IDC RIOP - Industrial Machinery Forecasting

N-BEATS deep learning for predictive maintenance of Desfibradora and Picadora equipment.

---

🚀 Quick Start - User Interface

The recommended way to use the pipeline is through the graphical user interface.

1. Installation & Launch

# Create and activate virtual environment (if not already done)
python -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1

# Install dependencies
pip install -r requirements.txt

# Launch the UI
python launch_app.py

2. Workflow via UI

The application is organized into tabs for each step of the process:

🔄 1. SharePoint Sync (Data Ingestion)

• Go to the SharePoint tab.
• Click "Descargar datos crudos" to fetch the latest performance (EPI) and dimension data.
• Note: Raw sensor data (DATOS_DESF_CLEANED.csv, DATOS_PICADORA_CLEANED.csv) should be placed manually in the data/ folder if not available on SharePoint.

🧠 2. Training

• Go to the Training tab.
• Select a Machine (DESF, PICADORA, or ALL).
• Choose Aggregation:
  • Mean: Best for trends and general forecasting.
  • Max: Best for peak detection and alarm thresholds.
• Click "Iniciar Entrenamiento".
• Tip: Use "Forzar reentrenamiento" to ensure models use the latest data.

📈 3. Forecasting & Visualization

• Go to the Forecast tab.
• Select Machine and Horizon (e.g., 2_days, 5_days).
• Click "Generar Pronósticos".
• Local Visualization:
  • Click the "👁 Ver" button next to the horizon selector to see an interactive chart of data coverage vs. forecast horizon.
  • Detailed plots are saved to forecast_visuals/{MACHINE}/.

🔍 4. Anomaly Detection

• Go to the Anomaly tab.
• Detect and clean anomalies in sensor data before training/forecasting.
• Results are saved to anomaly_results/ and data/anomaly/.

---

📂 Data & Output Locations

Input Data

• Raw Sensor Data: Place your CSV files in data/ (e.g., data/DATOS_DESF_CLEANED.csv).
• EPI/Performance Data: Synced from SharePoint to data/epi/.

Outputs

Content	Location	Description
Forecast CSVs	data/forecasts/{MACHINE}/	Combined forecast files (e.g., FORECAST_DESF_2_DAYS.csv).
Visualizations	forecast_visuals/{MACHINE}/	Individual plots (.png) and interactive charts.
Anomaly Reports	anomaly_results/{MACHINE}/	Anomaly detection plots and summaries.
Power BI Data	data/anomaly/	CSV exports formatted for Power BI dashboards.

---

💻 Command Line Interface (Advanced)

For automation or server environments, you can still use the CLI.

Complete Pipeline

# Train and forecast all machines
python run_nbeats_pipeline.py

# Train and forecast specific machine
python run_nbeats_pipeline.py --machine PICADORA

Training Only

# Train all models (MEAN aggregation - recommended)
python -m nbeats.nbeats_train --batch --machine ALL --force --agg mean

# Train with MAX aggregation (for peak detection)
python -m nbeats.nbeats_train --batch --machine ALL --force --agg max

Forecasting Only

# 2-day forecast (48h)
python -m nbeats.nbeats_forecast --batch --horizon 2_days

# 5-day forecast (120h)
python -m nbeats.nbeats_forecast --batch --horizon 5_days

Anomaly Detection

# Run detection on a specific file
python -m anomaly.anomaly_detector --input processed/DESF/corriente_motor_a.parquet

---

🔧 Project Structure

IDC_RIOP/
├── launch_app.py           # UI Entry point
├── ui/                     # User Interface code
├── data/                   # Input/output data
│   ├── epi/                # Performance data (from SharePoint)
│   ├── forecasts/          # Generated forecast CSVs
│   └── raw/                # Raw sensor data
├── forecast_visuals/       # Forecast plots (PNG/HTML)
├── anomaly_results/        # Anomaly detection reports
├── models/                 # Trained N-BEATS models
├── nbeats/                 # N-BEATS core logic
├── common/                 # Shared utilities (SharePoint, Data Prep)
└── requirements.txt        # Python dependencies

---

📋 Requirements

• Python 3.10+
• Dependencies listed in requirements.txt
• SharePoint credentials (in .env) for data sync features.