Predictive Maintenance: GBdegradation

This model predicts machine failure events based on industrial sensor telemetry.

Artifacts Description

• model.joblib: The HistGradientBoosting champion model.
• preprocessing.py: Python function to transform raw sensor JSON/CSV into features.
• pop_stats.joblib: Baseline statistics for anomaly detection.

How to Use the Pipeline

To get a prediction from raw telemetry, you must use the preprocess_sensor_data function:

import joblib
import pandas as pd
from preprocessing import preprocess_sensor_data

# 1. Load Model and Stats
model = joblib.load('model.joblib')
stats = joblib.load('pop_stats.joblib')
features = joblib.load('feature_list.joblib')

# 2. Raw Data Input (Must have: temperature_c, vibration_mm_s, run_hours_since_maintenance)
raw_df = pd.read_csv('telemetry.csv') 

# 3. Preprocess (Requires 12h history for rolling windows)
processed_df = preprocess_sensor_data(raw_df, population_stats=stats)

# 4. Predict
risk_probs = model.predict_proba(processed_df[features])[:, 1]

Model Performance

• Recall: 1.0 (at 0.8 threshold)
• Primary Driver: Run-hours since maintenance (50% importance)