Predictive Maintenance System

An industrial IoT application that uses machine learning to predict equipment failures before they happen

---

Screenshots

Dashboard	Machine Alerts

Sensor Chart & Table	Upload Data	Model Metrics

Confusion Matrix

---

Overview

Unplanned machine downtime is one of the costliest problems in industrial operations. This system uses a trained Random Forest classifier on the AI4I 2020 Predictive Maintenance Dataset to analyze real-time sensor readings and predict the probability of machine failure before it happens.

The backend exposes a FastAPI REST API for both single and batch predictions. The frontend is a React + Vite dashboard with live health status cards, sensor charts, alert panels, and a CSV batch upload tool.

---

Features

• Real-time Failure Prediction — Submit live sensor readings and get instant failure probability, risk score, and health status
• Batch CSV Upload — Upload a CSV file of sensor readings and get predictions for every row at once
• Health Status Classification — Each machine is labelled Healthy, Warning, or Critical based on risk score thresholds
• Actionable Recommendations — Rule-based engine surfaces specific maintenance guidance (e.g. "Inspect cooling system and consider tool replacement")
• Interactive Sensor Charts — Recharts-powered visualizations of torque, temperature, tool wear, and rotational speed across machines
• Model Metrics Dashboard — Accuracy, Precision, Recall, F1 Score, ROC-AUC, feature importances, and confusion matrix — all served live from the model bundle
• Responsive UI — Clean React dashboard with gradient layout, card-based health indicators, and alert panels

---

Tech Stack

Backend

Technology	Badge	Purpose
Python 3.10+		Core language
FastAPI		REST API framework
Uvicorn		ASGI server
scikit-learn		Random Forest classifier
Pandas		Data processing & CSV parsing
NumPy		Numerical operations
joblib		Model serialization (.pkl)
Matplotlib		Confusion matrix plot generation
Pydantic		Request schema validation
python-multipart		CSV file upload handling

Frontend

Technology	Badge	Purpose
React 19		UI framework
Vite 8		Build tool & dev server
React Router DOM v7		Client-side routing
Recharts		Sensor data bar/line charts
Axios		HTTP client for API calls

---

Project Structure

predictive-maintenance-system/
│
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI app & all route definitions
│   │   ├── schemas.py           # Pydantic input schema (SensorDataInput)
│   │   ├── ml/
│   │   │   ├── predictor.py     # Inference, risk scoring & recommendation engine
│   │   │   └── model.pkl        # Pre-trained RF bundle (model + metrics + features)
│   │   └── static/
│   │       └── confusion_matrix.png
│   ├── train_model.py           # Training, evaluation & artifact generation script
│   └── requirements.txt
│
├── frontend/
│   ├── index.html
│   ├── vite.config.js
│   ├── package.json
│   └── src/
│       ├── App.jsx
│       ├── main.jsx
│       ├── pages/
│       │   ├── Dashboard.jsx    # Main health monitoring view
│       │   ├── UploadData.jsx   # Batch CSV prediction page
│       │   └── Metrics.jsx      # Model performance & feature importance
│       ├── components/
│       │   ├── Layout.jsx       # App shell & navigation bar
│       │   ├── HealthCard.jsx   # Summary stat cards
│       │   ├── AlertCard.jsx    # Warning & Critical alert items
│       │   ├── MachineTable.jsx # Full predictions table
│       │   └── SensorChart.jsx  # Recharts sensor visualization
│       └── services/
│           └── api.js           # Axios base URL configuration
│
├── data/
│   └── ai4i2020.csv             # UCI training dataset (10,000 rows)
│
└── screenshots/

---

Machine Learning Model

Detail	Value
Algorithm	Random Forest Classifier
Estimators	200 trees
Max Depth	10
Class Weights	Balanced (handles ~3.4% positive rate)
Train / Test Split	80% / 20% (stratified)
Features	Air temp, Process temp, Rotational speed, Torque, Tool wear
Target	Binary Machine failure label

Risk Score Thresholds

Risk Score	Status	Action
≤ 30	✅ Healthy	Continue monitoring
31 – 70	⚠️ Warning	Schedule inspection
> 70	🔴 Critical	Immediate action required

