Assessment of tube and tube-sheet life cycle in shell-and-tube heat exchangers by means of AI-machine learning algorithms at LAVAN oil refinery

Purpose Predicting the life age of tube and tube sheet for changing tube bundle in Corrective Maintenance Planned (CMP) can prevent extra cost and outage of plant service. Design/methodology/approach In this study, the life age of tubes and tube-sheets in different types of shell-and-tube heat exchangers are being determined based on machine learning (ML) by results of more than 45 years inspection data at Lavan oil refinery. This study reviewed that the major factors consist of exchangers’ operating conditions with different media, process condition and exchanger design parameters to arrange the assessing life cycle of different tube and tube-sheet ages. Different ML algorithms for predicting results with evaluating error control system are also explained in this study. Results showed that the “Decision Tree Regressor” and “Gradient Boosting Regressor” algorithms were the best ones in AI-ML methods, which predict the age of tube and tube sheet by high resolution close to real overhaul inspection data. Findings The main goal of this study is to predict life age of tubes and tube sheet in a typical heat exchanger at Lavan oil refinery by different design and process operating conditions. Each process and design data related to the life assessment of tubes and tube sheets were normalized and processed in ML algorithms for better regression results. The AI-ML prediction results were being validated by calculating different error estimating method between real values and predicted ones. Results revealed that the “Decision Tree Regressor” and “Gradient Boosting Regressor” algorithms in both tube and tube-sheet life assessment also in both crude distillation unit and catalytic reformate unit were the best algorithms in AI-ML, which predicted the age of tubes and tube sheets of heat exchangers in refinery by high resolution and minimum values of errors. The predicted life age results in DT and GBR algorithm graphs were close to real values, and the error-estimating methods showed that the less the “mean squared error”, the more accurate the results in predictions. Predicted value results showed that the life age of tubes and tube sheets estimated using DT and GBR algorithm were closer to real value results by higher resolution and higher accuracy. Research limitations/implications Prediction errors may arise due to lack of exact data in data set. Originality/value The entire dataset and all of the results were derived from Lavan refinery.