Time Series Classification (TSC) is a problem with many applications in science and engineering. In reliability engineering, the time series data are used for equipment reliability analysis, fault detection etc. In this work, we examined several deep learning models on long-sequence time series. With the emergence of new technology, such as connected sensors, more long sequence of time series data become available. We developed several deep learning models to classify the high frequency / long sequence time series. Our approaches avoided heavy feature engineering, and attempted to make it “end-to-end.” The models were tested on a data set coming from real world high frequency voltage sensors to detect whether a partial discharge (PD) fault exists. We discussed the performance of the deep learning models, and their advantage and disadvantages to the non-deep-learning benchmark.