**3.3 Haul profile**

Information that effectively estimates travel time, such as distance, vehicle weight, slope, and speed limit, is called a haul profile.

*Energy Efficiency Improvement in Surface Mining DOI: http://dx.doi.org/10.5772/intechopen.104262*

**Figure 2.** *Opportunities for energy conservation in the mining industry [41, 42].*

## **4. Identifying the most influential parameters**

A variety of variables influences the energy consumption of a truck. Because of the constraints of the project, it is not possible to model all of the parameters at this time. The model, therefore, includes the most important parameters. Mining energy savings opportunities can be categorized based on the latest government reports, staff operations, maintenance procedures, management systems, energy measurement, energy management parameters, and new technologies [41–43]. **Figure 2** represents the amount of energy saved and the percentage of total savings achieved by mining companies during the 2019–2020 period, based on the types of energy efficiency opportunities identified and implemented by the companies. The mining entities identified the most energy savings opportunities through energy management projects or 4.61 PJ. This accounts for 55 percent of the total potential savings determined by the mining companies.

Three main parameters have been identified as effective in reducing truck fuel consumption due to an online survey conducted for this research. The survey reached out to 60 industry professionals, who responded at a rate of 81 percent. According to the survey findings, the payload, truck speed, and the resistance of the road are the three most important factors influencing haul truck fuel consumption. Following identifying the primary effective parameters on haul truck fuel consumption in surface mines, a practical method for creating the model must be selected to predict the burnt fuel with the trucks in the mine site. ANN is the name of this method.

### **5. Artificial neural network (ANN)**

ANNs or neural networks, also known as a simulated neural network (SNN), or what is known as 'parallel distributed processing,' represents how the brain uses various methods to learn. The ANN is a collection of mathematical models intended to mimic a few of the common characteristics of natural neural networks. In some cases,

**Figure 3.**

*An example of a typical artificial neural network procedure [44].*

the unusual structure of the data processing system may be the most critical component of an ANN paradigm. **Figure 3** depicts an example of a neuronal model that consists of weighted connectors, an adder, and an activation function, among other components. These models are used in computer applications to solve complex problems that arise from user input. They do not require a mathematical description of the process-related phenomena, nor do they need any information to identify the factors that are associated with the process. Instead, they rely on acceptable errors and simple models [35, 36].

In neural networks, the node is the main component. Signals from various sources are summarized by biological nodes, which perform nonlinear operations on the results to produce output. When it comes to artificial neural networks, they are typically divided into three layers: an input layer, a hidden layer, and an output layer. According to its most basic configuration, each of the inputs and its associated weights is multiplied by the connected weight of its neighboring input. The resulting quantities and biases pass through activation functions to produce the output.

### **6. Proposed model**

Several different variables influence fuel consumption for haul trucks. The performance of a typical haul truck is illustrated in **Figure 4** by the key factors that influence it.

**Figure 4.** *Influential critical factors of performance of a typical haul truck.*

The results of this study examined the effects of the Payload (L), Truck Speed (S), and Total Resistance (TR) on fuel consumption. Burt et al. define the TR as the sum of the Rolling Resistance (RR) and the Grade Resistance (GR) [45].

$$\mathbf{TR} = \mathbf{RR} + \mathbf{GR} \tag{1}$$

When the characteristics of the tires and the haul roads are considered, this RR can be used to calculate the Rimpull Force (RF). As the truck tire rolls down the haul road, the RF measures the resistance to motion in the tire. The GR denotes the gradient of the haul road. When expressed in percentage, it is determined by the relationship between the rise of the road and the horizontal length. The truck's Fuel Consumption (FC) can be calculated with the help of Eq. (2) [46]:

Eq. (2) (Filas 2002) can be used to calculate FC.

