run_25k_tests.py.Anodic corrosion current scales with the cathode-to-anode area ratio:
| Cathode/anode area ratio | Relative anodic rate |
|---|---|
| 1 | 1 |
| 10 | 10 |
| 100 | 100 |
Useful. A stainless fastener in an aluminium structure is not "a galvanic couple" — it is a couple with a specific area ratio, and that ratio is the rate. The audit is geometric, not electrochemical: measure the exposed area of each metal before blaming the materials.
| Time (h) | Fraction of preload lost |
|---|---|
| 0.1 | 0.37 |
| 1 | 0.55 |
| 10 | 0.72 |
| 100 | 0.84 |
| 1,000 | 0.93 |
| 10,000 | 0.97 |
Useful. A 24-hour torque-retention test captures roughly two-thirds of the total loss. A joint that passes at 24 h can still fail at 1,000 h, and the curve says so before any long test is run.
| ΔT (K) | Relative cycles to failure |
|---|---|
| 30 | 111,111 |
| 60 | 27,778 |
| 120 | 6,944 |
Useful. With the standard Coffin–Manson exponent n = 2, halving the thermal excursion quadruples life. A qualification test that passes at one ΔT says nothing about another, and the scaling is steep enough to matter at ordinary design margins.
| Relative velocity | Relative erosion rate |
|---|---|
| 1.0 | 1.00 |
| 1.5 | 11.4 |
| 2.0 | 64.0 |
Useful. A 50% speed increase raises erosion elevenfold. Any pump or valve compared across duties must be compared at matched tip speed, or the material ranking is a speed ranking in disguise.
With an illustrative outgassing rate of 10⁻⁹ Pa·m³/s per m², 0.5 m² of surface in a 0.02 m³ volume gives:
| Time | Pressure rise (Pa) |
|---|---|
| 1 min | 2×10⁻⁶ |
| 1 h | 9×10⁻⁵ |
| 1 day | 2.2×10⁻³ |
Failure of my parameter. These pressures are far too small to matter. Either the rate is 1,000× too low or the geometry is wrong — a real vacuum outgassing problem needs rates near 10⁻⁶ Pa·m³/s per m², three orders above what I used. The scaling is right; the number I chose was not, and the calculation exposed it. This is the kind of error that survives if only the formula is checked and not the magnitude.
Two independent random series, 300 points: derived-variable check gives R² = 0.0001, not flagged. Passes. A technique that fires on everything is worthless; this one stays quiet when there is nothing to find.
| Result | Outcome | Practical use |
|---|---|---|
| Galvanic area ratio | useful | measure exposed areas before blaming materials |
| Creep preload loss | useful | a 24 h test is not a lifetime test; the curve says how far it gets |
| Thermal fatigue | useful | halving ΔT quadruples life; qualifications do not transfer across ΔT |
| Cavitation | useful | compare at matched tip speed or the ranking is a speed ranking |
| Outgassing | parameter wrong by 10³ | check magnitudes, not only formulas |
| Negative control | passes | the technique stays quiet on independent data |
Six computations do not validate a catalogue of 25,000. What they do is establish that the kinds of question in it are answerable with arithmetic and existing measurements — and that one of them, when actually computed, exposed an error in the person doing the computing.
Of six items tested, one failed on my own parameter and one was a control. That ratio — one real failure per four useful results — is the honest expectation for the rest of the catalogue.