The Welding Institute - TWI 4th International Conference ...J. R. McCardle 1997 The Welding Institute - TWI 4th International Conference "Computer Technology in Welding Cambridge,

J. R. McCardle 1997

The Welding Institute - TWI

4th International Conference

"Computer Technology in Welding

Cambridge, UK, June 3 - 4, 1992

Paper 35

J. R. McCardle 1997

THE USE OF NEURAL NETWORKS TO CHARACTERIZE PROBLEMATICARC SOUNDS

K.L. Taylor Burge B.Sc, MA

Brunel UniversityT.J. Harris B.Sc, GradWeldI

Brunel University

R.R. Stroud MTech, PhD

Brunel UniversityJ.R. McCardle B.Sc

Brunel University

ABSTRACTAutomation of electric arc welding has been at the centre of considerable debate and thesubject of much research for several decades. One conclusion drawn from all this effort isthat there seems to be no single system that can monitor all of the variables and subse-quently, fully control any welding process. To date there has been considerable successin the development of seam tracking systems employing various sensing techniques,good progress has been made in the area of penetration measurement and worthwhileuse has been made of the integration of expert systems and modelling software withinthese control domains.

Skilled welders develop their own monitoring and control systems and it has been ob-served that part of this expertise is the ability to listen subconsciously to the sound of thearc and to alter the electrode position in response to an adverse change in arc noise.

Attempts have been made to analyse these sounds using both conventional techniquesand more recently expert systems, neither have delivered any usable information. Thispaper describes a new approach involving the use of neural networks in the identificationof sounds which indicate that the welding system is drifting out of control.

INTRODUCTION

Artificial Neural Networks (ANNs) offer potential as an alternative to standard computertechniques in control technology and have attracted a widening interest in their develop-ment and application. Although the conception of ANN theory predates that of themodern digital computer, the commercial success of Von Neumann systems and theutilization of Boolean logic has over-shadowed their development.

Due to the proliferation of the digital computer the use of ANNs in 'real' applicationstend to be in the form of software neural simulators. The commercial availability ofdedicated neural hardware is very limited. Consequently, the techniques employed for theproposed research would involve the application of commercial simulators as well ascustom compiled software.

Research within the Department of Design at Brunel University has aimed at the devel-opment of intelligent control systems incorporating ANNs. One focus of attention hasbeen industrial welding processes. To date ANNs have been successfully implemented toprocess ultrasonic scans of a submerged arc welding process, Fig. 1, to achieve real timekinetic control of the welding head as well as monitoring weld penetration.(1,2,3)

1

J. R. McCardle 1997

It has been observed that skilled manual arc welders subconsciously change the angle ofelectrode and arc length in response to a variation in the sound produced from theprocess. Much evidence exists for the successful application of ANNs in systems whichimitate human attributes including the development of the "Silicon Ear" and "Retina" byCarver Mead (12), Kohonen's "Phonetic Typewriter" (13) and hand written characterrecognition (14,15). The goal of this work is to propose a feasible ANN oriented systemto automatically control an arc welding process and thereby imitate the biological exper-tise of a human operator.

THE WELD VARIABLES

The creation of an ideal weld is dependent upon the optimum settings for :-

i) weld plate preparation (fit up)ii) welding voltage,iii) welding current (determined by the rate of feed of the sacrificial electrode)iv) position of the electrode

andv) speed of travel of the welding head along the seam.

A sixth parameter which can drastically affect the quality of the weld is the feed ofgranular flux. This is generally gravity fed and hence not directly controllable. However asystem which detects the onset of a blockage as a diagnostic feature is desirable.

Following the development of a positional control system for submerged arc welding bymeans of ultrasonics (1,2,3), the preliminary aim of this research is the control of para-meters ii), iii) and v).

In an industrial scenario the submerged arc welder is used to join plates of between 6mmand 40mm. For initial research purposes under laboratory conditions plates of 25mmwould be used. Optimum settings for each parameter would be preset by a controllingPC to obtain ideal weld penetration.

