Evaluation of rope shovel operators in surface coal mining using a Multi-Attribute Decision-Making m

International Journal of Mining Science and Technology 24(2014)259-268Contents lists available at Science DirectInternational Journal of Mining Science and TechnologyELSEVIERurnalhomepagewww.elsevier.com/locate/ijmstEvaluation of rope shovel operators in surface coal miningusing a multi-Attribute Decision-Making modelVukotic Ivana, Kecojevic VladislavARTICLE INFOABSTRACTArticle historRope shovels are used to dig and load materials in surface mines. One of the main factors that influenceReceived 27 July 2013the production rate and energy consumption of rope shovels is the performance of the operator. ThReceived in revised form 26 August 2013aper presents a method for evaluating rope shovel operators using the Multi-Attribute Decision-MakingAccepted 21 September 2013Available online 12 February 2014(MADM)model Data used in this research were collected from an operating surface coal mine in thesouthern United States. The MADM model consists of attributes, their weights of importance, and alter-natives. Shovel operators are considered the alternatives. The energy consumption model was developedpe shovelwith multiple regression analysis, and its variables were included in the MADM model as attributesPreferences with respect to min/ max of the defined attributes were obtained with multi-objective optimization Multi-objective optimization was conducted with the overall goal of minimizing energy con-Imption and maximizing production rate. Weights of importance of the attributes were determinedAHPby the Analytical Hierarchy Process(AHP). The overall evaluation of operators was performed by oneof the MADM models, i.e., PROMETHEE IL. The research results presented here may be used by miningprofessionals to help evaluate the performance of rope shovel operators in surface miningo 2014 Published by Elsevier B V on behalf of China University of Mining Technology1 Introductionmotion are used to control the depth of the bucket penetrationPenetration that is too shallow will lead to longer hoist travel dRope shovels are used in large surface mining operations for the tance needed to fill the bucket, which will increase the fill time anddigging and loading of materials. Shovel productivity is highly influ- decrease the fill factor. However, applying excessive crowd motionenced by the skill and working practices of operators. Caterpillar will make the hoist motion slower The whole cycle time increasesdicates that one of the major factors for obtaining maximum pro- up to 50%, or even more, when the bucket is stalled in the bankductivity is a well-trained operator [1. According to Fiscor, better Therefore, to achieve an effective digging process, it is mandatoryoperator training and a consistent clean-up process around the load- for an operator to establish balance between the hoist and crowding area and face could save 20 min of cycle time per day [2. Several motion. In addition, unbalanced crowd and hoist forces lead toresearch studies have shown that mine operating conditions and jacking of the boom, which can cause serious damage to the shoveloperators'practices significantly affect energy consumption 3-71The swinging phase begins when the bucket is full of material,he most influential factors related to the overall working cycle and in this phase, the operator controls the movement of the bucketof the shovel that impact its productivity are cycle time and fillabil- though a defined swing path and dump height toward the haulle bucket. The basic rope shovel motions involve hoist, truck. An experienced operator can achieve smooth and continuouscrowd, swing, and propel. The shovel operating cycle consists of dig- acceleration, maximum speed, and deceleration Motion that is notging, swinging, dumping, returning, and positioning. Even though smooth can result in the increase of time and spillage, which canthese parts of the operation are operationally independent, the skill further damage the body of the truck. In addition the operator influ-and coordination of an operator is required for a smooth cycleences the propel function, or positioning of the shovel in order to getThe digging phase involves crowd motion of the bucket into the the highest number of filled buckets before moving again. Locatingbank, hoist motion to fill the bucket, and drawing from the bank the shovel too far from the bank can decrease available digging18. When the bucket is hoisted to the bank, crowd and retract power. Also, inappropriate shovel location leads to an increase in cy-cle time. Maximum efficiency is achieved when it is positioned closeto the toe of the bank. According to P&H Mine Pro Services, by savingcomes ponding wian tie icomailwy3ed2 min per operati中国煤化工 shovel could loadone additional 1CNMHGhttp://dx.doiorg/10.1016/j.ijmst.2014.01.0192095-2686 2014 Published by Elsevier B V on behalf of China University of Mining Technology50L. Vukotic, V. Kecojevic/ International Journal of Mining Science and Technology 24(2014)259-268Since operator performance significantly affects performance of techniques that operators practiced during the digging part ofthe shovel, it is important to analyze parts of the cycle that can be the cycle with respect to hoist/ crowd utilization widzyk-Capehartimproved. The evaluation of shovel operators will help mining pro- and Lever pointed out that the individual styles of operators have afessionals develop strategies for improving productivity and en- significant impact on shovel productivity, and the operator behavergy efficiency, and reducing the operating costior was examined through a real-time feedback