DEVELOPMENT OF A NEURO FUZZY MODEL FOR NOISE PREDICTION IN

September 24, 2009 15:57 WSPC/118-IJUFKS 00623

International Journal of Uncertainty,Fuzziness and Knowledge-Based SystemsVol. 17, No. 5 (2009) 729–745c© World Scientific Publishing Company

DEVELOPMENT OF A NEURO FUZZY MODEL FOR NOISEPREDICTION IN OPENCAST MINES

SANTOSH KUMAR NANDA∗,† and DEBI PRASAD TRIPATHY‡

Department of Mining Engineering, National Institute of Technology,Rourkela, Orissa 769008, India

∗santosh [email protected]†[email protected]

‡debi [email protected]

SARAT KUMAR PATRA

Department of Electronics and Communication Engineering,National Institute of Technology, Rourkela, Orissa 769008, India

[email protected]

Received 13 August 2008Revised 10 May 2009

The need of developing appropriate noise prediction models for finding out the accuratestatus of noise levels (>90 dBA) generated from various opencast mining machineriesis overdue. The measured sound pressure levels (SPL) of equipments are not accuratedue to instrumental error, attenuation due to geometrical aberration, atmospheric at-tenuation etc. Some of the popular noise prediction models e.g. VDI and ENM havebeen applied in mining and allied industries. Among these models, VDI2714 is simpleand less complex model. In this paper, a neuro-fuzzy model is proposed to predict themachinery noise in an opencast coal mine. The proposed model is trained with VDI2714and the model output is seen very closely to matching with VDI2714 output. The modelproposed has a mean square error of 2.73%. This model takes CPU time of nearly 0.0625sec where as it takes 0.5 sec for VDI2714 i.e. approximately twelve times faster.

Keywords: Noise prediction models; opencast mines, VDI2714; fuzzy system; neuro-fuzzymodel.

1. Introduction

Noise is generated from all most all the opencast mining operations from differ-ent fixed, mobile and impulsive sources; thereby becoming an integral part of themining environment. With increased mechanization, the problem of noise has gotaccentuated in opencast mines. Prolonged exposure of miners to the high levelsof noise can cause noise induced hearing loss besides several non-auditory healtheffects. The impact of noise in opencast mines depends upon the sound power level

729


730 S. K. Nanda, D. P. Tripathy & S. K. Patra

of the noise generators, prevailing geo-mining conditions and the meteorologicalparameters of the mines. The noise levels need to be studied as an integrated ef-fect of the above parameters. In mining conditions the equipment conditions andenvironment continuously changes as the mining activity progresses. Depending ontheir placement, the overall mining noise emanating from the mines varies in qualityand level. Thus for environmental noise prediction models, the noise level at anyreceiver point needs to be the resultant sound pressure level of all the noise sources.

The need for accurately predicting the level of sound emitted in opencast minesis well established. Some of the noise forecasting models used extensively in Europeare those of the German Draft Standard VDI-2714 Outdoor Sound Propagation andEnvironmental Noise Model (ENM). These models are generally used to predictnoise in petrochemical complexes and mines. The algorithm used in these modelsrely for a greater part on interpolation of experimental data which is a valid anduseful technique, but their applications are limited to sites which are more or lesssimilar to those for which the experimental data were assimilated.

A number of models were developed and were extensively used for the assessmentof sound pressure level and their attenuation around industrial complexes. Generallyin Indian mining industry, ENM (Environmental Noise Model) developed by RTAgroup, Australia is mostly used to predict noise. Pal et al. (1997) applied ENMmodel to predict sound pressure level in mining complexes at Moonidih Project inJharia Coalfield (Dhanbad, Inida).1 The applied model output was represented asnoise contours.

Tonin (1985) studied the applications of different noise prediction models in-cluding VDI2714 for various mines and petrochemical complexes and discussedthat VDI2714 model is the simplest and least complex as compared to other mod-els.2 Rabeiy et al. (2004) used VDI2714 and ISO (1996) noise prediction modelsin Assiut cement plant, Assiut cement quarry and El-Gedida mine at El-Bahariaoasis of Egypt to predict noise. They concluded that the prediction models can beused to identify the safe zones with respect to the noise level in mining and indus-trial plants. They also inferred that the VDI2714 model is the simplest model forprediction of noise in mining complexes and workplace.3 Pathak et al. (2000) devel-oped air attenuation model for noise prediction in limestone quarry and mines ofIreland.5 They developed the model to predict attenuation in air due to absorptionin air considering temperature and relative humidity.