To compliment the existing ultrasonic weld penetration monitoring system (dedicated tothe closed loop control of the welding current), it is considered that the prime directiveof this research would be the control of the welding voltage to maintain an optimumbalance and hence weld stability. However the nature of a raw acoustic emission is suchthat data acquired will contain information concerning other parameters which can beutilised within the diagnostic arena.

PROPOSED METHODOLOGY

Acoustic emissions, within the audible range, have been successfully employed to extractusable diagnostic information from systems sucliis pulsed laser welders (4) and ICengines (5). The acquisition of acoustic data creates little problem providing the trans-ducer and associated amplifier is tailored to suit the application.

2

J. R. McCardle 1997

It is proposed that for this work an omni directional Electret Condenser Microphone(ECM) and pre-amplifier, are adequately noise shielded and resiliently mounted in thevicinity of the welding head. The ECM exhibits a large bandwidth with a uniform fre-quency response within the audible range of 15Hz to 20kHz as well as rejecting anyinduced noise from the 'Pinch Effect'(11) of the arc.

The signals could then be passed through a bandpass filter set to isolate the audible rangefrom infra and ultra sound. An interim 14F1 analysis at this point would emphasise usablefrequency ranges in response to intentionally unstable inputs to the welding system.Future development in this area will involve the hardware filtering and isolation of usablefrequencies within the spectrum.

The signal could then be digitised at a frequency of 1MHz and a resolution of 8 bits anddownloaded to a PC for analysis by the software simulated ANN. The outcome of theanalysis will then interface the controlling software of the welding parameters.

The equipment proposed for the experimentation, Fig.4, comprises

i) SAF Devimatic submerged arc welding setii) IBM PS/2 70

iii) 1-T1' analyzeriv) Custom built transducer arrangement, filter sets and interfacing system.

Computer programs are to be compiled in 'C' and Prospero Pascal.

Neural Network Paradigms

Neural Networks are modelled on the architecture of the biological brain. The network isconstructed of discrete neurons each providing an output to a given stimulus dictated bya mathematical function, Fig.5. The collective response of a network is dependent uponthe topology of synaptic connections. The topology is subject to debate as various archi-tectures exhibit differing characteristics of cognition and recognition. A simple exampleis shown in Fig.6. This illustrates a multi-layer fully connected system. The input layerwhere data is presented to the network, an output layer to communicate a response and a"hidden layer" which increases the processing ability of the system.

ANNs have been proven successful in high speed signal processing, especially in noisy orerratic systems where conventional signal processing failed.(1,2,3,5). Advantages ofnetworks are further enhanced by their apparent ability to "generalise", which is torespond correctly to novel, erroneous or even incomplete data; an inability of expertsystems.

The trend in ANN applications has steered towards the development of hybrid systems(6,7) where the characteristics of certain network topologies are used in conjunctionwith others or in parallel with expert systems. Complex problems require the combina-tion of knowledge based and neural computing techniques to reach an optimum solution.

3

J. R. McCardle 1997

For the application to this work it was envisaged that two neural network architecturescould be employed:-

i) Kohonen self organizing feature maps (8)ii) Weightless or Logical Neural Networks (9)

The Kohonen paradigm possesses the inherent characteristic to self organize its topologyby a method of competitive, "unsupervised learning". That is, the final pattern of synapticconnections is optimized by a self iterative process known as a "learning algorithm". Theresult is a system that can identify the organizational relationships between input patternsand map the similarities into "closeness" groups. The advantage over standard patternrecognition is its speed, because of the parallel nature of the neural process and theprovision of a graphical representation of pattern relationships. This, therefore, canprovide a complex pattern recognition system.(10)

The Logical Neural Network was a system devised jointly by Brunel University andImperial College, London. It utilizes the contents of addresses within a random accessmemory (RAM) in its learning algorithm rather than the more usual iterative update ofsynaptic weights adopted by most paradigms. Hence the term a "weightless" network. Ifa summing device is used to process the responses from a number of RAMs a degree ofsimilarity or "discrimination" can be established between novel input patterns and previ-ously stored "learned" patterns. With manipulation of the neural threshold functionswithin the "discriminator" similar input patterns will produce similar responses andtherefore exhibit properties of generalization. This technique was used in the WISARDsystem. Operating on real time video data it is able to distinguish between facial imagesand is consequently being applied to facial recognition purposes in security systems(8).

With the application of a Kohonen paradigm as a complex pattern "classifier" or featureextractor and a Logical Neural Network as a pattern "discriminator" an optimum solu-tion for the real time processing of compound acoustic signals is proposed.

An ANN can be developed using either commercial software tools or custom compiledprograms. Various topologies have proven successful in specific applications howeverthe architecture has to be tailored to suit the application if an optimum solution is to bereached. Too many hidden layers or artificial neurons (sometimes known as Nodes) willlead to lengthy training periods and ultimately slower operating speeds.

CONCLUSIONS

It has been previously proven that ANNs have the ability to interpret chaotic ultrasonicsignals in real time and provide on line weld control.(1,2,3)

With the application of hybrid systems it is feasible that real time acoustic data, bothaudible and inaudible, may be used to predict the onset of weld instability.

Neural Networks provide a key to a higher degree of process automation.

Research and development in the area of weld process and kinetic control continues with

4

J. R. McCardle 1997

the aim of providing a fully integrated and hybrid automated welding system.

REFERENCES

(1) Stroud R.R., Harris T.J., Taylor Burge K.L., 'Neural Networks in Automated WeldControl', TWI, Gateshead, 1991. Paper 22.

(2) Harris T.J., Stroud R.R. & Taylor Burge K.L., 'Neural Networks in a Weld ControlSystem', AINN'90, Zurich, June 1990.

(3) Stroud R.R., Harris T.J.& Taylor Burge K.L., 'Applications of Neural Networks inControl Technology', Proc 1st ICANN, Helsinki, Finland, June 1991. Vol 2, 1703.

(4) Meyendorf N. et al, 'Monitoring of Pulsed Laser Welding by Acoustic Emissions',COMADEM '91, Southampton, July 1991. P515.

(5) Vu V.V, et al, 'Time Encoded Matrices as Input Data to Artificial Neural Networksfor Condition Monitoring Applications', COMADEM '91, Southampton, July 1991.P31.

(6) Fogelman Soulie, F., 'Neural Network Architectures and Algorithms: A Perspective',ICANN-91, Espoo, Finland, June 1991. Voll, P605.

(7) Picton P.D., Johnson J.H., Hallam N.J. 'Neural Networks in Safety-Critical Systems',COMADEM '91, Southampton, July 1991. P17.

(8) Dayhoff J, 'Neural Network Architectures An Introduction', Van Nostrand Reinhold,1990.

(9) Aleksander I., 'Connectionism Or Weightless Neurocomputing?', ICANN-91, Espoo,Finland, June 1991. Vol 2, P991.

(10) Yoh-Han Pao., 'Adaptive Pattern Recognition And Neural Networks', Addison-Wesley, 1989. P182.

(11) International Institute Of Welding, 'The Physics Of Welding', Pergamon Press,1984. Ch3

(12) Mead C., 'Analog VLSI And Neural Systems', Addison-Wesley, Reading MA,1989.

(13) Torkkola K., et al, 'Status Report OF The Finnish Phonetic Typewriter Project',ICANN-91, Espoo, Finland, June 1991. Vol 1, P771.

(14) Nellis J., Stonham T.J, 'A Fully Integrated Hand-Printed Characture RecognitionSystem Using Artificial Neural Networks', IEE Second International Conference OnArtificial Neural Networks, Conference Publication No. 349, Bournmouth, November1991. P219.

5

J. R. MeCardlc 1997

(15) Fukushima K., Imagawa T., 'Recognition And Segmentation Of Connected Charac-

ters In Cursive Handwriting With Selective Attention', ICANN-91, Espoo, Finland,

June1991. Vol 1, P 105.

FIGURE 1.SUBMERGED ARC WELDER

J. A1;6 -2. 16

FIGURE 2.ACOUSTIC EMISSION OF

AN OPTIMUM WELD

i -2-4%6 ki5- 2: 6 I C144

.1 "At 6/cA4

FIGURE 3.ERRONEOUS EMISSION DUE TO

A FLUX BLOCKAGE

6

FFT ANALYZER

NEURAL PARADIGMSt-Th ADC

1

1FILTER SET

t .n

CONTROLLING SOFTWARE

TRANSFER FUNCTIONINPUT NEURON

SYNAPTIC WEIGHTSck

OUTPUT

\

ARTIFICIAL NEURONSOR NODES

OUTPUT LAYER

HIDDEN LAYER

J. R. McCardle 1997

SUBMERGED ARC WELDER/ 'n

I

PRE AMPLIFIER

0---ECM

IBM PSI2 70

FIGURE 4.EXPERIMENTAL SETUP

FIGURE 5.ELEMENTS OF NEURAL CONNECTION

INPUT LAYER

FIGURE 6.SIMPLE MULTI-LAYER ARCHITECTURE

7

DX195937_1_0001.tifDX195937_1_0003.tifDX195937_1_0005.tifDX195937_1_0007.tifDX195937_1_0009.tifDX195937_1_0011.tifDX195937_1_0013.tifDX195937_1_0015.tifDX195937_1_0017.tifDX195937_1_0019.tifDX195937_1_0021.tifDX195937_1_0023.tifDX195937_1_0025.tifDX195937_1_0027.tifDX195937_1_0029.tifDX195937_1_0031.tifDX195937_1_0033.tifDX195937_1_0035.tifDX195937_1_0037.tifDX195937_1_0039.tifDX195937_1_0041.tifDX195937_1_0043.tifDX195937_1_0045.tifDX195937_1_0047.tifDX195937_1_0049.tifDX195937_1_0051.tifDX195937_1_0053.tifDX195937_1_0055.tifDX195937_1_0057.tifDX195937_1_0059.tifDX195937_1_0061.tifDX195937_1_0063.tifDX195937_1_0065.tifDX195937_1_0067.tifDX195937_1_0069.tifDX195937_1_0071.tifDX195937_1_0073.tifDX195937_1_0075.tifDX195937_1_0077.tifDX195937_1_0079.tifDX195937_1_0081.tifDX195937_1_0083.tifDX195937_1_0085.tifDX195937_1_0087.tifDX195937_1_0089.tifDX195937_1_0091.tifDX195937_1_0093.tifDX195937_1_0095.tifDX195937_1_0097.tifDX195937_1_0099.tifDX195937_1_0101.tifDX195937_1_0103.tifDX195937_1_0105.tifDX195937_1_0107.tifDX195937_1_0109.tifDX195937_1_0111.tifDX195937_1_0113.tifDX195937_1_0115.tifDX195937_1_0117.tifDX195937_1_0119.tifDX195937_1_0121.tifDX195937_1_0123.tifDX195937_1_0125.tifDX195937_1_0127.tifDX195937_1_0129.tifDX195937_1_0131.tifDX195937_1_0133.tifDX195937_1_0135.tifDX195937_1_0137.tifDX195937_1_0139.tifDX195937_1_0141.tifDX195937_1_0143.tifDX195937_1_0145.tifDX195937_1_0147.tifDX195937_1_0149.tifDX195937_1_0151.tifDX195937_1_0153.tifDX195937_1_0155.tifDX195937_1_0157.tifDX195937_1_0159.tifDX195937_1_0161.tifDX195937_1_0163.tifDX195937_1_0165.tifDX195937_1_0167.tifDX195937_1_0169.tifDX195937_1_0171.tifDX195937_1_0173.tifDX195937_1_0175.tifDX195937_1_0177.tifDX195937_1_0179.tifDX195937_1_0181.tifDX195937_1_0183.tifDX195937_1_0185.tifDX195937_1_0187.tifDX195937_1_0189.tifDX195937_1_0191.tifDX195937_1_0193.tifDX195937_1_0195.tifDX195937_1_0197.tifDX195937_1_0199.tifDX195937_1_0201.tifDX195937_1_0203.tifDX195937_