SOLUTIONS

What you can build
with Tinia

Wire up an analysis pipeline that spans time-series signals · spectra · anomalies · compliance. Every solution is clear at a glance — what it solves, how it works, what it produces; what you receive is a graph you can still edit freely after opening it — a starting point, not a black box.

VisibleReproducibleReusableAuditable
S1 — 01 / 07
Fault diagnosis · Predictive maintenance

Turn “Fan No.3 outer race has failed”
from a veteran's ear into a reproducible conclusion

For
Wind power · Chemical · Steel · Power-plant O&M · Asset management · PdM service providers
Reference standards
Algorithms align with common bearing-fault characteristic-frequency practice
Pain point

Early bearing and gearbox faults are judged by a veteran's ear, and that expertise disappears with staff turnover; by the time the noise is audible, the damage is usually well advanced.

How it works
  • 01Band-pass to lock onto the resonance band, then envelope demodulation to extract the fault component from the high-frequency carrier
  • 02The envelope spectrum auto-matches BPFO/BPFI/BSF/FTF bearing characteristic frequencies, with harmonic verification
  • 03Build a health fingerprint per machine, score the deviation quantitatively, and raise alerts automatically
  • 04Batch-replay historical recordings to build up severity trend curves
Output
  • Envelope spectrum with the four characteristic frequencies annotated
  • Per-measurement-point severity scores and trend curves
  • An anomaly event list (alerts can be pushed)
Key nodes
IIR Filteriir_filterEnvelope Demodulationenvelope_demodFFT Spectrumfft_spectrumIndicator Mathindicator_mathBaseline Statisticsbaseline_statsZ-Score Anomaly Detectionzscore_anomalySpectrum Viewerspectrum_viewer
S2 — 02 / 07
Production-line acoustic QC · Spec-limit judgement

Every production unit
gets a traceable PASS / FAIL

For
Appliance assemblies / OEM lines · Rotating-machinery shop-floor inspection · Third-party testing · Environmental monitoring
Reference standards
IEC 60704 · ISO 3744 · ISO 20816 · IEC 61672 approach
Pain point

The judgement logic is hard-coded into a script or a single instrument, so every new standard or product means rebuilding it — and the basis for the verdict still can't be explained.

How it works
  • 01Multi-channel split → A/C/Z weighting → sound level meter + 1/1 and 1/3 octave, all in one pipeline
  • 02Compute sound power, equivalent sound level and rotating-machinery RMS on the fly (IEC/ISO approach)
  • 03Threshold tables and Zone A/B/C/D traffic-light rules live inside the nodes — visible and editable
  • 04Results feed the dashboard and export as a traceable factory report
Output
  • Sound power / equivalent sound level + 1/3 octave spectrum
  • Zone A/B/C/D red-amber-green list + trend
  • PASS / FAIL with documented basis, ready to generate a report
Key nodes
Channel Splitchannel_splitFrequency Weightingweighting_filterSound Level Meterlevel_meterOctave Analysisoctave_analysisIndicator Mathindicator_mathSpec Limit Checkspec_limit_checkIIR Filteriir_filter
S3 — 03 / 07
Psychoacoustics · Subjective-perception quantification

“Premium feel” vs “cheap feel”
give subjective ratings a set of objective, comparable numbers

For
OEM sound quality (SQ) teams · Tier1 / Tier2 trim parts · HVAC · Door locks · Appliances
Reference standards
ISO 532-1/2 · ECMA-418-2 · DIN 45631/45692
Pain point

A low A-weighted level doesn't mean it “sounds pleasant.” Fullness, sharpness and roughness used to rely on jury panels — expensive and inconsistent.

How it works
  • 01Loudness + sharpness + roughness + tonality as a set, with TNR to flag prominent tones
  • 02Implemented to the ISO 532 / ECMA-418-2 public standards — closer to the human ear than A-weighting
  • 03First slice out transient events such as door slams and button clicks, then compute indicators per event
  • 04Align multiple indicator tracks for State A vs State B comparison
Output
  • Loudness / sharpness / roughness / tonality time series + TNR
  • Per-sound-event psychoacoustic indicator comparison
  • An overall sound quality score to back “premium feel” decisions
Key nodes
Audio Segmentationaudio_segment_splitActive Segment Detectionactive_segmentLoudnessloudnessSharpnesssharpnessRoughnessroughnessTonalitytonalityTNRtnrFeature Aggregationfeature_mergeChart Viewerchart_viewer
S4 — 04 / 07
Order / Modulation spectrum · Rotating machinery

“That hum at 3000 rpm”
which order is it really?

For
OEM powertrain / e-drive NVH · EV tri-power Tier1 · Vehicle-level NVH teams
Reference standards
Engineering practice (no mandatory standard)
Pain point

When the speed keeps changing, a fixed spectrum can't reveal the drifting whine; electromagnetic orders, mesh orders and inverter sidebands overlap and are even harder to separate.

