The short version
This page tests your own residual-stress measurements against a shared model, entirely inside your browser. Nothing you load is uploaded. If you choose to contribute afterwards, only summary sums leave your machine — never a row of your data.
What happens when you use the public page
The server only ever hands out the current model and receives summary sums back — it never sees your file.
YOUR BROWSER THE SERVER
┌────────────────────────────┐ ┌──────────────────────────┐
│ your CSV file │ asks for the │ │
│ ↓ │ ←── model ──── │ global model, version N │
│ read it, match the columns │ │ │
│ ↓ │ │ │
│ fit + score, locally │ │ │
│ ↓ │ │ │
│ TWO SCORES on screen │ │ │
│ ↓ │ sends the │ │
│ [contribute] ─── sums ───→ │ ── sums ─────→ │ waiting list │
│ (no raw measurements) │ │ ↓ │
└────────────────────────────┘ │ reviewed by a human │
your measurements stay here │ ↓ │
│ merged → version N+1 │
└──────────────────────────┘
Four steps on the public page: pick your file, check the column matching, read your scores, and optionally contribute.
The two scores
They answer different questions, and the pair is what is interesting.
| What is computed | The question | |
|---|---|---|
| Transfer | the published model applied as is, no refitting | does our model work outside the lab it came from? |
| Refit | the model refitted on your data alone, scored leave-one-out | are your measurements predictable at all? |
A weak transfer with a strong refit means clean data but a model that does not travel — usually the most useful thing the page can tell you, since it points at a regime the original study did not cover.
What your file needs
A CSV with one row per measurement and these columns:
| Column | Meaning | Unit |
|---|---|---|
gamma | rake angle | degrees |
r_beta | cutting edge radius | µm |
Vc | cutting speed | m/min |
h | depth of cut | mm |
SRS_L | residual stress, longitudinal | MPa |
SRS_T | residual stress, transverse (optional) | MPa |
Header names do not have to match exactly — step 2 of the public page lets you map your columns by hand, and it copes with semicolons instead of commas, decimal commas, title rows above the header, and Greek letters in the names. Extra columns are simply ignored. Excel files are refused with a message explaining how to save as CSV; there is a ready-made template at template/titanium_template.csv.
Contributing to the shared model
What is sent
- What you type in the contribute form: your name, laboratory, material, measurement method.
- The number of rows and the range (min–max) of each cutting parameter.
- Summary sums over your data —
D'D,D'y,y'y,n— which let us refit the model on the pooled data exactly, without moving any measurement. - The coefficients fitted on your machine, and your two scores.
What is never sent
- Your measurement table. No row, no individual value, at any point.
One honest caveat: those summary sums are not anonymous. On very few rows they reveal a lot about the dataset — with a single row they would give it away entirely. This is why contributions need at least 5 rows, and why we say so here rather than claiming a guarantee we cannot make.
Sending a contribution requires an access key, given only to people allowed to contribute. Without one, the site still works for scoring your own data — only the contribute step is turned down.
How a new version is published
Contributions never enter the published model automatically. They land in a waiting list; an administrator reviews what is there — material, measurement method, number of rows — before folding any of it in. Updating the shared model automatically, with no one checking, would make it trivial for a single bad upload to quietly degrade the model for everyone.
Publishing recomputes the model from scratch on everyone's pooled data combined — it is not an average of each lab's own model. This gives exactly the same result as if every contributor's raw table had been pooled into one big table and fitted once, even though no one's table ever left their computer. An administrator can also roll back to an earlier version at any time; nothing in between is deleted, it simply publishes that earlier version again as the newest one.
Current limits
- Titanium only. The model covers rake angle, edge radius, cutting speed and depth of cut for Ti-6Al-4V — a file for another material does not map onto it.
- CSV files only. Excel files (.xlsx) are refused with a message explaining how to save as CSV instead.
- The published model stays a simple, well-understood shape. More elaborate model types are explored offline by the lab, not on this page.
- One shared contributor key. Fine for a handful of known labs; revoking a single contributor is not possible without changing the key for everyone.