電気通信システムと管理

Smart Spatial Analyses in Land Levelling Development and Evaluation of Models for Tractor Performance Parameters

Abstract

JAMIL ASAAD

The objective of this research was to develop two methods of computational intelligent (CI) techniques, namely, artificial neural network(ANN) and adaptive neural fuzzy inference system (ANFIS).Furthermore, to develop mathematical model using Design Expert software for modeling and predicting performance parametersof theMassey Ferguson (MF-285) tractorunder various field conditions.In this study a MF-285 tractor was instrumented with a low cost and precise data logging system as a means of recording and monitoring the affectual parameters on performance of tractor such as forward speed and instant fuel flow rate during field operation. A moldboard plow was used as tillage tool during the experiments under various tillage depths, engine speeds, forward speeds, tire inflation pressures, moisture contents and cone indexes. Acquired data were used to develop accurate models for drawbar pull, rolling resistance, slippage, Temporal Fuel Consumption (TFC), Area-specificFuel Consumption (AFC), Specific Fuel Consumption (SFC), drawbar power, axle power, net traction ratio, tractive efficiency and power loss. 
 The results showed that all developed models (ANN, ANFIS and mathematical) had satisfactory performance for predicting aforementioned parameters of tractor in various field conditions. For drawbar pull, ANN technique achieved optimum model with topology 6-8-1 andLevenberg-Marquardt learning method with MSE of 0.000515 and R2 of 0.997.  ANFIS method produced the best model with indicators statistical MSE of 0.00541 and R2 of 0.979 for rolling resistance. The premium model for anticipating slippage achieved by ANN with topology 6-8-1 and Bayesian regulation with MSE of 9.3621e-08 and R2 of 0.9999.For drawbar power, the best result was obtained by the ANN with 6-7-1 topology and Bayesian regulation training algorithm with R2 of 0.995 and MSE of 0.00024. 
The obtained result showed that the 6-7-1 structured ANN with Levenberg-Marquardt training algorithm represented a good prediction of tractive efficiency with R2 equal to 0.989 and MSE of 0.001327.Also the model of ANN is overcome other models in predicting axle power (Levenberg-Marquardt withstructure 6-7-1, MSE of 0.0001683 and R2 of 0.996), power loss (Bayesian regulation with topology 6-9-1, MSE of 0.0001032 and R2 of 0.985) and net traction ratio (Levenberg-Marquardt with structure 6-9-1, MSE of 0.0006814 and R2 of 0.994).
The performance of predicting fuel consumption (TFC, AFC and SFC) is acceptable. The ANN model with 6-7-1 structure and Levenberg-Marquardt training algorithm had the best performance with R2 of 0.969 and MSE of 0.13427 for TFC prediction.The 6-8-1 topology showed the best power for prediction of AFC with R2 and MSE of 0.885 and 0.01348 with Levenberg-Marquardt training algorithm. ANFIS method achieved the best model for prognostication SFC with MSE of 0.01475 and R2 of 0.9454.
The obtained results confirmed that the ANN, ANFIS and mathematical modelsare able to learn the relationships between the input variables and performance parameters of tractor, very well.