This paper introduces a cutting-edge multimodal time series classification model, specifically designed to handle the complexities of analyzing data from various sources, such as images and sensor measurements, which are distributed over time and may not always be fully available. Applied to the critical task of failure prediction in oil wells with rod lift systems, our model achieved an accuracy of 61%, outperforming the 54% to 59% accuracy range of single-modality models. Utilizing a Time-Distributed framework, our approach effectively integrates various sub-models to manage incomplete datasets. The model's innovative design not only enhances reliability and maintenance scheduling in oil extraction operations but also marks a significant advancement in using multimodal data for operational optimization and preemptive failure detection.