface changes that are not observable by the naked eye.
This paper studied the feasibility of detecting coating damages on steel components using hyperspectral imaging. Two different samples of industrial components, provided by ENGIE Laborelec, were scanned with a hyperspectral camera: a part of a flame tube of a heavy-duty gas turbine as well as an artificially corroded sample of coated steel. The images were collected with a snapshot hyperspectral camera in the visible and Near Infrared range (manufactured by IMEC) and analyzed using a k-means clustering method. The outputs showed that despite the visual similarities between coating layers, corroded, and base material, hyperspectral imaging was capable to distinguish between them. Spectral plots showed that differences in spectral behavior of corroded and coated materials were more apparent when the wavelength shifts from the visible towards the near-infrared range. While further investigation is needed, the current results demonstrated that hyperspectral image analysis is a promising technology to complement or even replace digital image analysis within the visible range for characterizing steel surfaces in situ.

Research area: