Energy Efficiency

, Volume 9, Issue 1, pp 129–140 | Cite as

A method for data-driven evaluation of operator impact on energy efficiency of digging machines

Original Article

Abstract

Material handling (including digging) is one of the most energy-intensive processes in mining. Operators’ skills and practices are known to be some of the major factors that affect energy efficiency of digging operations. Improving operators’ skills through training is an inexpensive and effective method to improve energy efficiency. The method proposed in this work uses data collected by monitoring systems on digging equipment to detect the monitored parameters that lead to differences in energy efficiency of operators (responsible parameters). After data extraction, removing the outliers, and identifying the operators with sufficient working hours, correlation analysis can be used to find parameters that are correlated with energy efficiency. Regression analysis on pairs of operators is then used to detect responsible parameters. Random sampling is used to overcome missing data issues in the analysis. This statistics-based method is simple and adequately accounts for the high variability in data collected from these monitoring systems. The proposed method was illustrated using data collected on five operators working on a 64-m3 (85 yd3) Bucyrus-Erie 1570w dragline. The case study results show that dump height and engagement/disengagement position of the bucket are the most likely parameters to cause differences between energy efficiency of these operators. On the other hand, cycle time, payload, and swing in time are least likely to influence differences in operator energy efficiency.

Keywords

Operators’ skills and practice Operators’ performance Energy efficiency Mining and digging equipment Regression analysis 

Introduction

The Energy Information Administration (EIA) reported that, in 2010, the industrial sector consumed about 58.6 trillion kWh (one half of the world’s total delivered energy). It is expected that by 2040, this amount will increase to 90 trillion kWh. In 2010, the US Department of Energy (DOE) estimated that the US mining industry consumes 365 trillion kWh, annually, which is about 5.3 % of the total energy consumed by the industrial sector in the USA. This is in spite of the fact that mining directly contributes to less than 1 % of gross domestic product (DOE/EIA 2013; National Mining Association (NMA) 2013; U.S. DOE 2007).

A decreasing trend in mining operating cost has enabled mining companies to extract lower-quality material and bigger and deeper mines. Therefore, it is expected that energy consumption, and as a result energy cost and greenhouse gas emissions, of mining operations will increase significantly in the future for the same amount of production (Norgate and Haque 2010). Stricter regulations on emissions and increase in energy prices have forced the mining industry to improve energy efficiency of their operations in order to keep their operations viable. The US DOE estimates that there is a potential to reduce annual energy consumption of the mining industry to 169 trillion kWh, which is about 46 % of current annual energy consumption. This will result in 35.2 million ton reduction in CO2 emissions. Digging equipment, such as hydraulic shovels, cable shovels, draglines, continuous mining machines, and long-wall mining machines, consume about 23 trillion kWh, annually, and practically about 44 % of this can be saved (U.S. DOE 2007).

In the past, one of the main challenges in identifying the best strategies to improve energy efficiency of mining operations was the limited information on energy consumption (Mielli 2011). Installation of real-time monitoring systems, such as AccuWeigh™ (Drives& Controls Services, Inc., Tyler, TX, USA), on digging equipment has provided information to determine key operating parameters such as energy consumption, cycle time, and payload in real-time. The data created by such monitoring systems is a cornerstone of any analysis on performance and efficiency of digging equipment.

Different studies have identified equipment characteristics, operating conditions, mine plans and designs, and operator’s practices and skills as major factors that affect the performance of digging equipment (Awuah-Offei et al. 2011; Bogunovic and Kecojevic 2011; Hettinger and Lumley 1999; Kizil 2010; Lumley 2005). This work focuses on the effect of operator’s skills on energy efficiency of digging equipment. Operator behavior (practices) is one of the cheapest and simplest factors to improve, and it can greatly affect the efficiency and productivity of a mining operation. For example, simulation experiments conducted by Awuah-Offei (2005) on cable shovels suggest that an operator who operates near optimal, all the time, with a 44.3-m3 (58 yd3) bucket can save over 1.42 million kWh/year/shovel in electricity consumption for the digging cycle alone, when compared to an average operator. It has also been well documented that properly designed training programs can be used to affect operators’ practices (Bernold 2007). Many studies have investigated the significant role of operators in performance and efficiency of operation using different criteria and statistical methods (Abdi Oskouei and Awuah-Offei 2014; Awuah-Offei et al. 2011; Bernold 2007; Komljenovic et al. 2010; Patnayak et al. 2007; Vukotic and Kecojevic 2014)

Operators’ practices have been shown to affect different indicators of operators’ performance. Lumley (2005) used productivity (amount of material moved during a specific period) to study the differences between dragline operators’ performance in Australia. He detected a 35 % difference in productivity between the best and worst operators. Patnayak et al. (2007) used productivity as a performance indicator of cable shovel operators but took advantage of other parameters such as hoist power and average number of dig cycles to load a truck (limited to truck size and shovel bucket size). Their work observed significant differences in hoist and crowd motor power draws of four teams of operators. Using energy consumption as an indicator, when judging operators’ performance, has the advantage of reducing energy consumption and associated emissions per unit of product.

