AcoustR.
Diagnostics from sound.
A microphone is the cheapest, lightest, most easily installed sensor on a complex asset — and on most rotating machinery, the one with the highest diagnostic information per unit cost. AcoustR is the pipeline that turns that signal into a condition assessment. Listen with us, below.
● Bins above ~62% intensity are highlighted in deep rust as an illustrative fault contour. The contour is a visual cue, not a diagnosis — the production pipeline scores against learned signatures, not raw thresholds.
Investor memo available on request — email to request.
Three stages. One signal.
A short pipeline, drawn small. Each stage has its own representation; the final score carries explicit uncertainty rather than a label.
Microphone → time-frequency representation
Audio frames are windowed and converted to a short-time Fourier transform. The hero panel above is exactly this representation, streaming live.
Feature extraction → diagnostic signature
Per-band energies, harmonic ratios, transient rates, and a handful of cepstral coefficients collapse the spectrogram into a fixed-length signature ready for comparison.
NDCI scoring → fault probability ranking
Each candidate signature is scored against the asset's fault library. NDCI weights each contribution by its diagnostic value rather than its raw activation, producing a ranked list with calibrated probabilities.
Diagnostic value, not sensor count.
Most condition-monitoring stacks are sold by sensor count. The research underneath AcoustR — the Multi-Objective Sensor Optimisation Framework (MOSOF) and the Normalised Diagnostic Contribution Index (NDCI) — proves that a smaller, smarter sensor set can match or beat a larger one when you actually score what each sensor contributes to a diagnosis.
AcoustR takes that insight to its limit: in many machines, a single well-placed microphone carries enough diagnostic information to act on. The product is what falls out when you treat sensor selection as an optimisation problem rather than a procurement decision.
Case study: B737-800 ECS sensor reduction.
One concrete case, end-to-end. The Pareto front the algorithm actually produced, the recommended 12-sensor knee at 0.69 normalised diagnostic performance, and the peer-reviewed paper that documents it.
Candour about pilot results
This page does not quote customer numbers. AcoustR is in early pilots; the published evidence on the underlying sensor-optimisation methods is on the case-study page above and in the cited journal articles. Once pilot outcomes are agreed for public release, they will appear here with sample sizes attached — not before.
Pilot, partner, or invest.
If you operate machines that fail expensively, fund early industrial-AI ventures, or work in MRO where triage cost is the bottleneck — I'd like to hear from you.
[email protected] →