Abstract: A lifetime prediction method based on Bi-GRU (gated recurrent unit, GRU) is proposed to address the challenge of processing long-term signals in existing methods, which often results in poor predictive accuracy for the service life of check valve in diaphragm pumps. Firstly, we meticulously identify and extract critical degradation indicators from the vibration data collected by sensors throughout the entire life cycle of the check valve in the actual engineering environment. These indicators effectively reflect the health status of the equipment, enabling us to construct a dynamic curve illustrating the equipment's health degradation over time.Then, the life prediction model is used to learn the mapping relationship between the historical vibration data and the degree of degradation. Finally, the degradation characteristics of the measured vibration data were input into the prediction model, and the prediction model was used to predict the remaining usable life of the check valve. This method avoids the data dependence problem of recurrent neural networks through the gate recurrent unit structure, and also improves the learning ability of the model by learning the degradation laws of the previous and subsequent moments of the time series data in both chronological and anti-temporal bidirectional directions. The proposed method is verified on the life signal of the check valve measured in engineering, and the experimental results show that the effectiveness of the proposed method is effective.