In this study, energy efficiency is used as the indicator of operators’ performance. Energy efficiency is defined as the ratio of useful work done (energy output) to the input energy (Zhu and Yin 2008). In cases where either energy output or input cannot be measured easily, proxy parameters have been used in their place. A more detailed discussion of different proxy parameters used to calculate energy efficiency in mining applications can be found in Abdi Oskouei and Awuah-Offei (2014). An example of an energy efficiency indicator with a proxy is the operator performance indicator (OPI) suggested by Komljenovic et al. (2010) to evaluate operators’ performance. OPI can be calculated by dividing the total production of digging equipment (a dragline in this case) over a specified period by the energy consumption. This variable is essentially the inverse of energy efficiency, where material produced is a proxy for useful work, and it captures both input energy and output work and can be used on any size equipment.

With the widespread use of equipment monitoring systems, there is adequate data available to estimate energy efficiency performance of each operator and examine whether differences exist (i.e., detect statistically significant difference in energy efficiency of operators). The next step after determining significant differences between operators’ performance is to identify the factors that cause one operator to perform better (be more energy efficient) than the other (what the authors call “responsible parameters”) in order to facilitate “smarter” (data-driven) training programs. There are different methods of training operators such as best practice meetings, crew coaching, trial and error, and simulators (Bernold 2007; Norman 2011). In the best practice meetings, the steps of the digging sequence are discussed with the operator in detail to teach best practices. Crew coaching takes place between a more experienced (lead) and less experienced operator. The lead operator will observe the less experienced operator during operation and provide him/her with positive feedback. Knowledge of responsible parameters, among a particular operator population, and their influence on energy efficiency and performance of operators can help operators focus on the significant practices to improve energy efficiency. Besides, it will also decrease the losses caused by trial and error. Identifying responsible parameters can help simulator-based training systems to report these parameters in the session reports so the trainer can go over these indicators with the operator.

There is very little work in the literature that examines the effect of digging equipment operator practices on energy efficiency. Awuah-Offei (2005) used kinematics and dynamics models of a cable shovel to identify causes of differences in operators’ energy efficiency during digging. The work concentrated on the effect of crowd and hoist speeds (the crowd arm and hoist rope are the two actuators used in the digging motion of a cable shovel) on energy consumption of cable shovels. However, this method is limited to cable shovels and can only be extended to other digging equipment after modeling the kinematics and dynamics of such equipment to predict energy efficiency (Awuah-Offei 2005).

In this work, a method is introduced to determine parameters that are responsible for differences between energy efficiency of operators (responsible parameters). The method uses detailed data collected by real-time monitoring systems as input and uses statistical approaches to detect the responsible parameters. This method can be used on any digging equipment with a real-time monitoring system. The proposed method is presented in the next section, and a case study is presented in following section to illustrate the method. The case study illustrates the usefulness of the proposed approach. The method presented in this paper can only identify responsible parameters, if they are monitored by the equipment monitoring system.

Method

The proposed method in this work is designed to identify the monitored parameters that lead to differences in energy efficiency of operators. There are five main steps in the method: (1) data extraction, (2) preliminary data analysis, (3) correlation analysis, (4) responsible parameter evaluation, and (5) drawing inferences on responsible parameters. Extracting a portion of data from the database of an equipment monitoring system can help to reduce the size of the working data while keeping the conclusions unbiased. Preliminary data analysis includes removing the outliers from the data and defining the minimum duration an operator has to work in order to be included in the analysis. In the next step, the monitored parameters which are correlated with energy efficiency are identified, and then linear regression analysis based on pairwise comparison of operators will be used to detect responsible parameters. Conclusions are made after removing the effect of random sampling and assessing the results across all pairs of operators in the final step.

Step 1: data extraction

Generally, monitoring systems on digging equipment collect and store a variety of operating parameters such as cycle times (total and sub-cycles), energy consumption, and payload in each cycle of operation depending on the system setup. This data is a great resource for assessing equipment productivity and operator performance. The data can also be mined for process improvement including identifying the best strategies to improve energy efficiency. The size of these databases can lead to data overload, which is a challenge in such data mining for process improvement. One can extract an adequate and relevant portion of the data from the database by considering the work schedule of the particular piece of equipment, number of operators, and location of the operation. The extracted data should be large enough to average out different operating conditions such as material type being mined and weather conditions for an unbiased conclusion on operators’ performance. The analyst should also consider which parameters (tables and fields in the database) are relevant to energy efficiency of the equipment.