All the noise models treat noise as a function of distance, sound power level, dif-ferent form of attenuations such as geometrical absorptions, barrier effects, groundtopography etc. Generally these parameters are measured in the mines and best fit-ting models are applied to predict noise. Mathematical models are generally com-plex and cannot be implemented in real time systems. Additionally they fail topredict the future parameters from current and past measurements. It has beenseen that noise prediction is a non-stationary process and soft-computing tech-niques like Fuzzy system, Adaptive network based fuzzy inference system (ANFIS)or (Neuro-Fuzzy), Neural Network etc. have been tested for non-stationary time-


Development of a Neuro Fuzzy Model for Noise Prediction in Opencast Mines 731

series prediction nearly for two decades. Fuzzy logic was introduced by Zadeh (1965)as a mathematical way to represent vagueness in linguistics and can be consideredas a generalization of classical set theory.6 This great innovation had supplementedconventional technologies in many scientific applications and engineering applica-tions.There is a scope of using soft-computing techniques viz. Fuzzy system, Ar-tificial neural networks, Radial basis function etc. for noise prediction in miningindustry. The authors had earlier investigated the use of fuzzy model for predictingnoise induced hearing loss in mining industry.4 The results have encouraged themto investigate noise prediction in opencast mine using Neuro-Fuzzy model.

In this paper, an attempt has been made to develop a neuro-fuzzy model fornoise prediction in Balaram opencast mine of Talcher, Orissa, India. The data as-sembled through surveys, measurement or knowledge to predict sound pressure levelin mines is often imprecise or speculative. Since fuzzy system is a good predictedtool for imprecise and uncertainty information, hence fuzzy approach would be themost appropriate technique for modeling the prediction of sound pressure level inopencast mines.

2. VDI-2714 Noise Prediction Model

In 1976, the VDI (Verein Deutscher Ingenieur) draft code 2714 on Outdoor SoundPropagation was issued by the VDI Committee on Noise Reduction.2 The soundpressure level at an environmental point is calculated form the following equation:

LpdB(A) = Σlogall sources[LW +K1 − 10 log(4ΠR2)

+ 3dB −K2 −K3 −K4 −K5 −K6 −K7] (1)

whereLw = source power level re 10−12 wattsK1 = source directivity index

−10 log(4ΠR2) + 3dB = geometric spreading term including infinite hard planecoinciding with the source

R = source to receiver distanceK2 = atmospheric attenuation = 10 log(1 + 0.0015R)dB(A)K3 = attenuation due to meteorological conditions

= [(12.5/R2) + 0.2]−1 dB(A)K4 = ground effects = 10 log[3 + (R/160)]−K2 −K3dB(A)K5 = barrier value (0-10) = 10 log(3 + 20d) dB(A)d = barrier path difference

K6 = attenuation due to woodland areasK7 = attenuation due to built-up areas.

3. Fuzzy Expert System An Introduction

This section provides an introduction to fuzzy systems. Detailed analysis on fuzzysystem can be found in numerous literature [6–9]. Block diagram of a typical fuzzylogic system is presented in Fig. 1. As outlined in Fig. 1, a fuzzy rule based system



Fig. 1. Structure of fuzzy rule based system.

consists of four parts: fuzzifier, knowledge base, inference engine and defuzzifier.These four parts are described below:

• Fuzzifier: The real world input to the fuzzy system is applied to the fuzzzi-fier. In fuzzy literature, this input is called crisp input since it containsprecise information about the specific information about the parameter.The fuzzifier converts this precise quantity to the form of imprecise quan-tity like ‘large’, ‘medium’, ‘high’ etc. with a degree of belongingness to it.Typically, the value ranges between 0 to 1.

• Knowledge base: The main part of the fuzzy system is the knowledge basein which both rule base and database are jointly referred. The databasedefines the membership functions of the fuzzy sets used in the fuzzy ruleswhere as the rule base contains a number of fuzzy if-then rules.

• Inference engine: The inference system or the decision-making unit per-forms the inference operations on the rules. It handles the way in whichthe rules are combined.

• Defuzzifier: The output generated by the inference block is always fuzzyin nature. A real world system will always require the output of the fuzzysystem to the crisp or in the form of real world input. The job of thedefuzzifier is to receive the fuzzy input and provide real world output. Inoperation, it works opposite to the input block.

