Thermoelectric waste heat recovery in copper smelting: where the physics meets the economics
Copper smelters operate at 1200 to 1300 degrees Celsius. The matte conversion, slag tapping, and off-gas handling stages all reject enormous quantities of thermal energy. Exhaust gases leave the stack at 200 to 400 degrees Celsius. Cooling water circuits carry away heat at 80 to 150 degrees Celsius. Slag radiates at 600 degrees Celsius or more before it is quenched. Most of this energy is simply discarded.
Thermoelectric generators offer a way to recover a fraction of it. A thermoelectric module converts a temperature difference directly into electrical current via the Seebeck effect. No turbine, no working fluid, no moving parts. A solid-state device sits between a hot surface and a cold sink, and electricity comes out the other end.
The figure of merit
The efficiency of a thermoelectric material is governed by the dimensionless figure of merit, ZT. The formula is straightforward: ZT equals the square of the Seebeck coefficient multiplied by electrical conductivity and absolute temperature, divided by thermal conductivity. In symbols: ZT = S squared times sigma times T, divided by kappa.
Materials with ZT above 1 are considered efficient. The best commercially available materials (bismuth telluride at low temperatures, lead telluride at mid-range) achieve ZT values between 1.0 and 1.5 in narrow temperature windows.
The computational challenge is that the numerator and denominator are coupled. Electrical conductivity and the electronic component of thermal conductivity both depend on carrier concentration. Increasing one typically increases the other. The Seebeck coefficient moves in the opposite direction. Optimizing all three simultaneously requires precise control of the electronic band structure and phonon spectrum, which is where computational physics becomes essential.
Computational screening
Brute-force experimental synthesis and characterization of thermoelectric candidates is slow. Growing a single crystal, measuring its transport properties across a temperature range, and iterating on composition can take months per candidate. Density functional theory and Boltzmann transport calculations can evaluate thousands of candidate compositions computationally before a single sample is synthesized.
The workflow starts with DFT to compute the electronic band structure and phonon dispersion of a candidate material. From these, the BoltzTraP or EPA methods calculate the Seebeck coefficient, electrical conductivity, and electronic thermal conductivity as functions of temperature and carrier concentration. Phonon calculations give the lattice thermal conductivity. The result is a predicted ZT curve across the full operating temperature range.
The right materials for smelter waste heat
For copper smelter applications, the relevant temperature range is 200 to 500 degrees Celsius. This range favors three material families: skutterudites (CoSb3-based), half-Heusler alloys (TiNiSn-based), and higher manganese silicides (MnSi-based). Each presents different computational challenges. Skutterudites have complex cage structures where filler atoms scatter phonons. Half-Heuslers have high power factors but also high lattice thermal conductivity. Silicides are earth-abundant but structurally complex.
The economic threshold
At current electricity prices in Zambia and the broader Southern African Power Pool, thermoelectric waste heat recovery becomes economically viable when module conversion efficiency exceeds roughly 5 to 8 percent for heat sources above 300 degrees Celsius. Current commercial modules based on bismuth telluride achieve 4 to 6 percent, but they are optimized for lower temperatures. Materials computationally optimized for the 300 to 500 degree range can push conversion efficiency toward 8 to 10 percent, crossing the viability threshold.
This is the kind of problem where computational physics pays for itself. One validated material prediction that achieves ZT above 1.2 in the target temperature range is worth more than years of trial-and-error synthesis. The DFT calculation takes days. The experimental validation of a single promising candidate takes weeks. Without computational screening, finding that candidate could take years.