Latent Neural Network For Recognition of Annomalies in 3D-Print of a Scale Model for Wind Tunnel Measurements

The paper concerns a new method of diagnosing and monitoring 3D printing processes based on measurements of kinematic parameters describing the movement of the 3D printer head. The aim of the research was to create a neural network model using unsupervised learning methods, whose task is to map signal characteristics (accelerometer and gyroscope data) to latent space. The measurement of significant quantities describing the kinematics of printer head movement was carried out during the execution of the given 3D print. The proposed latent artificial neural network algorithm was implemented in the microprocessor of the IMUMETER measuring device and tests were carried out during which the measuring system collected data and the neural network model mapped the features in such a way that it separated signals with anomalies from typical signals. The results allowed for the mapping of features and the analysis of the occurrence of anomalies on a sample printout of a wing section model for testing in a wind tunnel.