How it works
  • 01Order tracking resamples into the angle domain, producing waterfall and Campbell diagrams
  • 02Track mechanical and electrical orders separately and overlay them to identify sidebands and mesh orders
  • 03WOT full-throttle sweeps auto-annotate resonance peaks and output 1st/2nd/4th order slices
  • 04Speed × frequency-band heatmap for cross-scale observation of faint whines
Output
  • Order-RPM waterfall + Campbell diagram
  • Per-order slice curves + automatic resonance-peak annotation
  • Sideband / mesh-order localization to pin down the fault source
Key nodes
Order Trackingorder_trackingChannel Selectchannel_selectSpectrum Smoothingspectrum_smoothModulation Spectrummodulation_spectrumMulti-scale Spectrumscale_space_spectrumPivot Matrixmatrix_viewIndicator Viewerindicator_viewer
S5 — 05 / 07
AutoML · Capturing expertise

Turn a senior engineer's judgement
into a reusable discriminant function

For
OEM sound quality teams · Tier2 trim parts · Algorithm teams · Cross-domain cases such as AI-assisted diagnosis / ecological monitoring
Reference standards
Replaces ear / veteran judgement; the model stays interpretable and auditable
Pain point

The judgement of “good vs bad sound” is tied to a few experts: the criteria can't be articulated, and they're lost with staff turnover.

How it works
  • 01Convert audio into a set of interpretable indicators (loudness, sharpness, spectral features…)
  • 02After scale alignment, AutoML searches parameters automatically and distils a discriminant function
  • 03Regress against subjective ratings, or classify “normal / abnormal,” yielding a score predictor
  • 04The whole chain remains an editable node graph: transparent, auditable and re-trainable
Output
  • A batch-runnable discriminant function / score predictor
  • Subjective-to-objective regression curve or classification result
  • An interpretable model that can be re-distilled as data grows
Key nodes
LoudnessloudnessSharpnesssharpnessSpectral Feature Extractionfbank_extractStructure Tensor Featuresst_featuresFeature Normalizationfeature_normalizeIndicator Mergeindicator_mergeCluster Explorationcluster_explore
S6 — 06 / 07
Generalized time-series signals · Cross-domain

One set of nodes
covering acoustic · vibration · electrical · ultrasonic

For
Power quality testing · Smart water / gas-leak detection · Medical heart sounds · Ecological acoustic monitoring
Reference standards
IEEE 519 · GB/T 14549 · general time-series methods
Pain point

Current harmonics, pipe leaks, heart-sound screening, birdsong recognition — buying a dedicated instrument for each is costly and fragmented, with no unified foundation.

How it works
  • 01Current, ultrasonic and physiological signals are all treated as time series, reusing the same operators
  • 02Power quality: FFT to extract each harmonic + THD, judged against a spec-limit table (IEEE 519 approach)
  • 03Pipe leaks: ultrasonic band-pass + envelope demodulation down-shift + baseline / Z-Score alerts
  • 04Medical / ecological: active-segment detection + spectral features + AutoML classification screening
Output
  • Harmonic amplitudes + THD% + spec-limit compliance records
  • Leak / anomaly event lists with trend alerts
  • Classification and rhythm analysis for heart sounds, species and similar cases
Key nodes
FFT Spectrumfft_spectrumOctave Analysisoctave_analysisIIR Filteriir_filterEnvelope Demodulationenvelope_demodTime-domain Statisticstime_statsBaseline Statisticsbaseline_statsZ-Score Anomaly Detectionzscore_anomalySpec Limit Checkspec_limit_checkSpectral Feature Extractionfbank_extract
S7 — 07 / 07
Large datasets · Streaming real-time

From “a single recording”
to compute-as-you-capture, whole-library batch runs

For
On-line production-line monitoring · Edge / on-line diagnosis · Algorithm-team batch replay · Large-scale data governance
Reference standards
SDK streaming + resident execution + GPU acceleration (platform capability, v1.35)
Pain point

A validated graph needs to go live, tens of thousands of historical recordings can't be clicked through by hand, and the line needs compute-as-you-capture, low-latency results.

How it works
  • 01The same graph: run a single record in the editor, or feed the whole library in batch and replay in parallel via the Python SDK
  • 02SDK streaming session: push data continuously and retrieve results in real time — ideal for on-line scenarios
  • 03Resident execution: the process stays ready and libraries load only once, reusing warm processes to save startup overhead
  • 04FFT / spectral nodes support GPU compute, with a shared sidecar centralizing scheduling for speed
Output
  • Whole-library batch results in a single run, feeding dashboards and reports
  • Low-latency streaming results for on-line / edge diagnosis
  • Monitorable call volume / success rate / latency / top nodes
Key nodes
FFT Spectrumfft_spectrumOctave Analysisoctave_analysisEnvelope Demodulationenvelope_demodIndicator Mathindicator_mathBaseline Statisticsbaseline_statsZ-Score Anomaly Detectionzscore_anomalyAttribute Extractionattribute_extract
WHY A GRAPH
001
A template is a starting point, not a black box

Open it and see every step; tweak a parameter, add a branch or swap a node at will, never locked into a fixed flow.

EditableBranchableUnlocked
002
Judgement logic is visible and traceable

Thresholds, weightings and formulas all live inside the nodes, so the basis for PASS / FAIL can be explained to customers and auditors alike.

VisibleReproducibleAuditable
003
Expertise is captured and reusable

A veteran's workflow is packaged into a template for the whole team to reuse, AutoML distils the judgement into a function, and expertise survives staff turnover.

ReusableDistillableRetained
Get Started

Turn your scenario
into a graph

Tell us your industry and the acoustic / vibration / time-series signal problem you need to solve, and together we'll turn it into a runnable, reproducible Tinia workflow.

Request a Trial