MateriaLabMaterials R&D, permanent magnets

A magnet for 200 °C, without dysprosium

An industrial R&D centre needs permanent-magnet candidates for 200 °C service that do not depend on heavy rare earths, and has forty experiments of budget. This case follows the twelve-week pilot from the target to the first two lab batches.

See the product page

Organisation: An industrial R&D centre with its own synthesis lab
Decision: Which ten experiments to run first
Duration: Twelve weeks, two lab batches
Runs in: A private tenant, keys in the centre's own KMS

> The starting point

Millions Computed,Forty To Spend

The target is clear: a magnet that holds its performance at 200 °C without dysprosium or terbium. Open databases list thousands of candidate compositions, and the budget covers forty experiments.

Today the list is built by hand from literature and experience. Stability is screened at absolute zero, synthesis routes are guessed, and the centre's own measurements, which part from public calculations in known ways, never reach a model.

Who: Two senior researchers
On what: Literature, open databases, experience
How long: Several weeks per campaign plan
Supply risk: Checked late, if at all

# What could not move

What The LabWould Not Give Up

Agreed in the discovery workshop. Each one is a setting in the platform, not a promise in a slide.

Measurements stay inside
Past measurements are the centre's most valuable data and may not leave its infrastructure.
So the system runs in a private tenant, encrypted with the centre's own keys.
Service at 200 °C, not at 0 K
A candidate that is stable only at absolute zero is not a candidate.
So the system screens stability at the operating temperature, with a range on every prediction.
Makeable in the lab it has
There is no robotic lab, and the furnaces and routes are fixed.
So the system proposes routes from synthesis records of similar compounds, matched to that equipment.
Supply that can be sustained
Heavy rare earths are the dependency the project exists to remove.
So the system weighs supply risk into the ranking, not after it.

% The architecture

From Candidate SpaceTo A Ranked Batch

Everything before the prescription builds the plan; everything after it tests the plan and corrects the models, so each batch is aimed better than the last.

Bind
- Open calculation records, one ID per material
- The centre's past measurements, through LIMS or CSV
Check
- Stability at 200 °C, with p05–p95 ranges
- Synthesis routes from records of similar compounds
- Supply risk of every element
Prescribe
- Ten experiments, solved against budget and risk
- The reasoning behind each candidate
Learn
- Lab results returned after each batch
- Models recalibrated before the next

Runs inA private tenant reserved for the centre, encrypted with its own keys.

= The first twelve weeks

A Pilot OnYour Own Target

MateriaLab is best judged against a real R&D target of yours. By week twelve you hold the results of the first lab batches and a measured decision quality.

Discovery workshopWeek 0You leave with: Pilot goal and success criteria
Your team
R&D target, constraints, existing data
Our team
Use-case design
SetupWeeks 1–2You leave with: A working environment
Your team
Infrastructure access, cloud or on-premise
Our team
Deployment, identity integration, resource profile
Data bindingWeeks 3–4You leave with: First predictions calibrated to you
Your team
Past measurements and experiment records
Our team
LIMS or CSV binding, first calibration model
First portfolioWeeks 5–6You leave with: Batch 1 prescription
Your team
Budget, capacity, risk appetite
Our team
Candidate pool, portfolio solve, rationale report
Lab batchesWeeks 7–10You leave with: Batch 2 prescription, updated models
Your team
One or two batches run, results entered
Our team
Re-optimisation after feedback
ReviewWeeks 11–12You leave with: Pilot report and a scale-up decision
Your team
Internal review of the results
Our team
Comparative analysis, roadmap proposal

Success criteria, set together

Share of portfolio recommendations the lab accepts
Experiments needed for the first successful synthesis, against your past campaigns
Error reduction of calibrated predictions against your own measurements
Gap between predicted and realised successes

= The decision, as delivered

The PrescriptionFor Batch One

Week six: ten experiments, ranked, with supply risk weighed in.

Portfolio prescription

Week 6

Target: Permanent-magnet candidates for 200 °C service, free of heavy rare earths (Dy, Tb)
Budget: 40 experiments in 4 batches
Resource profile: Light rare earths and boron, domestic and abundant

Batch 1 candidates, ranked
Rank	Candidate	Proposed route	Success probabilityMean, with the p05–p95 range	Supply risk	Risk flag
1	(Nd,Ce)₂Fe₁₄B, 30% Ce	Arc melting → 1050 °C homogenisation → rapid solidification	0.710.58–0.82	Low	Anisotropy falls as the Ce share rises
2	(Nd,La,Ce)₂Fe₁₄B	Arc melting → strip casting → 650 °C anneal	0.540.40–0.67	Low	Possible secondary phase (CeFe₂)
3	CeFe₁₁Ti	Arc melting → 1000 °C anneal, 72 h under Ar	0.470.31–0.62	Very low	Phase-separation risk at high temperature
4	τ-MnAl(C)	Induction melting → 1100 °C quench → 500 °C phase transformation	0.420.28–0.57	Very low	High chance of a disordered phase
5	Fe₂B-based doped phase	Solid-state reaction, 900 °C, Ar	0.290.17–0.43	Very low	Low binding confidence; new route

10experiments in batch 1
4.1expected successes
96%chance of at least one success
70%lower supply risk than the heavy-rare-earth reference

Why candidate 1

Synthesis records for 23 similar compounds show a high success rate for arc melting plus homogenisation. Stability probability at 200 °C is 0.93. Ce substitution cuts supply risk markedly, and magnetisation calibrated to institutional measurements sits 8% above target.

In the product, every line opens onto the literature records, calculations and, where they exist, institutional measurements it rests on.

~ What gets measured

Against The Centre'sPast Campaigns

The baselines are the centre's own, so the comparison is with how it already works. Results are published with the centre's approval.

Measure	Baseline	How it is measured	Result
Experiments to the first successful synthesis	Past campaigns on similar targets	Count per campaign, with the same definition of success	Published on approval
Prediction error after calibration	Uncalibrated predictions against the centre's data	Error on measurements held out of calibration	Published on approval
Prescriptions the lab runs	A new measure, with no prior figure	Share of prescribed experiments actually run	Published on approval
Predicted against realised successes	Expected successes in the batch summary	Realised successes, per batch	Published on approval

What stays with the centre

A calibration model fitted to its own measurements
Two batch prescriptions, each line linked to its sources
Every result stamped with data, model and configuration versions

Close to yours if

You have a target and a fixed experiment budget
Your measurements part from public calculations in known ways
Critical-element supply is part of the requirement
Your data cannot go to a shared cloud

Your operation, in the same format

The technical briefing is where your specific constraints get discussed, including which of these four your operation resembles. Results are published with each customer's approval, never before.

Book a technical briefing

Millions Computed,Forty To Spend

What The LabWould Not Give Up

Measurements stay inside

Service at 200 °C, not at 0 K

Makeable in the lab it has

Supply that can be sustained

From Candidate SpaceTo A Ranked Batch

Bind

Check

Prescribe

Learn