Investigating the Effect of Tractor’s Tire Parameters on Soil Compaction Using Statistical and Adaptive Neuro-Fuzzy Inference System (ANFIS) Methods

Abstract

Many factors contribute to soil compaction. One of these factors is the pressure applied by tires and tillage tools. The aim of this study was to study soil compaction under two sizes of tractor tire, considering the effect of tire pressure and traffic on different depths of soil. Additionally, to predict soil density under the tire, an adaptive neuro-fuzzy inference system (ANFIS) was used. An ITM70 tractor equipped with a lister was used. Standard cylindrical cores were used and soil samples were taken at four depths of the soil inside the tire tracks. Tests were conducted based on a randomized complete-block design with three replications. We tested two types of narrow and normal tire using three inflation pressures, at traffic levels of 1, 3 and 5 passes and four depths of 10, 20, 30 and 40 cm. A grid partition structure and four types of membership function, namely triangular, trapezoid, Gaussian and General bell were used to model soil compaction. Analysis of variance showed that tire size was significant on soil density change, and also, the binary effect of tire size on depth and traffic were significant at 1%. The main effects of tire pressure, traffic and depth were significant on soil compaction at 1% level of significance for both tire types. The inputs of the ANFIS model included tire type, depth of soil, number of tire passes and tire inflation pressure. To evaluate the performance of the model, the relative error (ε) and the coefficient of explanation (R²) were used, which were 1.05 and 0.9949, respectively. It was found that the narrow tire was more effective on soil compaction such that the narrow tire significantly increased soil density in the surface and subsurface layers.

1. Introduction

Soil is a resource with a potentially rapid degradation rate and a very slow regeneration and formation process [1]. Thus, sustainable soil use is the only solution to global problems such as food security, energy and water demands, climate change and biodiversity [2]. In order to increase productivity in the agricultural sector, farms should be a large size, in which productivity can be increased by the use of more efficient machines [3]. For many years, the agricultural sector has tended to increase the size and weight of the tractor, which increases the risk of severe soil compaction [4].

One of the causes of soil compaction in mechanized agriculture is the passage of agricultural implements and tools. Soil compaction as a form of reducing the pore space between soil particles can significantly reduce soil production capacity [5]. According to data from seven countries in Europe and North America, a 14% drop in crop yield was reported in the first year after machine traffic [6]. Soil compaction is a form of physical degradation of soil that causes compacted soil particles to be closer to each other and the disappearance of voids in the soil, reducing porosity and soil permeability. This manifests as an increase in soil bulk density [7]. One method of assessing soil compaction is to measure its bulk density. When soil compaction occurs, the bulk density of the soil increases as the mass remains constant and the volume decreases. Density reduction decreases hydraulic conductivity and thus causes surface runoff and soil erosion by water. This induces forced currents in the soil pores, thus facilitating the transport of absorbable colloidal nutrients and pesticides to deeper horizons and groundwater, thereby reducing root growth and causing a loss of nitrogen and production of greenhouse gases through nitrogenization by anaerobic processes. As a result, soil compaction is one of the main causes of environmental and agricultural problems, and as a result causes significant economic damage to society and agriculture [8].

Increasing the load on the wheels of agricultural machinery is the most important cause of soil compaction created in deeper soil layers. The topsoil of arable lands is loosened every year by plowing operations, however, the subsoil often remains intact, which means that whenever the maximum load capacity is exceeded, additional soil compaction will occur deep in the soil. In addition, these effects are lasting, and the results of natural and artificial soil loosening are disappointing [9]. Excessive movement of agricultural machinery on the field and plowing at a constant depth gradually causes soil compaction, and this compaction restricts plant root growth and reduces crop yield [10]. To remove this hard layer created in the subsoil, which limits the longitudinal growth of roots and proper soil drainage, deep plowing is required, which greatly increases the required energy [11]. Another solution is to use energy-efficient combined (active-passive or passive-passive) tillage implements in place of conventional ones to reduce soil compaction, labor, and fuel cost as well as to save time in preparing the seedbed [12]. Another aspect affecting the protection of soil structure from compaction is the evolution of tires [13]. In this article we compare the effects of five tractor tires introduced between the 1970s and 2017 on soil stress and soil structure. Soil physical properties before and after one pass of each of the five tires were measured on undisturbed soil. The results of stress measurements showed the lowest mean normal stress at all depths with newer tires, both below the center of the tire and at the edge, even though the effect decreased with increasing depth. The results indicate that tire design can play an important role in reducing soil stress for a given wheel load. Tire design might then also help to reduce the compaction risk for larger wheel loads and other soil conditions which were tested in the study.

