<p>This study proposes an alarm sequence pattern analysis technique to detect and predict anomalies in automated manufacturing processes by utilizing programmable logic controller (PLC) alarm data. While traditional sensor-based anomaly-detection methods struggle to fully account for interactions between devices or temporal dependencies, this study analyzes complex alarm patterns across the entire process by considering both alarm event grouping and time interval encoding. Specifically, after data preprocessing, alarm events were grouped based on state transition or occurrence time criteria, and four experimental conditions were designed according to alarm event grouping and the inclusion or exclusion of time interval encoding. The generalized sequential pattern (GSP) algorithm was then applied to identify frequent alarm sequences, and a Bayesian probability-based approach was used to calculate the likelihood of subsequent alarm occurrences, thereby implementing a model to predict alarms with a high probability. The experimental results based on eight months of alarm logs (235332 records) from an automotive battery module assembly process demonstrated that the model incorporating state transition and time interval encoding achieved the highest prediction performance, with an F1 score of 89.1 %. This suggests that simultaneously considering alarm occurrence patterns and temporal relationships is effective in improving the prediction accuracy. The proposed technique is expected to be applicable to real-time prediction model development and integrated analysis with process variables in various industries.</p>