Eleven Forty-Seven When the Wire Disappeared

The twisted bar from Monday’s forge session sat on my bench next to a coffee can full of photodiodes and an old bench power supply. Cold fracture, crystalline structure visible in the break. The instructor said it snapped because I’d let the temperature drop below working range—maybe 850°C, still glowing dull red, but too cold for the steel to flow plastically.

The question I kept coming back to: what temperature was it? “Dull red” is a description, not a measurement. Different lighting conditions change what you see. Colour vision varies between people. Scale buildup darkens the surface. Reading forge heat by eye is a trained skill that accumulates error every single time.

Optical pyrometry measures temperature by comparing the colour of thermal radiation to a known reference. The classic instrument—invented in 1901 by Holborn and Kurlbaum in Germany—uses a heated tungsten filament placed in your line of sight to the hot object. Adjust the filament current until it disappears against the background glow. At that moment, both filament and workpiece emit identical blackbody spectra, which means they’re at the same temperature. Read the temperature from the filament current dial.

Ordered a used Leeds & Northrup disappearing-filament pyrometer from an estate auction. Arrived yesterday: brass telescope body, red filter eyepiece, current adjustment knob calibrated 600–2500°C. Weight: 1.8 kilograms. Built in 1964 for foundry use, still has the original calibration certificate from a steel mill in Pennsylvania.

Calibration Against Nothing

First problem: I don’t have a temperature standard. The pyrometer’s calibration is sixty-two years old and may have drifted. To check it, I’d need a reference blackbody source at known temperature—essentially a cavity radiator with emissivity approaching 1.0 at a verified temperature. Industrial labs use graphite tube furnaces. I have a propane torch and a jar of lamp black.

Made a makeshift cavity from a chunk of firebrick left over from the forge. Drilled a 12 mm hole 8 cm deep, coated the interior with carbon soot by burning acetylene inside it with insufficient oxygen. The soot layer raises effective emissivity to roughly 0.95, turning the cavity into an approximate blackbody. Not NIST-traceable, but better than bare metal.

Heated the brick with a MAP-Pro torch until the cavity glowed orange. Pointed the pyrometer at the hole, adjusted the filament current until the wire vanished against the background glow. Temperature reading: 1040°C.

Checked it with a type-K thermocouple (bare bead, 1.5 mm diameter, rated to 1370°C). Inserted the bead into the cavity. Thermocouple read 987°C.

53°C error. Either the pyrometer calibration drifted low, the thermocouple is reading cold due to conduction losses through the lead wires, or the cavity’s emissivity isn’t as close to 1.0 as I assumed. Probably all three.

The pyrometer manual addresses this. Page 14: “Brightness temperature measured assumes emissivity ε = 1.0. For non-blackbody targets, true temperature T = T_brightness / ε^0.25.” Real metals have emissivity between 0.3 (polished steel) and 0.8 (heavily oxidized). Same temperature, vastly different readings.

Disappearing Act

Tried it on the forge Wednesday evening. Heated a fresh bar of mild steel to yellow-orange—what I’d previously call “forging heat” based purely on colour. Sighted through the pyrometer. The glowing bar appeared as a bright field behind a dark filament wire.

Turned the current knob slowly. The filament began to glow red, then orange. At 1147°C on the dial, the filament brightness matched the bar and it seemed to disappear—same effect as placing a grey card against a grey wall.

Pulled the bar, made four hammer blows. Sound was bright and clear, metal moved easily. Checked temperature again before the fifth blow: 924°C. Below the “disappearing” point now, but still visibly glowing orange.

One more heat cycle, this time pulling the bar at 980°C instead of waiting for yellow. It moved under the hammer but more sluggishly. The ring tone was duller. Checked again after three blows: 831°C, dull red, metal resisting deformation.

The acoustic feedback I’d been using to judge working temperature correlates with a 100°C transition zone. Above ~920°C: clear ring, plastic flow. Below ~830°C: dead thunk, work-hardening faster than I can move it. The pyrometer gives me the numbers behind the sound.

The Emissivity Trap

Second bar, deliberately polished with 80-grit sandpaper before heating. Bright steel, no scale. Heated to the same yellow-orange colour as the first bar. Pyrometer reading: 894°C.

Wait. Same colour, 250°C lower reading?

The pyrometer measures radiance, not temperature. A polished surface reflects more and emits less at the same temperature. The manual’s correction formula requires knowing emissivity in advance, but emissivity changes during heating as oxide scale forms. A clean forging reads differently than a scaled one. The measurement system couples to the very process it’s trying to measure.

This is why two-colour ratio pyrometers were developed in the 1930s. They measure intensity at two wavelengths—typically 650 nm red and 900 nm near-infrared—and divide one by the other. If emissivity is wavelength-independent (the “grey body” assumption), it cancels out in the ratio and you get true temperature regardless of surface condition.

Except metals aren’t grey bodies. Steel’s emissivity at 650 nm isn’t the same as at 900 nm, especially during oxidation. The error persists, just differently.

Dug through amateur radio parts bins and found three photodiodes: one silicon (peak response 900 nm), one with a red filter (effective 650 nm), one UV-enhanced (300–400 nm). Built a crude three-wavelength comparator on a breadboard—three amplifier channels feeding an Arduino for ratiometric calculation. Total cost in scavenged parts: maybe $12.

Pointed it at the forge. Ratios calculated from the three channels didn’t converge to a consistent temperature using Wien’s displacement law. The math assumes blackbody radiation; real steel at 1000°C emits a heavily distorted spectrum dominated by molecular emission bands rather than smooth Planck curves.

The DIY multi-wavelength pyrometer is theoretically sound and practically useless without either a spectral emissivity database for the exact material or a way to measure emissivity independently in real-time.

What Gets Measured

Went back to the single-wavelength pyrometer with a different strategy. Stopped trying to measure absolute temperature. Started using it to measure relative temperature consistency across multiple heats.

Forge heat varies with coal bed depth, air flow rate, and how long the bar’s been buried in the fire. Pull ten pieces “by eye” and you’ll get ten different working temperatures. But if I set the pyrometer target to 1120°C ±20°C and don’t pull until the filament disappears, every bar comes out of the forge in the same thermal state.

Repeatable heat cycles matter for tool steel. High-carbon blades require specific temperatures for austenitizing (above 800°C to dissolve carbides), quenching (rapid cool to form martensite), and tempering (reheat to 200–400°C to reduce brittleness). Miss the austenitizing temperature by 50°C and your blade won’t harden properly. Overshoot by 100°C and grain growth ruins the edge geometry.

The pyrometer doesn’t tell me the truth. It tells me the same calibrated lie every time, and repeatability is what matters.

Crystal radio work last March taught a similar lesson. Tuning a cat’s whisker detector involves finding the “hot spot” on a galena crystal where the metal-semiconductor junction rectifies RF signals most efficiently. You’re not measuring absolute rectification efficiency in amps per volt—you’re listening for maximum loudness in the headphones and marking that position. The measurement is relative, the feedback is audio, and precision comes from consistency rather than accuracy.

Forge temperature control works the same way. The pyrometer’s brightness reading depends on emissivity, surface condition, and oxide thickness. But if I always measure through the same red filter at the same wavelength looking at the same material, those errors become constant offsets. The drift terms cancel out across repeated measurements of the same process.

Not truth. Precision. Close enough to keep the steel from snapping cold.