Step 2: preliminary data analysis

Preliminary data analysis includes removing outliers and erroneous records and identifying which operators would be included in the analysis based on the total time they worked within the period of study. Anomalous operating cycles and errors must be removed prior to data analysis to prevent inaccurate inferences. Removing mild outliers helps to eliminate cycles with irregular operating parameters.

Often, mines do not operate with similar operator schedules, especially on the largest and most energy-intensive units. For example with draglines, it is common to have a lead operator and a backup operator, which leads to significantly different operating hours for the two operators over the same time period. Also, normal labor issues related to absenteeism and operators with multiple skills in a shift-based working environment affects how many hours each operator runs a particular machine during a defined period. Including operators with insufficient working hours in the analysis can result in inaccurate inferences. The mean standard error statistic is an acceptable representation of the amount of support that each operator provides in the analysis (Biau 2011). It can be concluded that only operators with mean standard error higher than a specified cutoff value have sufficient working hours to be included in the analysis. In order to have a desired operator in the analysis, the period of data collection can be extended and more data points (cycles) can be retrieved from the database. However, in cases where there is a large disparity in schedules of operators, increasing the period of data collection to obtain enough data for the least scheduled operators will still result in large disparities in working hours, which may also not be desirable.

Step 3: correlation analysis

Monitoring systems record different operating parameters based on their setup. The goal of this step is to identify parameters that affect energy efficiency. Correlation analysis can be used to detect parameters that are correlated with energy efficiency and are, therefore, eligible for further analysis to determine whether they are the cause of differences in operator performance. Correlation coefficients, such as the Pearson (ρ), Spearman (r), and Kendall (τ) coefficients, measure the strength of the relationship between two variables. They can take values between −1 and 1: A value of 1 indicates a perfect positive correlation and a value of −1 indicates a perfect negative correlation. When the variables are independent, then correlation coefficient will be zero. The parameters that are correlated with energy efficiency (correlated parameters) can be identified based on the value of correlation coefficients and the desired confidence level. The p value of the null hypothesis (e.g., H0: ρ = 0) can be estimated using the Student’s t distribution. This allows one to make the inference, at a particular confidence level, whether to accept or reject the null hypothesis of no correlation between the two random variables under consideration. In cases where the correlation coefficient is too small (say below 0.15), the correlation may be of no practical meaning, even if the null hypothesis is rejected.

Once this correlation analysis is completed and all correlated parameters have been identified, data matrices containing energy efficiency and the correlated parameters are created for each operator from the database. Figure 1 illustrates the data matrix of operator i with energy efficiency (η) in the first column, n correlated parameters (par) in the next n columns, and ci rows (cycles).
Fig. 1

Data matrix of operator i with ci cycles and n correlated parameters

Step 4: responsible parameter evaluation

Handling the differences in number of cycles

The proposed method is based on pairwise comparison of all operators using the data matrices (Fig. 1) of the operators. For nOp operators, there exist \( np=\left(\underset{2}{nOp}\right) \) pairs of operators. The approach suggested here requires an equal number of cycles for each pairwise comparison. In reality, because of high variability in cycle times, the chance of getting equal number of cycles for two operators even in equal working hours is very low. Also, as explained earlier in step 2, other conditions like varying operator schedules result in different working hours for the operators in the same study period. This results in a situation where there is “missing data,” an issue common in many scientific and engineering research. Assuming that the number of cycles for operator i is greater than number of cycles for operator j (ci > cj), the pattern of the data matrices and the missingness for this pair (p) can be illustrated as in Fig. 2. In this example, ci–cj cases (cycles with energy efficiency and all the correlated parameters) are “missing” from the Opr j matrix.
Fig. 2

Pattern of data matrices for operator i and operator j (pair p). In this case operator j is missing ci–cj cycles

The literature identifies various response mechanisms, and the attendant analysis, for explaining the cause of missing data. Missing at random (MAR) is a response mechanism in which the probability of missingness for a data point (Xik) does not depend on the missing data but may depend on the observed data (Little 1992; Schafer and Graham 2002). Considering the nature of the data (the missing cycles and the probability that those cycles are not captured do not depend on the values of observed or unobserved cycles), it is assumed that the response mechanism is MAR.

Complete case analysis (CAA) is a robust and unbiased approach that can be used when the missingness is at random. In CAA, a case will be deleted from the data matrix if any of the parameters in that case are missing (Little 1992; Schafer and Graham 2002). When pairing operator i and operator j (pair p), with the assumption of MAR, ci–cj cycles will be considered as cases with missing data in the data matrix of operator i and will be removed from this matrix. The deletion should be at random to avoid bias in the conclusions.

Identifying responsible parameters in pairwise comparisons

For each pair of operators, a difference matrix is created by subtracting the two “treated” data matrices (data matrices with similar number of rows). Linear regression is used to fit a linear model to the difference matrix where Δη (difference between energy efficiency vectors of paired operators) is the dependent variable vector and Δpar matrix contains the independent variable (predictor variable) matrix.