system [ 5]. WeissThere is a lack of quantitative models to evaluate shovel opera- and Shanteau indicated that application of measurement instrutor performance. Particularly, there is a lack of models that con- ments eliminates inconsistencies related to fatigue and bias associ-[23 One of such examples would beenergy consumption, and the relationship between shovel operator using radar/ultrasound system as a position control system thatperformance and the elements of cycle time, production rate, and would bring the bucket into the right position [2]energy consumption. The objective of this research was to developVarious systems have been developed to provide feedbacka methodology to evaluate rope shovel operators with the goal of about different shovel parameters. One of the most advanced sho-maximizing production rate and minimizing energy consumption. vel monitoring systems is Tritronics ShovelPro Monitor, developedby Thunderbird mining Systems, and it is being used worldwide2. Literature review[24. This real-time monitoring system measures different shovelh as bucket payload, coordinates of each bucketSage indicates that a skilled performer is one who produces a fast gaged and disengaged, and dump point. The system gives the oper-and accurate output with a high consistency [7. However, according ator feedback about delays, positioning, and digging time, alongto Bernold, this approach, which is based on behavioral observation, with total production information(cycles, tonnes in chronologicalhas two main shortcomings [3. One of them is that performance is order. The Accuweight by Drivers Control Services, Inc is a simnot clearly related to skill since other aspects such as fatigue moti- ilar system that measures these parameters [25vation, boredom, temperature, and noise can also affect the performance. The second shortcoming is that this approach is qualitativein nature. Bernold suggests that motor skills of the operator are 3. Methodologysential component for everyday operations [ 3]. Robbins also statesthat the aim is to understand how people differ in abilities and useIn order to achieve the objective of this research, a Multi-Attrithat information to improve their performance [10]bute Decision-Making(MADM)model was used. The MADM modelell-trained operator contributes to increased productivity. consists of attributes(criteria), their weights of importance(if nec-Automated equipment involves electronically sophisticated com- essary), and alternatives. The first step was to derive energy conmands and accessories that ease the operation tremendously. sumption and production rate models. the energy consumptionTherefore, introducing controls that require minimal effort has model was derived using a multiple regression tool. Productionchanged the training requirements for operators. Work experience rate has an established method of formulation, based on the volis also positively correlated with skill. According to Harrel and ume of material in the bucket and number of cycles. After definingDaim, motivation is particularly important for good performance: these models, multi-objective optimization was performed usingin fact, unmotivated employee contributes to decreased productiv- an evolutionary algorithm Optimizationrformed with rey and quality [11. Likewise, according to Peterson et al., good spect to the minimization of energy consumption and maximiza-attitudes are important for high productivity [12]tion of the production rate, using measured data on four ropeOperators are usually evaluated by being observed in the perfor- shovel operators with arbitrary names(Operators A, B, C, and D)mance of equipment-related tasks [13, 14, 15. Evaluations are peIn the next phase, the significant variables and their values obformed by observation of operators engaging in duties such as tained from the previous analyses were used for establishing theequipment inspection prior to operation, excavating operations, criteria for the MADM. the alternatives of the MADm were thesafety practices, etc. Also, operators usually have to pass not only rope shovel operators, while weights of the criteria were obtainedpractical exams but also written exams. The evaluator's judgment from the Analytical Hierarchy Process(AHP). Finally, the overallis based on how often the task is performed and how critical it is ranking of rope shovel operators was performed using one of thewhen it is irregularly performed. Both of these criteria are qualitative MADM models from the outranking family -the Preference Rankin nature and dependent on subjective judgment by the evaluator. ing Organization Method for Enrichment of Evaluations(PROMThere are various simulators that are used to train operators [16ETHEE)IL. An outline of the methodology is presented in Fig. 121]. These simulators expose operators to a virtual rope shovel workData used in this research were collected in a surface coal mine lo-ing in the common mining site with trucks. Through each simulation cated in the southern part of United States. The mine uses a truck andmodule, productivity and quality of work are measured. Also, while shovel fleet composed of a P&H 2800 XPB (30.6 m bucket )rope shousing simulators, operators learn from their mistakes, as the simula- vel and Caterpillar 789 dump trucks(payload of 163 t). Usually, thetors identify the least effective parts of their performanceshovel removes up to 15 m of overburden before reaching the coalBernold used a backhoe shovel simulator for analyzing the mo- Caterpillar 789 trucks transport overburden to spoil dump areastor skills of operators [3]. Another example of simulating opera-The shovel has an integrated AccuWeight