Modelling of a corrosion detection and monitoring platform using Machine Learning
Inspection of corrosion has been a bottleneck process in many industries, especially in the marine industry, due to the sheer size of the structure that has to be inspected. The current corrosion detection methods are labour-intensive and only cover a small area.
The proposed approach uses a combination of weak classifiers such as machine learning and image processing techniques (GLCM, colour thresholding, quantization) to attain a robust global performance. This helps accelerate the corrosion detection process and provides overall information, for example, percentage, location and the severity of corrosion on the surface. For this study, MATLAB is used to do all the machine learning and image processing. Together, the overall process is named the “Corrosion detection and analysis software (CDAS).” While tested, the developed Machine Learning and GLCM platforms showed 90% and 80% accuracy, respectively. However, the overall accuracy of the developed CDAS is much better much compared to those individual processes. The user is able to capture the test subject using any camera-equipped personal communication device and upload it to the software. Thus, providing a quick and accurate method for the users to analyze the images.
However, the developed CDAS is still in its infant stage, where some of the steps need to be done manually on MATLAB. The flow of the processes could be automated, where users can upload a large number of images, and the software would be able to proceed according to the conditional path automatically. This could generate a more accurate testing result of CDAS as well. An easy-to-use user interface in a mobile phone application could be done for ease of use for the users.