For high quality steel products it is essential to have specific understanding of the underlying steel production process such as the electric slag remelting process (ESR). To assist the currently manual assessment, there is a high need for objective quality measures and standardized evaluation methods. Based on texture segmentation and ridge detection a computer-vision based automated evaluation of the pool profiles was achieved. The derived parameters yield valuable insight to and improve the overall steel production process.
The field of quality management and improvement in high quality steel production is one of the deciding reasons whether a steel producer remains competitive or not. In the production of high quality steel products it is essential to remelt conventional produced ingots. In order to yield specific understanding of the remelting process as well as to improve the process, there is a high need for an objective and comparable evaluation of remelted blocks.
In preparation for assessment, plates cut from a remelted block, are smoothed, polished and etched to reveal the inner crystalline solidification structure. Those structures provide information directly linked to the remelting parameters and as a consequence are essential for optimizing these parameters.
Significant parameters for the quality of steel can be derived from so-called pool profiles. With the aid of those pool profiles it is possible to determine certain quality attributes within the whole steel block. Therefore, the equality of the individual pool profile lines with their surroundings is taken into account. Figure 1 shows manually derived pool profiles of an example steel plate.
Fig. 1: Manually derived pool profiles.
Steel Specimen Segmentation
The first step of the automated quality assessment is the segmentation of globular and trans-crystalline solidification areas, where trans-crystalline solidification areas provide representative information and globular areas are basically unstructured and as a result do not provide meaningful information for the pool profiles.
The main idea for automated segmentation is based on the different textural appearance (regular and irregular patterns) of the different solidification regions. An evaluation of various algorithms for the description of the surfaces showed that texture analysis based on local binary patterns (LBP) performed best. LBP are used to describe the surrounding of a pixel. This is done by comparing a pixel to each of its neighbours (which are defined by radius and number of points on the consequential circle). Given eight neighbours, LBP result in an eight digit binary number where each digit gives information about whether the centre point value is greater or smaller than the neighbour. To retrieve information about an area, LBP for each pixel in that area are summed up in a histogram. Figure 2 shows a coloured image on the left hand side where each colour matches a bin (feature) from the outlined histogram and the final binary segmentation output on the right.
Fig. 2: LBP dominating feature/bin Output.
Pool profile derivation
The best performing method in pool profile derivation for different types of steel qualities is based on a combination of scale-space and ridge detection. The ridge detection is similar to a biometric fingerprint recognition approach with the difference that in this application regions with constant directions are important, whereas in fingerprint recognition characteristics like crossing points or ridge ends are relevant.
The pool profile itself comprises of trace lines derived from ridge orientations. The trace line is calculated from a given start point by calculating a normal on the underlying orientation to the consequential next one and so forth. The calculation begins either from the outer borders (left and right) to the middle or vice versa.
Figure 3 displays a whole steel block with final segmentation and automatically derived pool profiles.
Impact and effects
The automated quality assessment is currently under evaluation by metallurgists on additional steel blocks. First feedback indicates that the method for segmentation and pool profile generation is applicable for a wide range of steel products.
The objectively derived parameters enhance the overall quality assessment process by allowing for a larger number of steel blocks to be processed and wider knowledge on steel types to be gained. This valuable insight leads to improvements on the overall steel production process and yields significant advantage for the Austrian steel industry.