TAGRA 2024

The Detecting vibration anomalies in 3D printers is critical for maintaining print quality and increasing efficiency in production processes. Early awareness reduces costs by preventing faulty production and contributes to longer device life. Artificial intelligence applications using classification and anomaly detection models can detect these errors at an early stage by analyzing the data obtained with signal processing techniques. In this study, data collected from a 3D printer using vibration sensors were used to evaluate the performance of machine learning and deep learning algorithms in anomaly detection. The analyzed dataset consists of 7,967 vibration data (405 anomalies and 7562 normal data). In this analysis, eight machine learning algorithms such as Isolation Forest, K-Means, Single Class SVM and Spectral Clustering, among others, and two deep learning models, namely Autoencoder and the proposed Residual Convolutional LSTM model. In the data preprocessing process, dimensionality reduction and normalization were performed using PCA (Principal Component Analysis). The study also presents a new model (RConvLSTM4AD) that can detect anomalies by hybridizing Residual Techniques and Convolutional Long-Short Term Memory method. The hybrid model performed the best with 98.77% accuracy compared to the others. This was closely followed by Spectral Clustering with 94.95% accuracy and Agglomerative clustering with 94.82% accuracy. These findings emphasize the effectiveness of the proposed hybrid approach for vibration anomaly detection in 3D printers.