In the automotive industry, machinery failures of the resistance spot welding (RSW) guns would interrupt the manufacturing lines and cause unplanned downtime, potentially resulting in a significant loss of production and reliability. Predicting the machinery failures of the RSW gun can provide more scientific strategies for predictive maintenance and decision-making. However, fault prediction of RSW guns has become increasingly challenging due to their complex behavior and data variability. In this paper, we created a benchmark dataset and proposed welding gun fault prediction benchmarks to aid in the development of machine learning approaches toward welding gun fault prediction. The dataset was collected at the Body-Shop (BS) of BMW Brilliance Automotive Ltd. from different components of hundreds of RSW guns to capture the patterns and trends before welding errors with historical data. Then we provide state-of-the-art machine learning (ML) benchmarks on time series forecasting methods in a welding gun fault prediction use case. This study will provide insights for time series forecasting while enabling ML researchers to contribute towards the fault prediction of the RSW guns.