EDGE REST / PORT 80
/api/calibration · /api/dsp · /api/meterCalibration, electrical context and runtime DSP configuration.Built on SIIP / SPICA Industrial Intelligence Platform
PumpSpectra™ / Rotating Equipment Intelligence
See what the motor sees.
PumpSpectra is the first condition-intelligence product built on SIIP. It turns non-invasive synchronized electrical signals into physics-gated MCSA-EDA, stable asset state and explainable maintenance intelligence for motors, pumps and the systems they drive.
Developed through the I+D+I program of Industrias Metalúrgicas EXCEL, C.A., on the SIIP architecture.
32 kS/sPer-channel synchronized acquisition
73 voltage · 3 current · 1 neutral
262,144High-resolution global FFT points
24/7Autonomous collection workflow
One evidence chain
From cabinet current to a maintenance decision.
Built on the SIIP signal-to-decision architecture, PumpSpectra keeps the complete path traceable. Every conclusion can be followed through context, fault gates and temporal state to the synchronized waveform.
- 01
Acquire
Capture IA, VA, IB, VB, IC, VC and neutral together at 32 kS/s, aligned to a measured grid zero crossing.
- 02
Contextualize
Bind the waveform to calibration, meter data, DSP context, motor poles, speed and exact bearing geometry.
- 03
Decide
Run deterministic C++ FFT and MCSA with physics gates, noise evidence and multi-signal corroboration.
- 04
Persist & explain
Store the evidence, stabilize state over time and let bounded local AI communicate the approved maintenance action.
Intelligence at the asset
Built for the industrial edge.
The reference acquisition design routes three voltage channels, three phase-current channels and one neutral-current channel through RF and anti-alias filtering into the ADE9000. The ESP32-S3 edge node performs real-time DSP and streams calibrated context and synchronized waveforms; the host CPU/GPU core combines MCSA-EDA, full-band FFT, exact bearing geometry, rotor and stator sidebands, pump evidence and power-quality context without reducing the result to a black-box score.
- Non-invasive CT clamps or Rogowski coils at the motor control cabinet
- Seven simultaneous channels: 3 voltage, 3 current and neutral
- Zero-crossing aligned 32 kS/s capture on every channel
- Drive-end, non-drive-end and impeller context
- Deterministic urgency, action and rationale
WAVEFORM TCP / PORT 8080
INFO · CAPTUREVersioned 3,610-byte frames with 128 synchronized sets across seven channels.ANALYSIS CORE
motor_fft_cpu · motor_fft_gpuGlobal FFT, MCSA, fault competition, per-bearing and pump evidence in schema v2 JSON.SIIP DATA + STATE
MongoDB · health · historyMeasurements, spectra, alerts, temporal state and maintenance recommendations.Industrial deployment
Built for remote fields and on-site decisions.
PumpSpectra combines cabinet-level sensing, local processing, durable evidence and flexible communications in one field-ready workflow.
Non-invasive retrofit
CT clamps or Rogowski coils acquire supply-side evidence without accelerometers on the machine or physical access to the rotating shaft.
On-site history
Local MongoDB preserves synchronized captures, spectra, events and reports when external connectivity is limited or unavailable.
Bounded local AI
Ollama runs on site to translate deterministic C++ results into clear maintenance guidance; it cannot change the physics verdict.
Telemetry-flexible
Field deployments can use existing 3G/4G/LTE, VSAT or fiber links from the asset to the control room.
MCSA-EDA / Diagnostic map
One electrical view of the complete machine.
The revised engineering map connects five evidence families to the physical asset: pump and impeller, shaft and coupling, stator, bearings, and supply-side electrical quality.
English MCSA-EDA diagnostic mapPump · coupling · motor · bearings · electrical supply
Open the interactive MCSA simulator Verified diagnostic scope
Fault evidence organized by physical domain.
PumpSpectra does more than label a spectrum. It resolves competing signatures, records contamination and preserves the exact features that support—or prevent—a diagnosis.
Rotor & stator electrical faults
Broken-bar and eccentricity sidebands are evaluated against refined f0, speed, slip validity and phase corroboration. Stator end-winding evidence is tracked through odd current sidebands rather than one isolated peak.
