Development of a prediction model for estimating tractor engine torque based on soft computing and low cost sensors. (June 2018)

• Record Type: Journal Article
• Authors: Rajabi-Vandechali, Majid; Abbaspour-Fard, Mohammad Hossein; Rohani, Abbas
• Is Part Of: Measurement. Volume 121(2018)
• Page Start: 83
• Page End: 95
• Publication Date: 2018-06
• Subjects: ANFIS -- RBF -- Engine torque -- Tractor; PES primary engine speed (rpm) -- IES instantaneous engine speed (rpm) -- FCMF fuel consumption mass flow (g·s−1) -- EGT exhaust gas temperature (°C) -- MAEO maximum exhaust opacity (m−1) -- MEEO mean exhaust opacity (m−1) -- RPM revolutions per minute (rpm) -- ANN artificial neural network -- RBF radial basis function -- CRD completely randomized design -- LSD least significant difference -- ANFIS adaptive neuro fuzzy inference system -- GP grid partition -- SC subtractive clustering -- FCM fuzzy c-means -- RMSE root mean squared error -- R2 coefficient of determination -- TSSE total sum of squared error -- EF model efficiency -- ANOVA analysis of variance -- MF membership function -- IC internal combustion -- SI spark ignition -- CI compression ignition; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2018.02.050
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A model was proposed to estimate ITM285 tractor engine torque using soft computing. RBF and ANFIS were compared using the combination of three statistical methods. Measuring PES, IES, FCMF and EGT were sufficient for proper torque estimation. The RBF had a better performance than the ANFIS for engine torque estimation. Low cost and accessible sensors were used to estimate ITM285 tractor engine torque. Abstract: Torque estimation needs intensive efforts and costly sensors. In this research, a model was proposed based on soft computing to estimate the ITM285 tractor engine torque using some low cost sensors. To this end, two models including the radial basis function (RBF) neural network and adaptive neuro fuzzy inference system (ANFIS) were used. Thirteen training algorithms were examined to train the RBF. These algorithms were compared using three statistical methods, namely k-fold cross validation, completely randomized design (CRD) and least significant difference (LSD). Moreover, three methods, namely grid partitioning (GP), sub-clustering (SC) and fuzzy c-means (FCM), were used to construct the fuzzy inference system (FIS). However, the FCM was the most suitable method. The sensitivity analysis showed that only measuring engine speed, fuel mass flow and exhaust gas temperature was sufficient for proper engine torque estimation. The RBF had a better performance (R 2 = 0.99, RMSE = 0.5 and EF = 0.99) than the ANFIS and hence, was suggested for estimating the engine torque.