MLP and optimized FCM-ANFIS models proposed for inlet turbulent flow under ultrasonic vibration

In this research, two methods, the multi-layer perceptron (MLP) neural network and the adaptive neuro-fuzzy inference system (ANFIS), were employed to find the best relationship between inputs and outputs of the inlet turbulent flow under ultrasonic vibration extracted from experimental data. During the experimental tests, the Reynolds number of the inlet flow ranged from 3500 to 12,000, while the ultrasonic power levels were set at 75 and 100 watts. The experiments were conducted under various inlet temperature levels of 30 °C, 35 °C, 40 °C, and 45 °C. Since conducting experiments or numerical solutions might be usually problematic in terms of time, cost, or both, models presented in this study could be helpful. While the performance of the MLP was improved by using the best structure found by some trials and errors, the quality of the ANFIS model was optimized using the simulated annealing (SA) optimization algorithm. The average relative errors obtained by applying the MLP model for heat transfer coefficient (h), outlet temperature (T o ), heat flow rate (q), and pressure drop ( Δ P ) are 0.82%, 0.086%, 2.24%, and 3.4%, respectively. On the other hand, applying the ANFIS model yields the relative errors 0.32%, 0.072%, 0.44%, and 1.8% for h, T o , q, Δ P , respectively. According to the results, both methods are effective for predicting the characteristics of the process. While the final results of the ANFIS might be much better, optimizing its performance takes more computation time compared to the MLP. Therefore, choosing an approach is dependent on several factors like the expected computation time, the expected final error values and so on.