The significance of coefficient test, at desired confidence level, identifies parameters with regression coefficients that are significantly different from zero. These parameters are designated as parameters that are responsible for the differences in energy efficiencies of operator i and operator j (pair p). The output of the coefficient test is saved as a binary variable output vector, b = (b1, b2, ⋯, bn)T: The output is 1 (bk = 1) if the coefficient is significant (the regression coefficient is non-zero at the desired confidence level) and 0 (bk = 0) if the coefficient is not significant.

This pairwise comparison using linear regression of the difference matrices is a unique and critical component of the proposed method. Here, it is assumed that if two operators have significant differences in a particular parameter due to their preferences or operating styles and this difference is a significant contributor to their different energy efficiency profiles, then the regression coefficient of the difference matrices will be non-zero. There is always going to be some disparity between the monitored (correlated) parameters of two operators. It is known that these correlated parameters affect energy efficiency (otherwise those parameters will not be included in step 4). The question is to identify those correlated parameters that are responsible for observed differences in energy efficiency. By conducting regression analysis on the difference matrices, the analyst determines whether the differences in correlated parameters explain the differences in energy efficiency.

Obviously, this approach is limited by the use of linear regression. The relationship between energy efficiency and some of the parameters may not be linear. This requires non-linear regression analysis. However, the reason for using linear regression is the computational expense of the entire algorithm. As explained in the next paragraph, the regression analysis needs to be repeated several times for each pair of operators (to prevent bias from deleting some data). Non-linear regression results in non-linear optimization problems that are relatively computationally expensive. Linear regression has worked well in the test case shown in this work (even though some parameters may not be linearly related to energy efficiency), and the authors believe that it will be effective in identifying the most responsible parameters. It must be noted, however, that in the test case presented in this work, the authors used Pearson’s linear correlation coefficients to identify all correlated parameters.

To draw correct inferences from the regression analysis, it is critical to consider the effect of randomly selecting equal number of cycles from operators when building the difference matrix. To reduce the effect of random sampling error, the process of selecting (ci–cj) cycles from operator i for deletion (CCA) and regression analysis is repeated nr times. For each run of the process, one output vector b is created. The output matrix of a selected pair of operators (pair p) contains all nr output vectors (Fig. 3). Final conclusion for the selected pair is made based on vp and the desired significant level. The elements of vp for each parameter are calculated by summing the binary output of coefficient test (bk) of that parameter over the all nr runs of the process. A parameter is deemed responsible for the difference between energy efficiency of two operators if at significance level of α, the value of vp is greater than (1−α) (nr). Figure 4 demonstrates the overall discussed algorithm.
Fig. 3

Output matrix of pair p

Fig. 4

Linear regression analysis algorithm

Step 5: drawing inferences on responsible parameters

The result at the end of step 4 is limited to the individual pair (pair p) under study. To expand the result to all the operators in the particular sample, the inference should consider all the np pairs of operators. Considering the output vector (vp) of all pairs and number of times that a parameter is recognized as responsible parameter, at a specific significant level, the probability of being a responsible parameter can be calculated for each parameter.

Case study

Data collected from a Bucyrus-Erie 1570w dragline with bucket capacity of 64 m3 (85 yd3) during 1 month is used here to illustrate the suggested method. Draglines are one of the most energy critical machines in strip mines, commonly used for removing the overburden to expose coal seams for extraction. Some properties of dragline operation include simple and low-cost operation, high production rate, simple mine planning, and high capital and maintenance cost.

Dragline operation, not including the walking process, is a cyclic process. A cycle of a dragline operation consist of filling the empty bucket by dragging it on the (blasted) material, hoisting the bucket, swinging out to the dumping pile, dumping, returning (swinging in) to the digging spot, and positioning the bucket to start the next cycle. The drag and hoist drive mechanisms enable the bucket to move horizontally and vertically using electrical motors, gear reductions, wire ropes, and wire rope drums. Swing units (each consists of a vertically mounted electric motor and gear reductions), in the swing mechanism, are mounted to a rotating frame and drive the main swing shaft. These units swing the dragline boom in order to position the bucket properly for loading or dumping (Humphrey 1990). Table 1 shows the detailed list of motors and generators in the dragline drive for the Bucyrus-Erie 1570w in this study.
Table 1

Electrical configuration of the study dragline drive system

Quantity

Motors/generators

2

1.49 MW (2000 hp) - 4 unit MG sets (motor generator sets)

2

2.24 MW (3000 hp) - 5 unit MG sets (motor generator sets)

6

1 MW (1300 hp) hoist motors

4

1 MW (1300) hp drag motors

4

0.78 MW (1045 hp) swing motors

4

0.37 MW (500 hp) propel motors

Step 1: data extraction

AccuWeigh™ monitoring system is installed on the dragline used in this study. The monitoring system was modified to record and store energy consumption of drag, hoist, and swing motors for each cycle. Energy efficiency in each cycle was calculated after data extraction by finding the ratio of payload (proxy for useful work done) to total energy consumption of the swing, drag, and hoist motors (Eq. 1):
$$ \eta (i)=\frac{P(i)}{E_t(i)}=\frac{P(i)}{E_s(i)+{E}_d(i)+{E}_h(i)} $$
(1)
where η(i) is energy efficiency in cycle i, P(i) is payload in cycle i, Es(i) is swing energy in cycle i, Ed(i) is drag energy in cycle i, and Eh(i) is hoist energy in cycle i.