Maintaining proper ratios between solid, liquid and gas phases is of particular importance, because increasing the solid phase (i.e., reducing the liquid and gas phases) increases the mechanical strength of the soil and consequently reduces the growth and development of plant roots inside the soil. Reducing the development of root growth causes less water and nitrogen to be available to the plant. When the machine enters the field, the solid phase of the soil increases by a certain volume and the gas phase decreases. Soil density is one of the indicators of the destruction of the physical structure of the soil, which is defined as an increase in soil bulk density (solid phase mass to total volume), or a decrease in porosity [14]. Wheel load, tire type, and inflation pressure can increase soil bulk density and play an important role in soil compaction [15]. Almost all tires significantly increase soil compaction along the wheel track, while only some increase soil compaction near the track [15]. As the distance from the wheel path increases, a general decrease in soil density occurs, especially in the subsoil. Many researchers have reported that working with low-pressure tires can significantly reduce soil compaction and increase crop yield [16,17], while high tire pressure increases soil compaction [18]. Tire pressure control is an important factor in the control of and in soil compaction. A new tire pressure control system in forest machines was developed using a PressurePro solution. It was found that the installation of an automatic tire pressure control system leads to compaction reduction by 20%. Additionally, reduced tire pressure and its automatic control contribute to a minimal reduction in humus content and soil compaction over time [19]. Another researchers have investigated the effect of different tire inflation pressures of 100 and 200 kPa on selected physical soil properties by conducting field tests. The undisturbed samples were collected both in and between the tracks at depths of 0 to 0.5 m. The results indicated that fewer negative changes were found in the variant with a lower inflation pressure for all of the soil properties. Change in soil physical properties caused by the passage of the tractor were statistically significant for both tire pressures only at depths ranging from 0 to 0.1 m [20]. The effect of three types of wide, low-inflation pressure tires with similar dimensions on mean normal stress on the soil profile was also investigated. The results revealed a very limited effect of tire construction on mean normal stress. No differences were measured beneath the center, and differences at +0.3 m were found only at 0.2 m depth for the tires at the rear axle. The effect of low tire inflation pressure was limited to the upper parts of the soil profile for beneath the center of the tire, which was significant at depths of 0.2 m and 0.4 m. It was found that in order to reduce soil stress, tire design should allow for a large contact area and low inflation pressure [21].

Machine traffic plays an important role in soil compaction. Experiments have been carried out to study the effect of multiple passes of tillage operation on soil compaction beneath the tillage depth with three different tillage systems. The authors observed that active tillage machinery generally caused more compaction compared to passive tillage machinery for the same number of passes. However, active machinery required fewer vehicular passes [22]. In plowed soils, the first pass can dramatically increase soil compaction and cause soil degradation, while in firm soil and deep areas soil compaction occurs over time and with large numbers of tire passes [23]. Research has also shown that if the traffic exceeds 10 times, the benefits of using light tractors are lost and soil compaction is greatly increased. These studies have suggested that heavy tractors with less traffic should be used as much as possible because high traffic leads to compaction in the subsoil [24,25].

One method of assessing soil compaction is to measure its bulk density. When soil compaction occurs, the bulk density of the soil increases as the mass remains constant and the volume decreases. In this method, after determining soil bulk density at different depths and, based on the soil texture, measured data are compared with the standard values to determine the level of soil compaction. For example, soil compaction is considered significant for loamy clay soil when bulk density reaches more than 1.5 g/cm³, and for sandy and loamy soil more than 1.6 g/cm³ [26].

