← Volver a industrias

Mantenimiento predictivo y monitorización de estado

Detectar desgaste de rodamientos, degradación de engranajes y filtros obstruidos a partir de datos en tiempo real — a menudo con nada más que un micrófono y procesado de señal.

Reservar llamada

De reactivo a predictivo

Equipment failures are rarely sudden. Bearings degrade over weeks — their acoustic signature shifts as surface pitting develops. Gearboxes give measurable warnings months before seizure. Filters restrict flow gradually, visible in pressure and acoustic data long before the machine trips.

Reactive maintenance means unplanned downtime, emergency sourcing, and production losses. Scheduled maintenance is wasteful — replacing components that still have service life. Predictive maintenance is neither: act when the data says to, not on a calendar and not in a crisis.

Qué se puede monitorizar

Wear Monitoring — Bearings, Gears & Impellers

Bearing defect frequencies (BPFO, BPFI, BSF, FTF) appear as narrow spectral peaks long before audible noise develops. Gear mesh harmonics shift as tooth wear progresses. Envelope analysis isolates these indicators from broadband background noise — reliably, at signal-to-noise ratios where the human ear hears nothing unusual.

Abnormal Condition Detection — Filters, Pumps & Motors

A clogged fuel or hydraulic filter raises differential pressure and alters the acoustic signature of the fluid path. Pump cavitation produces characteristic broadband noise in the 1–10 kHz range. Motor current signature analysis (MCSA) detects eccentricity, broken rotor bars, and bearing defects from the supply current alone — no mechanical access required.

Enfoque técnico

Signal Acquisition

Low-cost sensors are usually sufficient: a microphone for airborne acoustics, a MEMS accelerometer for structure-borne vibration, or a current clamp for motor signature analysis. Often a single sensor positioned at the housing is enough.

Feature Extraction

FFT, envelope analysis, cepstral analysis, and time-domain metrics — RMS, kurtosis, peak-to-peak — isolate fault-characteristic frequencies and track their amplitude over time against a healthy baseline.

Baseline & Anomaly Detection

Establish a healthy-machine reference, then monitor deviation using threshold rules, statistical process control, or a lightweight anomaly detection model. No cloud infrastructure or specialist hardware required.

Integration & Alerting

Results feed into existing SCADA, PLC HMI, or a standalone dashboard. Configurable alarm thresholds, trend plots, and plain-language fault descriptions — not just a red light.

Por qué funciona

No Expensive Sensors Required

A USB microphone or low-cost MEMS accelerometer costing a few euros can detect bearing faults, gear damage, and filter blockage with the right signal processing. The value is in the analysis, not the hardware.

Scheduled on Your Terms

Act on data, not guesswork. Schedule the replacement during the next planned production window — not during an emergency shutdown at 2 a.m.

Non-Invasive Installation

Sensors are surface-mounted or clamped. No machine modifications, no PLC changes, no production interruption during installation or commissioning.

Interpretable Alarms

Fault-characteristic frequencies have physical meaning. When the system alarms, the data shows exactly which component and why — not just an anomaly score.

Stack tecnológico

Sensor Types

Accelerometers (MEMS or piezoelectric) for structure-borne vibration. Microphones for airborne acoustics. Current transformers (CT clamps) for motor current signature analysis (MCSA). Ultrasonic thickness sensors for corrosion/erosion monitoring. Thermal cameras for hot-spot detection. Oil debris sensors for lubrication system monitoring. In most cases a single well-placed low-cost sensor is sufficient.

Signal Processing

Fast Fourier Transform (FFT) decomposes the vibration spectrum into frequency components. Bearing defect frequencies (BPFO, BPFI, BSF, FTF) are calculated from geometry and appear as predictable spectral peaks. Envelope analysis demodulates high-frequency resonance to isolate impulsive fault energy. Cepstral analysis separates source and transmission path effects in gear diagnostics.

Machine Learning Methods

Anomaly detection: isolation forest, autoencoders, LSTM autoencoders track deviation from a healthy-machine baseline without requiring labelled failure data. Classification models (SVM, random forest, gradient boosted trees) identify fault type once labelled examples exist. Remaining Useful Life (RUL) prediction: survival models, LSTM networks trained on run-to-failure data estimate time to next intervention.

Integration & Alerting

Results fed to existing SCADA / PLC HMI via OPC UA, MQTT, or REST API. Edge deployment (on-site compute) for latency-sensitive applications. Cloud deployment for fleet-wide cross-asset models trained on data from multiple sites. Configurable alarm levels: early warning (trend), advisory (schedule soon), urgent (inspect this week).

Métricas de ROI documentadas

Metric Improvement Source
Unplanned downtime −30 to −50% Deloitte / DOE studies
Maintenance costs −12 to −25% MarketsandMarkets 2024
Emergency repairs −60 to −80% Industry aggregate
Overall Equipment Effectiveness (OEE) +10 to +25% Grand View Research
Spare parts inventory −30 to −50% Industry average
Payback period 6–36 months Process-dependent; high-criticality assets 6–18 mo
Adopters reporting positive ROI 95% Grand View Research 2024

Reactivo vs. Preventivo vs. Predictivo

Aspect Reactive Preventive Predictive
When action taken After failure On fixed schedule When data indicates need
Cost driver Emergency repairs + downtime Over-maintenance + wasted parts Monitoring system investment
Downtime Unplanned, long Planned, sometimes unnecessary Planned, minimal
Parts replaced When failed By time interval By actual condition
Operational risk High Medium Low
Typical savings vs reactive Baseline 10–25% cost reduction 25–50% cost reduction

Patrones de diseño relevantes

Reservar llamada

Si sospecha un problema de fiabilidad o quiere configurar la monitorización antes de la próxima parada — una llamada de 30 minutos es suficiente para evaluar qué señales recoger.

Reservar llamada →