Foundry · Induction melting
Your ERP knows how many heats ran and how much metal came out. It does not know that heat 14 held forty minutes on a crane, that heat 22 went back for chemistry, or that the furnace has been drawing more for the same tonnage all week. Munshi joins the furnace’s own readings to the operator’s heat log, heat by heat.
Your ERP logs the heat: tonnage out, kilowatt-hours in.
It won't tell you the furnace sat holding while the crane finished another bay.
Why the gap existsWhere the losses hide
None of these stop the furnace. They show up as a slightly longer cycle, a slightly higher draw, an extra charge — and they average away by the time the month closes.
Lining wear, charge mix, cold or damp scrap and power factor all move the energy a heat takes. The tonnage still books; the kilowatt-hours behind it quietly climb.
Metal at temperature with nowhere to go still draws power. Ladle not ready, crane committed elsewhere, moulding line behind — the furnace waits, and the wait is invisible in the output figure.
A heat that comes back off-spec is charged, melted and heated twice. In the ERP it is one heat with a correction. In energy terms it is close to two.
For scale: a US Department of Energy study put best-practice batch induction melting at 530 kWh per ton of melt, with typical current-technology figures around 500 kWh/ton — down from roughly 800 kWh/ton for 1950s equipment.1 Those are 2004 figures and an order-of-magnitude anchor, not a benchmark to grade your plant against. The point is the spread: the difference between a good heat and an ordinary one is real money, and it is not in the output number.
The join
Same heat, same clock, same asset — three streams that only mean something together.
Heat number, charge weight, metal out, energy booked, the grade produced. Accurate, and silent on cause.
Furnace power and coil temperature sampled continuously; the operator’s heat log with delay and rework reasons tagged at the furnace; spectro results tied to the heat they came from.
Energy per tonne per heat rather than per month, with the delay or rework that explains an outlier already attached to it.
What Munshi surfaces
Teal is normal, maroon is flagged. Illustrative data, ranged against real furnace readings.
Two heats sit well above the run. Both carry an operator tag — one a crane wait, one a re-charge after chemistry.
The shaded stretch is power held with no tap. The furnace is working; the foundry is not.
Tagged by the operator at the furnace, totalled across the month. Not inferred.
When an outcome moves and a signal moves with it, Munshi shows both and says they moved together. It does not tell you one caused the other. The operator confirms the cause, and the confirmed cause is written back to that furnace’s register — so the pattern is already named the next time it appears.
See this on your own furnace.
Bring a month you already understand. The useful test is whether Munshi finds what you already know — and then what you didn’t.