Industrial systems increasingly rely on automatic and predictive diagnostics to ensure operational reliability and efficiency. In this paper, we investigate the application of deep learning techniques to time series data for anomaly detection in a surface treatment line, with a particular focus on the gantry system responsible for dynamic transfers. We evaluated six state-of-the-art models: DLinear, MOMENT, LSTM, N-BEATS, N-HiTS, and NBEATSx/N-HiTSx—on a large dataset collected from an industrial immersion process. Our findings indicate that N-BEATS consistently outperforms the other models in terms of predictive accuracy (MAE, MSE, RMSE) and robustness to process variations. The results highlight the suitability of interpretable N-BEATS-based models for diagnosing cyclic industrial operations. We conclude with a discussion of real-time deployment perspectives and potential improvements through hybrid modeling and integration of exogenous process knowledge.