Refractory Raw Materials for Steel Casting Brick Products

Table 1: SCL raw materials, true-color RGB based on VNIR image. Not all grain-sizes for each material is represented here.

Figure 1: SCL raw material spectra in the VNIR (Left) and SWIR (Right). Spectra plotted in reflectance. X-axis: Wavelength in nanometers, from 940 to 2500m; Y-axis: Reflectance [%] scaled from 0-1.

Based on these raw materials a PLS-DA classification model was trained using Prediktera’s Breeze software. The VNIR model was trained based on 9 subsamples per brick. The balanced training dataset comprises 40 samples while testing was done on 59 samples. For the VNIR training dataset the overall class was predicted with 82.5%accuracy, whereas the testing dataset was predicted at an overall accuracy of 86.4%. In the VNIR, the most challenging samples for accurate prediction are MgO-C recycled, graphite, brown fused alumina and bauxite.

A PLS-DA classification model was trained on the raw material spectra both for the VNIR and SWIR separately. In the VNIR case, both the classification based on the pixel-class majority and the object average spectrum show acceptable results. Comparing the VNIR- and SWIR-based prediction, the VNIR case seems superior for predicting the recycled MgO-C and brown fused alumina classes. For both sensors, the classification of pure graphite powder does not seem possible. An excerpt of the VNIR prediction results of the different raw materials is shown in Table 2.

Table 2: VNIR-based prediction results for some of the raw materials for SCL bricks. From left to right: the sample name and grain size, where relevant; the true-color RGB, the pixel-based classification, the class predicted based on pixel-class majority (pixel count) and the class predicted based on the object’s average spectrum.

Based on these results, a classification model based on VNIR hyperspectral data for SCL bricks’ raw materials seems possible. The model could possible be improved further by adding more samples to cover the full range of variation expected in these input materials. Derived from this model, one of the used SCL bricks was classified showing an abundance of classified white fused alumina on the surface as well as identifying MgO-recycled and fused MgO in addition to graphite (Figure 2). Taking into account that graphite is prone to misclassification, this compares to the high value of Al2O3 found in the sample (92%) via X-Ray Fluorescence (XRF), along with the MgO content of around 5%.

Figure 2: Used brick type 5, alumina concrete, classified on a pixel-basis using the SCL raw materials classification model

This feasibility study is part of Horizon-Europe funded ReSource project, showing how hyperspectral imaging can help identify refractory bricks and classify them spatially. Though more research is needed to give conclusive results, the initial ability to map the three main CRK brick composition types using hyperspectral imaging is promising. It is also important to highlight, that hyperspectral imaging enables to map of small-scale changes in used bricks caused by mineralogical changes originating from changing temperature and pressure as well as infiltration of gases during the brick's service. This will change the surface and give indications of the brick's composition and re-use potential.