From Anomaly Detection to Defect Classification

Jaromír Klarák,Robert Andok,Peter Malík,Ivan Kuric,Mário Ritomský,Ivana Klačková,Hung-Yin Tsai

DOI: https://doi.org/10.3390/s24020429

IF: 3.9

2024-01-11

Sensors

Abstract:This paper proposes a new approach to defect detection system design focused on exact damaged areas demonstrated through visual data containing gear wheel images. The main advantage of the system is the capability to detect a wide range of patterns of defects occurring in datasets. The methodology is built on three processes that combine different approaches from unsupervised and supervised methods. The first step is a search for anomalies, which is performed by defining the correct areas on the controlled object by using the autoencoder approach. As a result, the differences between the original and autoencoder-generated images are obtained. These are divided into clusters using the clustering method (DBSCAN). Based on the clusters, the regions of interest are subsequently defined and classified using the pre-trained Xception network classifier. The main result is a system capable of focusing on exact defect areas using the sequence of unsupervised learning (autoencoder)–unsupervised learning (clustering)–supervised learning (classification) methods (U2S-CNN). The outcome with tested samples was 177 detected regions and 205 occurring damaged areas. There were 108 regions detected correctly, and 69 regions were labeled incorrectly. This paper describes a proof of concept for defect detection by highlighting exact defect areas. It can be thus an alternative to using detectors such as YOLO methods, reconstructors, autoencoders, transformers, etc.

engineering, electrical & electronic,chemistry, analytical,instruments & instrumentation

What problem does this paper attempt to address?

The paper attempts to address the problem of detecting defects on gear teeth in visual data and classifying these defects. Specifically, the paper proposes a novel defect detection system design method aimed at accurately identifying damaged areas through visual data containing images of gear teeth. The main advantage is that the system can detect various defect patterns present in the dataset.

The method in the paper combines unsupervised learning (anomaly detection and clustering) and supervised learning (classification) with the following specific steps:
1. **Unsupervised Learning - Anomaly Detection**: Use an autoencoder to define the correct areas on the controlled object and generate the differences between the original image and the autoencoder-generated image.
2. **Unsupervised Learning - Clustering**: Use the DBSCAN algorithm to cluster these differences into different groups, thereby determining the Regions of Interest (RoI).
3. **Supervised Learning - Classification**: Use a pre-trained Xception network to classify these RoIs to identify specific types of defects.

The main outcome of the system is a system capable of focusing on exact defect areas, employing a series of methods from unsupervised learning to supervised learning (U2S-CNN). Experimental results show that 177 regions were detected in the test samples, with 205 actual damaged regions occurring, of which 108 regions were correctly detected, and 69 regions were incorrectly marked. This indicates that the method has some potential in defect detection and can serve as an alternative to existing methods such as YOLO, reconstructor, autoencoder, and transformer.