Prediction Method for Aging Life of Rubber O-Rings Based on GRU-LSTM Neural Network

Abstract

This paper proposes a prediction model for rubber aging life based on the GRU-LSTM neural network to address the issue of sealing failure caused by material aging of rubber O-rings in the petrochemical industry. The model aims to overcome the limits of existing methods, which have low prediction accuracy and poor stability. In this study, we conducted rubber aging experiments to get data on rubber compression set at different temperatures. After preprocessing the data, we experimentally verified and compared the GRU-LSTM model with the two-layer LSTM model, two-layer GRU model, and BP neural network. Finally, we carried out an example analysis and application of high-temperature flange sealing in a chemical equipment. The results revealed significant differences between the two types of rubber, NBR-26 and HNBR. When using the aging data at 60, 80, and 100°C to predict the aging data at 70 and 90°C, the R2 values for NBR-26 and HNBR were 0.9968 and 0.9945, respectively. These findings highlight the superior predictive performance of the proposed model. These results show significantly better prediction performance compared to the other three models. Further, the accuracy of the prediction results increased with the number of temperature levels and samples. We recommend using at least three temperatures for prediction, and a larger sample size is preferable. The example analysis showed that this prediction model can help decide the replacement cycle of high-temperature flange rubber O-rings at different storage temperatures, reducing maintenance costs.