In recent years, the artificial intelligence approach has shown that it has great power to model and predict complex systems, thus it can be an alternative method to physical models that do not respond well to a large number of input variables [26]. Because agricultural systems such as soil are so complex, most researchers have focused on artificial intelligence to model different components of agricultural systems. The fuzzy neural adaptive inference system (ANFIS) is a multi-layered adaptive network consisting of the main elements and functions of fuzzy logic systems developed by Jang [27]. Each of the fuzzy systems and artificial neural networks (ANNs) has advantages and disadvantages. The fuzzy system is able to use human language and can use human experiences and experts while it is not able to learn [28]. In other words, the fuzzy system cannot be trained using observational data and it should be noted that this method does not provide good results in unpredictable conditions [29]. However, neural networks are self-learning using data sets. At the same time, neural networks are implicit and unable to use human language. They require a wide range of tested input and output data for successful modeling [30]. Although ANN is a powerful technique for modeling various problems in the real world, it has its own weaknesses. If the input data is vague or the uncertainty is relatively high, a fuzzy system such as ANFIS may be a better option [31].

In order to evaluate the potential of ANFIS in predicting the energy efficiency indicators of the drive wheel, experiments were performed in the soil reservoir. Input parameters included tire load, wheel speed and slip, each at three different levels. ANFIS was used with a combined method of descending gradient and least squares to find the optimum learning parameters using different membership functions (MFS). The results of this study showed high accuracy (MSE = 0.0166 and R² = 0.98) in prediction [31]. A fuzzy logic model was developed to describe soil fragmentation for seedbed preparation in a combination of primary and secondary tillage tools from subsoil, moldboard plow and disc. In this research, an intelligent model based on the Mardani fuzzy modeling principles approach was developed to predict soil fragmentation during tillage operations. Model inputs were: soil moisture, tractor speed and sampling depth. The fuzzy model consisted of 50 rules in which the three parameters of mean root mean square error (RMSE), relative error (e), and coefficient of determination (R²) were used to evaluate fuzzy models. These parameters were calculated to be 0.17%, 3.95% and 0.99%, respectively. According to the results of this study, the fuzzy model can be used as a method to predict soil fragmentation during tillage operations [32]. In the study, the ANFIS model was used to predict soil compaction under tractor wheels. Model inputs included four factors: manure rate, number of passes, moisture and depth. The network structure was selected from the network segmentation type. The number of membership functions for each input was considered to be three and the membership functions of bell, triangle, trapezoid and Gaussian were tested for best performance. The results of this study showed that the ANFIS model has a higher accuracy in predicting soil density than the regression model [33].

The main objectives of this research are as follows:

• Investigating the density created in different depths of the soil due to the passage of tractor wheels of different size;
• Investigating the effect of inflation pressure changes of drive wheel and machine traffic on the soil density created at different depths;
• Modelling the effect of tire size, tire pressure, and machine traffic on soil density at different depths using statistical and adaptive neuro-fuzzy inference system (ANFIS) methods

2. Materials and Methods

Field experiments were performed to measure the bulk density of the soil under two common and narrow radial tires. The first type of tire was a standard 8 layer tire with a size of 14–38. Three inflation pressures of 80, 120 and 160 kPa were used to evaluate the effect of tire pressure on soil compaction. The second type of tire was a 10 layer radial narrow tire with a size of 9.5–48. Inflation pressures of 240, 275 and 310 kPa were investigated. For each of these tire pressures, three traffics of 1, 3 and 5 were tested. In all tests, soil samples were obtained from four depths of 10, 20, 30 and 40 cm. The farm had been plowed already, and an ITM70 tractor equipped with a bedder-ridger on a three-point hitch was used for conducting tests. In this case, the weight measured on the rear axle was 2740 kg, and on the front axle 770 kg. In all experiments, the tractor speed was constant, equal to 3 km/h. The process of adjusting the desired speed was done in such a way that the tractor was moved to a distance of 30 m on the field to achieve a speed of 3 km/h with two gears and at engine speed of 1400 and 1600 rpm. the desired speed was obtained using a calibration diagram.

A hydrometric test (ASTM D-422) was used to determine the soil texture by determining the relative amount of sand, silt and clay. It was found that there were 43.92, 32.06 and 24.02% sand, soil and clay, respectively. According to the soil texture triangle, the soil texture was loamy. A standard proctor test was used to determine the soil critical moisture. According to Figure 1, a moisture level of 25% is the critical moisture content, and the highest value of compaction occurs at this humidity.

