<p>High-cost equipment demands the identification of malfunction signals and the time required for preventive maintenance on production lines more accurately to maintain production quality and performance and improve production efficiency. To make unmanned factories a reality, the present study adopted the Taguchi Method to design <i>L</i>₉ (3⁴) orthogonal arrays, after which experimental parameters were determined to perform robotic grinding experiments. The vibration signals captured during the experiment were used to create recurrence plots, whose features were employed to process the signals during the grinding using the Recurrence Quantification Analysis (RQA). Then, the Gaussian Mixture Model (GMM) was utilized to train data, and the Bayes Maximum Likelihood Classifier method was employed to train and identify abnormal vibration signals. Lastly, two models, i.e., RQA-GMM and RP-VGG16, were established for testing and comparison. The research outcomes showed that RQA-GMM achieved 96.7% accuracy in 12&#xa0;min of training using the independent dataset testing method; it is a highly accurate prediction regarding time efficiency. The research outcome suggests that RQA-GMM maximizes the performance in production lines with limited processing time, so it suffices to serve as a monitoring and prediction system.</p>