Abstract
In the context of Industry 4.0, multi-sensor data plays a pivotal role in monitoring, analyzing, and optimizing product quality in real time. The ability to capture and process data from various sensors allows manufacturers to identify deviations, detect anomalies, and improve overall production efficiency. However, raw data collected during the injection molding process often contains redundant, irrelevant, or highly correlated features that can introduce noise and reduce the efficiency of predictive models. Without proper preprocessing, such data can lead to increased computational complexity and diminished model performance. To address these challenges, effective feature extraction techniques are essential for refining the dataset, minimizing prediction errors, and enhancing the interpretability of machine learning models. In this study, we compare two widely used feature extraction methods: Principal Component Analysis (PCA) and an Autoencoder (AE). The primary objective of this research is to assess the effectiveness of these feature extraction methods in monitoring the injection molding process and predicting product quality on an advanced machine learning model LSSVM. The experimental results presented in this study are useful in determining the suitability and disadvantages of each method, with the prediction accuracy of up to 99.62% for the extracted deep feature.