To measure the bulk density after the tire passes, standard cylinders were placed in groups of 3 at different depths of 10, 20, 30 and 40 cm measured from the center of the cylinders to the soil surface and in line with tire path direction in the center line of the tire track (Figure 2). In order that the cylinders were placed exactly at the desired height and in one direction inside the profile, a string was drawn from the middle of the profile and from the beginning to the end. Figure 3 shows how the cylinders were exerted inside the soil vertically in the direction of the string. Each group of cylinders was about 40 cm apart so that there were no problems when removing the cylinders. It should be noted that the soil surface appears loose in Figure 3; this is because the soil had been dug to place the cylinder at a predetermined depth.

To implement the fuzzy neural adaptive model, MATLAB 2014a software toolbox under Windows was used. According to the performed experiments, soil density was defined as system output, while tire type, tire inflation pressure, traffic and soil depth were independent inlet parameters (Figure 4). The data is usually divided into three categories: training data, validation data, and test data. The training data were used in the training process to calculate the gradient and optimize the parameters. One of the points that should be considered during the training is aiming to prevent the network from becoming specialized and the occurrence of over-fitting phenomena. For this purpose, a series of data was considered for validation during the training of the network. These data are actually part of the training data. In this way, in the regular intervals of the optimization process, the data obtained from the network were checked using the validation data. In this case, network training continued until the optimization error related to the evaluation data began to increase, and as soon as this error increased to a certain value or a certain number of repetition, network training was stopped. Finally, the test data was used to test the performance of the network after training. The test data was not used during network training, but it was used to compare the error rate. Accordingly, in this research, 70% of the data were considered for training, 15% for evaluation, and 15% for testing the network.

A network separation structure was considered to create the network. Four commonly used membership functions, namely triangular, trapezoidal, Gaussian, and bell, were considered as membership functions to represent the inputs (Figure 5), and four different models were constructed using these membership functions.

The number of membership functions for each variable was considered proportional to the levels tested for that variable.

Table 1. Membership functions of input variables.

Input Variable	Membership Function
Tire Type	Narrow		Wide
Traffic	max	ave	min
Tire pressure	high	normal	low
Depth	subsoil	mid soil	topsoil

shows the membership functions for each variable.

The output membership function in this network is linear. A hybrid optimization method was used for network training. The number of rules created by the network was 54. Given that the number of rules created was very high, only a few of them are presented in

Table 2. Some rules created in the ANFIS model.

Input Parameters					Output Variable
Rules	Tire Type(T)	Traffic (P)	Tire Inflation (IP)	Soil Depth (D)	Special Mass
1	Wide	Min	Low	Topsoil	BD = 0.09035T + 0.06862P + 0.09035IP + 0.09035D + 0.09035
8	Wide	Min	High	Middle soil	BD = 0.01578T + 0.06565P + 0.07889IP + 0.03156D + 0.01578
16	Wide	Ave	High	Topsoil	BD = −0.00517T + 0.04506P − 0.02585IP −0.00517D − 0.00517
24	Wide	Max	Normal	Subsoil	BD = 0.00329T + 0.0256P + 0.009871IP + 0.009871D + 0.00329
32	Narrow	Min	Normal	Middle soil	BD = 0.05436T + 0.07231P + 0.08154IP + 0.05436D + 0.02718
40	Narrow	Ave	Normal	Topsoil	BD = −0.02717T+ 0.04873P − 0.04075IP − 0.01358D − 0.01358
48	Narrow	Max	Low	Subsoil	BD = 0.009582T + 0.02666P + 0.004791IP + 0.01437D + 0.004791
56	Narrow	Max	High	Subsoil	BD = 0.004228T + 0.02739P + 0.01057IP + 0.006341D + 0.002114

.

Finally, the performance of the models was evaluated based on the statistical criteria presented in Equations (1) and (2).

$$\mathrm{MAPE}=\frac{100\%}{N}{{\displaystyle \sum }}_{i=1}^{\mathrm{n}}\frac{{\mathrm{Y}}_{\mathrm{measured}}-{\mathrm{Y}}_{\mathrm{predicted}}}{{\mathrm{Y}}_{\mathrm{measured}}},$$

(1)

$$R^2=1-\frac{{{\displaystyle \sum }}_{i=1}^N{\left({\mathrm{Y}}_{\mathrm{measured}}-{\mathrm{Y}}_{\mathrm{predicted}}\right)}^2}{{{\displaystyle \sum }}_{i=1}^N{\left({\overline{\mathrm{Y}}}_{\mathrm{measured}}-{\mathrm{Y}}_{\mathrm{predicted}}\right)}^2},$$

(2)

where:

• MAPE: the absolute mean percentage of system error
• R²: is the coefficient of determination
• N: the number of samples
• Y_predicted: the predicted values
• Y_measured: the measured values
• $\overline{\mathrm{Y}}$_measured: the average of measured values.