real-time informationtions of a rope shovel was carried out by Awuah-offei and monitoring system, from which data were retrieved The systemFrimpong [4 The task was carried out with the purpose of finding contains sets of programmable logic controllers(PLcs)that morhoist rope and crowd arm speeds for optimal performance of the tor shovel parameters such as cycle time, fill time, payload, energshovel. The authors pointed out that hoist rope and crowd arm to load a bucket, etc. This system simultaneously samples 20-50speeds present fundamental actions for assessing of operator prac- times per second and stores in database average values of 20tices. Patnayak used the average hoist and crowd power consump- parameters per shovel cycle. Data flow involves cycle detection,tion of different teams of operators as a parameter for assessing the payload weight, andect of operator practice on the performance of the shovel [22]. events for a new cyclIV中国煤化工8t:meamAccording to Widzyk-Capehart and Lever, operator digging the bucket is dumpeeCNMin this researchtechnique has the direct influence on the stalling of the hoist dur- are cycle time fill time,allu ciieigy Lu lvad a bucket. En-dig and slacking of the hoist ropes [5]. The performance was ergy to load a bucket is recorded in a unit-less number format,examined by analyzing productivity, operator cycle time, and which must be adjusted to obtain energy consumption in kWh.1. Vukotic, V, Kecojevic/ International Journal of Mining Science and Technology 24(2014) 259-268Minimize Z1(x)=Z1(x1, x2, X3, x4)DefiningMADM modelMaximize Z2(x)=Z2(x3, X2)where Z,(x)represents the energy consumption; Z2(x) the producrate: x, the fill time: x2 the cycle time: X3 the volume of thDefining criteria of the MADMmaterial in the bucket, and x4 the number of working hours.model with preThe process of multi-objective optimization was performedand maximizing production rate:with the evolutionary algorithm. One of the evolutionary(genetic)Develeon modealgorithms that represent a standard approach to solving multiof energy consumption andobjective optimization is Non-dominated Sorting Genetic Algo-roduction rate modelrithm-II(NSGA-II), which was used in this research. OptimizationOptimization of energyconsumption and production ratewas performed with the GANetXL add-in for Microsoft Excel. Forthe given setting, multi-objective optimization provided the following results: minimum value of fill time, minimum value of cycle time maximum value of volume of material in the bucket, andDefining weights for themaximum value of number of working hours. This information wascriteria in the modelfurther used to set up the MADM modelproblems where multiple criteria represent peralternatives, Multi-Criteria Decision-Making(MCDM)can be con-sidered. Usually, there is no optimal solution for these problems,and the preferences of the decision maker characterize the differmodel to obtain ranking ofence between solutions the mcdm methods are divided in twomajor groups of methods: Multi-Objective Decision-Making andPROMETHEE II computationsMADM. For the ranking of operators from"the best "to"the weak-est, "the MADM is a more appropriate method because of thefollowing characteristics: explicit attributes, finite number of alterMaking the decisionnatives (discrete), and selection and evaluation of the solutioOutranking of operators from thethat are already knownbest to the weakest performanceDecision-making processes have several steps in the process ofsolving the particular problem: problem identification, set-up ofFig. 1. Outline of the methodologypreferences, evaluation of alternatives, and determination of thebest alternative[27. The flow of solving a decision-making problem related to this research has the following steps: classificationFor that purpose, conversion coefficients exist for different shovels of rope shovel operators based on given attributes(problem idenand different applicationstification); setting up preferences of measured parameters ob-The multiple regression analysis was used for developing the en- tained in the optimization part, and weights obtained with AHP;ergy consumption model Measured variables were used as regres- evaluation of alternatives(Operator A, Operator B, Operator Csors. These variables are: fill time FT(s), cycle time CT(s), volume of and Operator D)based on the given preferences and weightsthe material in the bucket VB(m), and number of working hours determination of the best operator, as well as ranking of otherWH (h). The correlation between the independent variables was operatorsperformed, and selection of the set of variables that would explainThe mathematical model for solving MADM prethe highest variability of the energy consumption was performed. lated as followsFor this purpose, the best subset analysis and stepwise regression Max or Min i(x), 2(x), ..,Sn(x), n>2), xE A(a1, a2, .. am)analysis were used. Stepwise regression uses iterations to make aseries of regression models by adding or removing variables atevery step [26]. Criteria used for evaluating and comparing regres- where fm()represents criteria(attributes): n the number of criteriasion models are coefficient of determination R, adjusted coefficientattributes ) a; the alternatives in the model; a the set of all alterna-ered to be a suitable candidate for the best regression model. The Cp represents one alternative, and every column one ciion(attristatistic defines the total MSE for the regression model, and there-ore the model with a minimum Cp statistic is considered to be the alternative a; considering the criterion f For n criteria and m alternabest regression model. Likewise, this statistic should