Rule-Based Recommendation Engine

Sensor Condition	Recommendation
Process temp > 310 K and Tool wear > 200 min	Inspect cooling system and consider tool replacement
Torque > 50 Nm and Rotational speed < 1400 rpm	Check motor load and shaft alignment
Air temp > 305 K and Torque > 55 Nm	Inspect overheating and vibration-related components
Default	No immediate action required. Continue monitoring

---

Getting Started

Prerequisites

---

1. Clone the Repository

git clone https://github.com/Dhanush-1213/predictive-maintenance-system.git
cd predictive-maintenance-system

---

2. Backend Setup

cd backend

# Create and activate virtual environment (recommended)
python -m venv venv
source venv/bin/activate          # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# (Optional) Retrain the model — a pre-trained model.pkl is already included
python train_model.py

# Start the API server
uvicorn app.main:app --reload

API → http://127.0.0.1:8000
Swagger Docs → http://127.0.0.1:8000/docs

---

3. Frontend Setup

Open a new terminal:

cd frontend
npm install
npm run dev

App → http://localhost:5173

Make sure the backend server is running before launching the frontend.

---

📖 Usage

Dashboard

Click "Run Sample Predictions" to run the model against 4 built-in machines with varied sensor profiles:

• View health summary cards (Total / Healthy / Warning / Critical)
• Inspect per-sensor readings in an interactive bar chart
• Review real-time alerts for Warning and Critical machines
• Browse the full prediction table with risk scores and recommendations

Upload Data (Batch Prediction)

Navigate to Upload Data to submit a CSV with multiple sensor readings at once.

Required CSV columns:

Column Name	Alternate (auto-renamed)	Description
Air temperature [K]	Air_temperature_K	Ambient temperature in Kelvin
Process temperature [K]	Process_temperature_K	Process temperature in Kelvin
Rotational speed [rpm]	Rotational_speed_rpm	Spindle speed
Torque [Nm]	Torque_Nm	Applied torque
Tool wear [min]	Tool_wear_min	Cumulative tool wear in minutes

The API accepts both bracket-style and underscore-style column names — export directly from any SCADA or historian system.

Model Metrics

Navigate to Metrics to view live model performance fetched from the model bundle: Accuracy, Precision, Recall, F1 Score, ROC-AUC, feature importance rankings, and the confusion matrix.

---

API Reference

GET /

Health check. Returns { "message": "Predictive Maintenance API is running" }.

---

GET /metrics

Returns model evaluation metrics and feature importances directly from the saved model bundle.

---

POST /predict

Single machine prediction.

Request:

{
  "Air_temperature_K": 300.0,
  "Process_temperature_K": 310.5,
  "Rotational_speed_rpm": 1500,
  "Torque_Nm": 45.0,
  "Tool_wear_min": 120
}

Response:

{
  "failure_prediction": 0,
  "failure_probability": 0.0821,
  "risk_score": 8.21,
  "health_status": "Healthy",
  "recommendation": "No immediate action required. Continue monitoring."
}

---

POST /predict-batch

Batch prediction from a CSV file upload.

Request: multipart/form-data with a .csv file in a field named file

Response:

{
  "total_rows": 100,
  "predictions": [
    {
      "row_id": 0,
      "failure_prediction": 1,
      "failure_probability": 0.8342,
      "risk_score": 83.42,
      "health_status": "Critical",
      "recommendation": "Inspect cooling system and consider tool replacement."
    }
  ]
}

---

Dataset

AI4I 2020 Predictive Maintenance Dataset — UCI Machine Learning Repository

• 10,000 data points simulating industrial milling machine sensor behavior
• 5 features used for training after dropping metadata and sub-failure-type columns
• ~3.4% positive (failure) rate — handled via class_weight="balanced"

Matzka, S. (2020). Explainable Artificial Intelligence for Predictive Maintenance Applications. IEEE AIKE.

---

Future Improvements

• WebSocket support for live streaming sensor data
• Time-series anomaly detection model (LSTM / Isolation Forest)
• Docker + Docker Compose for one-command deployment
• Authentication & multi-user support
• Historical prediction logging with a database backend
• Email / Slack alert notifications for Critical machines
• Export batch prediction results as CSV

---

Author

Dhanush K
PES University | CSE (AIML)
LinkedIn: @Dhanush K

---

⭐ If you found this project helpful, give it a star!

Built with ⚙️ FastAPI · ⚛️ React · 🌲 scikit-learn