Methods & validation
The method, shown rather than claimed.
Six worked examples on published, open data, and the sequence each report goes through before it is signed. Move the controls; the physics responds. These are demonstrations of method, not casework.
How I check my own work
Audits first, answers second.
Every analysis I produce carries its audits: momentum, energy, sanity checks against known physics, and a source for every input. A result that fails its audit is not softened; it is stopped and the reason is stated. That is what an opinion needs behind it before it enters a courtroom.
The six examples below are built on published, open data so that anyone can operate them and check them. They are demonstrations of method, not casework, and no casework result is published on this site.
Worked examples
See the analysis working.
Simple models illustrating the kind of checks, audits and stated ranges used in the practice's reports, each built on published data. Everything runs in your browser and stores nothing.
01 · Offset collision
Why struck cars spin
Plan view: side impact with offset
Inputs
- momentum
- angular momentum
- energy ≥ 0
Planar impulse mechanics with restitution and friction, checked against momentum, angular momentum and energy at every setting. This is the check applied to a reconstruction already in a case: do its stated speeds, masses and contact geometry balance. Worked example on published staged-collision data, not casework.
02 · Fracture
How small can a fatal crack be?
A component fails “without warning”. Fracture mechanics says how big the hidden flaw needed to be.
The 1967 Silver Bridge collapse began at a crack about 2.5 mm deep (NTSB). The formula shows why a flaw that size was critical for that steel in that cold. A textbook check, not an investigation; the same mechanics applies to a vehicle suspension or steering component. σ = K / (1.12 √(πa)), edge crack.
03 · Severity
Same speed, different outcome
Impact severity follows stopping distance, not speed alone. “It was only 30 mph” answers nothing.
a = v² / 2s. The order-of-magnitude spread from one uncertain input is why honest reports state ranges, never single numbers.
04 · Testing a reconstruction
Do the reconstruction's own inputs support its speed?
Take the mark length and the friction value from the reconstruction already in the case. The physics gives the range those inputs actually support, and the surface decides how wide it is.
v = √(2μgs), locked-wheel model. A minimum: energy already lost to impact or yaw is on top. Mark length and surface are the reconstruction's own inputs, taken from the collision investigator's report; the engineer's job is to test whether its conclusion sits inside the range they support. No scene measurement is made.
05 · Component failure
Corrosion, the quiet stress multiplier
A component does not need to look bad to be dangerous: lost wall thickness raises stress faster than it looks.
Generic box-section bending model: section loss from all four walls, so stress rises with the square of fractional loss. Pits add a local stress concentration on top, the fatigue-initiation mechanism behind many “sudden” failures. The same reasoning applies to a corroded vehicle chassis member or subframe.
06 · Simulation
Does the mesh converge? Finite elements against the exact answer
A cantilever beam in steel or aluminium, tapered from root to tip. Three answers compared: the hand formula (uniform-beam assumption, the class of calculation behind a screening report), the finite-element solution (beam elements, each using averaged section properties), and the exact analytical integration (the reference a commercial code converges toward).
Euler–Bernoulli beam elements (cubic Hermite shape functions, 2 DOF per node). Each element uses the second moment of area at its midpoint; the taper is the simplification the mesh resolves. Fixed end: v = θ = 0. Tip load P = 10 kN on a 1 m span, 50 mm wide. Comparing a finite-element result with an exact solution is a verification exercise; it shows the mesh is adequate, not that the model matches reality.
Before a report is signed
Eight stages, in order.
The same sequence is followed on each instruction, scaled to the matter, whether the matter is a small-claims damage dispute or a Crown Court case. The order matters: nothing is analysed until the papers are proved readable and indexed, and nothing is served until the draft has been attacked.
Intake
Regime (CPR 35 or CrimPR 19), conflicts, data protection, funding and the rate basis. The matter file and an open-points register are opened before anything else.
Ingest
Every page of the bundle made readable and proved readable. Scans are OCR-checked; video and photographs are catalogued with their metadata.
Document checks
Each PDF is checked as a file as well as a document: hidden text, incomplete redactions and mark-up layers are looked for before they can mislead.
Index and gaps
Every exhibit referred to is reconciled against every exhibit produced. What is missing is listed, with what it would be worth obtaining.
Triage
Each instructed question is tested against what the papers can actually support. Questions the evidence cannot answer are said to be unanswerable, early.
Analysis
Closed-form calculations first, parameter sweeps on each uncertain input, and a cross-check of each result by a second method.
Adversarial pass
The draft is attacked as a hostile expert would attack it. Anything that does not survive is removed or qualified before it is published.
Build and gates
The report is built to house style with working internal references, then passed through a fixed set of pre-service checks before it is signed.
Enquiries
A question worth testing?
Send an outline of the matter and the question the court needs answered. I will say whether the engineering can answer it, and how sure the answer can be.
Prefer email? rogernewbyconsulting@gmail.com
Roger Newby
Vehicle Engineer & Expert Witness