>
XIMETRICS
What it does Data Architecture Post-incident analysis Demo Request a demo
EN | ES | RU
Predictive analytics

Detect hidden losses before they become real costs

XIMETRICS detects early signals leading to breakdowns, downtime, defects, and overconsumption.

Request a demo How it works
Runs on top of your existing infrastructure
Data stays on-premise
From a faint signal to an action
Anomaly

Changepoint detected

Incident opened

Recommendation sent

What problems and risks the system catches

Catching risk early is cheaper than dealing with the fallout

Equipment-level anomalies

  • Unusual vibration profiles
  • Slow drift of parameters toward failure zones
  • Early forecast of breakdowns and critical states

Process anomalies and sabotage

  • Operating outside the corridors of the SOP
  • Differences between formally identical modes
  • Defect «checkerboard» across shifts, products, and lines

Behavioral anomalies in telemetry

  • Recurring sensor outages and failures
  • Unusual gaps and «holes» in telemetry
  • Repeating patterns tied to a specific scheme

Data and reporting anomalies

  • «Perfect» metrics where they shouldn't be possible
  • Mismatches between facts, telemetry, and manual reports
  • Signs of after-the-fact adjustments

What data it works with

No industry is hard-wired into the product.

What matters is the data format: signals, events, metrics, and related parameters

Sensor signals

Temperature, pressure, current/voltage, flow, power, load, and other numeric signals.

Supported types
thermocouple RTD Pt100 Pt1000 thermistor IR pyrometer thermal camera pressure gauge differential pressure vacuum gauge Coriolis vortex flowmeter ultrasonic flowmeter electromagnetic radar level capacitive hydrostatic accelerometer proximity probe tachometer encoder strain gauge torque LVDT laser rangefinder inclinometer CT power analyzer energy meter acoustic emission ultrasound O₂ CO CO₂ CH₄ H₂S NOx pH conductivity ORP turbidity humidity oil quality dust/aerosols machine vision

Multi-signal processes

Linked parameters where an anomaly shows up through the dynamics of several signals.

Typical pairings
temperature + pressure vibration + motor current flow + pressure drop RPM + vibration voltage + current power + RPM feed + pump pressure temperature + coolant flow process line KPIs process mode stationary / transient start-ups & shutdowns mass balance heat balance energy balance recipe PID loops control setpoints input–output linkage signal correlations process chart nominal point deviation from mode multivariate anomaly multivariate statistics unit dependencies

Event and alarm logs

Logging streams of all kinds, operator actions, and event sequences.

Log sources
SCADA alarms PLC alarms OPC alarms IEC 61850 GOOSE sequence of events (SOE) operator log shift log downtime log start-ups & shutdowns mode transitions setpoint changes recipe changes acknowledgements escalations MES events LIMS records quality log maintenance log repair history work permits shift reports equipment failures warnings incidents syslog Windows Event Log auditd manual rounds handwritten logs

Infrastructure metrics

Ready-made KPIs from your systems: efficiency, energy use, consumption, logistics.

What we collect
OEE line throughput good-output yield defect rate product quality downtime MTBF / MTTR cycle time equipment utilization run-hours motor hours MTBM specific raw-material consumption electricity consumption active / reactive power cos φ grid quality harmonics efficiency output grid losses transformer load fuel consumption steam consumption compressed-air consumption room temperature & humidity water consumption gas consumption HVAC status UPS OTIF delivery SLA fleet utilization fleet downtime fleet mileage order processing time inventory turnover

Event streams

Access events, passages, actions, and status changes.

Stream sources
access control turnstile passages checkpoint passes gate & door openings operator logins access denials security alarm events fire system events video analytics vehicle recognition line start-ups & shutdowns operator shifts recipe change equipment power on/off power source switching valve & gate state manual operation confirmations site rounds loading / unloading barcode scans RFID tags GPS track geofences & geo-events route deviations speeding vehicle movement personnel movement orders shipments receivings inventory status changes

When needed, we plug in intelligent recognition of handwritten records.

On-premise and under your control

Your data never leaves the company perimeter

On-premise deployment

Runs inside your own infrastructure. No mandatory cloud. External providers are connected only when you say so.

Data control

Read-only. Minimum required access rights.

AI on customer's terms

AI is off by default. OpenAI, Anthropic, Telegram, Slack and other external services are enabled explicitly. For local AI we use Ollama.

No mandatory cloud

External integrations are turned on only at your decision.

CSV, Excel, SQL, Kafka, MQTT, Prometheus, etc.

Plug-in mechanism for connecting sources

Read-only mode

The preferred way to access your data

Results → BI, reports, alerts

Data flows to where it's actually needed

How it works

From data to action: four layers of the system

First layer

Data sources

Sensors, PLC, SCADA, MES, ERP, logs, CSV/Excel/Parquet files, SQL, Kafka/MQTT, Prometheus. The system runs on top of your existing infrastructure.

