The bearing FFT race.
Industry baseline vs Validiti Maths.
Same signal. Same answer. Different CPU cost.
An inner-race bearing fault on a frac-pump quintuplex. Same 1-second snapshot at 12 kHz (12,000 samples) into both pipelines. Both compute the same three operations: dominant frequency, RMS, peak-to-peak. Both produce identical outputs. One uses 10× less CPU than the other.
No cd commands.
No clone. Just click.
Side-by-side, on the same single core.
Three operations a frac-pump bearing monitor runs every second.
Dominant Frequency
RMS
Peak-to-Peak
These three values get computed every second on every bearing on every pump on every pad. A typical frac pad has 12 pumps × 5 bearings = 60 bundles/sec. The CPU cost of running this bundle is the cost of doing this kind of monitoring at all.
99.6% CPU headroom vs 95.5%.
At 60 bundles/sec on one pad:
industry baseline
Validiti Maths
A VAM running V.L., Chronicle, Argus, Maths, Pacta, Reflex and Titus together — on commodity CPU — needs every cycle. Validiti Maths returning 95% of the CPU back to the rest of the stack is what makes the whole VAM possible without GPU.
Why this matters
industry baseline is the right answer for offline analysis on a workstation. It's the wrong answer at the asset, on a sealed box, sharing CPU with seven other Validiti components, replicated across every pump on every pad in every basin.
Validiti Maths is the kernel that ships inside the VAM. Deterministic. Signed. Replayable for regulatory review. And 10× cheaper to run on the hardware it actually runs on.
The bearing-fault signature in this bench is modeled on the published CWRU Bearing Data Center drive-end accelerometer data — the canonical dataset for rotating-machinery fault diagnostics, cited in 4,000+ papers.
- Dataset: CWRU Bearing Data Center · engineering.case.edu/bearingdatacenter
- Specific file shape modeled: CWRU drive-end inner-race fault recording (Inner Race fault, 0.007" defect, 1730 RPM, 12 kHz drive-end accelerometer)
- Why synthetic here: the bench needs deterministic numbers anyone can reproduce; CWRU access is straightforward but adds a download step. The shape, the fault frequencies, and the SNR ranges all match the published CWRU dataset.
Numbers run against real CWRU signals come out within a few percent of the synthetic case shown here.
Four roles, four wins.
Vibration analyst (CAT II/III)
The signature recognition you already do, made structural and continuous — on commodity CPU at the asset.
Reliability engineer
Every bearing on every pump on every pad scanned every second, no GPU, no cloud. The watch you wish you had time for.
Service-company technician
The bearing flag on the tablet matches what your trained ear hears at the pump — before the failure becomes a workover.
CFO / commercial
10× less CPU per pad-second means one box does the work of a small cluster. Material savings at scale, no GPU line item.
In numbers.
10× less CPU
Same operations, same outputs, single CPU core. Pad-scale: 4.4 ms used per second vs 44.4 ms.
99.6% CPU headroom
What’s left over runs every other Validiti capability on the same box. The reason CPU-only is possible at all.
Deterministic outputs
Same answer on every replay. Signed and replayable for regulatory review.