The above description shows the behavior of a fuzzy expert system. Let X be theuniverse of discourse and x is the elements of X . A fuzzy set A in a universe ofdiscourse X is characterized by a membership function µA(x) which has a valueranging from 0 to 1. If there are n fuzzy sets associated with a given input x,then fuzzifier would produce n fuzzy sets as A1(x), A2(x) . . . An(x) with n numberof membership function µAi(x), i = 1, 2 . . . n. This process is called the fuzzifica-tion. After fuzzification the information goes to knowledge base which comprises adatabase and rule base. Membership functions of the fuzzy sets are contained inthe data base. The rule base is a set of linguistic statements in the appearance of



IF-THEN rules with antecedents and consequents, correspondingly, with AND orOR operators. Fuzzy rule based system with multi-inputs single-output(MISO)canbe represented in the following manner:10

IF X1 is B11 AND X2 is B12 AND .. AND Xr is B1r THEN Y1 is D1

ALSO . . .ALSOIF X1 is Bm1 AND X2 is Bm2 AND . . .AND Xr is Bmr THEN Ys is Ds

Where X1, X2 . . . Xr are the input variables and Y1, Y2 . . . Ys are the output vari-ables, Bij(i = 1, ..m; j = 1, ..r) and Di(i = 1, ..s) are the linguistic labels defined asreference fuzzy sets over the input space X1, X2 . . .Xr and output space Y1, Y2..Ys

of the MISO system. After formation of rule base the last unit block defuzzifierconverts the fuzzy output obtained by inference engine into a non-fuzzy output realnumber domain and this process is called defuzzification. Among several methodsof defuzzification, the Center of Area (COA) is the most widely used method.

In general two most popular fuzzy inference systems are available i.e. Mamdanifuzzy model and Sugeno fuzzy model depending on the fuzzy reasoning and formu-lation of fuzzy IF-THEN rules. Mamdani fuzzy model11 is based on the collectionsof IF-THEN rules with both fuzzy antecedent and consequent predicts. The benefitof this model is that the rule base is generally provided by an expert and henceto a certain degree it is translucent to explanation and study. Because of its ease,Mamdani model is still most commonly used technique for solving many real worldproblems. Sugeno fuzzy model was proposed by Takagi and Sugeno12 ; Sugeno andKang13 in an attempt to build up a methodical approach to generate fuzzy rulesfrom a given input-output data. These models are build up with the help of if-thenrules that have fuzzy antecedent and functional consequent. Typically the conse-quent is a polynomial function of the desired input variables. When the order of thepolynomial function is first order, then the resulting fuzzy expert system is calledthe first order Sugeno or TSK fuzzy model, which was originally, proposed by Tak-agi and Sugeno;12 Sugeno and Kang.13 When the order of the polynomial functionof zero or the consequent part is totally constant, then the system is called zeroorder Sugeno and TSK fuzzy model, in which each rule’s consequent, is specifiedby a fuzzy singleton. To represent this constant output, a single spike is used whichalso known as singleton output membership function. The main advantage of theTSK model is its computational efficiency.

In general fuzzy systems work with a set of heuristic rules. Sometimes if, a setof training data (prototype output for given inputs) is available, fuzzy system canbe updated iteratively by training. These systems are generally termed as ANFIS(adaptive network based fuzzy inference system) or Neuro-Fuzzy systems [15–17].Here the fuzzy parameters consisting of rule base, defuzzification and fuzzy inferencerules are iteratively updated to provide optimal performance.



4. Adaptive Fuzzy Model for Predicting Sound Pressure Level

In this present study, an attempt has been made to use adaptive fuzzy modelto predict sound pressure level by using sound power level and distance as inputparameters. This model can be treated as a MISO (multi inputs and single output)system. Figure 2 shows the general architecture of the adaptive fuzzy system forthis application. Here the fuzzy model is designed so as to predict the output asgiven by VDI2714. The system adapts itself during the training period using a setof training samples. As per Fig. 2, the primary model is VDI2714 and secondarymodel is ANFIS (Adaptive network based fuzzy inference system) model. Soundpower level (LW ) and distance (R) are the two input parameters to the system.The output of the VDI-2714 is Sound pressure level,denoted as (y) where as (y)denotes the predicted output of the system.

The architecture of the fuzzy prediction model is analyzed here and representedin Fig. 3. The methodology for the development of the adaptive fuzzy noise predic-tion model involves the following steps:

(1) Selection of input and output variables;(2) Selection of membership functions;(3) Formation of linguistic rule base;(4) Defuzzification and(5) Training of parameters of the fuzzy model.

4.1. Selection of input and output variables

The first step in system modelling is the identification of input and output variablescalled the system’s variables. Here, sound power level and distance from the sourceare used as the input parameters. These parameters are also inputs to the VDI-2714 model. The output of the system is denoted as (y). This output is expectedto match the output (y) of VDI-2714 model.

4.2. Selection of membership functions for inputs

Linguistic values are expressed in the form of fuzzy sets. A fuzzy set is usuallydefined by its membership functions. In general, triangular membership functionis used to normalize the crisp inputs because of its simplicity, mathematicallydescribed in the following manner:14

triangle(x ; a, b, c) =

0, x ≤ a

x− a

b− a, a ≤ x ≤ b

c− x

c− b, b ≤ x ≤ c

0, c ≤ x

(2)



Fig. 2. Block diagram for adaptive fuzzy system of noise prediction model.

triangle(x; a, b, c) = max(

min(x− a

b− a,c− x

c− b

), 0

), (3)

where a, b, c are the parameters of the linguistic value and x is the range of theinput parameters. This triangular membership function as described in the aboveexpressions (2) and (3) convert the linguistic values in a range of 0 to 1.

In the proposed model, shown in Fig. 3, the two inputs are sound power level LW

and distance from the source (R) respectively. Each input has five membership func-tions e.g. L(1)

W , L(2)W . . . L

(5)W corresponding to (LW ) and R(1), R(2) . . . R(5) for R. The

process of fuzzification of input (LW ) and R is done with triangular membershipfunction. This is shown in Figs. 4 and 5 for input LW and R respectively. The mem-bership functions are represented as “L(1)

W = low (80–100 dB)”, “L(2)W = medium

(90–110 dB)” . . . “L(5)W = very high (120–140 dB)” for the input LW . The member-

ship functions corresponding to this input variables are µL1 , µL2 , . . . µL5 . Similarlythe membership functions for R are represented as “R(1) = medium low (0–10meter)”, “R(2) = low (1–15 meter)” . . . “R(5) = high (20–30 meter)” for the inputR and the membership corresponding to these input variables are µR1 , µR2 , . . . µR5

respectively.



Fig

.3.

Adapti

ve

fuzz

ysy

stem

arc

hit

ectu

refo

rnois

epre

dic

tion.



80 90 100 110 120 130 1400

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1Low Medium Medium High High Very High

Sound power level in dB (Lw)

Mem

ber

ship

gra

de

Fig. 4. Membership function of sound power level.

5 10 15 20 25 300

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1Medium Low Low High

Distance from the source in meter (R)

Mem

ber

ship

gra

de

Medium Medium High

Fig. 5. Membership function of distance.



4.3. Formation of linguistic rule-base

The relationship between input and the output are represented in the from of IF-THEN rules. The membership function L(1)

W , L(2)W . . . L

(5)W and R(1), R(2) . . . R(5) are

the inputs to the rule-base. Let the output, Sound pressure level is expressed as Z.Since there are two inputs and each input has five possible fuzzy sets. The systemcan have at most 52 =25 rules. Here in this proposed model, product inferencewas considered as discussed by Wang and Mendel.17 Each of these rules receivethe membership from each of input variable fuzzy sets. Hence rules are formed infollowing manner;

R1 : IF LW is L(1)W AND R is R(1) THEN sound pressure level (Z) is Z =

(µL1(L(1)W )×µR1(R(1))×w1;


(µL1(L(1)W )×µR2(R(2))×w2;

...


(µL5(L(5)W )×µR5(R

(5))×w25;

Since the system has 25 rules, each rule is associated with a weight. The rulesR1, R2 . . . R25 are associated with weights (w1, w2 . . . w25) respectively. Theseweights are initialized to random values at the beginning. Since this model wasproduct inference the fuzzy system can be considered as Takagi-Sugeno-Kang (TSK)fuzzy system.

4.4. Defuzzification

In the proposed model, Centroid of area (COA) method of defuzzification has beenused for determining the output. Centroid of area (COA) was expressed as (4).14

Centroid of areaz COA =

∫z µA(z)zdz∫z µA(z)dz

(4)

Here µA(z) is the membership value of set A and z is the output variable. In thismodel (Fig. 3), the estimated output is determined as

y =∑25

i=1 wi × ψi∑25i=1 wi

(5)

where ψi= µLk×µRm. where k = 1, ..5,m = 1, ..5 corresponding to number of fuzzysets in Lw and R.

4.5. Training of parameters of the fuzzy model

In the model presented at Fig. 3, weights wi, (i = 1, 2, . . . 25) are unknown. Theseare initialized to random values at the beginning. Subsequently these weights are



updated using LMS algorithm which was first proposed by Widrow and McCool in1976.18 This is similar to the adaption used by Patra and Mulgrew.19 The weightscan be updated iteratively by;

e(k) = y(k) − y(k) (6)

W (k + 1) = W (k) + 2αe(k)ψ(k) (7)

Where k is the time index, W(k+1)refers to the new weights of the system andW(k) is the existing weight. W=[(w1, w2 . . . w25)]T is the weight vector and ψ =[(ψ1, ψ2 . . . ψ25)] is the output of the inference engine.

5. Simulation Result and Discussion

The proposed model is simulated using MATLAB software. The flowchart for theANFIS system is shown in Fig. 6.

The training data set was derived from the VDI-2714 noise prediction model(Eq. (1)). A set of 3200 data set was generated for different values of input pa-rameters LW and R.Using these data in VDI model SPL was determined. Thisconstituted y for training. The fuzzy network was trained with 3000 sets out ofthe total data generated. Remaining 200 data set was used for testing the model.The performance of training was validated using mean square error(MSE)as per-formance index. The efficiency and simplicity of the fuzzy system was validatedusing the CPU time. Figure 7 shows the error update curve during training of thesystem. Figure 8 shows the performance of the system with 200 samples and it wasobserved that the mean square error for prediction is 2.73%. The adaptive fuzzyalgorithm took CPU time of 0.0625 sec approximately as compared to 0.5 sec forthe VDI2714 model.

To test the stability of the model, validation data is essential. The validationdata is collected from Balaram Opencast coal mines, Mahanadi Coalfield Limited(MCL), Talcher (Orissa, India). The test data or the field data was measured usingBruel & Kjaer 2239 (Denmark) precession sound level meter. From the measuredparameter, VDI-2714 gives prediction by calculating all the sound attenuations in‘dB (A)’ not in octave frequency band. In general, numerous machineries are usedin opencast mines for production, so it is difficult to show the noise predictionof all the machineries using the proposed model here. The machineries ex. Shovel(10m3 bucket capacity), Dozer (410HP), Tipper (10T-160HP), Grader (220 HP)and Dumper (85T) were selected to predict the sound pressure level (SPL) by usingVDI2714 and adaptive fuzzy system. Prediction results of the two models (VDI-2714 and Neuro-Fuzzy) for Dozer machine were graphically represented in Fig. 9(a).These models were compared by using measured distance (R) and sound powerlevel (SWL). The prediction results were also represented in Table 1. Similar plotfor other machineries were represented in Figs. 9(b), (c), (d) and (e) respectively.Table 1 shows the measured field data (validation data) and prediction results of



Fig. 6. Flowchart of the neuro-fuzzy noise prediction model.

the two models (VDI-2714 and Neuro Fuzzy) of all the selected machineries. It isseen that the Neuro-Fuzzy noise prediction model closely matches with VDI-2714model results in noise prediction.



Fig. 7. Square error of the neuro-fuzzy model.

0 20 40 60 80 100 120 140 160 180 20060

70

80

90

100

110

120

130

Samples

SP

L in

dB

desired outputestimated output

Fig. 8. Matching with 200 samples.

5.1. Advantages of neuro-fuzzy model

From the present research study, it is found that the Neuro-Fuzzy model takes0.0625 sec CPU time vis-a-vis 0.5 sec CPU time for VDI2714 noise predictionmodel. Some of the advantages of the model are enumerated below:

• The noise generated from industrial machineries is generally represented bymathematical models. One such model is VDI-2714. In this model, the input



5 10 15 20 25 3070

75

80

85

90

95

100

105

110

115

120

Distance from the Source in meter

So

un

d P

ress

ure

Lev

el in

dB

Neuro Fuzzy Noise Prediction of Shovel Noise

FuzzyVDI2714

(a)

5 10 15 20 25 3070

75

80

85

90

95

100

105

110

115

120


So

un

d P

ress

ure

Lev

el in

dB

Neuro Fuzzy Noise Prediction of Dumper Noise

FuzzyVDI2714

(b)

5 10 15 20 25 3070

75

80

85

90

95

100

105

110

115

120


So

un

d P

ress

ure

Lev

el in

dB

Neuro Fuzzy Noise Prediction of Grader Noise

FuzzyVDI2714

(c)

5 10 15 20 25 3070

75

80

85

90

95

100

105

110

115

120


So

un

d P

ress

ure

Lev

el in

dB

Neuro Fuzzy Noise Prediction of Heavy Truck(Tipper) Noise

FuzzyVDI2714

(d)

5 10 15 20 25 3070

75

80

85

90

95

100

105

110

115

120

Distance from the Source in (meter)

So

un

d P

ress

ure

Lev

el in

(d

B)

Neuro Fuzzy Noise Prediction of Bull Dozer Noise

FuzzyVDI2714

(e)

Fig. 9. Neuro-fuzzy noise prediction for machineries in the study area.


Development of a Neuro Fuzzy Model for Noise Prediction in Opencast Mines 743Table

1.

Neu

ro-fuzz

ynois

epre

dic

tion

for

diff

eren

tm

ach

iner

ies.

Shovelnois

ein

dB

Dum

per

nois

ein

dB

Gra

der

nois

ein

dB

Tip

per

nois

ein

dB

Dozer

nois

ein

dB

Dis

tance

from

the

sourc

ein

mete

r

Measu

red

field

data

(dB

A)

Pre

dic

tion

resu

lt(d

BA

)

Measu

red

field

data

(dB

A)

Pre

dic

tion

resu

lt(d

BA

)

Measu

red

field

data

(dB

A)

Pre

dic

tion

resu

lt(d

BA

)

Measu

red

field

data

(dB

A)

Pre

dic

tion

resu

lt(d

BA

)

Measu

red

field

data

(dB

A)

Pre

dic

tion

resu

lt(d

BA

)

N-fuzzy

VD

IN

-fuzzy

VD

IN

-fuzzy

VD

IN

-fuzzy

VD

IN

-fuzzy

VD

I

1102.3

000

90.2

051

95.6

919

102.4

000

90.2

853

95.7

919

105.3

000

92.6

115

98.6

919

100.9

000

89.0

821

94.2

919

100.5

000

88.7

612

93.8

919

2102.1

000

92.9

455

95.4

828

101.3

000

92.2

722

94.6

828

103.4

000

94.0

081

96.7

828

99.7

000

90.9

254

93.0

828

100.2

000

91.3

463

93.5

828

398.6

000

91.1

476

91.9

738

98.2

000

90.7

961

91.5

738

101.2

000

93.2

710

94.5

738

98.6

000

91.1

476

91.9

738

98.2

000

90.7

961

91.5

738

498.2

000

91.1

557

91.5

648

97.7

000

90.7

761

91.0

648

98.7

000

91.5

560

92.0

648

97.5

000

90.6

297

90.8

648

97.5

000

90.6

297

90.8

648

597.5

000

90.5

989

90.8

559

97.2

000

90.3

440

90.5

559

97.2

000

90.3

440

90.5

559

96.5

000

89.7

812

89.8

559

96.7

000

89.9

376

90.0

559

697.5

000

90.5

077

90.8

469

96.8

000

89.8

325

90.1

469

95.5

000

88.7

188

88.8

469

96.2

000

89.2

976

89.5

469

95.4

000

88.6

398

88.7

469

796.7

000

89.4

805

90.0

380

94.2

000

87.2

589

87.5

380

94.3

000

87.3

358

87.6

380

95.8

000

88.6

022

89.1

380

94.8

000

87.7

336

88.1

380

895.2

000

87.7

212

88.5

291

94.1

000

86.7

376

87.4

291

94.1

000

86.7

376

87.4

291

94.8

000

87.3

480

88.1

291

94.2

000

86.8

217

87.5

291

993.3

000

85.4

354

86.6

202

93.6

000

85.6

965

86.9

202

93.7

000

85.7

857

87.0

202

94.3

000

86.3

448

87.6

202

93.6

000

85.6

965

86.9

202

10

92.4

000

83.6

372

85.7

113

93.2

000

84.3

542

86.5

113

93.2

000

84.3

542

86.5

113

93.7

000

84.8

404

87.0

113

92.5

000

83.7

230

85.8

113

11

91.5

000

83.2

331

84.8

025

93.2

000

84.8

830

86.5

025

92.6

000

84.2

596

85.9

025

92.8

000

84.4

620

86.1

025

91.8

000

83.4

994

85.1

025

12

91.5

000

83.6

705

84.7

937

92.5

000

84.7

308

85.7

937

91.8

000

83.9

743

85.0

937

90.6

000

82.8

242

83.8

937

89.6

000

81.9

843

82.8

937

13

91.3

000

83.9

048

84.5

848

92.2

000

84.9

497

85.4

848

90.4

000

82.9

754

83.6

848

89.5

000

82.2

313

82.8

848

89.3

000

81.9

700

82.5

848

14

90.4

000

83.2

946

83.6

760

90.6

000

83.5

142

83.8

760

88.6

000

81.5

435

81.8

760

88.4

000

81.3

709

81.6

760

88.8

000

81.7

200

82.0

760

15

88.8

000

81.9

558

82.0

672

89.7

000

82.8

950

82.9

672

88.5

000

81.6

649

81.7

672

86.8

000

80.1

925

80.0

672

88.2

000

81.3

840

81.4

672

16

88.4

000

81.7

271

81.6

585

88.3

000

81.6

275

81.5

585

88.2

000

81.5

292

81.4

585

86.2

000

79.7

832

79.4

585

87.9

000

81.2

411

81.1

585

17

87.9

000

81.2

228

81.1

497

88.2

000

81.5

252

81.4

497

87.9

000

81.2

228

81.1

497

85.8

000

79.3

791

79.0

497

87.4

000

80.7

425

80.6

497

18

87.1

000

80.3

335

80.3

410

87.6

000

80.8

185

80.8

410

87.3

000

80.5

240

80.5

410

85.2

000

78.7

269

78.4

410

86.6

000

79.8

765

79.8

410

19

86.7

000

79.7

007

79.9

323

87.1

000

80.0

842

80.3

323

86.5

000

79.5

157

79.7

323

85.2

000

78.4

125

78.4

323

85.5

000

78.6

528

78.7

323

20

86.3

000

78.7

986

79.5

236

86.8

000

79.2

704

80.0

236

85.8

000

78.3

542

79.0

236

84.7

000

77.4

620

77.9

236

85.5

000

78.0

998

78.7

236

21

85.7

000

78.1

231

78.9

149

86.5

000

78.8

310

79.7

149

85.4

000

77.8

747

78.6

149

84.5

000

77.1

786

77.7

149

84.8

000

77.4

028

78.0

149

22

85.2

000

77.7

547

78.4

063

86.2

000

78.6

088

79.4

063

85.1

000

77.6

748

78.3

063

83.8

000

76.7

150

77.0

063

84.3

000

77.0

677

77.5

063

23

85.3

000

77.9

729

78.4

976

85.8

000

78.4

024

78.9

976

84.6

000

77.4

136

77.7

976

83.5

000

76.6

210

76.6

976

84.2

000

77.1

139

77.3

976

24

85.5

000

78.0

839

78.6

890

85.6

000

78.1

699

78.7

890

84.2

000

77.0

572

77.3

890

83.5

000

76.5

662

76.6

890

83.8

000

76.7

719

76.9

890

25

85.5

000

77.9

792

78.6

804

84.8

000

77.4

094

77.9

804

83.8

000

76.6

758

76.9

804

83.2

000

76.2

752

76.3

804

83.5

000

76.4

721

76.6

804

26

85.3

000

77.5

712

78.4

718

84.2

000

76.7

400

77.3

718

83.2

000

76.0

750

76.3

718

82.8

000

75.8

299

75.9

718

83.5

000

76.2

664

76.6

718

27

84.7

000

76.8

183

77.8

632

84.2

000

76.4

671

77.3

632

82.9

000

75.6

479

76.0

632

82.6

000

75.4

758

75.7

632

82.8

000

75.5

899

75.9

632

28

84.2

000

76.1

314

77.3

547

83.7

000

75.8

165

76.8

547

82.5

000

75.1

337

75.6

547

82.2

000

74.9

773

75.3

547

82.5

000

75.1

337

75.6

547

29

83.8

000

75.4

846

76.9

461

83.4

000

75.2

580

76.5

461

82.1

000

74.5

923

75.2

461

82.2

000

74.6

400

75.3

461

82.4

000

74.7

371

75.5

461

30

82.7

000

74.4

604

75.8

376

82.8

000

74.5

073

75.9

376

81.8

000

74.0

617

74.9

376

82.2

000

74.2

338

75.3

376

82.4

000

74.3

228

75.5

376

Mean

square

err

or

0.9

738

0.9

512

0.9

167

0.8

236

0.8

597



parameters are distance (R) and sound power level (SWL (LW )); the modelpredicts the sound pressure level (SPL (LP )). Sound pressure level (SPL) wasdetermined for each set of measured distance (R) and Sound power level (SWL).The process has to be repeated for each machinery. The calculation was com-plex. The fuzzy models need to be trained once for any specific machinery. Oncethe fuzzy model was trained, SPL can be determined for any input condition.The model predicts SPL with very little CPU time (∼ 12%) compared to VDI-2714. In general, the network can be trained to work for any standard model.The prediction would correspond to the model for which it was trained. In thisspecific case the authors have considered only VDI-2714.

• The model of Neuro-Fuzzy remains fixed as long as input and output remain thesame. This can be implemented as a fixed hardware. The training data is basedon actual measurement. This information can be used to train the network.Hence same network with different training sets can provide approximate resultfor different mines/working conditions.

• Neuro-Fuzzy model can be built using DSP hardware available in market. Thiscan be used in instrument for measurement owing to low CPU time. HigherCPU time of VDI2714 model will prohibit its usage in portable instrument.

• The Neuro-Fuzzy model can be used to predict noise of machineries for othermodels also. This can be done by using a different training data sheet. In aninstrument, this can be implemented easily, where the instrument can provideprediction for different models.

6. Conclusion

The neuro-fuzzy noise prediction model output for five different machineries usedin opencast mines such as dozer, shovel, tipper or dumper matches closely withVDI2714 prediction result. This model takes CPU time of nearly 0.0625 sec whereas it takes 0.5 sec for VDI2714 i.e. it is approximately twelve times faster. Thiscomparison of CPU time shows the computational stability of the model. It ishoped that the developed model will be quite useful in prediction of machinerynoise in opencast mines.

Acknowledgement

The authors like to thank the anonymous referees for their valuable comments andwords of encouragement. Their comments helped to improve the clarity of thispaper.

References

1. A. K. Pal, Y. Mishra, S. Mitra and N. C. Saxena, Noise model for mining complex —a case discussion, in Second World mining environment congress, Katowice, Poland,1997, pp. 1–10.



2. R. Tonin, Estimating noise level from petrochemical plants, mines and industrialcomplexes, Acoustics Australia 13(2) (1985) 59–67.

3. R. E. Rabeiy, M. Mohamed, A-K. and W. Gomma, Study on the noise prediction inmining and industrial plants, Ass. Uni. Bull. Environ. Res 7(2)(2004) 77–89.

4. S. K. Nanda, D. P. Tripathy and S. K. Patra, A sugeno fuzzy model for noise inducedhearing loss in the mining industry, Noise and Vibration Worldwide 39(10) (2008)25–36.

5. K. Pathak, S. Durucan and A. Korre, Development of an air attenuation model fornoise prediction in surface mines and quarries, Int. J. Surface Mining Reclamationand Environment 14(3)(2000) 223–234.

6. L. A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338–353.7. L. A. Zadeh, A fuzzy-algorithm approach to the definition of complex or imprecise

concept, Int. J. Man-Machine Studies 8 (1976) 249–291.8. J. M. Mendel, Fuzzy logic systems for engineering: a tutorial, IEEE Proc. 83 (1995)

345–377.9. E. A. Cox, Fuzzy fundamentals, IEEE Spectrum 29(10) (1992) 58–61.

10. R. R.Yager and D. P. Filev, Essential of Fuzzy Modelling and Control (John Willey& Sons, Inc., New York, 1994).

11. E. H. Mamdani and S. Assilian, An experiment in linguistic synthesis with a fuzzylogic controller, Int. J. Man-Machine Studies 7 (1975) 1–13.

12. T. Takagi and M. Sugeno, Fuzzy identification of systems and its application to mod-elling and control, IEEE Trans. System, Man, Cybernetics 15 (1985) 116–132.

13. M. Sugeno and G. T. Kang, Structure identification of fuzzy model, Fuzzy Sets andSystem 28 (1988) 15–33.

14. J.-S. R. Jang, C.-T. Sun and E. Mizutan, Neuro-Fuzzy and Soft Computing, (PrenticeHall of India Private Limited, New Delhi, 2005).

15. J.-S. R. Jang, ANFIS: Adaptive-network-based fuzzy inference system, IEEE Trans.System, Man, Cybernetics 23 (1993) 665–685.

16. L.-X. Wang and C. Wei, Approximation accuracy of some neuro-fuzzy Approaches,IEEE Trans. Fuzzy Systems 8 (2000) 470–478.

17. L.-X. Wang and J. M. Mendel, Fuzzy adaptive filetrs, with application to nonlinearchannel equalization, IEEE Transaction on Fuzzy Systems 1 (1993) 161–170.

18. S. Hykin, Adaptive Filter Theory (Pearson Education, Inc., New Delhi, 2002).19. S. K. Patra and B. Mulgrew, Efficient architecture for bayesian equalization

using fuzzy filters, IEEE Trans. Circuits and Systems-II: Analog and Digital SignalProcessing 45 (1998) 812–820.

Documents

DEVELOPMENT OF A NEURO FUZZY MODEL FOR NOISE PREDICTION IN