The advent of Industry 4.0 has stimulated an emergence of Cyber-Physical Systems (CPS) where physical and computational components are tightly conjoined via the Internet and computer networks. The manufacturing industry can take advantage of CPS in various respects. Specifically, Additive Manufacturing (AM) has been explored by researchers to improve the productivity of a manufacturing system in terms of flexibility, reliability, functionality, usability, and efficiency.

However, critical security concerns are lagging a practical implementation of AM in the CPS. The enlarged usability can increase the attack surface of the system in which large amounts of manufacturing information are being exchanged between machines. Thus, it can easily become a target for intruders. Also, a cross-domain connection between digital data and physical machines can lead to deteriorating particular attacks, such as Cyber-Physical Attack, which attack is initiated from the digital format and results in physical damages. This type of attack may usher in unique vulnerabilities where infill defectives can be injected without affecting the exterior. The infill defectives can cause unexpected changes in product properties and natures, which can reduce product safety and functionality without being detected.

To detect infill defectives, various machine learning algorithms—such as Naïve Bayes, J48 Decision Trees, Random-Forest, K-Nearest Neighbors, and Convolutional Neural Networks—have been attempted to classify 3D printing images layer by layer. However, such algorithms use a supervised learning approach, which requires predefined defective image data for labeling and training. This means that those algorithms are prone to misclassify unlabeled defective type images. Furthermore, it is not only impractical to predict and label a large number of defective types but also time-consuming. To solve this problem, a semi-supervised learning method can be used. Unlike the supervised learning method, the semi-supervised learning method needs a comparably small amount of labeled data with a large amount of unlabeled data. Also, only a normal class can be trained to identify anomalies from unlabeled data, and the anomalies can be used to generate more label examples for predicting other unlabeled data: self-training.

This paper presents a novel Infill Defective Detection System (IDDS) augmented by the semi-supervised learning method. IDDS was developed based on a multilayer perceptron neural network, which is modified to realize self-training. A heuristic approach is applied to construct the semi-supervised learning algorithm by using unlabeled data within a supervised learning framework. To validate the system, IDDS was tested with several 3D printing examples under various conditions. A comparative analysis with the classification accuracy is provided to evaluate the system's performance.


Infill Defective Detection System Augmented by Semi-Supervised Learning

Division / Sponsors: 
Aerospace, Bioengineering, Applied Mechanics, Design Engineering Division, Advanced Energy Systems, Manufacturing Engineering Division, Noise Control and Acoustics, Nondestructive Evaluation, Diagnosis, and Prognosis (NDPD), Nuclear Engineering Division, Process Industries Division, Fluids Engineering, Heat Transfer, Materials, Microelectromechanical Systems, Safety Engineering and Risk Analysis, Computers & Information in Engineering


Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2020/volume/84478

IMECE2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.