Date of Completion
Qian Yang; Jinbo Bi
Artificial Intelligence and Robotics
Consistently jetting different materials from the print head of a 3D printer is a key, yet challenging task in manufacturing processes. By using active machine learning, we can efficiently predict complex diagrams that illustrate the region of printing conditions under which “desirable jetting”, “jetting”, and “no jetting” of ink occurs for different substances. However, labeling the images of printed ink droplets that are fed to the active learning model can be time intensive. Therefore, it is ideal to use computer vision to automate the classification of this image data. This classification can be broken down into two steps. In the first step, we are interested in distinguishing between images that exhibit any jetting and those that show no jetting at all. To perform this binary classification, a convolutional neural network (CNN) is trained on a dataset consisting of both “jetting” and “no jetting” images. Because there is a relative lack of data exhibiting “no jetting”, data augmentation is applied in order to create a balanced number of images between the two classes and prevent bias towards the “jetting” class. In the second step, the differentiation between “desirable/consistent” and “inconsistent/satellite” jetting is made. “Inconsistent/satellite” jetting can present itself in many different ways, so the variation between images of this “class” becomes troublesome for binary classification. Instead, autoencoders can be used for anomaly detection, where “desirable/consistent” jetting is the normal class, and “inconsistent/satellite” images would be anomalies. The autoencoder is trained on only consistent jetting images, compresses the image data into a latent representation, and then attempts to reconstruct the original image data using the latent representation as an input into a decoder. The decoder has trouble reconstructing the original image when given the latent representation of an “inconsistent/satellite” image, yielding a high reconstruction error. Jetting images that generate a reconstruction error higher than a specified threshold are deemed anomalies, and thus not “desirable” jetting.
Chandy, Alexander, "Automatic Identification of Jetting Behavior in 3D Printing with Binary Classification and Anomaly Detection" (2023). Honors Scholar Theses. 979.