EAF Scrap Mix Optimization with AI for Steel Melters Guide

By James Smith on October 12, 2026

eaf-scrap-mix-optimization-with-ai-for-steel-melters-guide

Every melter knows the feeling of a charge that looked fine in the yard and behaved badly in the furnace. The bucket was too light and needed an extra charge, the copper crept above the grade limit, or the cheapest scrap on the plan pushed the power-on time up. Scrap is the largest cost in an electric arc furnace shop, yet the mix is often decided by habit, a static recipe and whatever is in the pile that morning. This guide explains how AI balances density, chemistry and cost together to lower kWh per tonne while still hitting the steel grade, and if you want to see it on your own heats, you can review your scrap mix strategy with our team.

EAF energy optimization

EAF Scrap Mix Optimization with AI: A Guide for Steel Melters

How dynamic scrap blending balances bulk density, residual chemistry and cost so every charge melts efficiently and lands on grade.

A charge bucket, built from the bottom up
Top: light scrap such as shred, to cushion the arc
Middle: dense heavy melt and bundles, for bulk
Base: light layer to protect the hearth
The core idea

Three Inputs Pulling in Three Directions

A good scrap mix is never one perfect recipe. It is a balance, and each input pushes against the others.

Density
Dense scrap packs more steel per bucket and cuts the number of charges.
Pulls toward fewer buckets
Chemistry
Residuals like copper and tin must stay under the limit for the target grade.
Pulls toward cleaner scrap
Cost
Cheaper grades often carry more residuals and lower yield.
Pulls toward lowest price

Humans balance this by experience. AI balances it by calculation, for every heat, using what is actually in the yard today.

Know your materials

Scrap Types Compared: Density, Residual Risk and Melting Behavior

The table shows indicative planning values. Real figures depend on your supplier, processing and yard handling.

Scrap typeBulk densityResidual riskMelting behavior
ShreddedMediumMedium to high copperFast melt, good for cushioning
Heavy meltingMedium to highMediumSteady, reliable bulk
BundlesHighMedium to highDense, slower to start
BushelingLow to mediumLowClean, fast, premium priced
Home scrapVariesKnown and lowPredictable, limited supply
Pig iron and DRIHighVery lowDilutes residuals, costs more
Chemistry watchlist

Residual Elements: How Close Is Each One to the Limit

This illustrative chart shows how much of the grade limit a loose mix might use up. A bar near full means little room for error.

Copper

90%
Tin

70%
Nickel

55%
Chromium

40%
Molybdenum

25%

Copper is the usual constraint because it cannot be removed in the furnace, only diluted. That is why the mix has to be right before the charge, not after tap.

Static versus dynamic

Why a Fixed Recipe Falls Behind a Dynamic Mix

A static recipe assumes every pile and every price is the same as last month. A dynamic mix does not.

Static recipe
Same split every heat
Copper checked after the melt
Price changes ignored for weeks
Bucket loading by feel
Dynamic AI mix
Mix rebuilt for each heat
Chemistry predicted before charging
Live price and stock inputs
Bucket order and weight suggested
Test it against your own heats

See What a Smarter Scrap Mix Could Do on Your Furnace

Share your grades, scrap sources and current kWh per tonne, and we will walk through how a dynamic mix would be set up.

The engine

How the AI Builds a Mix for Each Heat

The method follows the same path every time, so melters can see why a recommendation was made.

1
Read the grade
Target chemistry and tap specs for the next heat.
2
Read the yard
Available scrap, assayed residuals and bunker levels.
3
Solve the balance
Minimize cost and energy inside chemistry and density limits.
4
Plan the buckets
Split the mix across charges and set layer order.
5
Learn from the heat
Compare predicted and actual results to sharpen the next one.
What it can change

Where the kWh per Tonne Savings Come From

Energy savings do not come from one trick. They come from small gains that stack, shown here as an illustrative split of a typical improvement.

Fewer charges
Better layering
Less rework
Tighter chemistry
Fewer charges
Denser, better-packed buckets cut the door-open and power-off time between charges.
Better layering
Light cushion and dense core improve arc stability and melting speed.
Less rework
Hitting chemistry first time avoids extra refining and reheating.
Tighter chemistry
Less alloy over-correction and fewer off-grade heats.

Results vary by shop, furnace and scrap supply, so treat these as areas to measure rather than promised numbers. You can ask support how savings are tracked heat by heat.

What the AI needs

Data Inputs That Make the Recommendation Trustworthy

Scrap assays
Residual levels per supplier and lot.
Yard stock
Tonnage by type, updated as it moves.
Price list
Live cost per tonne by grade.
Power profile
kWh, tap-to-tap time and electrode use.
Heat results
Final chemistry and yield per heat.
Bucket records
Actual loading order and weights.

Most shops already hold this data in separate places. The value comes from bringing it into one model that the melter can trust.

Avoid these

Common Scrap Mix Mistakes That Waste Energy

Chasing the lowest price
Cheap scrap with high copper can cost more in dilution and off-grade heats.
Ignoring bulk density
Light buckets add charges and open the furnace more often.
Trusting old assays
Supplier quality drifts, so stale data hides risk.
Overriding without notes
Manual changes that are not recorded cannot teach the model.
Frequently asked questions

What Melters Ask Before Trusting an AI Scrap Mix

Does the AI replace the melter's judgment?
No. It proposes a mix and shows the reasoning, while the melter keeps the final call and can override it. Those overrides become useful feedback. See how recommendations are presented in a live demo.
How accurate is chemistry prediction before charging?
Accuracy depends on the quality of scrap assays and how well heat results are fed back. It usually improves as more heats are recorded. Ask support about data requirements for your shop.
Can it work with several steel grades in one week?
Yes, because the mix is rebuilt for each heat against that heat's target grade and the scrap actually available. Grade changes are handled as inputs. Test your grade schedule with our specialists.
Do we need new sensors on the furnace?
Often not. Many shops already log power, tap times and chemistry, and the platform can start from that data and add sensors only where gaps matter. Check what your plant already provides with the support team.
How do we measure whether the new mix is working?
Compare kWh per tonne, tap-to-tap time, number of charges and off-grade heats against a baseline period on similar grades. Set up a fair before and after comparison on a call.
Melt smarter, not just harder

Bring Dynamic Scrap Mix Optimization to Your Melt Shop

Book a session with iFactory AI to see how density, chemistry and cost can be balanced on every heat to lower kWh per tonne and hit grade.


Share This Story, Choose Your Platform!