not exceed tives, the form of the matrix is shown in the Eq (4). Values w, repre-k+ 1 regressors in the model, where k is number of regressors in sent weights of given criteria defined by a decision maker or in somehe model. These criteria were used for both analyses -the best other way, with the rule that their summation must be equal to onesubset as well as stepwise regression. In order to check for multicolma maxmaxflation factors vif were calculated forvariables in the model Multicollinearity represents dependencyf f2 fnamong the regressor variables, which has a significant effect oncoefficients of the regression as well as appropriateness of the de-rived model. Some authors suggest that the multicollinearity ashould not be more than 4 or 5[26]. Interpretation of the parame-11中国煤化工ters in the regression output was performed using standard statisf21THCNMHGtical procedure( F-test and t-test for individual regressors).The set-up for multi-objective optimization can be presented as am Ifxfollows1m2L Vukotic, V Kecojevic/ International Journal of Mining Science and Technology 24(2014) 259-268AHP was proposed as a part of this research to set up weights (1) Generalization of the criteria, which involves introduction offor the criteria used in the MADM model. The major steps of AHPthe preference function. The preference function is formucan be defined as follows:lated as follows [28(1)Deconstruction of the goal (problem) into a hierarchy ofjff(a)≤∫(a)interrelated elements, thus constructing a hierarchicalP(ai, aj)Pf(ai)-f(a,)), if f(ai)>f(aj)0≤P(aa reciprocal matrix by comparing theweights between the attributes of the model. The degree of where the difference between two alternatives is calculated aselative priority of one criterion to another is calculated by d(a;, ai)=fai)-fai. The authors of this method proposed six typesassigning weights to each of them from 1 to 9, where 1 indi- of preference functions, which should cover the most of the casescates that both criteria are equally important, and 9 indi- that appear in the practice. the decision maker selects suitable prefcates extreme importance of one criterion over another. a erence function for each criterion and defines additional parametersreciprocal value is given to the other criterion in the pair. depending on the type of function. These parameters are differentFurthermore, the ratings are normalized and averaged with thresholds defined by the decision maker, with which one couldthe goal of obtaining an average weight for each criterion give strong or weak dominance of one alternative to another. In thisin the modelresearch, the preference function of Type I-Usual Criterion is usedwhich means that the decision maker gives the strict preferenceIn order to establish the consistency of the subjective judgment to the action that has the higher valueand consistency of comparative weights, the consistency index(2)When the preference function is determined for every crite(C L )and the consistency ratio(CR )should be calculated. Matherion, all comparisons between alternatives are performed formatical expressions for these two indices can be written as:all criteria in the model. The multi-criteria preference indexis defined to globally compare every pair of alternativesCL-(max-n)CR.RIMathematical definition of the preference index is definedy Brans and Mareschal [28here imax shows the largest eigenvalue; n the number of attrir(a,a)=∑wPa,a)butes, and r l the random consistency index. The r l is made fromrandomly generated reciprocal matrices. Since this case involves where I(a ai )represent preference index and wi are criteria3x3 matrix, R L. value used in further calculations is 0.52For evaluation of operators and their ranking, a 3x 3 matrixweights. P reference index is from 0 to 1was created. The criteria in the matrix were measured rope shovel(3)Evaluation of the alternatives of the set A is performed byparameters relevant to production rate and energy consumption,using the outranking relation, through preference flowsi.e. cycle time CT, fill time Ft, and volume of material in bucketThe positive preference flow p(a)shows how a certain alterVM. The criterion" working hours"was omitted from the modelnative is outranking all the other alternatives while negativesince all operatorseled on an hourly basis. The AHP matrixrformance flow d(a) shows how a certain alternative isused in this research is presented in Table 1outranked by all the other alternatives. Finally, the net flowNext step involves summation of all elements in each of thep(a) represents the difference between positive and negativeumns, and then dividing elements of every column with the valueflow, and it is used for determination of the total ranking ofthat represents the sum of that column Summation of the numalternatives In PROMETHEE methods, the higher the posibers calculated in the latter way must be equal to 1. If that numbertive flow and lower the negative flow, the better the alternais different, the calculation is not valid. Finally, summation of alltive mathematical formulation of flows is defined with theelements in each row should be calculated. and then the mean vaEq.