The monitoring system recorded 44 parameters in each cycle. Fourteen parameters (Table 3) were extracted from the database as relevant parameters for this analysis (for the full list of parameters, see Abdi Oskouei (2013))

Step 2: preliminary data analysis

In this step, outliers and erroneous records were removed. During the data collection period, 13 operators worked on this dragline to remove the overburden from the coal seam in one pit. Mean standard error of energy efficiency was used to identify operators with sufficient working hours. Five operators with working hours greater than 32 h were included in the analysis (Abdi Oskouei 2013) (Table 2).
Table 2

Mean energy efficiency of five operators with sufficient working hours

Operator

No. of cycles

Total operating time (h)

Mean energy efficiency per cycle (t/kWh)

A

3897

56.91

11.23

B

3611

54.62

10.37

C

3350

49.60

11.14

D

3058

45.64

10.60

E

2211

32.77

11.91

Step 3: correlation analysis

Pearson correlation was used to examine the linear correlation between relevant parameters and energy efficiency. Table 3 shows the Pearson correlation coefficient and p value for each parameter. At 95 % confidence level, all tests rejected the null hypothesis of no linear correlation except for swing out time (time elapsed from the moment the bucket is detected to be full to the moment the bucket is detected to be empty). It can be concluded that all the parameters except swing out time are linearly correlated with energy efficiency.
Table 3

Pearson correlation coefficient

Parameter

Symbol

ρ

p value

Dump height

Dh

−0.6560

<0.001

Hoist energy

Eh

−0.5857

<0.001

Drag distance (vertical)

DDv

−0.5089

<0.001

Drag energy

Ed

−0.4569

<0.001

Drag distance (horizontal)

DDh

−0.4807

<0.001

Load bucket (digging) time

lbt

−0.4548

<0.001

Dump time

Dt

−0.3050

<0.001

Cycle time

Ct

−0.3755

<0.001

Swing energy

Es

−0.2724

<0.001

Swing in time

Sit

−0.3362

<0.001

Spot time

St

−0.1725

<0.001

Angle swing out

θo

−0.1556

<0.001

Swing out time

Sot

0.0123

0.0913

Payload

P

0.2429

<0.001

Step 4: responsible parameter evaluation

Considering the result of correlation analysis, a linear model of energy efficiency difference for each pair can be written as Eq. 2:
$$ \varDelta \eta ={k}_0+{k}_1\varDelta {D}_h+{k}_2\varDelta D{D}_v+{k}_3\varDelta D{D}_h+{k}_4\varDelta l{b}_t+{k}_5\varDelta {D}_t+{k}_6\varDelta {C}_t+{k}_7\varDelta S{i}_t+{k}_8\varDelta {S}_t+{k}_9\varDelta {\theta}_0+{k}_{10}\varDelta P $$
(2)
Data matrices for all five operators (leading to a total of ten pairs of operators) were created and served as input into the analysis in this step. The varying number of cycles (data points; Table 2) resulted in varying number of rows in the data matrices. Therefore, equal numbers of samples were selected at random for each pair. Linear regression was used to fit Eq. 2 to the difference matrix data. Testing for significance of coefficients (k) was carried out at 95 % confidence level. The result of this test was then recorded in output vector b which contains binary variables based on the output of the significance of coefficients test. The random cycle selection and linear regression analysis was repeated 30 times (nr = 30) for each pair in order to reduce the effect of random sampling. Then, v for each pair vector was created as illustrated in Fig. 3. Table 4 shows vp for all ten pairs in this study.
Table 4

The result of 30 runs regression analysis on all pairs of operators

  

Dump height

Drag distance (vertical)

Drag distance (horizontal)

Load bucket time

Dump time

Cycle time

Swing in time

Spot time

Angle swing out

Payload

Pair 1

v1

30

30

30

4

30

3

3

30

14

14

Pair 2

v2

30

30

30

30

30

16

16

30

30

3

Pair 3

v3

30

30

30

26

8

6

28

30

4

15

Pair 4

v4

30

30

30

30

30

2

10

30

8

12

Pair 5

v5

30

30

30

0

30

2

16

30

18

2

Pair 6

v6

30

30

30

8

8

26

29

30

30

13

Pair 7

v7

30

30

30

20

30

0

30

30

30

1

Pair 8

v8

30

30

30

5

30

10

3

21

22

6

Pair 9

v9

30

30

30

22

30

5

6

30

30

8

Pair 10

v10

30

30

30

30

30

27

7

30

2

30

A parameter is recognized as a responsible parameter, in each pairwise comparison, if the number of times it has a non-zero regression coefficient (at 95 % confidence), in 30 runs, is more than 28 (confidence level of 95 %).

