Module · Predictive Maintenance

Q-PdM

AI predictive maintenance — configurable in minutes, explainable by design.

Map your assets, point at the MQTT tags you already have, and let Q-PdM watch for the failure patterns hidden in your signals. Nine complementary detection methods, declarative analysis pipelines, peer-to-peer comparison, offline replay and full maintenance lifecycle — without ML PhDs and without per-asset license fees.

Request Information See it in action

What it does

A full predictive maintenance stack — asset health, anomaly detection, root-cause hints, maintenance workflow and offline replay — wired straight into the same MQTT signals your plant already publishes.

Health Score 0–100

Per-asset score with Healthy / Warning / Degrading / Critical levels and explainable contributors.

9 Detection Methods

Z-score, slope, EWMA, CUSUM, volatility, time-at-threshold, activeTooLong, stuckState, transition frequency.

Peer-to-Peer Comparison

Compare each asset against its group — relative ratio, peer z-score, rank, deviation score.

Declarative Pipelines

No-code analysis pipelines: rolling stats, filters, gates, correlations, conditional sampling.

Failure Modes

Map each evaluation rule to a named failure mode for actionable diagnostics.

Offline Replay

Re-evaluate historical data from CSV, QUANTUM historian or direct data API to validate tuning.

Maintenance Lifecycle

Log maintenance, reset accumulators, full history per asset, manual or tag-triggered.

AI-Assisted Tuning

Recommends thresholds, windows and weights based on historical behavior of your assets.

MQTT Publish & Annotations

Computed outputs published back to MQTT; anomaly events posted as annotations on charts.

Usage Requirements

Runtime-hours and cycle-count thresholds — trigger inspection windows without needing a sensor.

Configuration-Driven

Define assets, signals and rules in the UI — no code, no model retraining cycles.

Status-Derived Features

Auto-compute fault counts, time-in-state ratios, short-cycle ratios from a process state tag.

Health, scored and explained

Every asset gets a 0–100 health score and a health level — with top contributors and the rule that triggered them.

Healthy 80–100
Warning 60–80
Degrading 40–60
Critical 0–40

Nine complementary detection methods

Different failure modes need different math. Q-PdM ships with nine methods you can enable per signal — each one tuned to a specific failure pattern, and each one explainable.

Z-score Spikes

Catches unexpected deviations: z = (value − mean) / std over a rolling window. Sensor glitches, sudden overloads.

Slope Trend

Linear regression vs. sample index. Bilateral, increasing-only or decreasing-only. Catches gradual degradation Z-score misses.

EWMA Drift

Exponentially weighted moving average. Detects slow baseline shifts, emphasizing recent data.

CUSUM Persistent shift

Cumulative sum control. Detects small, sustained shifts — classic industrial SPC technique.

Volatility Instability

Ratio of recent std vs. baseline std. Many machines get noisier before they fail, even before the mean moves.

Time-at-threshold Sustained

"Above X for Y minutes". Accumulative since last maintenance, or consecutive uninterrupted run.

activeTooLong Digital

For 0/1 signals: equipment ran continuously beyond a duration limit.

stuckState Digital

For 0/1 signals: equipment stuck off for too long — expected to be active and isn't.

Transition frequency Flapping

Counts 0↔1 transitions in a period. Excessive cycling is a classic precursor to relay/contactor failure.

Peer comparison Cross-asset

Compares each asset against its group: relative ratio, peer z-score, rank, deviation score. Spots outliers that look fine on their own.

How it works

From asset definition to live monitoring — every step in one UI.

— Live monitoring —

Q-PdM dashboard with asset cards showing health scores and anomaly flags

Dashboard at a glance

Every asset on one screen, grouped by line or area. Health score, anomaly state and top contributors update live.

  • Cards grouped by line / area
  • One-click maintenance logging from the card
  • Drill-down to evaluation detail and history

— Configuration —

General configuration tab for an asset

Define the asset

Name, tag, line/area, asset type, criticality, nominal sampling, peer group. The basics that drive everything else.

