An AI-powered Predictive Maintenance System that predicts industrial machine failure using Machine Learning and real-time sensor data.

The system analyzes machine operating conditions and predicts whether the machine is healthy or has a possible failure risk. It helps industries perform preventive maintenance before unexpected breakdowns occur.

---

Dashboard Preview

---

Project Overview

Industrial machines continuously generate sensor information such as temperature, speed, torque, and tool usage.

Unexpected machine failure can lead to:

• Production downtime
• Increased maintenance cost
• Reduced efficiency

This project applies Machine Learning techniques to predict machine failure conditions and provides maintenance recommendations through a web-based monitoring dashboard.

---

Features

• Real-time machine failure prediction
• Industrial sensor monitoring dashboard
• Failure probability calculation
• AI-based maintenance recommendation
• Interactive slider control
• Manual sensor value input
• Light and Dark theme support
• Machine status monitoring
• FastAPI backend integration

---

Dataset

Dataset:

AI4I 2020 Predictive Maintenance Dataset

Dataset Size:

Total Records : 10000

Normal Machine Records : 9661

Failure Records : 339

---

Input Features

The machine learning model uses:

Feature	Description
Air Temperature	Machine surrounding temperature
Process Temperature	Operating temperature
Rotational Speed	Machine RPM
Torque	Machine load
Tool Wear	Tool usage duration

---

Machine Learning Workflow

Dataset

   ↓

Data Cleaning

   ↓

Exploratory Data Analysis

   ↓

Feature Selection

   ↓

Train Test Split

   ↓

Model Training

   ↓

Model Comparison

   ↓

Best Model Selection

   ↓

Model Deployment

   ↓

FastAPI + React Integration

---

Data Preprocessing

Performed preprocessing steps:

• Removed unnecessary columns
• Checked missing values
• Selected important sensor features
• Feature scaling using StandardScaler
• Stratified train-test splitting

---

Class Imbalance Handling

The dataset contains imbalance:

Healthy Machines : 9661

Failure Machines : 339

Accuracy alone is not enough for this problem.

Methods analyzed:

• SMOTE Oversampling
• Class Weight Balancing
• Stratified Splitting

Final approach:

RandomForestClassifier(
class_weight="balanced"
)

Reason:

Class weight balancing improved failure detection without creating artificial synthetic failure records.

---

Machine Learning Models Compared

Algorithm	Accuracy
Random Forest	98.5%
XGBoost	98.4%
Decision Tree	97.8%
SVM	97.2%
Logistic Regression	96.85%

---

Final Selected Model

Random Forest Classifier

Random Forest was selected because it achieved:

• High accuracy
• Better failure identification
• Good precision and recall balance
• Stable performance on sensor data
• Lower false predictions

---

Model Performance Evaluation

Evaluation metrics used:

• Accuracy
• Precision
• Recall
• F1 Score
• Confusion Matrix

---

Random Forest Classification Report

Class	Precision	Recall	F1 Score	Support
Healthy Machine (0)	0.99	1.00	0.99	1932
Failure Machine (1)	0.87	0.60	0.71	68

---

Overall Performance

Metric	Score
Accuracy	98%
Macro Precision	0.93
Macro Recall	0.80
Macro F1 Score	0.85
Weighted Precision	0.98
Weighted Recall	0.98
Weighted F1 Score	0.98

---

Confusion Matrix

The confusion matrix shows the Random Forest model performance on machine failure prediction.

Explanation:

• Correct Healthy Predictions : 1926
• Correct Failure Predictions : 41
• False Failure Alerts : 6
• Missed Failure Cases : 27

---

System Architecture

React Dashboard

        |
        |
        ↓

FastAPI Backend

        |
        |
        ↓

Machine Learning Model

        |
        |
        ↓

Prediction Result

        |
        |
        ↓

Maintenance Recommendation

---

Backend API

Backend Framework:

FastAPI

Machine Data API

GET /machine

Returns:

• Machine ID
• Temperature
• RPM
• Torque
• Tool Wear

---

Prediction API

POST /predict

Returns:

{
Prediction Status,

Failure Probability,

Maintenance Recommendation
}

---

Frontend Dashboard

Developed using:

• React
• TypeScript
• Axios
• CSS

Dashboard includes:

• Sensor input panel
• Prediction result card
• Machine status table
• Theme switching
• Real-time API communication

---

Technology Stack

Machine Learning

• Python
• Pandas
• NumPy
• Scikit-Learn
• XGBoost
• Matplotlib
• Seaborn
• Joblib

Backend

• FastAPI
• Uvicorn
• Pydantic

Frontend

• React
• TypeScript
• Axios
• CSS

---

Project Structure

AI-Predictive-Maintenance

│
├── backend
│
│   ├── main.py
│   ├── model.pkl
│   ├── scaler.pkl
│   └── requirements.txt
│
│
├── frontend
│
│   └── src
│
│       ├── App.tsx
│       └── App.css
│
│
├── ml
│
│   ├── dataset
│   └── model_training.ipynb
│
│
├── screenshots
│
│   └── dashboard.png
│
└── README.md

---

How To Run

Backend

Move to backend folder:

cd backend

Install dependencies:

pip install -r requirements.txt

Run server:

python -m uvicorn main:app --reload

Server:

http://127.0.0.1:8000

---

Frontend

Move to frontend:

cd frontend

Install packages:

npm install

Start React:

npm run dev

Application:

http://localhost:5173

---

Sample Prediction Output

Machine Status:

Healthy Machine


Failure Probability:

4.62%


Recommendation:

Machine condition is normal.
Continue monitoring.

---

Future Improvements

• IoT sensor integration
• Real-time industrial data streaming
• Cloud deployment
• Alert notification system
• Deep learning based prediction

---

Conclusion

The AI Predictive Maintenance System successfully predicts machine failure risks using industrial sensor data.

The integration of Machine Learning, FastAPI, and React provides a complete end-to-end intelligent monitoring solution for preventive maintenance.

---