- Broken bar
- Eccentricity
- Stator sidebands
Shaft & mechanical condition
The 1× RPM fundamental and its 2×–4× rotational harmonics are evaluated with competing evidence so shaft misalignment, coupling issues, unbalance and looseness are not collapsed into one label.
- Rotor unbalance
- Misalignment
- Looseness
Drive-end & non-drive-end bearings
BPFO, BPFI, BSF and FTF are calculated per bearing, with x2/x3 harmonic evidence, confidence and grid-contamination proximity.
- BPFO / BPFI
- BSF / FTF
- DE / NDE
Pump & impeller
Impeller blade count establishes fBPF; its harmonics, load imbalance and broadband electrical noise provide corroborating evidence for impeller imbalance and cavitation.
- Blade pass
- Impeller imbalance
- Cavitation evidence
Power-quality context
Current and voltage THD, phase imbalance, low power factor, harmonic order and frequency variation separate supply-side conditions from asset behavior.
- THD
- Phase imbalance
- Power factor
Time & time-frequency evidence
Synchronized voltage and current waveforms expose startup transients, abrupt mechanical overloads and phase asymmetry; STFT, wavelets and joint V-I tracks then reveal persistence without replacing global FFT gates.
- Transients
- Phase asymmetry
- STFT / wavelets
Trust logic
A diagnostic engine designed to refuse weak conclusions.
- 01No valid physics, no rotor claim
Broken-bar and eccentricity flags require valid f0, poles, speed and slip, plus resolvable sidebands.
- 02No isolated impeller peak
Impeller imbalance requires at least two corroborating signals across blade-pass, rotor asymmetry and current imbalance.
- 03No false bearing localization
When DE and NDE characteristic frequencies are too close, the system reports ambiguity and requests physical inspection.
Two evidence paths
Diagnose the capture. Then understand the history.
Immediate and historical diagnoses are separate services with separate evidence boundaries. Both use the same deterministic facts; only the historical path adds persistence and temporal context.
CAPTURE MODE
Immediate diagnosis
POST /diagnose_fftConsumes an FFT result in JSON and returns a deterministic, explainable report without reading or writing MongoDB. It is suitable for a live capture or an isolated engineering review.- Stateless
- Same physics pipeline
- Exact capture evidence
HISTORY MODE
Trend diagnosis
POST /diagnose_historyAnalyzes a selected days, hours or last-N window, measuring fault persistence, health movement, bearing trends, pump evidence and maintenance urgency.- Windowed history
- Persistence & deltas
- Maintenance context
Local AI explains these results after the deterministic assessment. It is not permitted to convert a rejected flag into a confirmed fault.
The measured asset
What the current signal is actually carrying.
A marked bearing race, a misaligned coupling, an unbalanced impeller or a shifting hydraulic load all change the mechanical effort at the shaft, and that effort modulates the stator current. These are the parts of the machine that leave a signature, and how PumpSpectra reads each one.
01ACQUISITION POINT
The measurement happens at the cabinet, not at the machine.
Three voltage channels, three phase currents and neutral are taken from the motor control cabinet through CT clamps or Rogowski coils. Nothing is mounted on the rotating equipment and the motor never has to be stopped to install it. The ADE9000 front end and its ESP32-S3 host stream calibrated context over REST and versioned 3,610-byte WaveFrames over TCP.
- No accelerometers
- No shutdown to install
- Cabinet or MCC mounting
02MOTOR · COUPLING · PUMP
The motor, the coupling and the pump are one mechanical system.
Bearing wear, a misaligned coupling, an unbalanced impeller or a change in hydraulic load all alter the mechanical effort the motor has to produce. That effort modulates the stator current. PumpSpectra reads the whole driven train through the one signal that already passes through every part of it.
- Rotor and shaft
- Flexible coupling
- Driven pump and impeller
03BEARING GEOMETRY
Four defect frequencies, computed from the actual bearing.
BPFO, BPFI, BSF and FTF follow from rolling-element count, ball and pitch diameter and contact angle, so every bearing is configured by geometry rather than assumed. The field configuration carries a 6309 at the drive end and a 6206 at the non-drive end, and each is evaluated on its own. BSF and FTF are reported but kept out of the severity score, because field data showed them picking up unrelated mechanical and slip components.