(8)lue of every row should be determined. These mean values represent normalized eigenvectors. In this way, weight of each criterionot(0-=n1;∑a,a)()=n1∑anais determinedIn order to accomplish the goal of ranking the alternatives, thep(a)=φ(a)-φ(a)outranking MADM methods can be used. Outranking methodscompare alteres based on their preference relations. There where o'(a)represents the positive flow; d(a)the negative floware several different types of outranking methods(ELECTRE, p(a)the net flow, and n the number of alternativesPROMETHEE, etc. ) For this research, the Promethee ll was usedItking of alternateThe dimension of the decision matrix is 4x3. where the alterConsider a MADM problem defined with Eq(3).At first, pair- natives are operators, criteria are the same as those used in settingwise comparisons of alternative a and alternative b should be per- up the AHP model, and preferences are obtained with optimizationformed, and if the result is such that fj(a)>f(b), for j-1 to n; then Weights of each criterion were integrated in the set-up of thea dominates b. The basic concept of this family of methods has PROMETHEE ll model. Considering type of the preference function,the score"is given to the dominated alternative, while the score1"is given to the one that dominates The scores were then multiplied with the weight assigned to that attribute. values relevantparison matrix of the criteria in the model.to the particular pairwere then added中国煤化工pea6opomparison matrixCT(s)T(s)VB(m)tive flow ) Finally, caCNMH Alternative was(s)performed. The alterLhC llglIcSL let flow is clearlyVB(m)the best, and the other values of flow allow the ranking of rest ofalternatives, by decreasing value.1. Vukotic, V Kecojevic/International Journal of Mining Science and Technology 24 (2014)259-2684. Results and discussionDistribution of cycle time for shovel operators is shown in Fig. 2The spread of the data ranges from 27.72 s to 87.27 s. The boxplot36.27graph indicates that the median of the raw data for cycle time isabout 36.63 S, with 25% of the observations falling at or below3328S, and 75% of the data falling at or below 42.84 s After the re-30moval of outliers from the data, the corresponding means and standard deviations of cycle time for each operator are shown inTable 2. The boxplot of cycle times for individual operators is presented in Fig 3Fig 3. Mean cycle time for each operatorThe overall mean value for cycle time is 35.69 s with standarddeviation of 3.69 s. Operator b has the lowest mean cycle time(34.71 s)with standard deviation of 3. 54 s. Operator a has the largMean cycle timeest mean cycle time (36.27 s) with standard deviation of 3.69 s Then Mean production rate0.55smallest standaeviation is characterized by Operator d withmean cycle time of 34.82 S. The largest standard deviation is char-cterized by Operator C. This operator has the statistical dispersionequal to the difference between 75% and 25% of the observations of5. 48s. It can be noticed that operators slightly differ in the meanvalues of cycle time, as well as in consistency from cyohCycle time is one of the most influential parameters for production rate, along with fill factor. Fig. 4 shows the dependency ofthese three parameters values for production rates and fill factorsfor each operator represent mean values of those parameters. Val- Fig 4. Relationship between mean production rate, mean fill factor and mean cycleles are transformed with vector normalization Operator D has the time for indivoperatorsighest value of mean production rate, highest value of mean fillfactor for the second lowest value of mean cycle time. OperatorB, who has the lowest mean cycle time, has the second best valueof mean production rate, as well as for mean fill factor. Operator Aas the highest mean cycle time the lowest mean production rateand the lowest mean fill factor, followed by the operator C. It canbe concluded, that considering production rate, Operators d and Blave the most preferred performance, followed by Operator C, andthe weakest performance was observed for Operator AFig 5 shows boxplot for the fill-time data. The spread of thedata is large, from 3. 23 s to 24.41 s. Outliers are assumed to beFig. 5. Boxplot of fill time.faulty data, and were removed from the further analysis. The box-plot indicates the median value of 10.2 S, with 25% of the observa-tions that fall at or below 9,27s, and 75% of the data that fall at or deviations of fill time were determined and these are shown inbelow 11.45 s. As in the case with cycle time the latter range is the table 3. The mean fill time for individual operators is shownconsidered as the range of the majority of data. After the removal in Fig. 6.of outliers from the data, the corresponding means and standardThe overall mean value for fill time is 10.33s with standarddeviation of 1.60 s. Operator b has the lowest mean fill time(10.02 s)with standard deviation of 1.54 s. Operator d has thehighest mean fill time(10.67 s)and the lowest standard deviationof 1.28 S Operator A has the highest standard deviation(1.73 s70This operator has the statistical dispersion equal to the differencebetween 75% and 25% of the observations of 1.39 s. Although thereare differences in the mean values of the mean fill time as well asin standard deviations, their values are very small, and it can beconcluded that operators are somewhat consistent in theirperformanceFig 7 represents boxplot of payload data. The spread of the datis from 11.08 t to 88.4 t. There were not any outliers in data. TheFig. 