Step 5: drawing inferences on responsible parameters

The final conclusion was made based on the number of times a parameter is determined to be a responsible parameter across all pairs of operators. For example, payload is found to be a responsible parameter when comparing pair 10 (30 times out of 30). Yet, when comparing all the other pairs, payload is a responsible parameter in only a few runs (less than 15 runs out of 30 for pairs 1 to 10). There needs to be a means to identify those parameters that are consistently important. In this work, the probability of being a responsible parameter is calculated for all pairs (Table 5). In Table 5, dump height and drag distances (vertical and horizontal) have 100 % probability of being responsible parameters. This means that differences in dump height and drag distances (vertical and horizontal) caused the differences between operators’ energy efficiency in all 10 pairs. On the other hand, cycle time was not recognized as the reason for significant differences between operators’ energy efficiency in any pairs (in Table 4, number of times cycle time is found to be responsible is less than 28 in all pairs).
Table 5

Final conclusion across all pairs of operators

 

Total

Probability

Dump height

10

100 %

Drag distance (vertical)

10

100 %

Drag distance (horizontal)

10

100 %

Spot time

9

90 %

Dump time

8

80 %

Load bucket time

4

40 %

Angle swing out

4

40 %

Swing in time

2

20 %

Payload

1

10 %

Cycle time

0

0 %

Discussion

The final result shows that there is a high chance that dump height, drag distances (vertical and horizontal), spot time, and dump time are responsible parameters. On the other hand, cycle time, payload, and swing in time have low probability of being responsible parameters. This means that among this group of operators, dump height, drag distances (both vertical and horizontal), spotting time, and dumping time are the most likely causes of differences in energy efficiency and should be targeted for improvements.

Previous studies have shown that payload, cycle time, digging time (load bucket time) and digging energy, fill factor, and engagement and disengagement position affect dragline productivity and energy consumption and, consequently, energy efficiency (Bogunovic and Kecojevic 2011; Bogunovic 2008; Erdem and Düzgün 2005; Lumley 2005; Williams 2005). The case study confirms these parameters as important explanatory variables of dragline energy efficiency (Table 3). Dumping height is shown to be highly correlated to energy efficiency (ρ = −0.6560, p < 0.001). This has never been shown with experimental data, to the best of these authors’ knowledge. However, the work done to hoist the load from the digging location to the top of the stockpile should be directly proportional to the height through which the load is lifted (dumping height). It must be noted, though, that the fact that these parameters are correlated to energy efficiency does not necessarily mean that they are responsible for differences in operator performance. Any of the parameters, that energy efficiency is sensitive to, can cause differences in energy efficiency, if it varies significantly between operators.

In this case study, dumping height, vertical and horizontal drag distances, and spotting and dump time are shown to be the primary parameters driving differences in energy efficiency (Table 5). Surprisingly, digging time (load bucket time), which has been identified by many researchers as a key discriminator between operators (Bogunovic and Kecojevic 2011; Erdem and Düzgün 2005; Rai et al. 2000; Torrance and Baldwin 1990; Williams 2005), was not found to be a significant factor between the five operators included in this study. The result of this work shows that there is only a 40 % probability that energy efficiency of cycles from these operators is significantly different because of differences in digging time. This probability is less than other cycle time components such as spotting and dumping time. This shows that operator performance evaluation, which is solely based on digging time or other parameters of the digging cycle (e.g., digging energy), can be misleading (Bogunovic 2008; Komljenovic et al. 2010), if the goal is to improve energy efficiency. On the other hand, drag distances (vertical and horizontal) have a high chance of being a responsible parameter, in this sample of operators. This confirms the point that material engagement and disengagement parameters are important parameters and affect dragline performance (Hettinger and Lumley 1999). These engagement and disengagement parameters include the relative position (relative to the dragline boom) of the bucket when the bucket engages (starts digging) and disengages (stops digging) the overburden as well as the distance the bucket is dragged during the digging process. The relative position of the bucket at engagement and disengagement is not directly measured by the vertical and horizontal drag distances. However, it is likely that differences in drag distances are a symptom of where (relative to the boom) the operator is engaging the material as it affects how much dragging is required to fill the bucket.

