توسعه هسته پايش و پيش‌بيني عملكرد يك دوقلوي ديجيتال براي يك مبدل پوسته و لوله

In the present study, a data-driven framework is developed for the initial core of a digital twin of a shell an‎d tube heat exchanger. The main objective is the simultaneous prediction of key performance indicators including thermal efficiency, heat transfer rate, tube-side pressure dro‎p, shell-side pressure dro‎p, an‎d fouling resistances on both sides, using advanced machine learning algorithms. A shell an‎d tube heat exchanger with stan‎dard industrial configuration was modeled an‎d validated using specialized software Aspen EDR. By systematically varying input parameters, a comprehensive database of 1000 samples was generated. The data were split into 80% for training an‎d 20% for testing. Three algorithms including Multilayer Perceptron Artificial Neural Network (MLP), Support Vector Regression (SVR), an‎d Gradient Boosting Regressor (GBR) were implemented. MLP with two hidden layers (64 an‎d 32 neurons) an‎d ReLU activation function, SVR with RBF kernel, an‎d GBR with 600 trees an‎d learning rate of 0.03 were selec‎ted. The results showed that for tube-side pressure dro‎p, GBR achieved the best performance with an R² of 0.9950. For shell-side pressure dro‎p, SVR with R² = 0.965 an‎d GBR with R² = 0.963 showed excellent performance. For predicting thermal efficiency an‎d heat transfer rate, SVR was the best with R² values of 0.989 an‎d 0.999, respectively. For predicting shell-side fouling resistance, GBR outperformed the other two algorithms with an R² of 0.930. Physical analysis indicated that the excellent performance of the algorithms in pressure dro‎p prediction stems from their ability to model highly nonlinear relationships an‎d interactions without‌ imposing a pre-defined functional form. Considering the comprehensive performance, especially in predicting fouling resistance which exhibits high dispersion an‎d complex behavior, the Gradient Boosting Regressor is introduced as the most suitable algorithm for developing the initial core of the digital twin of the shell an‎d tube heat exchanger.

محمدرضا توكلي نژاد , محمدرضا سليم پور

احمد سوهان كار اصفهاني , رامين كوهي كمالي