3. Results and Discussion

Table 3. Analysis of variance of soil bulk density change due to tire type, traffic and depth.

Source of Variation	Df	Sum of Squares	Mean Squares	F
Soil depth	3	0.054	0.018	715.6 **
Traffic	2	0.015	0.008	302.1 **
Tire type	1	0.113	0.113	4490 **
Depth × pass	6	0.001	0.00016	5.5 **
Depth × tire type	3	0.002	0.00067	25.6 **
Pass × tire type	2	0.001	0.0005	13.5 **
Depth × pass × tire type	6	0	0	1.6 ns
Error	46	0.001	0	-

**: Highly significant, ns: Not significant.

shows the analysis of variance of soil bulk density for two types of ordinary and narrow tire. Data were analyzed in a 4 × 3 × 4 factorial design based on a randomized complete block design. The data coefficient of variation (CV) was 0.44%. Average initial soil bulk density was 1.017 g/cm³, and based on data of initial bulk density it was uniform in all plots. In comparison with initial bulk density, statistical analysis showed that the main effects of tire type, traffic and depth on soil density change were significant at the level of 1% probability. The interactions of depth with traffic, depth with tire type, and traffic with tire type on soil bulk density were found to be significant at the 1% probability level. However, the combined interaction of depth, tire type and traffic on soil bulk density was not found to be significant at the 1% probability level.

3.1. The Binary Effect of Tire Type and Depth on Soil Bulk Density

The binary effect of tire type and depth on soil bulk density was evaluated using a Duncan test at 1% probability level, and is presented in Figure 6. According to the results, the maximum value of bulk density was 1.217 g/cm³ for narrow tire at a depth of 10 cm, and the lowest value was 1.061 g/cm³ which occurred for common tires at a depth of 40 cm. For both tires, a significant decrease in the soil bulk density was seen with increasing depth, but the intensity of these changes in the narrow tire and especially in shallow depths was greater than for the common tire. The bulk density of soil for the narrow tire, even at a depth of 40 cm, was greater than the bulk density of soil for a normal tire at a depth of 10 cm. The effect of tire size on stress applied to different soil layers was examined [34]. Their results showed that a narrow tire created much more stress in different layers of soil than a common tire, and as a result, it compacted the soil more. Stress for a narrow tire at a depth of 50 cm was higher than the stress of a common tire at depth of 30 cm.

3.2. The Binary Effect of Tire Type and Traffic on Soil Bulk Density

The results of comparing the mean interaction of tire type and depth on soil bulk density using a Duncan test at 1% probability level are presented in Figure 7. In both tires, the soil bulk density increased with increasing traffic. According to the results, the highest mean value of soil density was found for the narrow tire after five passes with a value of 1.185 g/cm³ and the lowest for a normal tire after one pass with a value of 1.112 g/cm³. It is noteworthy that the bulk density of soil for a narrow tire, even at one pass, is more than the bulk density of soil for a normal tire in five passes.

3.3. Results of Modeling Soil Bulk Density Using ANFIS

Four different models were developed to predict soil bulk density.

Table 4. Specifications and evaluation results of adaptive fuzzy-neural inference system models (ANFIS).

Network Structure	Membership Function Type		Number of Membership Functions		Optimization Method	Test Result
	Input	Output	Input	Epochs		Coefficient of Determination (R2)	Relative Error (e)
Grid Partition	Triangular	Linear	2-3-3-3	20	hybrid	0.992	1.5
Grid Partition	Trapezoidal	Linear	2-3-3-3	20	hybrid	0.99	1.75
Grid Partition	Gaussian	Linear	2-3-3-3	20	hybrid	0.982	2.2
Grid Partition	Bell-shaped	Linear	2-3-3-3	20	hybrid	0.995	1.05

presents the structural parameters of the models along with their statistical criteria to evaluate their performance. According to the results presented in Table 4, it is clear that all models have a high ability (R² ≥ 0.99 and ε (%) ≤ 2) to predict. The best model for predicting soil bulk density is a model that uses the bell membership function (Gbellmf) (R² = 0.9949 and ε (%) = 1.05).

