/Deep-Learning-and-Computer-Vision-for-Structural-Crack-Detection-And-Classification

Incorporating Inductive Bias into Deep Learning: A Perspective from Automated Visual Inspection in Aircraft Maintenance

Primary LanguagePython

Application of Deep Learning (CNN's) for the Detection and Classification of Cracks Present in an Aircraft Panel

Contents

  1. General Information
  2. Abstract
  3. Usage Guide

General Information

Abstract

Narrow artificial intelligence, commonly referred as ‘weak AI’ in the last couple years, has developed due to advances in machine learning (ML), particularly deep learning, which has currently the best in-class performance among other machine learning algorithms. In the deep learning framework, many natural tasks such as object, image, and speech recognition that were impossible in the previous decades using classical ML algorithms can now be done by a typical home personal computer. Deep learning requires a rapid collection of a large amount of data (also known as ‘big data’) to create robust model parameters that are able to predict future occurrences of certain event. In some domains, large datasets such as the CIFAR-10 image database and the MNIST handwriting database already exist. However, in many other domains such as aircraft visual inspection, such a large dataset of damage events is not available, and this is a challenge in training deep learning algorithms to perform well to recognize material damage in aircraft structures. As many computer science researchers believe, we also think that in order to achieve a performance similar to human-level intelligence, AI should not start from scratch. Introducing an inductive bias into deep learning is one way to achieve this human-level intelligence in the aircraft inspection for damage. In this paper, we give an example of how to incorporate domain knowledge from aerospace engineering into the development of deep learning algorithms. We demonstrate the suitability of our approach using data from fatigue testing of an aerospace grade aluminum specimen to build a deep convolutional neural network that classifies crack length according to the crack propagation curve obtained from fatigue test. The results of this network were then compared to the same network that was not trained with domain knowledge and the biased learning achieved a validation accuracy of 97.55% on determining crack length, while unbiased network selected the unwanted parameter of sunlight intensity, however with 99.45% accuracy.

Usage Guide

The very first thing to do in order to test out and witness the prediction-making performance, presented and discussed in the conference research paper, is as simple as just running the OOP_Predicter.py script!