1_0205.tifDX195937_1_0207.tifDX195937_1_0209.tifDX195937_1_0211.tifDX195937_1_0213.tifDX195937_1_0215.tifDX195937_1_0217.tifDX195937_1_0219.tifDX195937_1_0221.tifDX195937_1_0223.tifDX195937_1_0225.tifDX195937_1_0227.tifDX195937_1_0229.tifDX195937_1_0231.tifDX195937_1_0233.tifDX195937_1_0235.tifDX195937_1_0237.tifDX195937_1_0239.tifDX195937_1_0241.tifDX195937_1_0243.tifDX195937_1_0245.tifDX195937_1_0247.tifDX195937_1_0249.tifDX195937_1_0251.tifDX195937_1_0253.tifDX195937_1_0255.tifDX195937_1_0257.tifDX195937_1_0259.tifDX195937_1_0261.tifDX195937_1_0263.tifDX195937_1_0265.tifDX195937_1_0267.tifDX195937_1_0269.tifDX195937_1_0271.tifDX195937_1_0273.tifDX195937_1_0275.tifDX195937_1_0277.tifDX195937_1_0279.tifDX195937_1_0281.tifDX195937_1_0283.tifDX195937_1_0285.tifDX195937_1_0287.tifDX195937_1_0289.tifDX195937_1_0291.tifDX195937_1_0293.tifDX195937_1_0295.tifDX195937_1_0297.tifDX195937_1_0299.tifDX195937_1_0301.tifDX195937_1_0303.tifDX195937_1_0305.tifDX195937_1_0307.tifDX195937_1_0309.tifDX195937_1_0311.tifDX195937_1_0313.tifDX195937_1_0315.tifDX195937_1_0317.tifDX195937_1_0319.tifDX195937_1_0321.tifDX195937_1_0323.tifDX195937_1_0325.tifDX195937_1_0327.tifDX195937_1_0329.tifDX195937_1_0331.tifDX195937_1_0333.tifDX195937_1_0335.tifDX195937_1_0337.tifDX195937_1_0339.tifDX195937_1_0341.tifDX195937_1_0343.tifDX195937_1_0345.tifDX195937_1_0347.tifDX195937_1_0349.tifDX195937_1_0351.tifDX195937_1_0353.tifDX195937_1_0355.tifDX195937_1_0357.tifDX195937_1_0359.tifDX195937_1_0361.tifDX195937_1_0363.tifDX195937_1_0365.tifDX195937_1_0367.tifDX195937_1_0369.tifDX195937_1_0371.tifDX195937_1_0373.tifDX195937_1_0375.tifDX195937_1_0377.tifDX195937_1_0379.tifDX195937_1_0381.tifDX195937_1_0383.tifDX195937_1_0385.tifDX195937_1_0387.tifDX195937_1_0389.tifDX195937_1_0391.tifDX195937_1_0393.tifDX195937_1_0395.tifDX195937_1_0397.tifDX195937_1_0399.tifDX195937_1_0401.tifDX195937_1_0403.tifDX195937_1_0405.tifDX195937_1_0407.tifDX195937_1_0409.tifDX195937_1_0411.tifDX195937_1_0413.tifDX195937_1_0415.tifDX195937_1_0417.tifDX195937_1_0419.tifDX195937_1_0421.tifDX195937_1_0423.tifDX195937_1_0425.tifDX195937_1_0427.tifDX195937_1_0429.tifDX195937_1_0431.tifDX195937_1_0433.tifDX195937_1_0435.tifDX195937_1_0437.tifDX195937_1_0439.tifDX195937_1_0441.tifDX195937_1_0443.tifDX195937_1_0445.tifDX195937_1_0447.tifDX195937_1_0449.tifDX195937_1_0451.tifDX195937_1_0453.tifDX195937_1_0455.tifDX195937_1_0457.tifDX195937_1_0459.tifDX195937_1_0461.tifDX195937_1_0463.tifDX195937_1_0465.tifDX195937_1_0467.tifDX195937_1_0469.tifDX195937_1_0471.tifDX195937_1_0473.tifDX195937_1_0475.tifDX195937_1_0477.tifDX195937_1_0479.tifDX195937_1_0481.tifDX195937_1_0483.tifDX195937_1_0485.tifDX195937_1_0487.tifDX195937_1_0489.tifDX195937_1_0491.tifDX195937_1_0493.tifDX195937_1_0495.tifDX195937_1_0497.tifDX195937_1_0499.tifDX195937_1_0501.tifDX195937_1_0503.tifDX195937_1_0505.tifDX195937_1_0507.tifDX195937_1_0509.tifDX195937_1_0511.tifDX195937_1_0513.tifDX195937_1_0515.tifDX195937_1_0517.tifDX195937_1_0519.tifDX195937_1_0521.tifDX195937_1_0523.tifDX195937_1_0525.tifDX195937_1_0527.tifDX195937_1_0529.tifDX195937_1_0531.tifDX195937_1_0533.tifDX195937_1_0535.tifDX195937_1_0537.tifDX195937_1_0539.tifDX195937_1_0541.tifDX195937_1_0543.tifDX195937_1_0545.tifDX195937_1_0547.tifDX195937_1_0549.tifDX195937_1_0551.tifDX195937_1_0553.tifDX195937_1_0555.tifDX195937_1_0557.tifDX195937_1_0559.tifDX195937_1_0561.tifDX195937_1_0563.tifDX195937_1_0565.tifDX195937_1_0567.tifDX195937_1_0569.tifDX195937_1_0571.tifDX195937_1_0573.tifDX195937_1_0575.tifDX195937_1_0577.tifDX195937_1_0579.tifDX195937_1_0581.tifDX195937_1_0583.tifDX195937_1_0585.tifDX195937_1_0587.tifDX195937_1_0589.tifDX195937_1_0591.tifDX195937_1_0593.tifDX195937_1_0595.tifDX195937_1_0597.tifDX195937_1_0599.tifDX195937_1_0601.tifDX195937_1_0603.tifDX195937_1_0605.tifDX195937_1_0607.tifDX195937_1_0609.tifDX195937_1_0611.tifDX195937_1_0613.tifDX195937_1_0615.tifDX195937_1_0617.tifDX195937_1_0619.tifDX195937_1_0621.tifDX195937_1_0623.tifDX195937_1_0625.tifDX195937_1_0627.tifDX195937_1_0629.tifDX195937_1_0631.tifDX195937_1_0633.tifDX195937_1_0635.tifDX195937_1_0637.tifDX195937_1_0639.tifDX195937_1_0641.tifDX195937_1_0643.tifDX195937_1_0645.tifDX195937_1_0647.tifDX195937_1_0649.tifDX195937_1_0651.tifDX195937_1_0653.tifDX195937_1_0655.tifDX195937_1_0657.tifDX195937_1_0659.tifDX195937_1_0661.tifDX195937_1_0663.tifDX195937_1_0665.tifDX195937_1_0667.tifDX195937_1_0669.tifDX195937_1_0671.tifDX195937_1_0673.tifDX195937_1_0675.tifDX195937_1_0677.tifDX195937_1_0679.tifDX195937_1_0681.tifDX195937_1_0683.tifDX195937_1_0685.tifDX195937_1_0687.tifDX195937_1_0689.tifDX195937_1_0691.tifDX195937_1_0693.tifDX195937_1_0695.tifDX195937_1_0697.tifDX195937_1_0699.tifDX195937_1_0701.tifDX195937_1_0703.tifDX195937_1_0705.tifDX195937_1_0707.tifDX195937_1_0709.tifDX195937_1_0711.tifDX195937_1_0713.tifDX195937_1_0715.tifDX195937_1_0717.tifDX195937_1_0719.tifDX195937_1_0721.tifDX195937_1_0723.tifDX195937_1_0725.tifDX195937_1_0727.tifDX195937_1_0729.tifDX195937_1_0731.tifDX195937_1_0733.tifDX195937_1_0735.tifDX195937_1_0737.tifDX195937_1_0739.tifDX195937_1_0741.tifDX195937