Scribd
Upload
14 views4 pages

DAN1

The DANCAST project aims to develop an automated system for detecting casting defects using deep learning, specifically autoencoders, to improve upon the limitations of manual inspection. The approach includes a data processing pipeline, model architecture, and a defect detection strategy that focuses on training with normal samples and real-time inference. The system promises cost reduction, quality improvements, and enhanced operational efficiency in manufacturing quality control.

Uploaded by

nishagarghnd825
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
14 views4 pages

DAN1

The DANCAST project aims to develop an automated system for detecting casting defects using deep learning, specifically autoencoders, to improve upon the limitations of manual inspection. The approach includes a data processing pipeline, model architecture, and a defect detection strategy that focuses on training with normal samples and real-time inference. The system promises cost reduction, quality improvements, and enhanced operational efficiency in manufacturing quality control.

Uploaded by

nishagarghnd825
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

DANCAST: Deep Autoencoder Network for Casting

Surface Defect Detection

Objective:
This project aims to develop an automated inspection system for casting defect detection using
machine learning techniques. The system addresses the limitations of manual inspection
processes in manufacturing, including inconsistent accuracy and high labour costs. By leveraging
deep learning architectures, particularly autoencoders, the system can identify various casting
defects.

Problem Statement:
Manufacturing quality control, especially in casting processes, traditionally relies on manual
inspection, which presents several challenges:
1. Inconsistent inspection accuracy due to human factors
2. High labour costs and me consumption
3. Limited scalability with production volume
4. Difficulty in detecting subtle defects
5. Fatigue-induced errors in prolonged inspection sessions

Technical Approach:
1. Data Processing Pipeline

• Image acquisition from the production.


• Preprocessing steps:

o Normalization and standardization

o Image augmentation for robust model training

o Noise reduction and enhancement

o Feature extraction

2. Model Architecture
The core of the system is built using autoencoder neural networks and CNN:
1. Encoder Network

• Compresses input images into a lower-dimensional

• latent space Captures essen al features and pa erns of


normal products
• Utilizes convolutional layers for spatial feature
extraction
2. Decoder Network

• Reconstructs images from the latent representation

• Learns to reproduce normal product characteristics

• Identifies anomalies through reconstruction error

Defect Detection Strategy:


• Training on normal (non-defective) samples only
• Establishing reconstruction error thresholds
• Real- me inference for production line integra on
• Adaptive threshold adjustment based on false positive/negative rates

Technology Stack:
• Primary Framework: TensorFlow/PyTorch
• Programming Language: Python
• Key Libraries:
o NumPy for numerical operations

o OpenCV for image processing

o Scikit-learn for preprocessing and evaluatin

o Pandas for data management

Methodology:
The project follows a systematic approach, as outlined below:

Step 1: Data Collection and Preprocessing

• Dataset: Images of casting products labelled as defective and non-defective.


• Data Preprocessing:
o Image resizing and normalization.
o Data augmentation (rotation, flipping, brightness adjustments) to improve model
generalization.
Step 2: Model Selection and Training

• Deep Learning Model: Convolutional Autoencoder


• Custom Layers: Fully connected layers added for classification.
• Training Strategy:
o Loss Function: Mean Squared Error (MSE).

o Optimizer: Adam optimizer.


o Evaluation Metrics: Accuracy, Precision, Recall, F1-score.

Step 3: Model Evaluation and Explainability

• Performance Metrics: The trained model is evaluated using a test dataset to measure:
o Accuracy
o Precision
o Recall
o F1-score

Technical Achievements:
1. Automated detection of major casting defects:
Blow holes and pinholes
Burr formation
Shrinkage defects
Mould material defects
Pouring metal defects
Metallurgical defects
2. Performance Metrics

o High detection accuracy


o Low false positive rate
o Real- me processing capability
o Real- me processing capability

Business Impact:
1. Cost Reduction
2. Quality Improvements

o Consistent inspection accuracy


o 24/7 operation capability
o Early defect detection
o Detailed defect analytics
3. Operational Efficiency

o Reduced inspection time


o Data-driven quality control
o Increased production throughput
Conclusion:
This defect detection system represents a significant advancement in manufacturing quality control.
By leveraging modern machine learning techniques, it addresses the limitations of traditional
inspection methods while providing a scalable, consistent, and cost-effective solution for industrial
inspection processes.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy