research article a dynamic health assessment approach for...

Research ArticleA Dynamic Health Assessment Approach for Shearer Based onArtificial Immune Algorithm

Zhongbin Wang1 Xihua Xu1 Lei Si12 Rui Ji1 Xinhua Liu1 and Chao Tan13

1School of Mechatronic Engineering China University of Mining and Technology Xuzhou 221116 China2School of Information and Electrical Engineering China University of Mining and Technology Xuzhou 221116 China3Xuyi Mine Equipment and Materials RampD Center China University of Mining and Technology Huairsquoan 223001 China

Correspondence should be addressed to Xihua Xu xuxihua cumt163com

Received 25 January 2016 Revised 3 March 2016 Accepted 7 March 2016

Academic Editor Cheng-Jian Lin

Copyright copy 2016 Zhongbin Wang et alThis is an open access article distributed under theCreativeCommonsAttribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

In order to accurately identify the dynamic health of shearer reducing operating trouble and production accident of shearerand improving coal production efficiency further a dynamic health assessment approach for shearer based on artificial immunealgorithm was proposed The key technologies such as system framework selecting the indicators for shearer dynamic healthassessment and health assessment model were provided and the flowchart of the proposed approach was designed A simulationexample with an accuracy of 96 based on the collected data from industrial production scene was provided Furthermore thecomparison demonstrated that the proposed method exhibited higher classification accuracy than the classifiers based on backpropagation-neural network (BP-NN) and support vector machine (SVM) methods Finally the proposed approach was appliedin an engineering problem of shearer dynamic health assessmentThe industrial application results showed that the paper researchachievements could be used combining with shearer automation control system in fully mechanized coal face The simulation andthe application results indicated that the proposed method was feasible and outperforming others

1 Introduction

Due to the randomicity and complexity of undergroundgeological conditions assessment of shearer health conditionwould present the characteristics of complexity fuzzinessand uncertainty and this may affect the coal production oreven endanger the operatorrsquos life Moreover because of thepoormining environment and complex component structureof shearer the shearer operator cannot accurately estimate theworking status of shearer which may lead to some problemsof poor coal quality and low mining efficiency Furthermorean increasing number of safety accidents in collieries arecaused frequently Therefore it is necessary to assess thedynamic health condition of shearer which has become achallenging and significant research subject [1]

Depending on the assessment of the health condition ofshearer this can reduce operating trouble and productionaccident of shearer and improve production efficiency fur-ther In recent years many researches have brought outsome achievement on shearer health condition diagnosis

The multiple fault classifier based on the improved supportvector machine theory is used to judge the fault types of coalshearer [2] In [3] a correct and timely diagnosis mechanismof shearer failures by knowledge acquisition through a fuzzyinference system is provided which can approximate expertexperience Althoughmany research achievements have beenproposed they have some common shortcomings summa-rized as follows Firstly most research cannot confirm thehealth degree clearly Moreover it costs long diagnosis timeand cannot be used in real-time health assessment

Dynamic health assessment was used in spacecraft prim-arily in the 1970s At present domestic and abroad researchershave worked on the modeling approaches for dynamic healthassessment and proposed several solutions The density-based spatial clustering of applications with noise has beenused for bearingsrsquo condition monitoring [4] and a novelonline method based on dynamic Bayesian networks (DBNs)for the estimation of the SOH of lithium- (Li-) ion bat-teries has been presented [5] and so on However due to

Hindawi Publishing CorporationComputational Intelligence and NeuroscienceVolume 2016 Article ID 9674942 12 pageshttpdxdoiorg10115520169674942

2 Computational Intelligence and Neuroscience

the complex component structure and bad working con-dition there has not been a health assessment concepton shearer and this paper tries to present it In the realmining condition some key index parameters have a strongrelationship with shearer health condition The relationshipis highly nonlinear in nature so that it is hard to developa comprehensive mathematic model The current methodsand mature assessment systems are hardly satisfied with theshearer health state assessment In this paper we try topropose a novel prediction approach for shearer dynamichealth assessment to identify the health state during coalmining

The first mathematical model in artificial immune systemwas proposed in 1974 which initiated subsequent researchesand discussions Artificial immune system (AIS) as a novelintelligent algorithm method inspired from the biologicalimmune system is an effective means for prediction [6ndash8] The AIS can acquire learning capability by learningthe biological protection principle According to the aboveanalysis a novel prediction approach for shearer dynamichealth assessment based on artificial immune algorithm isproposed and the assessment system is validated by thesample data of operating parameters from industrial pro-duction scene Moreover it will prove that artificial immunealgorithm is a better tool for classifying due to its classificationaccuracy than the classifiers based on back propagation-neural network (BP-NN) and support vectormachine (SVM)methods later

The remainder of this paper is organized as follows Somerelated works are outlined in Section 2 The key technologiessuch as system framework selecting the indicators for shearerdynamic health assessment and the proposed approach arepresented in Section 3 Section 4 provides a simulation exam-ple and an industrial application example for shearer dynamichealth assessment based on the proposed approach to specifythe application effect Our conclusions are summarized inSection 5

2 Literature Review

Recent publications relevant to this paper are mainly con-cerned with two research streams the dynamic health assess-ment methods and artificial immune algorithm In thissection we try to summarize the relevant literatures

21 The Dynamic Health Assessment Methods For the dyna-mic health assessment problem lots of research has beendone since the last decades In [9] Zhong-Bin et al developeda remotemonitoring platformof the shearer by usingVirtual-Prototype technology to realize the remote monitoring forthe shearer in the fully mechanized long-wall coal miningface In [10] Zhou et al proposed a novel approach basedon the coal floor height variation which is taken as asignificant factor and fuzzy optimization theory to improvethe implement precision of shearer memory cutting In [11]P W Tse and Y L Tse designed an innovative system thatis installed in a passenger car or a truck that is running onroad and provides instantaneous engine health evaluation

and diagnosis In [12] Black and Winiewicz provided amethod and apparatus for internal network device dynamichealth monitoring to increase network device availabilityIn [13] Vichare and Pecht presented the state of practiceand the current state of research in the area of electronicsprognostics and health management In [14] Pecht and Jaaipresented an assessment of the state of practice in prognosticsand health management of information and electronics-richsystems In [15] Yang et al proposed an accurate identifica-tion of the shearer late underground cutting coal and rockconditions and fault diagnosis by the method of vibrationanalysis In [16] Yin et al designed an embedded healthevaluation system to meet the requirement of continuousmonitoring of the mine special gear box In [17] Mascarenaset al investigated a vibrohaptic human-machine interface forstructural health monitoring In [18] Cerda et al exploredan indirect approach for structural health bridge monitoringallowing for wide yet cost-effective bridge stock coverageIn [19] Zubizarreta-Rodriguez and Vasudevan introduceda new multisensor measurement framework for conditionmonitoring of brushless DCmotors (BLDCM)with bearingsIn [4] Kerroumi et al introduced a dynamic classificationmethod inspired byDBSCAN clusteringmethod formachinecondition monitoring in general and for bearings in particu-lar In [5] He et al presented a novel online method for theestimation of the SOH of lithium- (Li) ion batteries based ondynamic Bayesian networks (DBNs) In [20] Herrmann et algave an introduction into the principle of structural healthmonitoring (SHM) basics of fatigue of fiber resin compositematerials and the possible application of these principles inthe automotive industry

22 Artificial ImmuneAlgorithm Theartificial immune algo-rithm was firstly proposed by Farmer in 1986 [21] It is ableto recognize novel shapes without preprogramming based onthe capacity of learning memory and pattern recognition In[22] Ishiguro et al proposed a new decentralized consensus-making system inspired from the biological immune systemand an adaptation mechanism that can be used to constructa suitable immune network for adequate action selection In[23] Tang et al described a new model of multiple-valuedimmune network based on biological immune responsenetwork In [24] Abbattista et al proposed the use of immunenetwork model for designing associative memories In [25]Deng et al proposed a fuzzy logic resource allocation andmemory cell pruning based artificial immune recognitionsystem (AIRS) to improve the resource allocationmechanismof AIRS and decrease the memory cells In [26] De Castroand Von Zuben proposed computational implementationof the clonal selection principle that explicitly takes intoaccount the affinity maturation of the immune responseIn [27] Chun et al presented a new method employingthe immune algorithm (IA) as the search method for theshape optimization of an electromagnetic device In [28]Endoh et al proposed an optimization algorithm based onimmune model and applied it to the 119899th agentsrsquo travellingsalesman problem called 119899-TSP In [29] Ishiguro et alproposed a new inferenceconsensus-making system inspiredby immune systems in living organisms and they apply

Computational Intelligence and Neuroscience 3

the proposed method to the behavior arbitration of anautonomous mobile robot as a practical example In [30]Harmer et al developed a self-adaptive distributed agent-based defense immune system based on biological strategieswithin a hierarchical layered architecture In [31] Pan et alpresented an immune dominance clonal selection multi-objective algorithm based on the artificial immune systemto further improve the performance of the optimizationalgorithm for locomotive secondary spring load adjustmentIn [32] Souza et al presented two new approaches to solvingthe reconfiguration problem of electrical distribution systems(EDS) using the Copt-aiNet (Artificial Immune Networkfor Combinatorial Optimization) and Opt-aiNet (ArtificialImmune Network for Optimization) algorithms In [33]Zhang et al proposed a novel fuzzy hybrid quantum artificialimmune clustering algorithm based on cloud model (C-FHQAI) to solve the stochastic problem In [34] Savsaniet al presented the effect of hybridizing Biogeography-Based Optimization (BBO) technique with artificial immunealgorithm (AIA) andAnt ColonyOptimization (ACO) in twodifferent ways In [35] Kuo et al were dedicated to proposinga cluster analysis algorithm which is integration of artificialimmune network (aiNet) and 119870-means algorithm (aiNet119870)

23 Discussion According to the above researches manyhealth assessment methods such as density-based spatialclustering and dynamic Bayesian networks have been appliedin the bearingsrsquo condition monitoring network devicedynamic health monitoring and so on But there are still norelevant studies on the dynamic health assessment methodsfor shearer Considering the superiority and universality ofartificial immune algorithm this paper prepares to use thisAI algorithm to predict the dynamic health status of shearerA simulation experiment and an application example arecarried out and the proposed approach is proved to be feasibleand efficient

3 The Dynamic Health Assessment ApproachBased on Artificial Immune Algorithm

31 The Framework of the Proposed Approach Some real-time running indicators of shearer are usually used to classifythe health condition of shearer since the signals can describeits dynamic characteristics In order to identify the dynamichealth status of shearer the following three processes arerequiredThese processes are assessment indicators selectingdata acquisition and initialization and multiclass classifierstraining and testing The proposed condition classificationapproach for shearer dynamic health state is shown inFigure 1 The approach mainly consists of three critical stepsindicators selecting data initialization and data training andclassification Firstly choosing the most effective indicatorsto assess the health condition of shearer is important sinceexcessive assessment indicators will reduce the impact ofmain indicators and cause an incorrect result Then all theobject data in the schema object set are normalized so theattribute value is within the unit interval [0 1] and the sampledata are divided into four types Finally the artificial immune

algorithm is used to classify the dynamic health status ofshearer

32 Selecting the Assessment Indicators The system of sheareris made up by many subsystems Establishing a scientificand reasonable evaluation system is the foundation of thehealth state evaluation for shearer Depending on the actualoperation situation of shearer and referencing other healthassessment systems the assessment consequences for shearerhealth can be divided into four typical modes normal modetransition mode abnormal mode and danger mode Thedefinition of each type of operation is given as follows

Normal Mode During the working process the health indi-cators of shearer change a little and are all in normal rangeThe shearer works normally

Transition Mode During the working process one or twohealth indicators of shearer have a wide range change occa-sionally and are not up to the danger line The shearer worksnormally andmeanwhile the worker of shearermust discoverthe problem and solve it

Abnormal Mode During the working process some of thehealth indicators of shearer have a wide range change persis-tently and are not up to the danger lineTheworker of shearershould stop coal production before returning it to normal

Dangerous Mode During the working process some of thehealth indicators of shearer have a sudden change and areup to the danger line The worker of shearer should stop coalproduction immediately

By setting malfunction threshold value depending onoperation situation four modes of shearer health situationdecrease progressively Four different healthmodes can guidecoal worker adopting corresponding operation respectively

The system of shearer is made up by many subsystemsThe data from historical recording and real-time monitoringof the subsystems reflect the health status of shearer moreor less However in practical application we must choosethe most effective indicators to assess the health situationof shearer and eliminate subordinate indicators as excessiveassessment indicators will reduce the impact of the mainindicators causing an incorrect result According to theexpert experience and actual working condition of shearerthe dynamic health condition depends on the real-timemonitoring data In this paper the key content is the real-time health assessment of shearerThus to assess the dynamichealth situation of shearer we choose nine real-time runningindicators the pulling speed 119901

1 the right cutting motor cur-

rent 1199012and the left cuttingmotor current 119901

3 the right pulling

motor current 1199014and the left pulling motor current 119901

5 the

right cuttingmotor temperature 1199016and the left cuttingmotor

temperature 1199017 and the right pulling motor temperature 119901

8

and the left pulling motor temperature 1199019 There are test data

showing that the pulling speed has a mapping relation withworking load of shearerMonitoring the change of the pullingspeed can reflect the working load in a degree Moreover asthemost important information on judging shearer operating


Data initialization

Training detector set

Testing detector set

Normal modedetector

Transition modedetector

Abnormal modedetector

Danger modedetector

Monitoring assessment

Data initialization

Health state assessment

Assessmentconsequences

History sample data

Generatingdetector set

Multiclassclassifiers

Selecting

indicators

assessmentindicators

Figure 1 The framework of the proposed approach

state the cuttingmotor current and the pullingmotor currentcan be influenced by the pulling speed the cutting drumheight the working load the coal-rock characteristic andso on The difference between the two currents is that thecutting motor current has a direct proportion with workingload while the pulling motor current can comprehensivelycharacterize the pulling resistance Finally the slow changeof the cutting motor temperature and the pulling motortemperature can represent the general state of working loadand pulling load over a period of time The nine indicatorscan reflect the shearer operation state clearly The assessmentindicators of dynamic health assessment model for shearercan be shown in Figure 2

33 Defining Detectors Set According to the nine indicatorsof shearer dynamic health assessment approach definedabove we can determine the unknown schema object 1199011015840for nine-dimensional attribute space shown in detail inDefinition 1 Corresponding to the four modes of shearerdynamic health assessment consequences multiclass classi-fiers are constituted by four detectors normal mode (119862

1)

transitionmode (1198622) abnormalmode (119862

3) and dangermode

(1198624) Any one of the non-self-class objects (the schema

object of classes 1198621 119862

119894minus1 119862119894+1 119862

4) can be recognized

by the 119894th detector (119877119894) excepting the self-class object (the

schema object of class 119862119894) In other words each detector only

cannot recognize the corresponding class object of particularassessment consequence modeThe immune classifier modelof dynamic health assessment for shearer is shown in Figure 3

Before establishing the dynamic health assessment modelbased on artificial immune algorithm the related definitionsof multiclass classifiers are given as follows

Definition 1 Each schema object can be represented as a 119896-dimensional vector (119901 119888) = (119901

1 1199012 119901

119896 119888) and 119875 is data

Health status ofshearer

Normalmode

Transitionmode

Abnormalmode

Dangermode

The right cuttingmotor current

The left cuttingmotor current

The left pulling

The right cuttingmotor temperature

The left cuttingmotor temperature

The right pullingmotor temperature

The pulling speed

The left pullingmotor temperatureEvaluation

indicators

Four health modes

The right pullingmotor current

motor current

Figure 2 The indicators of dynamic health assessment model forshearer

set of the schema object 119901 isin 119875 where 119896 = 9 is the numberof attributes of the schema object and 119888 is attribute class ofschema object

Definition 2 Detectors set119877 can recognize all the certain typedata in the schema object set Each member of the detectorsset is called receptor marked as 119903 (119903 119888) = (119903

1 1199032 119903

119896 119888)

The receptor has similar structure to the schema object


recognizingschema object of

Normal

Unknown


Abnormal DangerTransition

schema object p998400

class C1

Detector R1recognizing

schema object of class C2





Detector R4

recognize p998400

only if detector Ri cannotp998400 belongs to class Ci if and

Figure 3 Immune classifier model of dynamic health assessment for shearer

Definition 3 Receptor can recognize any one of the schemaobjects of a certain type The degree of similarity betweenreceptor and schema object can measure affinity

affinity (119903 119901) = 1 minus119863 (119903 119901)

119896

119863 (119903 119901) = radic

119896

sum

119894=1

(119903119894minus 119901119894)2

(1)

where 119877 is detectors set 119903 isin 119877119863(119903 119901) is the Euclidean distance of schema object 119901 and

detector 119903The function value of affinity lies between 0 and 1 The

more similar the value between schema object 119901 and detector119903 the greater the function value of affinity

Definition 4 120597selection is a choosing threshold for selectedreceptor and the value lies between 0 and 1 The choosingthreshold is very important as it selects which receptorsshould be removed from detectors set in the step of training

Definition 5 120597detection is a testing threshold for detectors setThe value of testing threshold is the key to correctly classifydetectors as it decides which detectors should be activated inthe step of testing

34 Establishing the Assessment Model In this section theflows for establishing the dynamic health assessment modelbased on artificial immune algorithm are provided in detailIt mainly includes three steps

341 Data Initialization This step can be regarded as a datapreprocessing stage Each schema object is represented as a9-dimensional vector (119901 119888) = (119901

1 1199012 119901

9 119888) 119901 isin 119875 All

object data of assessment indicators in the schema objectset are normalized so the attribute value is within the unit

interval [0 1] and 119901119894isin (0 1) To correspond to the four

patterns of shearer dynamic health assessment consequencesthe training data of schema object set are divided into fourtypes normal mode (119863

1) transition mode (119863

2) abnormal

mode (1198633) and danger mode (119863

4)

342 Training Detectors Set The purpose of training stage isgenerating an effective detector for each schema object Thesteps for generating detector are given as follows

Step 1 Take preprocessed 119863119894as self-data set 119863self 119863self =

119863119894 so non-self-data set 119863nonself was made up by the other

preprocessed data sets 119863nonself = 1198631 cup sdot sdot sdot cup 119863119894minus1 cup 119863119894+1 cupsdot sdot sdot cup 119863

119873 Initial detector 119877

119894is empty 119877

119894= 0

Step 2 Generate random alternative detectors set 1198771015840

Step 3 Calculate affinity between 1198771015840 and 119863self if affinity(119901119894

119903119895) gt 120597selection delete 119903119895 from 119877

1015840 (negative selection)

Step 4 Calculate affinity between 1198771015840 and 119863nonself ifaffinity(119901

119894 119903119895) lt 120597selection delete 119903

119895from 1198771015840 (positive

selection)

Step 5 Delete the individual from 119863nonself if it can berecognized by 1198771015840

Step 6 If 119863nonself = 0 detector 119877119894 is accomplished anditeration is finished 119877

119894= 119877119894cup 1198771015840 otherwise turn back to

Step 2

Using the same negative selection algorithmof generatingdetector repeat calculation four times from class object 119862

1

to class object 1198624to all detector sets until every detector

can distinguish self-class object and non-self-class objectThegeneration process of a detector is shown in Figure 4

343 Testing Detectors Set To distinguish self-class objectand non-self-class object all detectors are used for circulatory


Begin

Defining self-data Defining non-self-

Calculating affinity Calculating affinity

Choosingthreshold

match

Choosingthreshold

match

Negative selection

Positiveselection

Delete Delete

Detection setmatches entirely

End

NonselfSelf

No

Yes

YesYes

NoNo

Updating mature detector set Ri

between R998400 and Dnonself

nself

between R998400 and Dself

data set Dnoset Dself

Generating random alternative detector set R998400

Figure 4 The generation process of a detector

Begin

Detector 1 Detector 2 Detector 3 Detector 4

Calculating affinity

Setting test threshold

Threshold match

Self-class object

End

Unique onesdo not match

Two or more do not match

All match

Non-self-class object

DetectorsUnknown schema object p998400

Figure 5 Flowchart of negative selection test for a new sample

elimination for new sample in testing step The flowchartof negative selection test for a new sample is shown inFigure 5

1199011015840 is sample data of unknown schemaobject for inputting

Then calculate the value 120597detection between1199011015840 and all detectors

(1198771sim 119877119894) If data 1199011015840 can be recognized by detector 119877

119898(1 le

119898 le 119894) data 1199011015840 does not belong to the schema object 119862119898

Repeat this process until all detectors are tested The finalconsequence will be one of the following cases

Case 1 If schema object 1199011015840 only cannot be recognized bydetector 119877

119894 then schema object 1199011015840 belongs to class object

119862119894

Case 2 If schema object 1199011015840 cannot be recognized by two ormore detectors set a new test threshold value Calculate thevalue 120597detection between 119901

1015840 and nonactivated detectors Repeatthis process until only one detector remained nonactivatedThen schema object 1199011015840 belongs to the corresponding schemaobject of the last nonactivated detector

Case 3 If schema object1199011015840 can be recognized by all detectorsset a new test threshold value Calculate the value 120597detectionbetween 1199011015840 and all detectors Repeat this process until onlyone detector remained nonactivated Then schema object1199011015840 belongs to the corresponding schema object of the last

nonactivated detector

4 Simulation Examples and Application

41 Simulation Examples In this section some simulationexamples were put forward to verify the feasibility andefficiency of the proposed approach


Table 1 Normalized data of pattern objects for shearer

Number 1198751

1198752

1198753

1198754

1198755

1198756

1198757

1198758

1198759

Categories1 0226 0515 0485 0559 0539 0455 0568 0298 0338 119863

1

2 0194 0540 0485 0557 0535 0455 0565 0287 0328 1198631

3 0168 0577 0494 0486 0480 0458 0565 0291 0335 1198631

4 0167 0562 0485 0513 0496 0458 0565 0290 0338 1198631

185 0257 0509 0494 0461 0489 0458 0565 0296 0343 1198631

186 0258 0519 0506 0471 0497 0455 0565 0295 0343 1198631

187 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

188 0227 0503 0488 0417 0422 0458 0568 0281 0327 1198631

354 0350 0519 0491 0459 0450 0458 0565 0284 0329 1198631

355 0858 0552 0821 0531 0704 0458 0570 0301 0350 1198633

356 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

357 0408 0522 0497 0455 0445 0458 0568 0280 0327 1198631

587 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

588 0773 0556 0990 0659 0653 0458 0568 0304 0348 1198634

589 0850 0503 0907 0630 0619 0458 0568 0301 0348 1198634

590 0854 0540 0861 0587 0635 0458 0568 0303 0349 1198633

753 0943 0506 0509 0816 0783 0461 0570 0311 0359 1198633

754 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

755 0000 0481 0497 0445 0432 0458 0570 0313 0352 1198632

756 0000 0494 0497 0449 0431 0458 0570 0313 0352 1198632

893 0042 0491 0497 0386 0329 0458 0570 0294 0339 1198632

894 0151 0494 0491 0405 0419 0458 0570 0290 0342 1198632

895 0950 0494 0503 0708 0724 0461 0570 0296 0354 1198633

896 0399 0488 0491 0389 0393 0458 0570 0288 0338 1198632

1072 0950 0506 0506 0621 0783 0461 0570 0297 0358 1198633

1073 0948 0506 0500 0695 0658 0461 0570 0298 0358 1198633

1074 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

1075 0854 0506 0920 0587 0635 0458 0568 0303 0349 1198634

1200 0347 0500 0475 0535 0534 0464 0527 0298 0326 1198631

The sample data were acquired from the shearer in22210 fully mechanized coal face of Zhong Ping EnergyChemical Group No 6 Mine Depending on the assessmentmodel of prediction approach of shearer dynamic healthassessment the acquired data were normalized so that theobject data were represented as a 9-dimensional vector

(119901 119888) = (1199011 1199012 119901

9 119888) 119901 isin 119875 The data in the schema

object set was initialized so the attribute value was withinthe unit interval [0 1] 119901

119894isin (0 1) The training data of

schema object set were divided into four types normal mode(1198631) transitionmode (119863

2) abnormalmode (119863

3) and danger

mode (1198634) As shown in Table 1 1000 groups of data were


f = 1f = 2f = 3

f = 4Wrongly classified samples

20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes

Figure 6 Classification results of the classifier based on artificialimmune algorithm

randomly chosen to train the detectors set and the last 200groups were used to test the classification performance of thetrained detectors

After the assessment model based on artificial immunealgorithm was trained the multiclass classifiers of the assess-ment system were constituted by four detectors and eachdetector only could not recognize corresponding class objectof particular assessment consequence mode Actually if theinput schema object only could not be recognized by onedetector then the schema object belongs to this class object

After the training phase an assessment system could beobtained In order to verify the accuracy of the model theremaining 200 samples were utilized to test its performanceThe prediction consequence was given as in Figure 6 Asshown in Figure 6 only eight testing samples were mis-classified and circled in red The ordinate values 1 2 3and 4 corresponded to four assessment consequences ofshearer dynamic health modes and identification accuraciesof four detectors were 96 94 98 and 96 respectivelyThe overall average classification accuracy was 96 whichsatisfied the engineering requirement The testing resultsindicated that the proposed approach performed with lowerdeviation and could be applied in the assessment of shearerdynamic health

In order to indicate the meliority of assessment modelbased on artificial immune algorithm the assessment modelbased on back propagation-neural network (BP-NN) andsupport vector machine (SVM) was provided to solve theproblem of the above example The training samples andtesting samples were the same The configurations of simu-lation environment for three algorithms were uniform andthe relevant parameters were in common with the aboveexample The prediction consequence of the assessmentmodel based on BP-NN was given as in Figure 7 As shownin Figure 7 twenty-one testing samples were misclassifiedand some samples had a large deviation to real situationsThe classification accuracies for four health modes were90 86 90 and 92 and overall average classificationaccuracy was 895

20 40 60 80 100 120 140 160 180 2000Testing samples

f = 1f = 2f = 3


0

1

2

3

4

5

6

Hea

lth m

odes

Figure 7 Classification results of the classifier based on BP-NN

f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes

Figure 8 Classification results of the classifier based on SVM

The prediction consequence of the assessment modelbased on SVM was given as in Figure 8 As shown inFigure 8 twenty-three testing samples were misclassifiedand some samples had a large deviation to real situationsThe classification accuracies for four health modes were92 86 88 and 90 and overall average classificationaccuracy was 885 It was observed that the proposedmethod had a better classification capability and performancethan the competing methods With the benefits of artificialimmune algorithm in uncertain fields the proposed classifiercould obtain higher classification accuracy than single BP-NN and SVM classifiers

42 Further Discussion In order to further compare andanalyze the overall performance of SVM BP-NN and AI thesame 1200 samples are experimented with In this examplea certain number of samples denoted by training size (119879size)are randomly selected from the data as the training samplesand 200 samples are randomly selected from the remaining1200 minus 119879size samples as the testing samples Each learningalgorithm is then trained and tested 200 times and theclassification error rate is recorded as the final result In this


SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e

Figure 9 The changes of classification error rate with different training sizes

study the training size of the example varies over 119879size =100 120 140 160 1000 That is to say we run severaltrials over the algorithms with training size ranging from100 to 1000 The classification error rate 119898119899 (where 119898 isthe classification error times and 119899 is the total test times) ischosen as the metric to express the result as a proportion ofthe optimal solution

Figure 9 plots themeans of thismetric (classification errorrate) for each trial as a function of problem size 119879size It canbe seen that for all trials the classification error rate decreasesnonlinearly with 119879size and the classification accuracy of AIoutperforms BP-NN and SVM for all 119879size

From Figure 9 it is obvious that the classification errorrate descent velocity of AI is the fastest across different train-ing sizes and owns stronger generalization ability than BP-NN and SVM regardless of the training size What is morethe classification accuracy of AI is more supernal and robustTherefore the AI algorithm can obtain a relatively highaccuracy to provide an effective support tool for dynamichealth assessment for shearer

43 Industrial Application In this section a system based onthe proposed approach had been developed and applied inthe field of shearer dynamic health assessment as shown inFigure 10

As Figure 10 showed the ldquogateway controllerrdquo andldquoground monitoring centerrdquo were used to control and moni-tor the shearer running parameters The system based on theproposed approach was uploaded into the gateway controllerThe pulling speed the left cutting motor current the rightcuttingmotor current the left pullingmotor current the right

pulling motor current the left cutting motor temperaturethe right cutting motor temperature the left pulling motortemperature and the right pulling motor temperature werecollected every 1Hz from the shearer controller and thecollected data were transmitted to the gateway controllerThen the changes of shearer dynamic health assessmentconsequence were identified and showed on ldquomonitoringinterface for shearer dynamic health assessmentrdquo

In order to illustrate the application effect of the proposedapproach the shearer was running in fully mechanized coalface from 1350m to 1500m by the manual operation Thedynamic health assessment curve based on the proposedclassifier was shown in Figure 11 and the ordinate valuesof 1 2 3 and 4 denoted four health classes normal modetransition mode abnormal mode and danger mode Thecurve showed two obvious changes in segments A and BTheordinate values were leaped from 1 to 3 which means thatthe shearer health status had a sudden change from normalmode to abnormal mode The changes of some operationalparameters were plotted in Figure 12 The cutting motorcurrent had a noticeable increasewhen the right cutting drumcut the floor between 1350m and 1385m and the left cuttingdrum cut the roof between 1460m and 1480m The resultsof shearer dynamic health assessment based on the proposedsystem were almost completely consistent with the actualcutting status of shearer

5 Conclusions and Future Work

The main contribution of this paper was that a methodologybased on artificial immune algorithm for the assessment of


Ground LAN

Undergroundoptical networks

Airborne monitors for

running parameters

Crossheading remote

monitoring center

Ground monitoring center for shearerdynamic health assessment

Fully mechanized coal face wireless

MESH switched networks

Crossheading wireless MESH switched networks

Monitoring interface for shearer dynamichealth assessment

Gateway controller Ground monitoring center

Fully mechanized coal face wireless MESH switchednetworks

Airborne monitors for running parameters

Coal mining face

middot middot middot




Figure 10 Hardware construction in fully mechanized coal face

shearer dynamic health status was presented The detailedflows for the proposed approach were described includingthree critical steps that is assessment indicators selectingdata acquisition and initialization and multiclass classifierstraining and testing In order to verify the feasibility and effi-ciency of the proposed approach a simulation example wasprovided and some comparisons with other algorithms werecarried out The simulation results showed that the proposedapproach was outperforming others Finally the proposedapproach was applied to an engineering problem of shearerdynamic health assessmentThe industrial application resultsshowed that the paper research achievements could beused combining with shearer automation control system in

fully mechanized coal face and had obvious effectivenesson reducing operating trouble and production accident ofshearer and improving coal production efficiency furtherTheartificial immune algorithm could obtain a relatively highaccuracy to provide an effective support tool for dynamichealth assessment for shearer

In future studies the authors plan to investigate someimprovements for the proposed approach Possible improve-ments may include the combination of artificial immunealgorithm with other intelligent algorithms for better per-formance In addition the applications of the proposedapproach in dynamic health assessment domain are worthfurther study from the authors


A B

Figure 11 The dynamic health assessment curve of shearer based on the proposed system

The cutting height curve of right and left cutting drum The right and left cutting motors current curve

Figure 12 The operational parameters curve of shearer with the manual operation

Competing Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

The support of Joint Funds of the National Natural ScienceFoundation of China (no U1510117) National Key BasicResearch Program of China Key Fundamental Research onthe Unmanned Mining Equipment in Deep Dangerous CoalBed (no 2014CB046301) the Priority Academic ProgramDevelopment (PAPD) of Jiangsu Higher Education Institu-tions and the Innovation Funds of Production and ResearchCooperation Project in Jiangsu Province (BY2014107) incarrying out this research is gratefully acknowledged

References

[1] B H G Brady and E T Brown ldquoLongwall and caving miningmethodsrdquo in Rock Mechanics For Underground Mining pp430ndash483 Springer Amsterdam The Netherlands 2006

[2] X Zhang X Ma M and Z S Yang ldquoAnalysis and diagnosisof coal shearer machine fault based on improved supportvector theoryrdquo in Proceedings of the International Conferenceon Electrical Automation and Mechanical Engineering PhuketThailand July 2015

[3] F Gao L J Xiao W Y Zhong and W Liu ldquoFault diagnosisof shearer based on fuzzy inferencerdquo Applied Mechanics andMaterials vol 52ndash54 pp 1577ndash1580 2011

[4] S Kerroumi X Chiementin and L Rasolofondraibe ldquoDynamicclassification method of fault indicators for bearingsrsquo monitor-ingrdquoMechanics and Industry vol 14 no 2 pp 115ndash120 2013


[5] Z- W He M-Y Gao G-J Ma Y-Y Liu and S-X ChenldquoOnline state-of-health estimation of lithium-ion batteriesusing Dynamic Bayesian Networksrdquo Journal of Power Sourcesvol 267 pp 576ndash583 2014

[6] F-R Liu Q-L Wang and X-Z Gao ldquoSurvey of artificialimmune systemrdquo in Proceedings of the 1st International Sym-posium on Systems and Control in Aerospace and Astronautics(ISSCAA rsquo06) p 989 2006

[7] K-J Wang K-H Chen and M-A Angelia ldquoAn improvedartificial immune recognition systemwith the opposite sign testfor feature selectionrdquoKnowledge-Based Systems vol 71 pp 126ndash145 2014

[8] I Jenhani and Z Elouedi ldquoRe-visiting the artificial immunerecognition system a survey and an improved versionrdquo Arti-ficial Intelligence Review vol 42 no 4 pp 821ndash833 2012

[9] W Zhong-Bin N Wen-Feng and L Shu-Bin ldquoResearch onkey technologies of remote monitoring platform for shearerrdquo inProceedings of the International Conference on IEEE MeasuringTechnology and Mechatronics Automation (ICMTMA rsquo09) vol1 pp 316ndash319 Zhangjiajie China April 2009

[10] X Zhou ZWangC Tan R Ji andX Liu ldquoAnovel approach forshearer memory cutting based on fuzzy optimization methodrdquoAdvances in Mechanical Engineering vol 5 Article ID 3192722013

[11] P W Tse and Y L Tse ldquoOn-road mobile phone based auto-mobile safety systemwith emphasis on engine health evaluationand expert advicerdquo in Proceedings of the Technology Manage-ment for Emerging Technologies Conference (PICMET rsquo12) pp3232ndash3241 IEEE Vancouver Canada July 2012

[12] D Black and A K Winiewicz ldquoInternal network device dyna-mic health monitoringrdquo US Patent 7143153 2006

[13] N M Vichare and M G Pecht ldquoPrognostics and healthmanagement of electronicsrdquo IEEE Transactions on Componentsand Packaging Technologies vol 29 no 1 pp 222ndash229 2006

[14] M Pecht and R Jaai ldquoA prognostics and health managementroadmap for information and electronics-rich systemsrdquoMicro-electronics Reliability vol 50 no 3 pp 317ndash323 2010

[15] J-J Yang R Xi D Liu H Jiang and M Wu ldquoAnalysis ofshearer gear vibration in the no-load staterdquo in Proceedings ofthe 5th Conference on Measuring Technology and MechatronicsAutomation (ICMTMA rsquo13) pp 247ndash250 IEEE Hong KongJanuary 2013

[16] X-D Yin A-G Liu X-M Dong and W-S Hao ldquoStudy onhealth evaluation system for coal mine speed reducer based onembedded systemrdquo Applied Mechanics and Materials vol 105pp 660ndash663 2012

[17] D Mascarenas C Plont C Brown et al ldquoA vibro-haptichumanndashmachine interface for structural health monitoringrdquoStructural Health Monitoring vol 13 no 6 pp 671ndash685 2014

[18] F Cerda S Chen J Bielak J H Garrett P Rizzo and JKovacevic ldquoIndirect structural health monitoring of a sim-plified laboratory-scale bridge modelrdquo Smart Structures andSystems vol 13 no 5 pp 849ndash868 2014

[19] J F Zubizarreta-Rodriguez and S Vasudevan ldquoConditionmon-itoring of brushless DC motors with non-stationary dynamicconditionsrdquo in Proceedings of the IEEE International Instrumen-tation andMeasurement TechnologyConference (I2MTC rsquo14) pp62ndash67 IEEE Montevideo Uruguay May 2014

[20] S Herrmann J Wellnitz S Jahn and S Leonhardt ldquoStructuralhealth monitoring for carbon fiber resin composite car bodystructuresrdquo in Sustainable Automotive Technologies 2013 pp 75ndash96 Springer 2014

[21] J D Farmer N H Packard and A S Perelson ldquoThe immunesystem adaptation and machine learningrdquo Physica D Nonlin-ear Phenomena vol 22 no 1ndash3 pp 187ndash204 1986

[22] A Ishiguro Y Shirai T Kondo and Y Uchikawa ldquoImmunoidan architecture for behavior arbitration based on the immunenetworksrdquo in Proceedings of the IEEERSJ International Confer-ence on Intelligent Robots and Systems (IROS rsquo96) vol 3 pp1730ndash1738 November 1996

[23] Z Tang T Yamaguchi K Tashima O Ishizuka and K TannoldquoMultiple-valued immune network model and its simulationsrdquoin Proceedings of the 27th International Symposium on IEEEMultiple-Valued Logic pp 233ndash238 Antigonish Canada May1997

[24] F Abbattista G Di Gioia G Di Santo and A M FanellildquoAn associative memory based on the immune networksrdquo inProceedings of the IEEE International Conference on NeuralNetworks (ICNN rsquo96) vol 1 pp 519ndash523 1996

[25] Z-LDengG-Z Tan PHe and J-X Ye ldquoA fuzzy logic resourceallocation and memory cell pruning based artificial immunerecognition systemrdquo Journal of Central South University vol 21no 2 pp 610ndash617 2014

[26] LNDeCastro andF J VonZuben ldquoLearning andoptimizationusing the clonal selection principlerdquo IEEE Transactions onEvolutionary Computation vol 6 no 3 pp 239ndash251 2002

[27] J-S ChunM-KKimH-K Jung and S-KHong ldquoShape opti-mization of electromagnetic devices using immune algorithmrdquoIEEE Transactions on Magnetics vol 33 no 2 pp 1876ndash18791997

[28] S Endoh N Toma and K Yamada ldquoImmune algorithm for n-TSPrdquo inProceedings of the 1998 IEEE International Conference onSystems Man and Cybernetics vol 4 pp 3844ndash3849 October1998

[29] A Ishiguro T Kondo YWatanabe andYUchikawa ldquoDynamicbehavior arbitration of autonomous mobile robots usingimmune networksrdquo in Proceedings of the IEEE InternationalConference on Evolutionary Computation vol 2 pp 722ndash727Perth Wash USA November 1995

[30] P K Harmer P D Williams G H Gunsch and G B LamontldquoAn artificial immune system architecture for computer securityapplicationsrdquo IEEE Transactions on Evolutionary Computationvol 6 no 3 pp 252ndash280 2002

[31] D-F Pan M-G Wang Y-N Zhu and K Han ldquoAn optimiza-tion algorithm for locomotive secondary spring load adjust-ment based on artificial immunerdquo Journal of Central SouthUniversity vol 20 no 12 pp 3497ndash3503 2013

[32] S S F Souza R Romero and J F Franco ldquoArtificial immunenetworks Copt-aiNet and Opt-aiNet applied to the reconfigu-ration problem of radial electrical distribution systemsrdquo ElectricPower Systems Research vol 119 pp 304ndash312 2015

[33] R-L Zhang M-Y Shan X-H Liu and L-H Zhang ldquoA novelfuzzy hybrid quantum artificial immune clustering algorithmbased on cloud modelrdquo Engineering Applications of ArtificialIntelligence vol 35 pp 1ndash13 2014

[34] P Savsani R L Jhala and V Savsani ldquoEffect of hybridizingBiogeography-Based Optimization (BBO) technique with Arti-ficial Immune Algorithm (AIA) and Ant Colony Optimization(ACO)rdquo Applied Soft Computing Journal vol 21 pp 542ndash5532014

[35] R J Kuo S S Chen W C Cheng and C Y Tsai ldquoIntegrationof artificial immune network and K-means for cluster analysisrdquoKnowledge and Information Systems vol 40 no 3 pp 541ndash5572014

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



the complex component structure and bad working con-dition there has not been a health assessment concepton shearer and this paper tries to present it In the realmining condition some key index parameters have a strongrelationship with shearer health condition The relationshipis highly nonlinear in nature so that it is hard to developa comprehensive mathematic model The current methodsand mature assessment systems are hardly satisfied with theshearer health state assessment In this paper we try topropose a novel prediction approach for shearer dynamichealth assessment to identify the health state during coalmining

The first mathematical model in artificial immune systemwas proposed in 1974 which initiated subsequent researchesand discussions Artificial immune system (AIS) as a novelintelligent algorithm method inspired from the biologicalimmune system is an effective means for prediction [6ndash8] The AIS can acquire learning capability by learningthe biological protection principle According to the aboveanalysis a novel prediction approach for shearer dynamichealth assessment based on artificial immune algorithm isproposed and the assessment system is validated by thesample data of operating parameters from industrial pro-duction scene Moreover it will prove that artificial immunealgorithm is a better tool for classifying due to its classificationaccuracy than the classifiers based on back propagation-neural network (BP-NN) and support vectormachine (SVM)methods later

The remainder of this paper is organized as follows Somerelated works are outlined in Section 2 The key technologiessuch as system framework selecting the indicators for shearerdynamic health assessment and the proposed approach arepresented in Section 3 Section 4 provides a simulation exam-ple and an industrial application example for shearer dynamichealth assessment based on the proposed approach to specifythe application effect Our conclusions are summarized inSection 5

2 Literature Review

Recent publications relevant to this paper are mainly con-cerned with two research streams the dynamic health assess-ment methods and artificial immune algorithm In thissection we try to summarize the relevant literatures

21 The Dynamic Health Assessment Methods For the dyna-mic health assessment problem lots of research has beendone since the last decades In [9] Zhong-Bin et al developeda remotemonitoring platformof the shearer by usingVirtual-Prototype technology to realize the remote monitoring forthe shearer in the fully mechanized long-wall coal miningface In [10] Zhou et al proposed a novel approach basedon the coal floor height variation which is taken as asignificant factor and fuzzy optimization theory to improvethe implement precision of shearer memory cutting In [11]P W Tse and Y L Tse designed an innovative system thatis installed in a passenger car or a truck that is running onroad and provides instantaneous engine health evaluation

and diagnosis In [12] Black and Winiewicz provided amethod and apparatus for internal network device dynamichealth monitoring to increase network device availabilityIn [13] Vichare and Pecht presented the state of practiceand the current state of research in the area of electronicsprognostics and health management In [14] Pecht and Jaaipresented an assessment of the state of practice in prognosticsand health management of information and electronics-richsystems In [15] Yang et al proposed an accurate identifica-tion of the shearer late underground cutting coal and rockconditions and fault diagnosis by the method of vibrationanalysis In [16] Yin et al designed an embedded healthevaluation system to meet the requirement of continuousmonitoring of the mine special gear box In [17] Mascarenaset al investigated a vibrohaptic human-machine interface forstructural health monitoring In [18] Cerda et al exploredan indirect approach for structural health bridge monitoringallowing for wide yet cost-effective bridge stock coverageIn [19] Zubizarreta-Rodriguez and Vasudevan introduceda new multisensor measurement framework for conditionmonitoring of brushless DCmotors (BLDCM)with bearingsIn [4] Kerroumi et al introduced a dynamic classificationmethod inspired byDBSCAN clusteringmethod formachinecondition monitoring in general and for bearings in particu-lar In [5] He et al presented a novel online method for theestimation of the SOH of lithium- (Li) ion batteries based ondynamic Bayesian networks (DBNs) In [20] Herrmann et algave an introduction into the principle of structural healthmonitoring (SHM) basics of fatigue of fiber resin compositematerials and the possible application of these principles inthe automotive industry

22 Artificial ImmuneAlgorithm Theartificial immune algo-rithm was firstly proposed by Farmer in 1986 [21] It is ableto recognize novel shapes without preprogramming based onthe capacity of learning memory and pattern recognition In[22] Ishiguro et al proposed a new decentralized consensus-making system inspired from the biological immune systemand an adaptation mechanism that can be used to constructa suitable immune network for adequate action selection In[23] Tang et al described a new model of multiple-valuedimmune network based on biological immune responsenetwork In [24] Abbattista et al proposed the use of immunenetwork model for designing associative memories In [25]Deng et al proposed a fuzzy logic resource allocation andmemory cell pruning based artificial immune recognitionsystem (AIRS) to improve the resource allocationmechanismof AIRS and decrease the memory cells In [26] De Castroand Von Zuben proposed computational implementationof the clonal selection principle that explicitly takes intoaccount the affinity maturation of the immune responseIn [27] Chun et al presented a new method employingthe immune algorithm (IA) as the search method for theshape optimization of an electromagnetic device In [28]Endoh et al proposed an optimization algorithm based onimmune model and applied it to the 119899th agentsrsquo travellingsalesman problem called 119899-TSP In [29] Ishiguro et alproposed a new inferenceconsensus-making system inspiredby immune systems in living organisms and they apply
















3 the right pulling


5 the



8




Data initialization



Normal modedetector



Danger modedetector


Data initialization



History sample data



Selecting

indicators





1)


3) and dangermode



119894minus1 119862119894+1 119862







1 1199012 119901

119896 119888) and 119875 is data


Normalmode

Transitionmode

Abnormalmode

Dangermode



The left pulling




The pulling speed


indicators

Four health modes


motor current




1 1199032 119903

119896 119888)




Normal

Unknown




class C1







Detector R4

recognize p998400




affinity (119903 119901) = 1 minus119863 (119903 119901)

119896

119863 (119903 119901) = radic

119896

sum

119894=1

(119903119894minus 119901119894)2

(1)








1 1199012 119901

9 119888) 119901 isin 119875 All





2) abnormal


4)






119894is empty 119877

119894= 0







119895from 1198771015840 (positive

selection)




Step 2


1





Begin



Choosingthreshold

match

Choosingthreshold

match

Negative selection

Positiveselection

Delete Delete


End

NonselfSelf

No

Yes

YesYes

NoNo



nself





Begin




Threshold match

Self-class object

End



All match








119898(1 le





119862119894









Number 1198751

1198752

1198753

1198754

1198755

1198756

1198757

1198758

1198759

Categories1 0226 0515 0485 0559 0539 0455 0568 0298 0338 119863

1

2 0194 0540 0485 0557 0535 0455 0565 0287 0328 1198631

3 0168 0577 0494 0486 0480 0458 0565 0291 0335 1198631

4 0167 0562 0485 0513 0496 0458 0565 0290 0338 1198631

185 0257 0509 0494 0461 0489 0458 0565 0296 0343 1198631

186 0258 0519 0506 0471 0497 0455 0565 0295 0343 1198631

187 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

188 0227 0503 0488 0417 0422 0458 0568 0281 0327 1198631

354 0350 0519 0491 0459 0450 0458 0565 0284 0329 1198631

355 0858 0552 0821 0531 0704 0458 0570 0301 0350 1198633

356 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

357 0408 0522 0497 0455 0445 0458 0568 0280 0327 1198631

587 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

588 0773 0556 0990 0659 0653 0458 0568 0304 0348 1198634

589 0850 0503 0907 0630 0619 0458 0568 0301 0348 1198634

590 0854 0540 0861 0587 0635 0458 0568 0303 0349 1198633

753 0943 0506 0509 0816 0783 0461 0570 0311 0359 1198633

754 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

755 0000 0481 0497 0445 0432 0458 0570 0313 0352 1198632

756 0000 0494 0497 0449 0431 0458 0570 0313 0352 1198632

893 0042 0491 0497 0386 0329 0458 0570 0294 0339 1198632

894 0151 0494 0491 0405 0419 0458 0570 0290 0342 1198632

895 0950 0494 0503 0708 0724 0461 0570 0296 0354 1198633

896 0399 0488 0491 0389 0393 0458 0570 0288 0338 1198632

1072 0950 0506 0506 0621 0783 0461 0570 0297 0358 1198633

1073 0948 0506 0500 0695 0658 0461 0570 0298 0358 1198633

1074 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

1075 0854 0506 0920 0587 0635 0458 0568 0303 0349 1198634

1200 0347 0500 0475 0535 0534 0464 0527 0298 0326 1198631


(119901 119888) = (1199011 1199012 119901






3) and danger



f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes






20 40 60 80 100 120 140 160 180 2000Testing samples

f = 1f = 2f = 3


0

1

2

3

4

5

6

Hea

lth m

odes


f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes





SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e












Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in


















3 the right pulling


5 the



8




Data initialization



Normal modedetector



Danger modedetector


Data initialization



History sample data



Selecting

indicators





1)


3) and dangermode



119894minus1 119862119894+1 119862







1 1199012 119901

119896 119888) and 119875 is data


Normalmode

Transitionmode

Abnormalmode

Dangermode



The left pulling




The pulling speed


indicators

Four health modes


motor current




1 1199032 119903

119896 119888)




Normal

Unknown




class C1







Detector R4

recognize p998400




affinity (119903 119901) = 1 minus119863 (119903 119901)

119896

119863 (119903 119901) = radic

119896

sum

119894=1

(119903119894minus 119901119894)2

(1)








1 1199012 119901

9 119888) 119901 isin 119875 All





2) abnormal


4)






119894is empty 119877

119894= 0







119895from 1198771015840 (positive

selection)




Step 2


1





Begin



Choosingthreshold

match

Choosingthreshold

match

Negative selection

Positiveselection

Delete Delete


End

NonselfSelf

No

Yes

YesYes

NoNo



nself





Begin




Threshold match

Self-class object

End



All match








119898(1 le





119862119894









Number 1198751

1198752

1198753

1198754

1198755

1198756

1198757

1198758

1198759

Categories1 0226 0515 0485 0559 0539 0455 0568 0298 0338 119863

1

2 0194 0540 0485 0557 0535 0455 0565 0287 0328 1198631

3 0168 0577 0494 0486 0480 0458 0565 0291 0335 1198631

4 0167 0562 0485 0513 0496 0458 0565 0290 0338 1198631

185 0257 0509 0494 0461 0489 0458 0565 0296 0343 1198631

186 0258 0519 0506 0471 0497 0455 0565 0295 0343 1198631

187 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

188 0227 0503 0488 0417 0422 0458 0568 0281 0327 1198631

354 0350 0519 0491 0459 0450 0458 0565 0284 0329 1198631

355 0858 0552 0821 0531 0704 0458 0570 0301 0350 1198633

356 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

357 0408 0522 0497 0455 0445 0458 0568 0280 0327 1198631

587 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

588 0773 0556 0990 0659 0653 0458 0568 0304 0348 1198634

589 0850 0503 0907 0630 0619 0458 0568 0301 0348 1198634

590 0854 0540 0861 0587 0635 0458 0568 0303 0349 1198633

753 0943 0506 0509 0816 0783 0461 0570 0311 0359 1198633

754 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

755 0000 0481 0497 0445 0432 0458 0570 0313 0352 1198632

756 0000 0494 0497 0449 0431 0458 0570 0313 0352 1198632

893 0042 0491 0497 0386 0329 0458 0570 0294 0339 1198632

894 0151 0494 0491 0405 0419 0458 0570 0290 0342 1198632

895 0950 0494 0503 0708 0724 0461 0570 0296 0354 1198633

896 0399 0488 0491 0389 0393 0458 0570 0288 0338 1198632

1072 0950 0506 0506 0621 0783 0461 0570 0297 0358 1198633

1073 0948 0506 0500 0695 0658 0461 0570 0298 0358 1198633

1074 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

1075 0854 0506 0920 0587 0635 0458 0568 0303 0349 1198634

1200 0347 0500 0475 0535 0534 0464 0527 0298 0326 1198631


(119901 119888) = (1199011 1199012 119901






3) and danger



f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes






20 40 60 80 100 120 140 160 180 2000Testing samples

f = 1f = 2f = 3


0

1

2

3

4

5

6

Hea

lth m

odes


f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes





SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e












Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Data initialization



Normal modedetector



Danger modedetector


Data initialization



History sample data



Selecting

indicators





1)


3) and dangermode



119894minus1 119862119894+1 119862







1 1199012 119901

119896 119888) and 119875 is data


Normalmode

Transitionmode

Abnormalmode

Dangermode



The left pulling




The pulling speed


indicators

Four health modes


motor current




1 1199032 119903

119896 119888)




Normal

Unknown




class C1







Detector R4

recognize p998400




affinity (119903 119901) = 1 minus119863 (119903 119901)

119896

119863 (119903 119901) = radic

119896

sum

119894=1

(119903119894minus 119901119894)2

(1)








1 1199012 119901

9 119888) 119901 isin 119875 All





2) abnormal


4)






119894is empty 119877

119894= 0







119895from 1198771015840 (positive

selection)




Step 2


1





Begin



Choosingthreshold

match

Choosingthreshold

match

Negative selection

Positiveselection

Delete Delete


End

NonselfSelf

No

Yes

YesYes

NoNo



nself





Begin




Threshold match

Self-class object

End



All match








119898(1 le





119862119894









Number 1198751

1198752

1198753

1198754

1198755

1198756

1198757

1198758

1198759

Categories1 0226 0515 0485 0559 0539 0455 0568 0298 0338 119863

1

2 0194 0540 0485 0557 0535 0455 0565 0287 0328 1198631

3 0168 0577 0494 0486 0480 0458 0565 0291 0335 1198631

4 0167 0562 0485 0513 0496 0458 0565 0290 0338 1198631

185 0257 0509 0494 0461 0489 0458 0565 0296 0343 1198631

186 0258 0519 0506 0471 0497 0455 0565 0295 0343 1198631

187 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

188 0227 0503 0488 0417 0422 0458 0568 0281 0327 1198631

354 0350 0519 0491 0459 0450 0458 0565 0284 0329 1198631

355 0858 0552 0821 0531 0704 0458 0570 0301 0350 1198633

356 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

357 0408 0522 0497 0455 0445 0458 0568 0280 0327 1198631

587 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

588 0773 0556 0990 0659 0653 0458 0568 0304 0348 1198634

589 0850 0503 0907 0630 0619 0458 0568 0301 0348 1198634

590 0854 0540 0861 0587 0635 0458 0568 0303 0349 1198633

753 0943 0506 0509 0816 0783 0461 0570 0311 0359 1198633

754 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

755 0000 0481 0497 0445 0432 0458 0570 0313 0352 1198632

756 0000 0494 0497 0449 0431 0458 0570 0313 0352 1198632

893 0042 0491 0497 0386 0329 0458 0570 0294 0339 1198632

894 0151 0494 0491 0405 0419 0458 0570 0290 0342 1198632

895 0950 0494 0503 0708 0724 0461 0570 0296 0354 1198633

896 0399 0488 0491 0389 0393 0458 0570 0288 0338 1198632

1072 0950 0506 0506 0621 0783 0461 0570 0297 0358 1198633

1073 0948 0506 0500 0695 0658 0461 0570 0298 0358 1198633

1074 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

1075 0854 0506 0920 0587 0635 0458 0568 0303 0349 1198634

1200 0347 0500 0475 0535 0534 0464 0527 0298 0326 1198631


(119901 119888) = (1199011 1199012 119901






3) and danger



f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes






20 40 60 80 100 120 140 160 180 2000Testing samples

f = 1f = 2f = 3


0

1

2

3

4

5

6

Hea

lth m

odes


f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes





SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e












Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





Normal

Unknown




class C1







Detector R4

recognize p998400




affinity (119903 119901) = 1 minus119863 (119903 119901)

119896

119863 (119903 119901) = radic

119896

sum

119894=1

(119903119894minus 119901119894)2

(1)








1 1199012 119901

9 119888) 119901 isin 119875 All





2) abnormal


4)






119894is empty 119877

119894= 0







119895from 1198771015840 (positive

selection)




