Water injection into an oil well increases pressure in the reservoir, preventing its rapid decline in oil recovery. Failure of the injection pump can compromise oil production for as long as it is stopped. Therefore, anticipating failures in advance and adopting predictive maintenance policies will make their availability higher than current values. One approach to do this is by Condition-Based Maintenance (CBM), in which assets are continuously monitored to determine their health state. The obtained data is analyzed to identify and predict failures. Maintenance planning is done according to these results. In recent years, Machine Learning (ML) and Deep Learning (DL) techniques have been used in the context of CBM, since they can deal with large amounts of data. They have the ability to identify complex patterns in the data that provide relevant information regarding the state of the equipment. Amongst DL algorithms, Long Short-Term Memory Networks (LSTM) stand out for being used to deal with time dependencies within the data. This paper presents a model for predicting the operating state of a water injection pump using LSTM. When predicting failures within one day, all performance metrics (precision, recall and F1- score) reach values above 92.0%. Model is compared with an Artificial Neural Network (ANN). Future works include doing remaining useful life (RUL) predictions.