AbstractOff-line Galvanneal steel classification using chemical and mechanical processes, has matured to a level where they are now able to confidently assess the state of iron/zinc alloying within the galvanneal coating. However, the assessments require that the samples be extracted and analysed off-line. This restrictive requirement means that the techniques tend to be destructive in nature, leading to unnecessary wastage. Moreover, they are unable to provide immediate line control feedback to ensure that the alloying condition is consistently at the optimum level.
Application of image processing methods to the analysis of the steel surface morphology presents the ability to classify coating quality based on extracted texture information. Using images extracted from actual galvanneal steel coils, the primary objective is to investigate the link between the annealing phase of the steel surface and the digital optical image equivalent. This investigation offers the potential of developing and implementing an on-line real-time annealing phase classification system that is less intrusive and time consuming than the existing laboratory analysis techniques.
Known image processing procedures have been applied to off-line galvanneal steel image samples to determine the degree of correlation between the captured image texture and the level of surface alloying. The procedures include wavelet compression, first and second order statistical pattern recognition techniques, genetic algorithm and K-nearest neighbour classifiers (Knn) and the rank-conditioned (R-C) morphological transform. Two novel techniques have emerged from the investigative research, a grey level co occurrence image filtration procedure and a morphological template optimisation scheme. The combination of the R-C transform using spiral templates with the image filtration system produced classification rates of 100% and 80% for two sets of galvanneal data, despite the latter being affected by high levels of noise. The second order statistical approach with the Knn classifier produced results of 93% and 62% respectively.
Further galvanneal test images should increase the confidence in the applied morphological techniques prior to their application as an on-line classification system.
|Date of Award||Aug 2000|