Step 2


1





Begin



Choosingthreshold

match

Choosingthreshold

match

Negative selection

Positiveselection

Delete Delete


End

NonselfSelf

No

Yes

YesYes

NoNo



nself





Begin




Threshold match

Self-class object

End



All match








119898(1 le





119862119894









Number 1198751

1198752

1198753

1198754

1198755

1198756

1198757

1198758

1198759

Categories1 0226 0515 0485 0559 0539 0455 0568 0298 0338 119863

1

2 0194 0540 0485 0557 0535 0455 0565 0287 0328 1198631

3 0168 0577 0494 0486 0480 0458 0565 0291 0335 1198631

4 0167 0562 0485 0513 0496 0458 0565 0290 0338 1198631

185 0257 0509 0494 0461 0489 0458 0565 0296 0343 1198631

186 0258 0519 0506 0471 0497 0455 0565 0295 0343 1198631

187 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

188 0227 0503 0488 0417 0422 0458 0568 0281 0327 1198631

354 0350 0519 0491 0459 0450 0458 0565 0284 0329 1198631

355 0858 0552 0821 0531 0704 0458 0570 0301 0350 1198633

356 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

357 0408 0522 0497 0455 0445 0458 0568 0280 0327 1198631

587 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

588 0773 0556 0990 0659 0653 0458 0568 0304 0348 1198634

589 0850 0503 0907 0630 0619 0458 0568 0301 0348 1198634

590 0854 0540 0861 0587 0635 0458 0568 0303 0349 1198633

753 0943 0506 0509 0816 0783 0461 0570 0311 0359 1198633

754 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

755 0000 0481 0497 0445 0432 0458 0570 0313 0352 1198632

756 0000 0494 0497 0449 0431 0458 0570 0313 0352 1198632

893 0042 0491 0497 0386 0329 0458 0570 0294 0339 1198632

894 0151 0494 0491 0405 0419 0458 0570 0290 0342 1198632

895 0950 0494 0503 0708 0724 0461 0570 0296 0354 1198633

896 0399 0488 0491 0389 0393 0458 0570 0288 0338 1198632

1072 0950 0506 0506 0621 0783 0461 0570 0297 0358 1198633

1073 0948 0506 0500 0695 0658 0461 0570 0298 0358 1198633

1074 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

1075 0854 0506 0920 0587 0635 0458 0568 0303 0349 1198634

1200 0347 0500 0475 0535 0534 0464 0527 0298 0326 1198631


(119901 119888) = (1199011 1199012 119901






3) and danger



f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes






20 40 60 80 100 120 140 160 180 2000Testing samples

f = 1f = 2f = 3


0

1

2

3

4

5

6

Hea

lth m

odes


f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes





SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e












Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Begin



Choosingthreshold

match

Choosingthreshold

match

Negative selection

Positiveselection

Delete Delete


End

NonselfSelf

No

Yes

YesYes

NoNo



nself





Begin




Threshold match

Self-class object

End



All match








119898(1 le





119862119894









Number 1198751

1198752

1198753

1198754

1198755

1198756

1198757

1198758

1198759

Categories1 0226 0515 0485 0559 0539 0455 0568 0298 0338 119863

1

2 0194 0540 0485 0557 0535 0455 0565 0287 0328 1198631

3 0168 0577 0494 0486 0480 0458 0565 0291 0335 1198631

4 0167 0562 0485 0513 0496 0458 0565 0290 0338 1198631

185 0257 0509 0494 0461 0489 0458 0565 0296 0343 1198631

186 0258 0519 0506 0471 0497 0455 0565 0295 0343 1198631

187 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

188 0227 0503 0488 0417 0422 0458 0568 0281 0327 1198631

354 0350 0519 0491 0459 0450 0458 0565 0284 0329 1198631

355 0858 0552 0821 0531 0704 0458 0570 0301 0350 1198633

356 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

357 0408 0522 0497 0455 0445 0458 0568 0280 0327 1198631

587 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

588 0773 0556 0990 0659 0653 0458 0568 0304 0348 1198634

589 0850 0503 0907 0630 0619 0458 0568 0301 0348 1198634

590 0854 0540 0861 0587 0635 0458 0568 0303 0349 1198633

753 0943 0506 0509 0816 0783 0461 0570 0311 0359 1198633

754 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

755 0000 0481 0497 0445 0432 0458 0570 0313 0352 1198632

756 0000 0494 0497 0449 0431 0458 0570 0313 0352 1198632

893 0042 0491 0497 0386 0329 0458 0570 0294 0339 1198632

894 0151 0494 0491 0405 0419 0458 0570 0290 0342 1198632

895 0950 0494 0503 0708 0724 0461 0570 0296 0354 1198633

896 0399 0488 0491 0389 0393 0458 0570 0288 0338 1198632

1072 0950 0506 0506 0621 0783 0461 0570 0297 0358 1198633

1073 0948 0506 0500 0695 0658 0461 0570 0298 0358 1198633

1074 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

1075 0854 0506 0920 0587 0635 0458 0568 0303 0349 1198634

1200 0347 0500 0475 0535 0534 0464 0527 0298 0326 1198631


(119901 119888) = (1199011 1199012 119901






3) and danger



f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes






20 40 60 80 100 120 140 160 180 2000Testing samples

f = 1f = 2f = 3


0

1

2

3

4

5

6

Hea

lth m

odes


f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes





SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e












Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





Number 1198751

1198752

1198753

1198754

1198755

1198756

1198757

1198758

1198759

Categories1 0226 0515 0485 0559 0539 0455 0568 0298 0338 119863

1

2 0194 0540 0485 0557 0535 0455 0565 0287 0328 1198631

3 0168 0577 0494 0486 0480 0458 0565 0291 0335 1198631

4 0167 0562 0485 0513 0496 0458 0565 0290 0338 1198631

185 0257 0509 0494 0461 0489 0458 0565 0296 0343 1198631

186 0258 0519 0506 0471 0497 0455 0565 0295 0343 1198631

187 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

188 0227 0503 0488 0417 0422 0458 0568 0281 0327 1198631

354 0350 0519 0491 0459 0450 0458 0565 0284 0329 1198631

355 0858 0552 0821 0531 0704 0458 0570 0301 0350 1198633

356 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

357 0408 0522 0497 0455 0445 0458 0568 0280 0327 1198631

587 0000 0485 0497 0444 0432 0458 0570 0309 0329 1198632

588 0773 0556 0990 0659 0653 0458 0568 0304 0348 1198634

589 0850 0503 0907 0630 0619 0458 0568 0301 0348 1198634

590 0854 0540 0861 0587 0635 0458 0568 0303 0349 1198633

753 0943 0506 0509 0816 0783 0461 0570 0311 0359 1198633

754 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

755 0000 0481 0497 0445 0432 0458 0570 0313 0352 1198632

756 0000 0494 0497 0449 0431 0458 0570 0313 0352 1198632

893 0042 0491 0497 0386 0329 0458 0570 0294 0339 1198632

894 0151 0494 0491 0405 0419 0458 0570 0290 0342 1198632

895 0950 0494 0503 0708 0724 0461 0570 0296 0354 1198633

896 0399 0488 0491 0389 0393 0458 0570 0288 0338 1198632

1072 0950 0506 0506 0621 0783 0461 0570 0297 0358 1198633

1073 0948 0506 0500 0695 0658 0461 0570 0298 0358 1198633

1074 0950 0509 0707 0630 0619 0458 0568 0301 0348 1198634

1075 0854 0506 0920 0587 0635 0458 0568 0303 0349 1198634

1200 0347 0500 0475 0535 0534 0464 0527 0298 0326 1198631


(119901 119888) = (1199011 1199012 119901






3) and danger



f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes






20 40 60 80 100 120 140 160 180 2000Testing samples

f = 1f = 2f = 3


0

1

2

3

4

5

6

Hea

lth m

odes


f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes





SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e












Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes






20 40 60 80 100 120 140 160 180 2000Testing samples

f = 1f = 2f = 3


0

1

2

3

4

5

6

Hea

lth m

odes


f = 1f = 2f = 3


20 40 60 80 100 120 140 160 180 2000Testing samples

0

1

2

3

4

5

6

Hea

lth m

odes





SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e












Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




SVMBP-NNAI

400 600 800 1000200Training size

0

005

01

015

02

025

Clas

sifica

tion

erro

r rat

e












Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Ground LAN



running parameters

Crossheading remote

monitoring center









Coal mining face










A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




A B




Competing Interests


Acknowledgments


References












































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



research article a dynamic health assessment approach for...

Documents