Table 5. Statistical characteristics of stepwise regression model for predicting soil bulk density.

Model	Unstandardized Coefficients		Standardized Coefficients	t	Sig
	Beta	Std. Error	B
Constant	0.98	0.006		154.528	<0.0001
Tire type	0.08	0.003	0.705	30.916	<0.0001
Depth	−0.002	0.001	−0.458	−20.091	<0.0001
Inflation pressure	0.031	0.002	0.444	19.452	<0.0001
Traffic	0.008	0.001	0.245	10.748	<0.0001

presents the statistical characteristics of the stepwise regression model for predicting soil bulk density. The ANFIS models have poorer performance than the regression model due to a low coefficient determination of 0.96. However, regression models have valuable advantages, such a the fact that ANFIS models do not provide any information about the internal structure of the model and the relationships between independent and dependent variables, while the regression model directly deals with the impact of each factor and tries to provide a model in which the importance and impact of each factor is clear. According to Table 5 and the standard coefficients mentioned therein, type of tire, soil depth, inflation pressure, and finally the number of passes have the greatest impact on the bulk density.

Figure 8 shows the relationship between measured and predicted values using ANFIS models under different operating conditions. In addition, the deviation between the measured values and the predicted values was calculated and plotted by ANFIS and multiple regression models. Figure 9 shows that the deviation distance of the predicted values of the ANFIS model (−0.011 to 0.0085) was much less than the deviation distance of the predicted values of the multiple regression model (−0.021 to 0.033).

Figure 10 shows the three-dimensional variation of soil bulk density with the interaction of tire type and soil depth. It indicates that with increasing depth, the bulk density of the soil decreased, indicating that the compaction created by the tire traffic of the tractor in the topsoil was higher than in the sub layers. It was found that at all depths the density change by the passage of common tire was less than the density created by the narrow tire. Such a low pressure at the soil surface and soil behavior was investigated using a flat tire [3].

Figure 11 shows the binary effect of tire pressure and tire traffic on the soil bulk density. As the tire pressure decreases, the bulk density of the soil decreased. As the pressure decreased, the contact area between the tire and the soil increased, which reduces the pressure on the soil, resulting in decreased soil compaction and this was reported by other researchers [35,36,37]. With increasing traffic, the bulk density of the soil increased [33,38,39].

Figure 12 illustrates with a surface plot of the interaction of tire type and traffic on the bulk density of the soil. With increasing traffic, when using a narrow tire, the soil bulk density increased sharply, but using a common tire was less effective on soil bulk density increment. This indicates that narrow tires are very destructive on soil structure.

Figure 13 shows the interactive effect of tire type and tire inflation pressure on soil compaction. The soil bulk density changed strongly with increasing pressure when using a narrow tire, but less changes occurred when using a wide tire with increasing pressure.

4. Conclusions

• The results showed that using a narrow tire increased the soil bulk density in different layers of soil much more in comparison with a common tire. A narrow tire compacted the soil more and increased the soil density, even at a depth of 50 cm, more than a common tire at a depth of 30 cm. Additionally, the soil bulk density of soil for a narrow tire at a depth of 40 cm was greater than that of a normal tire at a depth of 10 cm.
• In both tires, the soil bulk density of the soil increased with increasing tire wheeling. According to the results, the highest mean value of soil density for narrow tire was after five passes, with a value of 1.185 g/cm³. The lowest value was for a common tire after one pass, with a value of 1.112 g/cm³. It is noteworthy that the bulk density of soil for a narrow tire after one pass was more than the bulk density of soil for a common tire after five passes.
• Using a narrow tire with tire wheeling increased the soil bulk density very sharply, but when using a wide tire, with increasing passes, fewer changes in the bulk density of the soil occurred. This indicates that using narrow tires should be considered to be very destructive on the soil structure. Additionally, with increasing tire pressure, and when using a narrow tire, the soil bulk density changed much more than when the common tire’s pressure changed.