HYBRID SARIMA-LSTM MODEL FOR PREDICTING EROSION IN BUTTERFLY VALVE

Abstract

In the oil and gas industry, butterfly valves often undergo erosion due to being placed in fluid flow with high pressure and high temperature.  Erosion on butterfly valves can result in huge losses so it requires early anticipation. To overcome these problems, this research proposes a hybrid SARIMA-LSTM model to predict the mass erosion of butterfly valves under several opening conditions. The results show that the SARIMA-LSTM model has superior performance compared to the conventional LSTM and SARIMA model with MSE values at valves opening 20  – 90  reaching 1E-06; 1E-06; 6.2E-05; 2.34E-04; 1.35E-07; and 1E-06 respectively. The hybrid SARIMA-LSTM model successfully identifies the non-linear characteristics of the erosion data by identifying the residual value resulting from the difference between the SARIMA model prediction and the actual data. This study also reveals that the combination of SARIMA and LSTM models significantly affects the performance of the LSTM model. This study also successfully used the SARIMA-LSTM model to predict the erosion mass value for the next 30 time-steps. Through this study, it is known that the SARIMA-LSTM hybrid model has the possibility to be applied to the oil and gas industry to help the process of observing the erosion mass on the butterfly valve.  

Keywords: butterfly valve, seasonal autoregressive integrated moving average (SARIMA), long-short term memory (LSTM), erosion mass, time-series forecasting.