2. Cycle time boxplotboxplot shows a median of 61 t, with 25% of the data falling at orbelow 47.98 t, and 75% of the data falling at or below 68.68t Meanvalues, as well as the standard deviation of the measured payloadare presented in Table 4. The mean payload value for every operaand standard deviation of cycle time for each operator(s)tor is shown in Fig 8The overall medeviation of 15.13中国煤化工 lue of thpayload(60.03 t)CNMH Geviation(12.93 tOperator A has thePayIuau (JJ.U/ t)with the higheststandard deviationd to all other operators(16.54 t). Thisoperator has the statistical dispersion equal to the difference64L. Vukotic, V. Kecojevic/ International Journal of Mining Science and Technology 24(2014)259-268Table 3Mean and standard deviation of fill time for each operator(s).OperatorStandard deviation10.808000000055670456191.541.59128Overall10.331.6OperatorFig 8. Mean payloads for each operator.160.841010312086420DFig. 6. Mean fill time for each operator.Fig 9. Boxplot of energy data90Table 5Mean and standard deviation of energy to load a bucket for each operator.OperatorMeanStandard deviation6013.697,932964.3Fig. 7. Boxplot of payload data.16344.00between 75% and 25% of the observations of 28, 85 t In regards toconsistency of operators performance, it is clear from the data thatoperators differ among themselves.22Fig. 9 represents boxplot of energy to load a bucket in unit-lessnumbers. The histogram of energy to load a bucket data shows that186013.7the distribution of the variable is roughly symmetric; the center ofenergy distribution is approximately 14,916(median 15,525). A9925% of the observations fall at or below 12,287 and 75% of the data12ll at or below 17,294. Mean values and standard deviations of thedata are shown in Table 5, and the boxplot of energy to load abucket for each operator is presented in Fig. 10. The overall meanvalue for energy to load a bucket is 14, 916. 11 with standard deviation of 3, 139. 10. Operator b has the lowest mean energy to load abucket(14, 262. 49 )with the lowest standard deviation of 2, 964.39Fig. 10. Energy to load a bucket for each operatoOperator d has the highest mean value of energy to load a bucket,while Operator C has the largest standard deviation of 3, 177.44.Itcan be seen that operators differ in their mean values for this consumed per fill time while Operators a and D, particularly latter,parameter while they are more or less consistent in performance. have higher energy consumed per fill time.The relationship between energy to load a bucket and fill time isThe relationship among mean production rate, mean fill factor,presented in Fig. 11. values are transformed with vector normali- and mean energy to load a bucket for individual operators is shownzation. It can be seen that Operators b and C have less energy in Fig. 12 Values are transformed with vector normalization Oper-ator d has the highest value of mean production rate, highest valueof mean fill factor and highest mean value of energy to load a bucket On contrary, Operator A has the second highest value of meanand standard deviation of payload for each operatorenergy to load a bucket for the lowest mean fill factor and lowestStandard deviationmean production ratTvL中国煤化工mmean energy to16.54load a bucket, withtion rate andan fill factor CCNMH Oean productionrate, mean energy to luau a burrel, allu lcan in factor, OperatorDOverall5.13B appears to have the best performance, while Operator A has1. Vukotic, V Kecojevic/International Journal of Mining Science and Technology 24 (2014)259-268a Mean energy to load a bucketTable 7055lean fill factorResults of the best subset regressionVariabR0.50.45688VB. CT85385.2VB. FTWH390.485.885.7Fig. 11. Energy to load a bucket vs fill time for every operator.Table 8a Mean energy to load a bucketa Mean production ratevariables0.55a Mean fill factor85.348584685884.680.50755.00434.105.000.45Maximum value of vif for the model is 1.55 and indicates that0.40multicollinearity is not anin this case. Comparison of PRESSstatistics with Sse is a way of informal judging of sensitivity of themodel fit. Value for sum squares of error (error SS in the Table 10rgy to load a bucket for eal p fil factor, mean production rate, and mean is close to the value of the PRESS statistic(Table 9),which indicatesFig. 12. Relationship between meanthat over fitting is not the issue in this model Over fitted modelswould give small residuals for observations(Ss error in the model,The correlation coefficients between the parameters of the gi- The coefficient of determination r2 shows that 86. 80% of the variven shovel are shown in Table 6. High correlations between the ation of the energy consumption is explained with the variables invariables are not observed, except for fairly high correlation be- the modelTable 7 shows the results of best subset regression. The boldedLikewise, validation of the model is performed, with the splitvalues in Table 7 represent the most preferable model. It can be sample validation method. The data were divided into two sepaamples, with one sample representing data for two operatorseen that all four variables participate in the explanation of energy and the other sample representing data for the other two operaconsumption variation. Table 8 represents the results of the step- tors. Next, one of the samples was used for building the model,wise regression analysis. The method gave the same preferred and the other sample was used for validation of the model.Themodel selection as best subset method. One can also see in the best subset regression as well as the stepwise regression analysis,utput that the variable volume of the material in the bucket has showed the same subset of the regressor variables for explanationhe highest correlation with the energy consumption response of the energy consumption varThe parameters of thend explains the highest portion of variability in it. Therefore, withariables suggested with best subset and stepwise regression analestimated model. as well as the validation model are shown inysis, the model of energy consumption was developed. The regresTable 11. Mathematical formulation for the validation model is repsion equation has the following formresented by Eq (11)Y=(100735+5622F-1768cT+976VB-88WH20Y=(94685+6524F-1683c7+84VB-786WH2(10)It is expected that differences in parameters for