In recent years, the applications of the smart factory are very popular. Predictive maintenance is one of the issues. Some research achieved the goal of predictive maintenance with Artificial Intelligence (AI). Here we focus on the local scrubber (LSR) system, a water purification and recycling system. This paper proposed a machine learning model to solve predictive maintenance problem. The device learns the pattern of input data through the RNN model and classify the different state of device. We can know the current situation of the device and judge whether it is about to be replaced. As far as we know, this is the first predictive task maintenance in the LSR system and has an accuracy of 84% in the datasets of different years. The smart factory will come true while the LSR system can be reduce cost, manpower, time and money with predictive maintenance.