3D printing technology has revolutionized the manufacturing industry by providing a flexible and efficient approach to constructing complex three-dimensional objects from digital blueprints, layer by layer. Despite the numerous benefits of this technology, such as reducing waste and streamlining production, it also presents unique challenges and potential drawbacks. Errors and inconsistencies can lead to defects, inaccuracies, or failed prints, making it crucial to address these obstacles to enhance the overall quality and reliability of 3D-printed objects.
The purpose of this dissertation is to analyze each layer to identify any potential printing errors. To accomplish this objective, a camera is integrated into the 3D printer, capturing real-time images of each layer created during production. The methodology focuses on unsupervised machine learning, mainly using the K-means Algorithm and Gaussian Mixture Model (GMM) to detect possible errors.
The analysis is conducted in two stages: an initial phase involving a first layer of basic models to ensure accuracy and a subsequent phase analyzing actual printed electric machines. The study indicates that K-means and GMM produce comparable precision, recall, and F-measure results. However, the complexity of model structures and the number of materials used on each layer can affect the algorithm’s accuracy. The error detection algorithm achieves a robust detection rate, ensuring the identification of errors in higher layers of multi-material structures.