Russian team used fuzzy clustering to sort seven high-temperature slags into three groups, finding steelmaking slag resource-valuable but moderately hazardous while copper and incinerator slags are high-hazard

The study proposes and tests a two-criteria quantitative method that separates resource indicators from environmental indicators and then applies fuzzy clustering to rank industrial slags, yielding three treatment groups: resource-inert (granulated blast-furnace slag, lump blast-furnace slag, ferrochrome production slag), resource-valuable with moderate environmental impact (steelmaking and electric steelmaking slags), and environmentally hazardous requiring deep processing (copper-bearing and incinerator slags). Existing classifications of industrial mineral facilities typically use a traditional hierarchical approach and usually describe either resource potential or environmental hazard; this work brings both independent indicator sets onto the same samples and ranks them by clustering. Based on measured chemical composition (Table 1) and physical properties (Table 2) of seven slag types, clustered with Statistica using c=3, m=2, ε=0.05, with the three-cluster model supported by a silhouette index S>0.5.

The cluster structures obtained from resource indicators and from environmental indicators do not coincide; for example, steelmaking slags fall in cluster 3 by resource indicators but cluster 2 by environmental indicators, showing that high resource value does not guarantee environmental safety, and vice versa. A traditional literature review usually discusses resource potential or environmental hazard separately and cannot reveal this mismatch for the same samples. The mismatch follows directly from two clusterings of the same dataset, and the authors present it as the main scientific result; the sample is seven slag types and the authors describe the results as illustrative.

Correlation analysis shows that most indicators have low, absent, or random correlation, with only a few pairs statistically significant: basicity modulus with metallic iron (r=0.88, p=0.009), Cu with FeO (r=0.78, p=0.039), crystallinity with the 5-70 mm grain content (r=0.88, p=0.009), Ni with Zn (r=0.76, p=0.047), and among environmental indicators Cu with Pb (r=0.94, p=0.001). The authors use this to show that over 85% of variable pairs have r<0.7, allowing the Euclidean metric to be used without strong multicollinearity distortion. Based on heat maps and correlation coefficients with p-values for n=7 samples; the authors explicitly state that with the limited sample size these interpretations should be regarded as preliminary.

The study gives treatment directions for each slag type: granulated blast-furnace slag, lump blast-furnace slag, and ferrochrome production slag can be used in construction materials without preliminary dressing or after sorting; steelmaking and electric steelmaking slags contain 13-21% metallic iron and can yield metal concentrate for recycling, with tailings usable in road construction and geotechnical barriers; copper-bearing and incinerator slags require deep metallurgical processing with non-ferrous metal recovery, and their tailings need stabilization and detoxification. It maps the cluster groups directly onto actionable treatment routes and notes that conventional uses such as making abrasives from copper-containing slags may hide environmental risks. The recommendations rest on the measured compositions and properties in Tables 1 and 2 and on the cluster grouping; the authors stress that products must still meet environmental safety requirements and require repeat analysis.

The results apply to the seven types of large-tonnage high-temperature process waste currently present in the Ural industrial region, for assigning slags to different treatment routes by resource value and environmental hazard, and they serve as a methodological starting point for a comprehensive resource and environmental assessment methodology for industrial mineral facilities and for digital waste management systems; the authors position the approach as a pilot methodological tool suitable under limited source data.

The sample is only seven slag types, and the authors state the cluster structures are illustrative and require confirmation on at least 20-30 types; the correlation interpretations are described by the authors as preliminary, and the physical meaning of the correlations awaits a larger database; the authors propose building an open database of slag physicochemical properties and using machine learning methods, which this paper does not yet complete.