$$\text{FC} = \frac{\text{SFC}}{\text{FD}} (\text{LF.P}) \tag{2}$$

Where SFC is the engine Specific Fuel Consumption at full power (0.213–0.268 kg/ kW hr) and FD is the Fuel Density (0.85 kg/L for diesel). The simplified version of Eq. (2) is presented by Runge [47]:

$$\text{FC} = \mathbf{0.3} (\text{LF.P}) \tag{3}$$

LF is the engine Load Factor and is defined as the ratio of average load to the maximum load in an operating cycle [48], p is the truck power (kW), and it is determined by:

$$\mathbf{P} = \frac{\mathbf{1}}{\mathbf{3.6}} (\mathbf{RF.S}) \tag{4}$$

The calculation mentioned above method does not work ideally in mine sites. The calculated consumed fuel by haul trucks using the simple formula same as Eq. (3) cannot help mine managers, operation team, and other related groups estimate fuel consumption. The accuracy of proposed straightforward approaches by researchers is not enough to allow the mine managers to make the correct decisions and improve the energy efficiency in surface mines. Based on the reasons mentioned above and to solve the business problem, this chapter introduces an innovative solution using ANN to predict truck fuel consumption based on the collected data for three effective parameters: payload, truck speed, and total resistance.

#### **6.1 Developed ANN model**

Biological nodes generate outputs by combining signals from various sources nonlinearly. A neural network is typically composed of three layers: an input layer, one or more hidden layers, and an output layer, among other things. In its most basic form, each input is multiplied by the weight of the connected input, and the result is passed through the activation functions to generate the output (see Eqs. (5)–(7)).

$$\mathbf{E\_{K}} = \sum\_{\mathbf{J=1}}^{\mathbf{Q}} \left( \mathbf{W\_{l,l,K}} \mathbf{X\_{l}} + \mathbf{B\_{l,K}} \right) \qquad \quad \text{K-1,} \quad \mathbf{2}, \quad \dots, \ \mathbf{M} \tag{5}$$

**Figure 5.** *Structure of ANN developed model.*

Where x is the normalized input variable, w is the weight of that variable, i is the input, b is the bias, q is the number of input variables, and k and m are the counter and number of neural network nodes, respectively, in the hidden layer.

**Figure 5** depicts a simplified representation of the structure of the model developed in this research. It should be noted that the hidden layer nodes are free to generate their output using any differentiable activation function they choose.

In general, the activation functions are made up of both linear and nonlinear equations, depending on the situation. Matrixes Wi,j,k, and bi,k are used to organize the coefficients associated with the hidden layer in the hidden layer. As an activation function between the hidden and output layers, Eq. (6) can be used to achieve the desired result (in this Equation, f is the transfer function).

$$\mathbf{F}\mathbf{k} = \mathbf{F}(\mathbf{E}\mathbf{k})\tag{6}$$

During the output layer's computation, the hidden layer's signals are weighted summed, and the coefficients associated with these weights are organized into three matrices: Wo,k, and Bo. The network's output can be calculated using matrix notation, as shown in Eq. (7).

$$\text{OUT} = \left(\sum\_{\mathbf{K}=1}^{\text{M}} \mathbf{W}\_{\text{O,K}} \mathbf{F}\_{\text{K}}\right) + \mathbf{B}\_{\text{O}} \tag{7}$$

It is presented in this chapter the results of a study in which different types of algorithms were investigated to determine the best back-propagation generating algorithm. First, let us compare the Levenberg-Marquardt (LM) back-propagation generating algorithm to other similar algorithms. It has the lowest mean square error (MSE), Root mean square error (RMSE), and Correlation Coefficient (R<sup>2</sup> ) of any of

*Energy Efficiency Improvement in Surface Mining DOI: http://dx.doi.org/10.5772/intechopen.104262*

the algorithms (see Eqs. (8)–(10)). In addition, network generation using the LM algorithm can be accomplished with the smallest possible Expanded Memory Specification (EMS) and a quick generating process by using the LM algorithm. The statistical criteria MSE, RMSE, and R2 are used to evaluate the accuracy of the results in accordance with the following Equations (Ohdar and Pasha 2003 and Poshal and Ganesan 2008), which are as follows:

$$\text{MSE} = \frac{1}{\mathbf{p}} \sum\_{\mathbf{r}=1}^{\mathbf{p}} \left( \mathbf{y}\_{\mathbf{r}} - \mathbf{z}\_{\mathbf{r}} \right)^{2} \tag{8}$$

$$\text{RMSE} = \left(\frac{1}{\text{P}} \sum\_{\mathbf{r}=1}^{\text{P}} \left(\mathbf{y}\_{\mathbf{r}} - \mathbf{z}\_{\mathbf{r}}\right)^{2}\right)^{\frac{1}{2}} \tag{9}$$

$$\mathbf{R}^2 = \mathbf{1} - \frac{\sum\_{\mathbf{r}=1}^{\mathbf{p}} \left(\mathbf{y}\_{\mathbf{r}} - \mathbf{z}\_{\mathbf{r}}\right)^2}{\sum\_{\mathbf{r}=1}^{\mathbf{p}} \left(\mathbf{y}\_{\mathbf{r}} - \overline{\mathbf{y}}\right)^2} \tag{10}$$