- BPFO · BPFI · BSF · FTF
- Per-bearing geometry file
- DE 6309 · NDE 6206
04MCSA-EDA
From a current spectrum to an inspection order.
A 262,144-point global FFT resolves the spectrum to roughly 0.12 Hz per bin, which is fine enough to separate a bearing line from a neighbouring grid harmonic. Deterministic C++ then applies physics gates, harmonic corroboration and competing-fault logic, and the reporting layer turns per-bearing severity into one of four actions: immediate, next scheduled stop, routine monitoring, or none.
- Δf ≈ 0.12 Hz
- Harmonic corroboration
- Four maintenance actions
The animations above are technical illustrations of the monitored assembly. Measured results appear in the field section and in the interface captures further down.
Field validation
53 captures on a 10 HP motor-driven pump.
The method was evaluated on a motor-pump test campaign at Industrias Metalúrgicas EXCEL in Valencia, using a bearing geometry file with a 6309 at the drive end and a 6206 at the non-drive end. Of the 53 captures, 45 carry a group label from the campaign field notes and are summarised below; the remaining 8 were recorded without one and are counted in the dataset but not in these groups. The full paper is available in the technical library below.
G0
Nut imbalance
5captures0bearings flagged
10 June baselineG1
Original bearings, no damage
8captures0bearings flagged
Healthy referenceG2
Original bearings + one damaged ball
22captures2bearings flagged
15 rotor-evidence capturesG3
New bearings + impeller imbalance
10captures0bearings flagged
4 rotor · 4 impellerWhat the campaign produced
Five maintenance-relevant captures, placed before any table.
The report leads with the actions so a reviewer can read the decision first and the numbers second. No capture reached the alarm threshold; the highest bearing evidence stayed just below it, and the report says so rather than rounding up to a fault.
- 2Inspect both bearingsNext scheduled stop
DE and NDE both showed elevated BPFO/BPFI evidence, and MCSA could not rule either one out.
- 2Inspect the drive endNext scheduled stop
Drive-end evidence crossed the warning band while the non-drive end stayed below it.
- 1Monitor the non-drive endRoutine monitoring
Monitor-level elevation that never reached the warning threshold.
Real product interfaces
One operational picture, from waveform to diagnosis.
These are working PumpSpectra views—not conceptual dashboards. Open any image to inspect the original interface at full resolution.
Technical library
Read the engineering behind the product.
Download the field paper, product brief, verified acquisition API reference and standalone CPU-engine manual directly from this page.
Field research · English
Field-Validated Multi-Bearing MCSA
A five-page paper covering synchronized acquisition, per-bearing evidence, harmonic corroboration and field results from a 10 HP motor-pump campaign.PDF · 5 pages · 262 KBDownload PDFTechnical brief · Spanish
PumpSpectra MCSA Technical Brief
A visual introduction to diagnosing mechanical faults from motor current—without installing vibration sensors on the machine.PDF · 4 pages · 91 KBDownload PDFFirmware reference · English
ADE9000 REST + TCP Endpoint Guide
A source-verified map of 30 firmware endpoints for measurement, calibration, DSP, protection and versioned 32 kS/s WaveFrame capture.PDF · 13 pages · 495 KBDownload PDFCPU engine reference · English
motor_fft_cpu Complete Technical Reference
A source-derived guide to the standalone C++23 CPU engine: its 65,536-point FFT baseline, seven-channel frame decoding, motor physics, bearing frequencies, hypothesis scoring and JSON output.PDF · 21 pages · 259 KBDownload PDFOfficial & independent sources
Supporting technical references.
Manufacturer documentation and independent research remain on their publishers' sites. These links provide engineering context and do not imply SPICA authorship or third-party endorsement.
Manufacturer referenceADE9000 Data SheetAnalog Devices · Product specifications↗Manufacturer referenceADE9000 Technical Reference ManualAnalog Devices · User Guide UG-1098↗Manufacturer application noteUsing the ADE9000 Waveform BufferAnalog Devices · Application Note AN-1483↗Independent researchTemporal Reasoning for Pump Fault DiagnosisProcesses 2025, 13, 2624 · Open access↗
Bring the signal into focus