these two mels exist. However, the following criteria were used for verificationTable 9 represents the analysis of variance table for the energy con- of the validation model: (a)the overall relationship of the depen-sumption model. It can be seen from Table 10 that hypothesis test dent variable and the regressors must be statistically significantof"all regression coefficients are equal to zero"(F statistic) yields for the estimation and the validation models. In this case, the bestp value that is less than 0.05, which indicates at least one of the subset and the stepwise analysis yielded the same results for bothariables is statistically significant to the model. However, one does models; (b)the value of r2 for both models can be different withinot have information which one is significant or not significant, and the range of +5%. As can be seen from Table 11, the difference bethus the t tests for individualors are performed. The hypoth- tween these two values is approximately 1%esis tests of the individual regression coefficients(with t statisticThe normalized comparison matrix for the given MADM modelof the variables are statistically significant in the model(Table 9). obtaining the normalized vector, the weights of the each criterionare given in Fig. 13. Criterion 1, 2 and 3 (Fig. 13)are volume of theTable 6material in the bucket, fill time, and cycle time respectively. TheCorrelations between shovel parameterscriterion of volumecriterion in the中国煤化工 -nd cycle time.Tocheck the consisCNMGistency index anconsistency rationuci lu calculate consis-0.136002tency index, the largest eigenvalue(2max) should be obtainedThe calculation is performed as follows66L. Vukotic, V. Kecojevic/ International Journal of Mining Science and Technology 24(2014)259-268Table gResults of the multiple regression model of energy consumptionTermCoef.TConstant1007.3547.705221.11614.061217.681.446312.22420.5512771153.34202.63590.000mmary of model8555%PRESS= 238459Table 13Analysis of variance.Input data for PROMETHEE IL.Criteria0.000B(m3)FTCTRegression1432779358195197251202179101816Preference function Type立,立Thresholds0.11Table 11Parameters for the regression and the validation model.Operator A10.836.27Validation modelOperator B阝o946.8500735Operator COperator D31857Table 14R2(%878286,81Comparisons Ibetween opOperatorcomparisonsMax VBMin ctTable 12Comparison matrix of the criteria in the model.0.31CT(sCT(s)0.100.310.300.00VB(m)0.60Operator DAfter obtaining the weights for the criteria, the decision tablefor the input data in the PRoMEthee II approach is formed(Table 13).All par-wise comparisons between operators areted inTable 14. For example, regarding criterion volume of material inthe bucket(which calls for maximization), comparing Operator Awith Operator B, Operator a did not dominate; thus,0"was placed in the a/B cell. Next, regarding the same criterion,Fig. 13. Weights of the critecomparing Operator B with Operator A, Operator b dominates;hus a score"1 multiplied with the weight of VB"(0.58)was placedin the b/a ce0330.21「0.1103291「0.329/0.1Comparing Operator d with Operator B for the criterion CT,0503109290929031Operator d did not dominate; thus score of0"was placed in D/5210581747L1.7470.58B cell, or comparing Operator C with Operator A for the criterion(11) FT, Operator C dominates; thus value of 0.31 was placed in the C/A cell the rest ofperformed in the same way3.001+3004+3.006Finally, preference3.004ator B are 1. 1 and 0V凵中国煤化工73:or9Pator d are 10.58CNMHG of preferenceCalculated value of CI is 0.002, and of CR is 0.0038(0.38%) Accord- function is selected, tlpal lULl scores are calto Saaty, consistency is satisfactory if the Cr is less than 10%. lated differently for each of them, and calculation can be found inTherefore, consistency doesn't need to be improved in this case.any literature source related to PROMETHEE method. The final1. Vukotic, V Kecojevic/International Journal of Mining Science and Technology 24 (2014)259-268Table 15Statistical analysis provided valuable insight about the differPaired matrix of alternative comparisons.ences in operator performance considering different parts of theOperator A B C D Sum of positive Net Finalcycle. It was determined that the most important parameter forevaluating the performance of operators is volume of the material0.0000001.000ⅣVin the bucket. This parameter is positively correlated with both100010000800613production rate and energy consumption. the second most important parameter for given preferences is fill time, and finally cycle10.5800.6900.750.513ⅢSum of negative 1 0.193 0.563 0.243After analysis of the data, the following can be concluded:(1)Operator A and Operator c have weaker performance con-sidering measured parameters. Operator A has the weakestTable 16rformance and is the least consistent operator overall. ThisOverview of mean production rate and energy consumption of shovel operatorsevaluation of performance was even obvious in the phase ofOperatorlean production(m/h) Mean energy consumption(10the statistical analysis of data, and it was confirmed with the2912.67ROMETHEE-Il ranking3136.99(2)Operator B and Operator d have the better performance con2959491.473282.47idering measured parameters, and they are fairly similar inerformance. Operator b has better performance in almostall analyzed parameters with the given preferences, exceptmean production rate. On the other hand, Operator d haspaired matrix for operator comparison which yields positive flows,the highest mean production rate, but also the highest meannegative flows, overall flow of each alternative, and final ranking ofenergy consumption. This operator is the most consistent inalternatives is