Payload and cycle time have been shown to affect productivity (Bogunovic 2008; Erdem and Düzgün 2005; Lumley 2005; Williams 2005). In this case study’s data set, the correlation coefficients between energy efficiency and payload and cycle time are low (0.2429 and −0.3755, respectively). The results in Table 5 show that, among the five operators, payload and cycle time have a low chance of being a responsible parameter and are not likely to cause differences in energy efficiency.

It can be concluded that, given a particular group of operators, not all monitored parameters that are correlated with energy efficiency are necessarily correlated to the difference of energy efficiency between operators. That is, not all correlated parameters are the cause of differences between energy efficiency of operators. Hence, the methods proposed in this work are necessary to find out which parameters are actually responsible for the differences in performance, so that operator training can focus on these responsible parameters. Often, operator training programs attempt to “correct” all operator practices that are known to affect performance (energy efficiency). This can overwhelm the operator with things to watch for and lead to sub-par performance. In this case study, even with traditional analysis of the monitored data, engineers and managers would most likely concentrate on 10 measured parameters (the 13 correlated parameters minus drag, swing, and hoist energy) that were found to be correlated to energy efficiency (Table 3). These parameters cover all parts of the dragline operating process (dragging, hoisting, and swinging) and involves varying parameters like engagement and disengagement practices, digging time, swing in time, and dumping time. It is difficult for an operator to try to improve all these practices at once. However, with the extra analysis proposed in this work, management can tailor training programs to focus on only four issues: dump heights, engagement/disengagement practices, spotting time, and dumping time. These can be discussed during weekly and morning meetings, and the experienced operators doing crew coaching can watch for these issues. Even with the use of simulators, reports on these parameters can then be evaluated more closely. Such a strategy is likely to yield higher and quicker improvements in energy efficiency than when all potential parameters/issues are addressed regardless of whether they are relevant in this group of operators or not.

Conclusion

Increase in energy consumption and energy price, and stricter greenhouse gas emission regulations have forced the mining industry to increase the energy efficiency of their operations. Material handling operations (including digging operations) are energy-intensive processes in mining with high potential for energy savings. Recent advances in real-time monitoring systems have facilitated the easy acquisition of data that is essential for energy efficiency analysis. This data has been used by many researchers to investigate and identify factors that can affect performance and efficiency of digging equipment. Operators’ practice is one of the most important factors that affect operational efficiency. Unlike other factors, operators’ skills can be improved through training programs. The method suggested in this work can be used to identify measured parameters that make an operator energy efficient (responsible parameters). Such parameters point to the areas an operator can watch to improve energy efficiency. This method uses statistical approaches such as correlation and regression analysis to identify these parameters when studying a pool of operators. This approach goes beyond identifying parameters that are correlated to energy efficiency to identify parameters that are responsible for differences in energy efficiency. To do this, the method combines missing data techniques with statistical random sampling as well as linear regression of difference matrices.

Data collected from a dragline was used to illustrate the method. However, this method is not limited to draglines and can be used on any digging equipment with a monitoring system. The analysis shows that, for this group of operators, four issues—dumping heights, engagement/disengagement practices, spotting time, and dumping time—are the main areas for potential improvement in energy efficiency. To the best of these authors’ knowledge, dumping height has never been shown with experimental data to be a discriminating factor in operator energy efficiency, although work done in hoisting is proportional to the height the load is hoisted through.

This method is not limited to draglines and can be used on any digging equipment with a monitoring system to identify the parameters that cause differences in operators’ performance. Depending on the equipment, monitoring system, and purpose of the research, a compatible performance indicator can be chosen (e.g., energy efficiency). Correlation analysis between variables in the dataset and performance indicator can be used to identify correlated parameters. Then, operators with significant differences between their performances can be paired and CAA can be used to handle the unequal number of cycles (missingness). Regression analysis on correlated parameters at desired significant level can then be used to identify responsible parameters and make the final conclusion.

