Surface Roughness Prediction using CNN + ML Ensemble

🔍 Project Summary

Domain: Machine Learning, Deep Learning, Computer Vision, Manufacturing Analytics Role Alignment: ML Engineer | AI Engineer | Data Scientist | Computer Vision Engineer

This project implements an end-to-end machine learning pipeline for predicting surface roughness (Ra) of Inconel material using a hybrid CNN + machine learning ensemble approach. The system combines image-based feature extraction from microscopic surface images with numerical machining parameters, followed by ensemble regression for accurate prediction.

The focus of this project is on robust ML system design, effective feature fusion, and reproducibility. With Real Time GUI based Deployement . Video: https://drive.google.com/file/d/1O0Fqksh8wsjKsUJNcq4yzmxeFYI2mqe0/view

---

🧠 Key Skills & Technologies (ATS Keywords)

• Machine Learning & Deep Learning
• Convolutional Neural Networks (CNN)
• Computer Vision (OpenCV)
• Feature Engineering & Feature Fusion
• Ensemble Learning
• Regression Modeling
• Error Analysis & Data Cleaning
• Python, TensorFlow, Keras
• Scikit-learn
• NumPy, Pandas
• Manufacturing Analytics / Industry 4.0

---

📌 Problem Statement

Surface roughness is a critical quality metric in machining processes. Traditional analytical and empirical models struggle to generalize under limited experimental data and complex surface textures.

This project addresses the problem by:

• Extracting deep visual features from surface images using a custom CNN
• Integrating machining parameters with image features
• Applying ensemble regression models for improved generalization

---

📂 Data Description

• Material: Inconel

• Data Type:

  • Microscopic surface images (PNG format)
  • Corresponding machining parameters (numerical)

• Image Naming Convention: SampleX_ImageY.png

• Each image is associated with a unique machining condition and measured surface roughness value

Note: Dataset is experimental and intended for academic/research use. With an objective of Industrial Level implementation.

---

🧩 System Architecture

1. Image Preprocessing

• Image resizing and normalization
• No Data augmentation as each surface patterns are unique.

2. CNN-Based Feature Extraction

• Custom CNN architecture trained from scratch
• CNN used strictly as a feature extractor
• High-level texture and morphology features extracted from surface images

3. Feature Fusion

• CNN-extracted features concatenated with machining parameters
• Numerical features scaled before fusion

4. Ensemble Regression

• Machine learning regressors trained on fused feature space

• Supports ensemble strategies such as:

  • Gradient Boosting
  • Random Forest
  • XGBoost
  • Stacking-based regression

---

📁 Repository Structure

├── CNN_Ensemble_V1.ipynb
├── Images/
│   ├── SampleX_ImageY.png
│   └── ...
├── best_feature_extractor.keras
├── ensemble_model.pkl
├── parameter_scaler.pkl
└── README.md

---

⚙️ Tech Stack

• Programming Language: Python
• Deep Learning: TensorFlow, Keras
• Computer Vision: OpenCV
• Machine Learning: Scikit-learn
• Data Handling: NumPy, Pandas
• Visualization: Matplotlib

---

▶️ How to Run

1. Clone the repository
2. Install required dependencies
3. Ensure image directory structure is preserved
4. Open and run CNN_Ensemble_V1.ipynb
5. The best-performing models are saved automatically

---

🔁 Reproducibility & ML Best Practices

• Controlled data splits
• Model checkpointing
• Feature scaling and versioning

Designed to ensure consistent and repeatable results, aligning with real-world ML engineering standards.

---

🧪 Applications

• Smart Manufacturing Systems
• Automated Quality Control
• Surface Integrity Prediction
• Industry 4.0–enabled analytics

---

📈 Key Contributions

• Designed a custom CNN feature extractor for surface texture analysis
• Implemented feature-level fusion of vision and numerical data
• Applied ensemble learning for improved robustness on small datasets
• Built a modular and extensible ML pipeline suitable for research and deployment

---

📄 License

This project is intended for academic and research purposes. Commercial usage requires prior permission.

• This model was further used as Pre-Trained Mode for Real Time Surface Roughness Predicition.
• Images can not be uploaded into Github. As it exceeds storage limit.
• Like the repository if you find it helpful.
• For any improvement in code base feel free to reach out.