Journal of Agricultural Machinery (Mar 2016)
Predicting the wheel rolling resistance regarding important motion parameters using the artificial neural network
Abstract
Introduction: Rolling resistance is one of the most substantial energy losses when the wheel moves on soft soil. Rolling resistance value optimization will help to improve energy efficiency. Accurate modeling of the interaction soil-tire is an important key to this optimization and has eliminated the need for costly field tests and has reduced the time required to test. Rolling resistance will change because of the tire and wheel motion parameters and characteristics of the ground surface. Some tire design parameters are more important such as the tire diameter, width, tire aspect ratio, lugs form, inflation pressure and mechanical properties of tire structure. On the other hand, the soil or ground surface characteristics include soil type; moisture content and bulk density have an important role in this phenomenon. In addition, the vertical load and the wheel motion parameters such as velocity and tire slip are the other factors which impact on tire rolling resistance. According to same studies about the rolling resistance of the wheel, the wheel is significantly affected by the dynamic load. Tire inflation pressure impacted on rolling resistance of tires that were moving on hard surfaces. Studies showed that the rolling resistance of tires with low inflation pressure (less than 100 kPa) was too high. According to Zoz and Griss researches, increasing the tire pressure increases rolling resistance on soft soil but reduces the rolling resistance of on-road tires and tire-hard surface interaction. Based on these reports, the effect of velocity on tire rolling resistance for tractors and vehicles with low velocity (less than 5 meters per second) is usually insignificant. According to Self and Summers studies, rolling resistance of the wheel is dramatically affected by dynamic load on the wheel. Artificial Neural Network is one of the best computational methods capable of complex regression estimation which is an advantage of this method compared with the analytical and statistical methods. It is expected that the neural network can more accurately predict the rolling resistance. In this study, the neural network for experimental data was trained and the relationship among some parameters of velocity, dynamic load and tire pressure and rolling resistance were evaluated. Materials and Methods: The soil bin and single wheel tester of Biosystem Engineering Mechanics Department of Urmia University was used in this study. This soil bin has 24 m length, 2 m width and 1 m depth including a single-wheel tester and the carrier. Tester consists of four horizontal arms and a vertical arm to vertical load. The S-shaped load cells were employed in horizontal arms with a load capacity of 200 kg and another 500 kg in the vertical arm was embedded. The tire used in this study was a general pneumatic tire (Good year 9.5L-14, 6 ply) In this study, artificial neural networks were used for optimizing the rolling resistance by 35 neurons, 6 inputs and 1 output choices. Comparison of neural network models according to the mean square error and correlation coefficient was used. In addition, 60% of the data on training, 20% on test and finally 20% of the credits was allocated to the validation and Output parameter of the neural network model has determined the tire rolling resistance. Finally, this study predicts the effects of changing parameters of tire pressure, vertical load and velocity on rolling resistance using a trained neural network. Results and Discussion: Based on obtained error of Levenberg- Marquardt algorithm, neural network with 35 neurons in the hidden layer with sigmoid tangent function and one neuron in the output layer with linear actuator function were selected. The regression coefficient of tested network is 0.92 which seems acceptable, considering the complexity of the studied process. Some of the input parameters to the network are speed, pressure and vertical load which their relationship with the rolling resistance is discussed. The results indicated that in general trend of changes, the velocity is not affected by rolling resistance. Rolling resistance increases when tire pressure decreases. This is due to energy consumption for creating deflection on the body of the tire at the lower levels of tire inflation pressure. Another variable parameter is the vertical load on the wheel and its logical relation with rolling resistance using neural network. The results showed that increasing the vertical load increases the rolling resistance. Conclusions: The major purpose of this study was the feasibility of using learning algorithms for interaction between wheel and soil. The parameters of the wheel when clashes with soil are not stochastic and in spite of their complexity follow a specific model, certainly. Artificial neural network trained with a correlation coefficient of 0.92 relatively had a good performance in education, testing and validation parts. To validate the network results, the impact of some factors on the extraction process such as velocity, load and inflation pressure was simulated. The main objective of this article is comparing the network performance with basic principles and other scientific reports. In this regard, the predictions by trained neural network indicated that rolling resistance is independent of the velocity of the wheel. On the other hand, rolling resistance decreases by increasing tire inflation pressure which is a general trend similar to other studies and reports in the same mechanical condition of the soil tested. Rolling resistance changes are directly proportional to load vertical variations on the wheel in terms of quantity and quality, similar to experimental models such as Wismer and Luth.
Keywords