  • Asset taxonomy & criticality 1–5
  • Peer groups for cross-asset comparison
  • Sampling expectations to detect missing data
Runtime-based maintenance configuration with thresholds and output options

Runtime-based requirements

Track running hours since last maintenance. Trigger an inspection window when accumulated runtime hits the threshold — no special sensor required.

  • Configurable runtime condition: >, >=, ==, etc.
  • Threshold with custom output label and severity
  • Reset on maintenance log or tag
Cycle-based maintenance configuration with threshold options

Cycle-based requirements

Count cycles from a monotonic counter or by counting transitions to a specific value. Trigger inspection when the cycle threshold is hit.

  • Cumulative counter or transition-based
  • Short-cycle ratio detection
  • Per-asset reset on maintenance
Analog signal configuration with multiple detection methods enabled

Analog signals

For each analog signal, enable any combination of the nine methods. Each evaluation row has its own failure mode mapping — so the diagnostic is meaningful, not just "anomaly".

  • Z-score, slope, EWMA, CUSUM, volatility per signal
  • Threshold rules (accumulative or consecutive)
  • Per-rule failure mode — on the same row
Discrete (digital) signal configuration with activeTooLong, stuckState, transitionFreq

Discrete signals

For 0/1 signals: activeTooLong, stuckState, transition frequency. Catch contactors that flap, valves stuck open, equipment that ran past its duty cycle.

  • Active too long — accumulative or consecutive
  • Stuck off — failed-to-start detection
  • Excessive cycling — relay/contactor wear precursor
Asset-wide evaluation settings: weights, thresholds, AI tuning toggles

Asset-wide evaluations

Tune the health formula: anomaly weight, drift, volatility, anomaly-rate, event frequency. Toggle annotations and AI tuning per asset.

  • Health weights & anomaly threshold
  • Anomaly-rate over 24h
  • Per-asset overrides for global defaults

— Runtime & diagnostics —

Runtime trigger view showing live signal and evaluation state

Live trigger view

See exactly what fired: which signal, which rule, which value, against which threshold. Anomaly is never a black box.

  • Live signal values overlaid on rule state
  • Rule-by-rule diagnostics
  • Direct link to evaluation detail
Per-asset evaluation detail with all methods and contributions

Evaluation detail

Open any asset and inspect every enabled method, its current value, its threshold and its contribution to the health score. Tuning becomes obvious.

  • Per-method live state
  • Contribution breakdown to overall score
  • Reason strings written for humans
Anomaly log showing chronological events

Anomaly & event log

Every state transition recorded with timestamp, contributors and the rule that fired. Audit-friendly, searchable.

  • Chronological event history
  • Filter by asset / signal / rule
  • Annotations published back to charts
Maintenance log per asset with user, shift, action and notes

Maintenance history

Every maintenance action logged with user, shift, type and description. Maintenance resets accumulators, clears anomaly state and feeds back into peer comparison.

  • Manual log from the card, or auto via reset tag
  • Per-asset 10 most recent actions one click away
  • Audit trail for compliance
AI-assisted tuning recommendations for thresholds and weights

AI-assisted tuning

Q-PdM looks at the asset's recent behavior and recommends thresholds, windows and weights. Accept, edit or ignore — you stay in control.

  • Recommendations grounded in real history
  • Apply per-signal or whole-asset
  • Always reversible
Offline replay panel with date range, source selection and progress

Offline replay

Re-run the whole evaluation engine against historical data — from CSV, QUANTUM historian or direct data API — to validate tuning before it goes live.

  • Continuous timeline with fill-forward
  • Per-asset progress & last-run summary
  • Optional state persist on completion

Where it's used

Built for messy industrial reality — gappy data, noisy sensors, real maintenance crews.

Motors & pumps Compressors & blowers Burners & furnaces Conveyors & drives Valves & actuators Fans & cooling towers Hydraulic systems Bearings & gearboxes Refractory monitoring Filtration & separation Lubrication / oil consumption Energy & utilities equipment

Predictive maintenance, without the platform tax

Q-PdM ships with the Industrial tier of QUANTUM Industrial Studio. No per-asset license, no per-tag fees. Tell us about your assets and we'll scope it with you.