Validation

Engineering software should show its work.

What has been checked, how, with what result — and what has not been checked. Every number below is reproducible in the tools themselves; the regression assertions ship with the site.

Pressure Pulse Inversion for Blockage and Deposition Analysis

Independent-solver comparison and synthetic recovery

TestTruthResultStatus
Synthetic blockage, 50 % area, 20 m long, 0.2 m sensor noiseface at 390 m from valve390.5 m · 50.3 % area remaining · 20.0 m extentpass
Synthetic leak700 m from valve700.1 m · leak flow within 8 %pass
Clean pipeno reflectorno echo with |r| ≥ 0.05pass
Spurious 400 m transient late in the recordclosure at 0.005 sclosure found at 0.0045 s; transient reported as an echo, not as closurepass
Independent solver (TSNet, WNTR-based): leak300 m from valve307 ± 9 m · 4.3 L/s vs 4.9 L/spass
Same trace: additional echoesnone physicaltwo additional strong echoes (~562 m, ~877 m) from the other solver's boundary treatment are reported, not suppresseddocumented
Sensitivity sweep, closure 10 ms, noise 0.1–0.8 m, 500–2000 Hzblockage at 390 m5 / 5 detections in every case, 0 false alarmspass
Sensitivity sweep, closure 50 ms / 200 msblockage at 390 m4 / 5 and 1 / 5 detections — slow closure degrades detectiondocumented

Not yet validated: against published field hammer-test data with ground truth; leak sizing on real lines; branched networks, column separation, viscoelastic pipe walls. The tool's Method panel lists every modelling assumption.

Bayesian Parameter Inversion

Convergence diagnostics and known-truth recovery

CaseResultStatus
Seismic benchmark, 3 chains × 3000all impedances inside 95 % CrI · max R̂ 1.006 · min ESS ≈ 440pass
Pipeline blockage, synthetica, f, x, blocked area inside 95 % CrI · R̂ 1.09pass
Pipeline leak, syntheticx = 700 ± 4 m (true 700) · orifice 0.0050 (true 0.0050) · R̂ ≤ 1.09pass
Well-log petrophysical, 7-zone synthetic log13 / 14 truths inside CrI · rms residual 1.0 σpass
Well-log petrophysical, sample_well.csv with default endpointsREVIEW REQUIRED · R̂ 1.02 · residual 2.3 σ — the default matrix endpoints do not match that log; the residual tracks show itdocumented
Independent TSNet trace, full-waveform fitNOT CONVERGED (R̂ ≫ 1): the solver cannot represent the other code's boundary artefacts; the diagnostics say so rather than reporting a confident wrong answerdocumented

Status CONVERGED requires split-R̂ < 1.01 and bulk and tail ESS > 400 on every parameter; every run is reproducible from its seed and reopens exactly from the saved record.

Tubular Lockup Prediction

Provenance and scope

Johancsik soft-string drag (JPT 1984) with build-rate curvature; effective submerged weight separating steel, internal and annular fluid; Mitchell post-helical contact; Dawson-Paslay, Chen-Lin-Cheatham and Lubinski buckling limits. Sign convention and lockup definition are stated on the page: required push is a positive magnitude, helical onset is reported separately from lockup, and lockup is the first depth at which required push reaches injector capacity or helical drag runs away. Not validated against commercial torque-and-drag software or field lockup records; pressure-area effects are not implemented. It is a screening tool and is labelled as one.