Sample Report — 42CrMo4 / AISI 4140 Quenched & Tempered

937 MPa

Yield

1043 MPa

Tensile

314 HV

Hardness

14.5 %

Elongation

±1.9 %

Model spread, yield

Download the full 20-page PDF 3.5 MB · no email required

The inputs, in full. Fe 97.28, Cr 1.05, Mn 0.80, C 0.42, Si 0.25, Mo 0.20. Hot forged at 870 °C, strain rate 2.88 s⁻¹, reduction 5.689:1. Austenitised 850 °C, oil quenched, tempered 570 °C for 2 h. Nothing else — the platform inferred the rest and says so where it did.

Where the 937 MPa comes from

The number is not a black-box output. The report itemises every strengthening mechanism, and shows how much of the answer is physics and how much is the machine learning correcting it.

Mechanism	MPa	share
Martensite (σ_MART)	385.9
Grain boundary (σ_GB)	185.3
Dislocation (σ_DISL)	100.0
Lattice friction (σ₀)	50.0
Solid solution (σ_SS)	48.6
Precipitation (σ_PPT)	32.4
Total, physics alone	802.2	86 % of the answer
ML correction	+135.0
Predicted yield	937.2

This ratio is the health check. Physics carries 86 % and the model corrects the last 14 %. When that gap opens up, the platform notices: on our other published case the model sat 75 % above its own physics, and the report dropped to screening only with a ±23 % spread instead of ±1.9 %. Same engine, same page, two very different verdicts — and it tells you which one you are looking at.

What it will and will not let you do

Every report gates itself against four levels of use. Both blocked levels name the specific condition that is blocking them — there is no way to read this page and think you have a certification-grade number.

✓ Screening. Feasibility OK. Acceptable for concept exploration and “is this direction plausible” triage.
✓ Process comparison. Strengthening quality is NORMAL; model uncertainty ≈ ±1.9 % on yield — self-reported spread, not measured error. Suitable for ranking alternative routes.
× Pre-production. Blocked by: feasibility status is WARNING (needs VALID); strengthening quality is NORMAL (needs HIGH).
× Certification. Blocked by: pre-production criteria not met; fatigue uncertainty ±0.34 decades > the ±0.2 threshold.

Note what is doing the blocking. Not a blanket disclaimer — a named, numeric condition in each case. ±0.34 decades against a ±0.2 threshold is a statement you can argue with. “For information only” is not.

Three things it flagged without being asked

1. It criticised the forging temperature

“Temperature 870 °C is below typical Fe forging window (900–1250 °C). Flow stresses will be high; cracking risk. → Consider raising to ≥900 °C for cleaner deformation.”

Nobody asked it about forgeability. It read the process temperature against the window for the alloy class and objected — and that objection is what holds the feasibility layer at WARNING, which is in turn what blocks pre-production above.

2. It computed weldability off the same composition

“Carbon equivalent CE(IIW) = 0.80 — hydrogen-induced HAZ cold cracking becomes likely above 0.45 without precautions. → Preheat ≈ 250 °C (section-dependent), low-hydrogen consumables, and slow cooling; qualify the procedure per ISO 15614.”

That is correct for 4140 and it is the kind of thing that is easy to discover too late. The same arithmetic is free and ungated on our carbon equivalent calculator.

3. It refused to call a low load “infinite life”

“The applied amplitude is 0.31 of the engine's textbook endurance-limit estimate, where classical S–N would call the loading run-out. In the Fe training corpus, of the 315 tests loaded below this estimate, 280 FAILED, at a median of 10^6.52 cycles, and only 36 % reached 10⁷ — so the estimate is not a limit the data respects.”

This is the passage we would point at first. The textbook answer here is “run-out, design freely”. The platform checked that against its own corpus, found the textbook wrong 89 % of the time, and said so — then explained that the finite life it shows comes from the ML residual fitted on exactly those rows, and should be treated as less certain than a life above the limit.

A real report, start to finish