References

  1. Abdi Oskouei, M. (2013). Methods for evaluating effect of operators on dragline energy efficiency. Missouri University of Science and Technology.Google Scholar
  2. Abdi Oskouei, M., & Awuah-Offei, K. (2014). Statistical methods for evaluating the effect of operators on energy efficiency of mining machines. Mining Technology, 123(4), 175–182. doi:10.1179/1743286314Y.0000000067.CrossRefGoogle Scholar
  3. Awuah-Offei, K. (2005). Dynamic simulation of cable shovel-formation interactions for efficient oil sands excavation. Missouri University of Science and Technology.Google Scholar
  4. Awuah-Offei, K., Osei, B., & Askari-Nasab, H. (2011). Modeling truck / shovel energy efficiency under uncertainty. Transactions of the Society for Mining, Metallurgy, and Exploration, 330, 573–584.Google Scholar
  5. Bernold, L. E. (2007). Quantitative assessment of backhoe operator skill. Journal of Construction Engineering and Management, 133(November), 889–899. doi:10.1061/(ASCE)0733-9364(2007)133:11(889).CrossRefGoogle Scholar
  6. Biau, D. J. (2011). In brief: standard deviation and standard error. Clinical Orthopaedics and Related Research, 469(9), 2661–2664. doi:10.1007/s11999-011-1908-9.CrossRefGoogle Scholar
  7. Bogunovic, D. (2008). Integrated data environment for analysis and control of energy consumption (IDE-ACE) in surface coal mining. Ph.D. diss., The Pennsylvania State University.Google Scholar
  8. Bogunovic, D., & Kecojevic, V. (2011). Impact of bucket fill factor on dragline production rate and energy consumption. Mining Engineering, 63(8), 48–53.Google Scholar
  9. DOE, U. S. (2007). Mining industry energy bandwidth study. Washington, D.C.: U.S. Department of Energy.Google Scholar
  10. DOE/EIA. (2013). International energy outlook 2013 (pp. 1–300).Google Scholar
  11. Erdem, B., & Düzgün, H. Ş. B. (2005). Dragline cycle time analysis. Journal of Scientific & Industrial Research, 64(January), 19–29.Google Scholar
  12. Hettinger, D., & Lumley, G. (1999). Using data analysis to improve dragline productivity. Coal Age, 104(9), 64–66.Google Scholar
  13. Humphrey, J. D. (1990). The fundamentals of the dragline (pp. 1–28).Google Scholar
  14. Kizil, M. (2010). Improving dragline productivity using a diggability index as an indicator. In: Society for Mining, Metallurgy & Exploration, Inc. (SME) (pp. 134–141). Phoenix, Arizona, USA.Google Scholar
  15. Komljenovic, D., Bogunovic, D., & Kecojevic, V. (2010). Dragline operator performance indicator. International Journal of Mining, Reclamation and Environmet, 24(1), 34–43.CrossRefGoogle Scholar
  16. Little, R. J. A. (1992). Regression with missing X’s : a review. Journal of American Statistical Association, 87(420), 1227–1237.Google Scholar
  17. Lumley, G. (2005). Reducing the variability in dragline operator performance reducing the variability in dragline operator performance. In: Coal operaotrs’ conference (pp. 97–106). Wollongong, Australia.Google Scholar
  18. Mielli, F. (2011). Three major mining challenges and their technological solutions. Mining Engineering, (september), 69–72.Google Scholar
  19. National Mining Association (NMA). (2013). The economic contributions of U.S. mining (2011).Google Scholar
  20. Norgate, T., & Haque, N. (2010). Energy and greenhouse gas impacts of mining and mineral processing operations. Journal of Cleaner Production, 18(3), 266–274. doi:10.1016/j.jclepro.2009.09.020.CrossRefGoogle Scholar
  21. Norman, S. (2011). Variability reduction in dragline operator performance. In: SME Annual Meeting (preprint 11-095). Denver, CO.Google Scholar
  22. Patnayak, S., Tannant, D. D., Parsons, I., Del Valle, V., & Wong, J. (2007). Operator and dipper tooth influence on electric shovel performance during oil sands mining. International Journal of Mining, Reclamation and Environment, 1–26. doi:10.1080/17480930701482961.
  23. Rai, P., Ratnesh, T., & Nath, R. (2000). Cycle time and idle time analysis of draglines for increased productivity–a case study. Indian Journal of Engineering & Material Sciences, 7, 77–81.Google Scholar
  24. Schafer, J. L., & Graham, J. W. (2002). Missing data: our view of the state of the art. Psychological Methods, 7(2), 147–177. doi:10.1037//1082-989X.7.2.147.CrossRefGoogle Scholar
  25. Torrance, A., & Baldwin, G. (1990). Blast performance assessment using a dragline monitor. In: Third International Symposium on Rock Fragmentation by Blasting (pp. 219–224). Brisbane, Australia.Google Scholar
  26. Vukotic, I., & Kecojevic, V. (2014). Evaluation of rope shovel operators in surface coal mining using a Multi-Attribute Decision-Making model. International Journal of Mining Science and Technology, 24(2), 259–268. doi:10.1016/j.ijmst.2014.01.019.CrossRefGoogle Scholar
  27. Williams, G. (2005). Achievements through the dragline improvement group (DIG) in Anglo coal. In: Institute of quarrying southern Africa (36th conference & exhibition). Boksburg, South Africa.Google Scholar
  28. Zhu, Y. Q., & Yin, Z. D. (2008). A new evaluation system for energy saving based on energy efficiency and loss ratio. In: IEEE International Conference on Sustainable Energy Technologies (ICSET) (pp. 121–124). Singapore.Google Scholar

Copyright information

© Springer Science+Business Media Dordrecht 2015

Authors and Affiliations

  1. 1.University of IowaIowa CityUSA
  2. 2.Missouri University of Science & TechnologyRollaUSA

Personalised recommendations