🧠 Machine Tool Condition Detection using Local Outlier Factor (LOF)

🚀 A Flask Web App for Predicting Machine Tool Health using Anomaly Detection

---

📘 Overview

This project demonstrates the end-to-end development and deployment of a machine learning application that predicts the health condition of a machine tool based on real-time sensor data. The model identifies whether the machine is in a "Normal" state or has a potential "Failure" condition.

The model was trained using the Local Outlier Factor (LOF) algorithm — a robust unsupervised anomaly detection method ideal for highly imbalanced datasets where failure events are rare.

The complete pipeline includes:

• Data preprocessing and feature engineering
• Model selection and training using scikit-learn
• Model serialization and deployment via Flask
• Cloud hosting on Render.com

---

🧩 Problem Statement

Industrial machine tools are equipped with various sensors that monitor their operational parameters. However, failure cases are rare, making supervised classification models ineffective due to class imbalance.

To address this, the project applies anomaly detection to identify deviations from normal operational behavior using features such as:

Feature	Description
Process Temperature	Temperature during operation
Air Temperature	Ambient air temperature
Rotational Speed	RPM of the machine tool
Torque	Applied torque value
Tool Wear	Cumulative wear on the tool
Type	Type of tool in use

---

⚙️ Machine Learning Approach

1. Exploratory Data Analysis (EDA)

• Visualized distributions and correlations of sensor readings
• Identified skewed data and outliers in operational parameters
• Examined imbalance between "Normal" and "Failure" cases

2. Data Preprocessing

• Standardized numeric features using StandardScaler
• Encoded categorical variable Type using one-hot encoding
• The data did not have noisy points or missing values.

3. Model Selection

Due to high data imbalance, anomaly detection was a suitable approach to train the model. Multiple anomaly detection models were tested such as Local Outlier Factor, Isolation Forest, and One Class SVM. Out these, LOF was the best performing model. It was selected for Hyperparameter tuning.

• LOF detects rare deviations in data without requiring class labels
• Provides a robust binary output: -1 (Anomaly/Failure) and 1 (Normal)

4. Model Training & Evaluation

• Trained LOF using features: ['Process temperature', 'Air temperature', 'Rotational speed', 'Torque', 'Tool wear', 'Type']
• Tuned hyperparameters such as n_neighbors and contamination
• Evaluated using:
  • Confusion matrix for binary classification performance
  • Precision and recall of failure detection

5. Model Serialization

• Model saved using Pickle for deployment
• Input preprocessing pipeline also serialized to ensure consistent predictions

---

💻 Flask Web Application

A lightweight Flask interface allows users to input machine parameters and receive instant predictions.

🔹 Web Features

• Simple input form for entering sensor data
• Model prediction displayed as:
  • ✅ Normal Condition
  • ⚠️ Failure Condition (Potential Anomaly)
• Backend built using Flask and scikit-learn

🔹 Deployment

• App deployed on Render.com for free testing
• GitHub repository integrated with Render for continuous deployment
• Requirements managed via requirements.txt

---

🧠 Skills Demonstrated

Category	Skills
Data Science & Machine Learning	Data preprocessing, Feature engineering, Model selection for imbalanced data, Anomaly detection, Scikit-learn pipeline
Software Engineering	Flask app development, REST API design, Modular code structure
Model Deployment (MLOps)	Pickle model serialization, Environment setup with requirements.txt, Render.com deployment
Version Control & CI/CD	Git, GitHub integration with Render, Automated build and deploy
Visualization & Reporting	Matplotlib/Seaborn EDA, Model evaluation metrics, Clear documentation

---

🧪 Example Usage

1. Go to the deployed app: 🔗 Live Demo on Render
2. Youtube Demo
3. Input the following values:

Feature	Example Input
Process Temperature	308.6
Air Temperature	295.0
Rotational Speed	1430
Torque	40.2
Tool Wear	120
Type	H

3. Click Predict → The app will output:
   ✅ Normal Condition or ⚠️ Failure Condition

---

🧰 Technologies Used

• Python 3.10+
• Flask
• scikit-learn
• pandas, numpy
• matplotlib, seaborn
• Render.com
• GitHub

---

🧾 Future Improvements

• Integrate real-time IoT sensor streaming for continuous monitoring
• Replace LOF with Isolation Forest or Autoencoder models for comparison
• Add visual dashboard using Plotly Dash or Streamlit
• Containerize the app using Docker for scalable deployment

---

👤 Author

Abrar Asghar
🔗 GitHub Profile
💼 Machine Learning Based Reliability Engineering
📧 abrar.asghar@gmail.com