Can a thermal camera replace the thermocouples?
A directional-solidification cuvette is normally fused quartz with embedded thermocouples. Swap the wall for zinc selenide — transparent in the 8–14 µm longwave-IR band — and an Optris PI-640i can read the fluid profile through the sidewall, with no wires in the melt. We forked the validated cuvette model to ZnSe (solver unchanged) and asked what the 11×-higher-conductivity wall does to the field. It is thermally benign and slightly favourable: a roughly 6× flatter interface, read faithfully by the camera to within microns. The costs are process recalibrations, not failures — a 31% shallower gradient and a 44% longer solidification dwell.
| property | fused quartz | ZnSe | ratio |
|---|---|---|---|
| conductivity k (W/m·K) | 1.39 | 15 | 11× |
| density ρ (kg/m³) | 2650 | 5270 | 2.0× |
| specific heat cp (J/kg·K) | 730 | 339 | 0.46× |
| volumetric heat ρcp (J/m³·K) | 1.94×10&sup6; | 1.79×10&sup6; | 0.92× |
| diffusivity α (m²/s) | 7.2×10-7 | 8.4×10-6 | 11.7× |
| LWIR (8–14 µm) | opaque | transparent | — |
The two walls have nearly equal volumetric heat capacity (ZnSe 0.92×): its lower cp and higher density cancel. So the material difference is driven almost entirely by diffusivity (11.7×), which comes from k — not heat capacity.
Sweeping the wall from quartz (k = 1.39) to ZnSe (k = 15) and beyond shows an asymmetric trade. Interface curvature flattens fast and then saturates — about 75% of the achievable flattening is already captured by k ≈ 6. The axial-gradient penalty, by contrast, keeps accruing roughly linearly. ZnSe sits deep in the saturated-flatness region: it buys about 85% of the flattening while paying a 31% gradient penalty.
| metric | fused quartz | ZnSe | change |
|---|---|---|---|
| interface bow across the bore (converged) | ~23 µm | ~4 µm | ~6× flatter |
| axial gradient G at the front | 1.79 K/mm | 1.24 K/mm | −31% |
| front position (from cold end) | 23.62 mm | 24.21 mm | +0.59 mm hotward |
| local cooling rate at Tm (latent on) | 0.589 K/min | 0.502 K/min | −15% |
| mushy (solidification) dwell, V = 5 µm/s | 170 s | 245 s | +44% |
| mushy-zone width | 0.85 mm | 1.23 mm | +44% |
DMSO absorbs strongly in the 8–14 µm band, so the Optris does not path-integrate through the full 3 mm bore. It reads a thin near-wall skin (about 150 µm deep) — the fluid temperature immediately inside the wall. Because the cross-section is near-isothermal, that skin tracks the bulk: the camera-reported interface matches the true centre interface to within 2–9 µm. And ZnSe's fast wall holds no thermal history that would bias the reading — a whole-assembly quench settles in 552 s for ZnSe vs 472 s for quartz (both DMSO-limited; the thin wall is not the bottleneck). Contactless LWIR is a faithful method here.
The interface gets flatter
The high-k wall isothermalises the perimeter, so the front bows only ~4 µm across the bore versus ~23 µm in quartz — a cleaner, more planar interface. This is a win, not a cost.
Contactless reads true
The camera sees a near-wall skin that tracks the bulk to within microns, and the fast window stores no lag. The thermocouples can come out without losing the measurement they were there to make.
The costs are recalibrations
A 31% shallower gradient, a 0.6 mm hotward front shift, and a 44% longer mushy dwell all point one way: a lower-gradient, longer-solidification regime. Since G/V and mushy residence set lamellar spacing, check the new regime against the target microstructure before adopting.
This is a conduction and phase-change model of the thermal field — not the detector response. The near-wall-skin result assumes an in-band absorption length of 0.05 mm (an estimated parameter, no measured DMSO spectral coefficient in hand); the conclusion holds for any value up to about 0.3 mm, which covers strongly-absorbing organic liquids. The radiometric chain is not modelled: ZnSe's refractive index n ≈ 2.4 gives ~17%/surface Fresnel reflection, so a real Optris deployment needs AR-coated windows plus emissivity and reflection calibration. The wall is a continuous shell; the four ZnSe plate seams carry contact resistance the monolithic model omits, which would nudge corner behaviour back toward quartz. Steady curvature is mesh-converged; the transient uses a coarser 0.3 mm grid, adequate for the cooling-history comparison.
Method: validated NovaThermal core (cuvette_model_v3 + nt_solve), solver unchanged; steady field plus a moving-boundary backward-Euler apparent-cp PCM driver with latent heat active. Open-literature physics, non-ITAR.
See the freezing front sweep the sensors → Run your own case in the console →