Where y denotes the target (actual), z denotes the output (estimated) of the model, (y denotes the average value of the targets, and p denotes the number of network outputs). To examine the error and performance of the neural network output, the MSE and R2 methods were used. In addition, the LM optimization algorithm was used to determine the optimal weights for the network.

The proposed ANN model for function approximation has the structure of a feedforward multi-layer perceptron neural network with three input variables and a single output. One or more hidden layers of sigmoid nodes are frequently found in the feedforward network, tracked by an output layer of linear nodes. Nodes with nonlinear activation functions are arranged in multiple layers, allowing the network to learn the linear and nonlinear connections between the input and output vectors over time. The linear output layer enables the network to generate values outside the [�1,+1] range using a linear function. The activation functions in the hidden layer (f) are the continuous differentiable nonlinear tangents sigmoid presented by Eq. (11).

$$\mathbf{f} = \tan^{-1} \text{sig}(\mathbf{E}) = \frac{2}{\mathbf{1} + \exp\left(-2\mathbf{E}\right)} - \mathbf{1} \tag{11}$$

When determining the optimal number of nodes in the hidden layer, MSE and R2 were calculated for various hidden layer densities to determine their optimal number of nodes. For 15 nodes in the hidden layer, the minimum MSE and the maximum R<sup>2</sup> (best performance) were discovered, resulting in the best overall performance (as shown in **Figure 6**).

To train the ANN model, 4600 pairing data points were randomly selected from the 6630 values of the site data that had been gathered for this study (A large surface mine located in central Queensland, Australia). The values of payload, Vmax, and TR were calculated from the site data and used to train the ANN model, which was then used to calculate the fuel consumption from the site data.

As shown in **Figure 7**, the variation of MSE occurs during the network training process: it can be seen that the error approaches zero after 25 epochs, which indicates that the desired network convergence was achieved during the training process.

**Figure 6.** *The performance of the network at different hidden nodes using the LM algorithm.*

**Figure 7.** *Neural network error diagram (MSE) during network training.*

Approximately 2030 independent samples were used to evaluate the accuracy of the network and validate the model. The test results of the synthesized network are depicted in **Figure 8**, where the vertical and horizontal axes represent the estimated fuel consumption values by the model and the actual fuel consumption values, respectively, and the vertical and horizontal axes represent the actual fuel consumption values.

**Figure 8** illustrates the accuracy of the developed model. The results show more than 85% accuracy, which is acceptable for a mining application using unstructured noisy data collected from a real mime site.

*Energy Efficiency Improvement in Surface Mining DOI: http://dx.doi.org/10.5772/intechopen.104262*

**Figure 8.** *Comparison of actual values with network outputs for test data (first quarter bisector).*

**Figure 9.**

*Correlation between Payload, S,T.R., and FCIndex based on the developed ANN model for CAT 793D.*

For a standard range of loads, **Figure 9** shows the correlation between Payload, Truck Speed, Total Resistance, and FCIndex created by the constructed ANN model for CAT 793D tested in a coal surface mine in central Queensland, Australia.

The results show that ANN could correctly predict the fuel consumed by haul trucks in different conditions. As a result, there are different ranges of consumed fuel for different haul road conditions. **Figure 9** also shows that there is the minimum area for consumed fuel in all tested scenarios. This minimum area is located close to the maximum recommended payload for the truck. It means that loading the truck with the recommended weight can help the mine managers to reduce fuel consumption.

**Figure 10.** *Correlation between gross vehicle weight, S,T.R., and FCIndex based on the developed ANN model for Komatsu HD785.*

The developed application also tested for a Komatsu truck (HD785) to validate the model for different truck's specifications. **Figure 10** shows the results of model testing for the Komatsu truck.

The minimum areas highlighted by the presented graphs in **Figures 9** and **10** illustrate the potential of deploying optimization algorithms aimed to improve energy efficiency in surface mines. This concept can be a title for further investigations and studies in the future.
