Why water changes cooling speed
Nitrate salt baths operate between roughly 150 and 400 °C. Pure molten salt has limited cooling capability in that range, dominated by conduction and convection. A small amount of water vaporises violently at the workpiece surface, collapsing the film-boiling stage and sharply raising the cooling rate in the high-temperature region.
That is the principle of water-added quenching: what is added is not a coolant, it is the high-temperature leg of the cooling curve.
Too little and too much
- Too little water — insufficient high-temperature cooling, incomplete bainite/martensite transformation, low hardness and excess troostite.
- Too much water — violent cooling, thermal and transformation stresses rising together, higher risk of distortion and cracking, plus splashing and fumes.
That is why water content is held in a narrow window, commonly 0.3–1.5 %, depending on material, section size and loading pattern.
It keeps escaping
Water cannot stay in salt above 150 °C. Evaporation, reaction with workpieces and carry-over with fumes all remove it continuously. Larger workpieces, higher bath temperature and stronger agitation all accelerate the loss.
Water content is therefore not a parameter you set once; it is a dynamic quantity that must be continuously replenished.
Why the traditional method lags
Gravimetric testing — sample, oven-dry, weigh, calculate — takes 1.3 to 4 hours. By the time the result exists, the bath has already changed. The number describes the salt of several hours ago, not the salt now. Deciding how much water to add from that number is still guessing.
From guessing to measuring
Salt density depends on both temperature and water content, and both act in the same direction: higher temperature lowers density, and so does more water. A buoyancy signal alone cannot tell “more water” from “hotter salt”.
Collecting buoyancy and temperature together solves it: look up the baseline force expected at the measured temperature, subtract it from the measured force, and what remains is the water-driven component. That is two-parameter differencing.
On top of that, segmented linear correction (ten segments for temperature, ten for indication), on-site zero and span calibration bring full-range error into a reproducible band.
After closing the loop
Once water content is displayed in real time, control becomes straightforward: compare with the setpoint, drive a make-up pump or solenoid with PID, add water only when needed, then keep watching.
The real value is not the automation but the traceability: every make-up action and every segment of the curve is recorded, so a quality dispute can be reviewed against the actual medium condition instead of a recollection.