Real-time anomaly detection
Continuous monitoring of sensor signatures. Detection of deviations from the learned nominal distribution, with a criticality score.
Isolation Forest · Autoencoder · LSTM · One-class SVM
Domain of expertise
Predictive maintenance: from sensors to failure prediction
Predictive maintenance pipelines built from your SCADA, historian or IoT sensor data: vibration, temperature, motor current, pressure. From time series ingestion to the alert in your CMMS.
Capabilities
What we build
Remaining useful life (RUL) estimation
Degradation modelling on critical components: bearings, seals, blades. Prediction horizon configurable to the degradation dynamics.
LSTM · Transformer · Parametric survival · Weibull
Failure mode classification
Identification of the likely failure type: imbalance, cavitation, abrasive wear, inner / outer race bearing fault.
FFT · envelope · Random Forest · multi-class SVM
Machine health dashboard
Criticality score per equipment, anomaly history, trends on key features, fleet view for the maintenance manager.
Grafana · Plotly · Real-time REST API
CMMS integration
Automatic work order creation when the threshold is crossed. Priority, symptom description and affected equipment pre-filled.
SAP PM · IBM Maximo · Infor EAM · REST API
PM interval optimization
Based on observed degradation, recalibration of existing preventive maintenance intervals to reduce unnecessary interventions.
Survival analysis · degradation models
Architecture
Reference pipeline
Time series ingestion → feature engineering → detection / estimation → alert → CMMS action. Designed to run on existing historians without interrupting production.
Sources
Sensors & historian
Accelerometers, motor current, temperature, pressure, flow. OPC-UA, Modbus, OSIsoft PI, InfluxDB, Ignition.
Feature eng.
Feature extraction
RMS, kurtosis, skewness, FFT, spectral envelope, frequency bands. Python / Apache Spark depending on volume.
Model
Detection / prediction
Models trained on the labelled history. Thresholds calibrated to the cost of a false alarm vs. an unplanned stoppage.
Action
Alerts & CMMS
Email/SMS/Teams notifications, automatic work order creation, escalation by criticality. Traceable logs for auditability.
Target equipment
What we instrument
Applicable as soon as a history of sensor data on past failures exists: even partial. We help you label the historical data.
Electric motors
Centrifugal pumps
Compressors
Industrial fans
Mills and crushers
Conveyors
Hydraulic presses
Industrial furnaces and ovens
Results
Observed orders of magnitude
Measured on comparable deployments. They depend on the quality of the historical data, the type of equipment and the level of documentation of past failures.
2 – 6 weeks
of lead time before the visible symptom on targeted failures
30 – 60 %
reduction in unplanned stoppages observed on instrumented fleets
+ 15 – 30 %
machine availability through PM interval optimization
< 5 %
false positives after threshold calibration by equipment criticality
Approach
How we proceed
01
Data audit
Inventory of sensor sources, quality and granularity of the historical data, labelling of past failures. 1 week.
02
Feature engineering
ETL pipeline, extraction of relevant features per equipment type, performance baseline. 2 to 4 weeks.
03
Prototype
Training and cross-validation on the history, calibration of alert thresholds, health dashboard. 4 to 6 weeks.
04
Deployment
SCADA / historian / CMMS integration, alerts, go-live on pilot equipment. 2 to 4 weeks.
05
MLOps
Drift detection on feature distributions, retraining on new failures, continuous improvement.
Get started
An equipment fleet to instrument?
Tell us about the critical equipment, the available data and the failure history. We reply with a feasibility analysis within 48 hours.