Large models have transformed various fields, particularly in time series forecasting, but their effectiveness in time series classification remains limited, especially for specialized domains like oil and gas drilling. This paper evaluates the performance of large models in time series classification tasks, highlighting their challenges in handling real-world univariate and multi-variates real-world time series data. Through comprehensive experiments, we show that these models, are outperformed by lightweight convolutional baselines in both accuracy and efficiency. While large models like Chronos and Moments demonstrate some success, they require significantly more computational resources to achieve optimal classification performance. Our results suggest that lighter CNN are better suited for time series classification in industrial applications, where both accuracy and computational efficiency are critical.