Second layer

Analytics

A profile of normal built on historical data. Anomaly detection: behavior changes, drift, alarm cascades, rate exceedance. Time alignment, gap filling, derivatives, rolling statistics, lags.

Third layer

Qualification

Every deviation is qualified by type and significance. Anomaly search, forecasting, signal-to-signal links, factor-impact explanations, changepoint detectors. An incident is built with context and timeline.

Fourth layer

Output

Alerts and notifications. Reports. Data for BI and dashboards. API/WebSocket, reports, 3D visualization. AI interpretation of the result and recommendations for staff.

Typical anomaly detection scenarios

What signals Ximetrics catches

Post-incident analysis follows the same principles as decoding aviation «black boxes»

Changepoint

Signal behavior change

An abrupt shift to a different behavior regime. The system pinpoints the moment when the signal's statistical properties changed.

Drift

Parameter drift

A slow, unnoticed drift away from normal. Every value is within tolerance, but the trend leads to a problem.

Envelope

Out of envelope

The signal leaves the dynamic statistical corridor built from its own behavioral history.

Alarm cascade

Alarm cascade

A burst of linked alarms within a narrow time window. The system chains them together and isolates the root cause.

Control-response

No expected response

A control action was applied, but the expected change in the parameter never followed.

Mode transition

Risky mode transition

A switch between operating modes shows up with atypical parameter behavior.

Methodology

Anomaly detection

For each specific task we tailor a stack of statistical, analytical, and predictive algorithms based on the nature of the process and the data.

Statistics and corridors
z-score, MAD, IQR, adaptive quantile ranges
Changepoint and decomposition
CUSUM, Bayesian online, PELT, STL, Fourier/wavelet
Forecast and deviations
Prophet, LSTM/GRU, ARIMA — divergence between actual and forecast
ML detectors and links
Isolation Forest, One-Class SVM, autoencoders, correlation analysis

XIMETRICS reconstructs how the incident unfolded.

Output: ranked critical points, timeline, incident phases, reconstruction, and data ready for review in BI or a report.

Timeline and phases
reconstructing the order of events, links to operating modes and setpoints
Root cause
alarm cascade analysis, impact propagation, cause-and-effect chains
Asset context
spec data, maintenance and repair history, adjacent and similar incidents
Analytical methods
changepoint, decomposition, causal analysis, actual-vs-forecast divergence

More than just detection

Forecasting, analytics, and visualization

Forecasting

Prophet and LSTM/GRU for signal forecasting. Spotting actual-vs-forecast divergence.

Relationship analytics

Correlations, lagged dependencies, operating-mode clusters, and explanations of factor impact.

Reports

Interactive HTML reports with charts. For post-incident analysis, audit, and handing off the result.

Monitoring

A web interface for live process observation and rapid response.

3D visualization

3D visualization of process trajectories for tasks with spatial parameter dynamics.

XIMETRICS forecasts, explains, and visualizes in the format that fits the task.

Some of the analyzers and visualizations depend on the configuration you choose.

A result in a single meeting

How the demo works

Request

You submit a demo request

Scheduling

We agree on a date and time for the demo

Data export

CSV/Excel/Parquet, or any pre-agreed export.

Demo session

We show the result on your data in a single meeting

Decision

You decide whether to take the next step

No infrastructure connection required — works on anonymized data.

Request a demo

If the demo confirms the value, the next step is:

A pilot project in 5 stages

1
2–3 days

Scope and scenarios

We define the area, process, or equipment. Agree on data sources and the expected outcome. Focus on the zones of largest losses or suspicions.

2
3–5 days

Data connection

Integration with existing systems and telemetry. Loading historical data. On-premise deployment inside your infrastructure.

3
3–5 days

Profile building

The system builds a profile of normal behavior on historical data. Calibration to the specifics of the pilot area. We surface deviations, suspicious patterns, and bottlenecks.

4
1–2 weeks

Monitoring and detection

The system runs live. We work with process engineers, quality, and security teams. Assessment of impact on defects, losses, downtime, and risk.

5
2–3 days

Results assessment

We review the detected incidents and assess accuracy. Build a vulnerability and impact map. Calculate the economics of scaling. Agree on next steps.

Demo on your own data

Get a result
on your own data

Demo in a single meeting

We'll show Ximetrics on your data export — no infrastructure connection required.

If the demo confirms the value, we'll discuss a pilot project.

On your real data, not a canned demo dataset
A single meeting, no upfront integration
Concrete findings about your processes, not product slides

Request a demo

No commitment — we'll just have a conversation if it sounds interesting.