Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/BhaveshBytess/PREDICTIVE-MAINTENANCE/llms.txt

Use this file to discover all available pages before exploring further.

​
Overview

Initiates the system calibration process by generating 1,000 healthy sensor samples, building baseline profiles, and training dual anomaly detection models (legacy 1Hz + batch 100Hz). This is the required first step before monitoring or fault injection.

​
Endpoint

POST /system/calibrate

​
Query Parameters

​
asset_id
string
default:"Motor-01"
Asset identifier to calibrate. Defaults to Motor-01 if not specified.

​
Calibration Process

The calibration runs as a background task with the following phases:

​
Phase 1: Burst Data Generation (Instant)

Generates 1,000 healthy sensor samples with timestamps spread across the last hour:
  • Voltage: 230V ± 2V (Range: 226-234V)
  • Current: 15A ± 1A (Range: 13-17A)
  • Power Factor: 0.92 ± 0.02 (Range: 0.88-0.96)
  • Vibration: 0.15g ± 0.03g (Range: 0.09-0.21g)

​
Phase 2: Baseline Profile Construction

Builds statistical profiles for each sensor signal:
  • Min/Max ranges
  • Mean and standard deviation
  • 3-sigma thresholds for anomaly detection

​
Phase 3: Legacy Model Training

Trains 1Hz Isolation Forest on 6 features:
  • voltage_rolling_mean_1h
  • current_spike_count
  • power_factor_efficiency_score
  • vibration_intensity_rms
  • voltage_stability
  • power_vibration_ratio

​
Phase 4: Batch Model Training

Trains 100Hz Isolation Forest on 16 batch features (mean, std, peak-to-peak, RMS) extracted from 100-point windows. This model is primary for inference.

​
Phase 5: Healthy Monitoring

Transitions to MONITORING_HEALTHY state and begins generating healthy data with batch-feature ML detection.

​
Response

​
status
string
required
Always returns "started" when calibration begins successfully
​
message
string
required
Human-readable confirmation message
​
state
string
required
New system state: "CALIBRATING"

​
Example Request

cURL
curl -X POST "https://predictive-maintenance-uhlb.onrender.com/system/calibrate?asset_id=Motor-01"
Python
import requests

response = requests.post(
    "https://predictive-maintenance-uhlb.onrender.com/system/calibrate",
    params={"asset_id": "Motor-01"}
)

data = response.json()
print(f"Status: {data['status']}")
print(f"State: {data['state']}")
JavaScript
const response = await fetch(
  'https://predictive-maintenance-uhlb.onrender.com/system/calibrate?asset_id=Motor-01',
  { method: 'POST' }
);

const data = await response.json();
console.log(`Status: ${data.status}`);
console.log(`State: ${data.state}`);

​
Example Response

200 OK
{
  "status": "started",
  "message": "Calibration started. Generating healthy data...",
  "state": "CALIBRATING"
}

​
Error Responses

{
  "detail": "Cannot calibrate in state 'MONITORING_HEALTHY'. Must be IDLE."
}
Cause: System is not in IDLE state. Resolution:
  1. Call POST /system/stop to return to IDLE
  2. Or call POST /system/purge for a deep reset

​
Monitoring Progress

Poll GET /system/state to track calibration progress:
MessageDurationAction
"Generating training data... 500/1000"~1-2sBurst generation in progress
"Building baseline profile from 1,000 samples..."~1sStatistical profiling
"Training anomaly model on 1,000 samples..."~2sLegacy model training
"Training batch-feature model (16-D statistical features)..."~2sBatch model training
"Calibration complete. Trained on 1000 samples."DoneState → MONITORING_HEALTHY
Total Duration: ~5-10 seconds

​
Validation Metrics

After calibration completes, monitor these metrics via GET /system/state: 
{
  "training_samples": 1000,
  "healthy_stability": 97.3,  // ✅ >95%
  "fault_capture_rate": 100.0
}
If healthy_stability drops below 85%, the baseline contains too much noise. Run POST /system/purge and recalibrate.

​
State Persistence

Calibration data is stored in:
  1. InfluxDB - 1,000 sensor samples with source=calibration tag
  2. In-Memory - Baseline profiles, trained models, degradation state (DI=0.0)
  3. Disk - Model artifacts are not persisted (retraining on restart)
Calibration state is lost on server restart. You must recalibrate after backend deployment or container restart.

​
Use Cases

First-Time Setup

Run calibration immediately after deployment to establish healthy baseline

After Purge

Required after POST /system/purge to rebuild models from scratch

Baseline Drift

Recalibrate weekly/monthly if operating conditions change (e.g., seasonal temperature shifts)

Model Retraining

Retrain models after hardware upgrades or sensor replacements

​
Performance Characteristics

  • Latency: 200ms (calibration runs in background)
  • Background Duration: 5-10 seconds
  • CPU Usage: High during training phase
  • Memory: +50MB for model storage
  • InfluxDB Writes: 100 points (every 10th sample)

Build docs developers (and LLMs) love

Get started for freeTalk to us