shown in Table 15. It can be seen that Operator Bperformance out of all other operators. With a statistical(net flow 0.613)and Operator D(net flow 0.513) have close netanalysis of the data, the clear separation in performance offlow values, and thus similar performances. On the other hand,these two operators could not be observed. PROMETHEE-lIOperator A(overall flow -1)has the weakest performance, far be-identified the operator who is the best in performance forlow all other operators. the operator who demonstrated the bestthe given preferences, which is Operator Bperformance, as defined by maximum production and minimumenergy consumption (through the analysis of measured volumehe advantages of using this approach, particularly PROmethEEof material in the bucket, fill time, and cycle time) was Operatornethod can be summarized as follows: (a) easy-to-understand,B, followed in descending order by Operator D, Operator C, andtional model; (b)straightforward comparison of pairs of alterOperator Atives in the model; (c)flexible allows addition of other criteria forIn the future evaluation of operators, the partial ranking with evaluation of operators-not only quantitatively in nature, but alsothe PROMETHEE I method, which gives comparisons on the posi- qualitatively. The disadvantage of the PROMETHEE II method istive and negative flows separately, may be used. This is because,that subjectivity is involved when selecting criteria weights as welleven though the output of the PRoMethee Il method is not diffi- as selecting a preference function. Also, if selected preference func-cult to explain, it provides less information since the differences volved as welltion requires determination of thresholds; there is subjectivity inbetween positive and negative flows are no longer apparent Nev-ertheless, from partial ranking, a decision maker can see which ofthe actions are different to compare, and focus on themReferencesTable 16 shows an overview of mean production rate and energy consumption for shovel operators. Out of all parameters ana11 Caterpillar. Caterpillar performance handzed, the one that is the most variable is the payload of theAvailableathttp://vmaterial.Consequently, operators significantly differ in mean val- 131 BRRtd2007: 133(11):889-99ues of production rate. The difference between the most productiveoperator(Operator D)and the least productive one(Operator A)on4) Awuah OK, Frimpong S. Cable shovel digging optimization for energyciency Mech Mach Theory 2007: 42(8): 995-1006.average is approximately 11% in production rate. In addition, theVidzyk-Capehard E, Lever P. Towards rope shovel automation operationfference between the most energy efficient and least energy effilation system. In: CRC mining conference, Queensland 2004cient operator(Operator B and Operator D)is approximately 12%[6] Awuah OK, Summers D Reducing energy consumption and carbon footprintEvaluation of operators in this research was based only on cerScience and technology: 2010.tain parameters that were available for analysis. Collection of more Knaicateo. inter Nationa journal of MiniNg. Reclaimation ran envirommentbetter insight into the performance of the shovel operators. The [ P&H Mine Pro Services. Peak performance practices-Dippers. P&H Miningadditional work of evaluating operator performance could involveevaluation based on changes in the material properties and varying 19 G H Sage, Motor learning and control: A neuropsychological approach.Brownweather conditions[10 Robbins SP. Organizatbehavior. New Jersey: Pearson EducationInternational: 2003t management with5 Conclusionsmployee motivation: the case of silicon forest. Eng Manage 2010: 22(1): 23-33[12 Peterson DJ. LaOperators and their principal role in performance of the shoveliews of critical to中国煤化工 i Corporation 1990are not frequently considered. A model that analyzes shovel opervailable at: htCNMH_Instructors- Notesator performance in different parts of the cycle, with respect toproduction rate and energy efficiency, was developed in this114 Vista TIInc, Equipment operator skill-based pay plan(SBPP). 2012,skill-based_pay_plan. pdf.58L. Vukotic, V. Kecojevic/ International Journal of Mining Science and Technology 24(2014)259-268[15 CareerTech. Heavy equipment operation: operator. Oklahoma Department ofwww.thoroughnd Technology Education Stillwater, Oklahomaexcavator-simulator html16| Simlog, The Standard in Cost-Effective Simulation for Training Heavy[22] Patnayak S. Key performance indicators for electric mining shovels and oiuipment Operators, 2012. Availablands diggability. Alberta: University of Alberta: 2006123 Weiss Deau J. Empirical assessment of expertise. Hum Factors[171 Vista, Electric Rope Shovel. VISTA Training, Inc. 721 Cornerstonnee Crossing003:45(1):104-14WaterfordWi53185,Usa.2012.Availableathttp://www.vista-training.com/[24]thunderbirdMiningSystemTritronicsShovelproMonitor.[brOchure].2012training/type/simulators/electric-shovelilableathttp://www.tbirdpac.com/shovelsmonitoring/docsshovelpro.p[18DoverSimulatorsElectricRopeShovel2012.Availableathttp://[25]Accuweigh.AccuweighproductionmonitoringsystemDrivers&controLServicesTexasUsa.2003.Availableat:http://www.drivesandcontrols.com[19 Immersive TechnoSimulators2012.Availableat:http://[26)montgomeryDc,RungerGc.Appliedstatisticsandprobabilityforwww.immersivetechnologies.comengineers. New York: John Wiley and Sons Inc 2003[20 Fifth Dimension TechnShovel/ Excavator Training Simulator. 2011. 271[21 Thoroughtec simtSimulators, Digger Simulators and ExcavatorSimulatorsThoroughbredTechnologies(pty)ltd.2010.Availableathttp://90.avAilableathttp://www.inf.unideb.hu/valseg/dolgozok/anett./DSS/ Promethee. pdf.中国煤化工CNMHG