XAI Manufacturing Inspector Pro

An Explainable AI (XAI) system for automated surface defect detection in manufacturing — providing transparent, human-readable decision-making for quality assurance.

---

🧠 What It Does

XAI Manufacturing Inspector Pro takes real-time sensor readings from industrial machines and predicts whether a machine is likely to pass or fail quality inspection. More importantly, it explains why — using SHAP and LIME to break down which sensor readings are driving the risk, and providing actionable maintenance recommendations.

---

✨ Features

• Real-time failure prediction — XGBoost classifier trained on 7 sensor inputs
• Explainable AI (XAI) — SHAP TreeExplainer + LIME for transparent predictions
• Sensor impact analysis — Ranked feature contributions with directional reasoning
• Actionable recommendations — Specific maintenance guidance based on sensor thresholds
• Dataset analytics — Correlation heatmap & distribution plots (Matplotlib + Seaborn)
• Interactive dashboard — D3.js & Chart.js powered frontend with KPI cards
• Multiple ML models — Linear/Logistic Regression, Random Forest, Decision Tree, XGBoost

---

🗂️ Project Structure

XAI-Manufacturing-Inspector-Pro/
├── backend/
│   ├── main.py              # FastAPI app — prediction, training & analytics endpoints
│   ├── models.py            # ML model training + SHAP/LIME explanation functions
│   ├── dataset.py           # Synthetic sensor dataset generation
│   ├── predict.py           # Standalone prediction utilities
│   ├── train_models.py      # Script to train and save the XGBoost model
│   ├── inspect_data.py      # Data inspection helpers
│   └── requirements.txt     # Python dependencies
├── frontend/
│   ├── src/
│   │   └── server.js        # Express.js server
│   ├── views/               # EJS templates
│   ├── public/
│   │   └── js/
│   │       └── charts.js    # D3.js / Chart.js dashboard visualisations
│   └── package.json
├── models/
│   └── xgboost_classifier.joblib   # Saved trained model (generated after training)
└── README.md

---

⚙️ Tech Stack

Layer	Technology
ML / XAI Backend	Python, FastAPI, XGBoost, scikit-learn, SHAP, LIME
Data & Analytics	Pandas, NumPy, Matplotlib, Seaborn
Frontend Server	Node.js, Express.js
Templating	EJS
Charts / Visuals	Chart.js, D3.js
API Communication	Axios

---

---

🚀 Getting Started

Prerequisites

• Python 3.12+
• Node.js 18+
• Docker (optional, for containerised deployment)

---

1. Local Development Setup

Backend Setup

cd backend

# Install Python dependencies
pip install -r requirements.txt

# Train the XGBoost model (saves to models/xgboost_classifier.joblib)
python train_models.py

# Start the FastAPI server on port 8000
uvicorn main:app --reload --port 8000

Frontend Setup

cd frontend
npm install
npm start

Go to: http://localhost:5000

---

🐳 Docker Deployment (Recommended)

Run the backend via Docker for a consistent, production-ready environment:

cd backend
docker build -t xai-backend .
docker run -p 8000:8000 xai-backend

---

🚢 Cloud Deployment

1. Backend (Railway [previously] => Render)

This project is configured for seamless deployment on Railway using the provided backend/Dockerfile.

2. Frontend (Vercel)

Deployed the Express/EJS frontend to Vercel:

Check Out: https://xai-manufacturing-inspector-pro.vercel.app

📱 Responsiveness

The dashboard is fully responsive and optimized for:

• 💻 Desktop: Full multi-chart grid view.
• 📱 Mobile: Collapsible tables and stacked KPI cards designed for screens as small as 320px (iPhone SE).
• 📊 Dynamic Scaling: Canvas-based charts (Chart.js) and SVG (D3.js) adjust size automatically.

---

📡 API Endpoints

Method	Endpoint	Description
GET	/	Health check
POST	/predict	Predict failure from sensor readings
POST	/train	Re-train the XGBoost model on fresh data
GET	/dataset-analytics	Returns correlation heatmap & distribution plots (base64 PNG)

/predict — Request Body

{
  "Temperature": 85.0,
  "Pressure": 95.0,
  "Speed": 1800.0,
  "Vibration": 3.2,
  "Humidity": 45.0,
  "Power_Consumption": 160.0,
  "Material_Hardness": 65.0
}

/predict — Response

{
  "status": "Fail",
  "failure_probability": 0.87,
  "prediction": 1,
  "model_confidence": 74.0,
  "top_risk_factor": "Vibration Level",
  "risk_factors_count": 3,
  "weaknesses": ["Vibration Level", "System Temperature", "Internal Pressure"],
  "improvement_plan": ["Inspect bearings for excessive wear..."],
  "explanation": {
    "features": [...],
    "recommendation": "..."
  },
  "detailed_summary": {
    "problem": "...",
    "cause": "...",
    "solution": "..."
  }
}

---

🔬 Sensor Thresholds

Sensor	Risk Threshold	Risk Reason
Temperature	> 90 °C	Thermal expansion / cooling failure
Vibration	> 4.0	Bearing wear / spindle misalignment
Power Consumption	> 180 W	Electrical fault / overload
Pressure	> 100 bar	Valve blockage / line leak
Speed	> 2000 RPM	Mechanical stress on moving parts
Humidity	> 60 %	Corrosion / moisture ingress risk
Material Hardness	> 75	Excessive tooling wear

---

🤖 XAI Explainability

• SHAP (SHapley Additive exPlanations): Uses TreeExplainer to attribute each sensor's contribution to the final prediction, returning a ranked list of feature impacts with direction (positive/negative).
• LIME (Local Interpretable Model-agnostic Explanations): Fits a locally linear model around the prediction point to produce intuitive, instance-level explanations.

---

📊 Analytics

The /dataset-analytics endpoint returns two charts:

1. Sensor Correlation Heatmap — shows which sensors correlate with each other and with failure
2. Temperature & Vibration KDE Distribution — overlaid density plots comparing Pass vs Fail distributions

---

📄 License

This project was developed as a Web Technology semester project (Sem 6).