Comparative analysis of Multiple linear Regression (MLR) and Adaptive Network-Based fuzzy Inference Systems (ANFIS) methods for vibration prediction of a diesel engine containing NH3 additive

• Published in: Fuel (Guildford) 2023-10, Vol.350, p.128686, Article 128686
• Main Authors: Çağıl, Gültekin, Nur Güler, Sena, Ünlü, Ayşe, Böyükdibi, Ömer, Tüccar, Gökhan
• Format: Article
• Language: English
• Subjects: ANFIS; Engine vibration estimation; MLR; NH3 additive rate
• ISSN: 0016-2361; 1873-7153
• DOI: 10.1016/j.fuel.2023.128686

•In this manuscript, the engine vibration of a diesel engine is estimated by Adaptive Network-Based Fuzzy Inference Systems (ANFIS) and Multiple Linear Regression (MLR) methods.•In this study, a total of 5 variables, namely NH3 additive rate, x-axis (m/s2), y-axis (m/s2), RMS (m/s2), and Engine Speed (m/s), were used as inputs in both models and z-axis (m/s2) output was estimated.•The relationship between the variables with the MLR method was examined and interpreted.•ANFIS and MLR methods were compared in terms of prediction performance. With the increase in population, the need for fuel has led researchers to search for alternative green fuels. The leading of these alternatives is Ammonia (NH3), which minimizes carbon emission as opposed to petroleum derivatives that contain carbon, primarily when used as fuel. In this study, NH3 was mixed with sunflower biodiesel in different volumetric ratios and burned at varying engine speeds using a diesel engine, recording experimental vibration data in the engine block. With these obtained data, Multiple Linear Regression (MLR) and Adaptive-Network Based Fuzzy Inference Systems (ANFIS) methods were used to compare the two methods by examining the effect of inputs on output and the factors affecting output. In this context, a total of 5 variables, namely NH3 additive rate, x-axis (m/s2), y-axis (m/s2), RMS (m/s2), and Engine Speed (m/s), were used as inputs in both models and the dependent variable z-axis (m/s2) was estimated. For this purpose, first of all, estimation was carried out with MLR method, then different models were created with ANFIS method, and the prediction performances of both methods were calculated. In the performance evaluation of the test data, R2 (certainty coefficient) value for MLR was found as 0.58, Mean Squared Error (MSE) was 10.61, Mean Absolute Error (MAE) was 2.43, and Root Mean Squared Error (RMSE) was 3.25, while R2 value for ANFIS was calculated as 0.86, MSE 3.85, MAE 1.25 and RMSE 1.96. When both methods are compared in terms of performance, it is seen that ANFIS gives better results than MLR.