Predictions & Anomaly Detection

AI-powered remaining useful life (RUL) estimation, anomaly detection, and predictive narratives.

Fleet predictions

GET /machines/predictions/fleet

Auth: Bearer JWT (any role)

Returns RUL predictions for all machines in the fleet.

{
  "predictions": [
    {
      "machine_id": "uuid",
      "machine_name": "CNC Mill #1",
      "health_score": 82,
      "estimated_days_to_failure": 45,
      "risk_level": "medium",
      "top_risk_metric": "vibration"
    }
  ]
}

Machine prediction

GET /machines/{machine_id}/prediction

Auth: Bearer JWT (any role)

Parameter	Type	Default
language	string	en

Detailed prediction for a single machine including per-metric health scores.

AI narrative

GET /machines/{machine_id}/prediction/narrative

Auth: Bearer JWT (any role)

Returns an AI-generated natural language summary of the machine's health status and predicted trends. Supports multiple languages.

{
  "machine_id": "uuid",
  "narrative": "CNC Mill #1 is showing early signs of bearing degradation. Vibration levels have increased 15% over the past 2 weeks, with the X-axis showing the most significant change. At current degradation rates, bearing replacement is recommended within the next 30-45 days to avoid unplanned downtime.",
  "language": "en",
  "generated_at": "2026-04-04T10:30:00Z"
}

Prediction history

GET /machines/{machine_id}/prediction/history

Auth: Bearer JWT (any role)

Parameter	Type	Default	Range
days	int	30	1-365

Returns historical prediction snapshots to track how health scores and RUL estimates evolve over time.

Anomaly summary

GET /machines/{machine_id}/anomalies/summary

Auth: Bearer JWT (any role)

Returns anomaly event counts for the last 7 days, broken down by metric.

{
  "machine_id": "uuid",
  "period_hours": 168,
  "anomalies": [
    {"metric_name": "vibration", "count": 12, "latest_at": "2026-04-04T09:15:00Z"},
    {"metric_name": "temperature", "count": 3, "latest_at": "2026-04-03T22:00:00Z"}
  ],
  "total_count": 15
}

Detection methods

Haltless uses three complementary anomaly detection methods:

Method	Description	Parameters
Static baseline	Z-score against 7-day mean/stddev	Threshold: 3.0 std devs
EWMA	Exponentially weighted moving average	Span: 20, warmup: 10 readings
Rate-of-change	First-difference velocity analysis	Window: 5, z-threshold: 3.5

Anomaly sub-scores are blended into the composite health score:

health_score = (0.7 * alert_score) + (0.3 * anomaly_score)