_1_0743.tifDX195937_1_0745.tifDX195937_1_0747.tifDX195937_1_0749.tifDX195937_1_0751.tifDX195937_1_0753.tifDX195937_1_0755.tifDX195937_1_0757.tifDX195937_1_0759.tifDX195937_1_0761.tifDX195937_1_0763.tifDX195937_1_0765.tifDX195937_1_0767.tifDX195937_1_0769.tifDX195937_1_0771.tifDX195937_1_0773.tifDX195937_1_0775.tifDX195937_1_0777.tifDX195937_1_0779.tifDX195937_1_0781.tifDX195937_1_0783.tifDX195937_1_0785.tifDX195937_1_0787.tifDX195937_1_0789.tifDX195937_1_0791.tifDX195937_1_0793.tifDX195937_1_0795.tifDX195937_1_0797.tifDX195937_1_0799.tifDX195937_1_0801.tifDX195937_1_0803.tifDX195937_1_0805.tifDX195937_1_0807.tifDX195937_1_0809.tifDX195937_1_0811.tifDX195937_1_0813.tifDX195937_1_0815.tifDX195937_1_0817.tifDX195937_1_0819.tifDX195937_1_0821.tifDX195937_1_0823.tifDX195937_1_0825.tifDX195937_1_0827.tifDX195937_1_0829.tifDX195937_1_0831.tifDX195937_1_0833.tifDX195937_1_0835.tifDX195937_1_0837.tifDX195937_1_0839.tifDX195937_1_0841.tifDX195937_1_0843.tifDX195937_1_0845.tifDX195937_1_0847.tifDX195937_1_0849.tifDX195937_1_0851.tifDX195937_1_0853.tifDX195937_1_0855.tifDX195937_1_0857.tifDX195937_1_0859.tifDX195937_1_0861.tifDX195937_1_0863.tifDX195937_1_0865.tifDX195937_1_0867.tifDX195937_1_0869.tifDX195937_1_0871.tifDX195937_1_0873.tifDX195937_1_0875.tifDX195937_1_0877.tifDX195937_1_0879.tifDX195937_1_0881.tifDX195937_1_0883.tifDX195937_1_0885.tifDX195937_1_0887.tifDX195937_1_0889.tifDX195937_1_0891.tifDX195937_1_0893.tifDX195937_1_0895.tifDX195937_1_0897.tifDX195937_1_0899.tifDX195937_1_0901.tifDX195937_1_0903.tifDX195937_1_0905.tifDX195937_1_0907.tifDX195937_1_0909.tifDX195937_1_0911.tifDX195937_1_0913.tifDX195937_1_0915.tifDX195937_1_0917.tifDX195937_1_0919.tifDX195937_1_0921.tifDX195937_1_0923.tifDX195937_1_0925.tifDX195937_1_0927.tifDX195937_1_0929.tifDX195937_1_0931.tifDX195937_1_0933.tifDX195937_1_0935.tifDX195937_1_0937.tifDX195937_1_0939.tifDX195937_1_0941.tifDX195937_1_0943.tifDX195937_1_0945.tifDX195937_1_0947.tifDX195937_1_0949.tifDX195937_1_0951.tifDX195937_1_0953.tifDX195937_1_0955.tifDX195937_1_0957.tifDX195937_1_0959.tifDX195937_1_0961.tifDX195937_1_0963.tifDX195937_1_0965.tifDX195937_1_0967.tifDX195937_1_0969.tifDX195937_1_0971.tifDX195937_1_0973.tifDX195937_1_0975.tifDX195937_1_0977.tifDX195937_1_0979.tifDX195937_1_0981.tifDX195937_1_0983.tifDX195937_1_0985.tifDX195937_1_0987.tifDX195937_1_0989.tifDX195937_1_0991.tifDX195937_1_0993.tifDX195937_1_0995.tifDX195937_1_0997.tifDX195937_1_0999.tifDX195937_1_1001.tifDX195937_1_1003.tifDX195937_1_1005.tifDX195937_1_1007.tifDX195937_1_1009.tifDX195937_1_1011.tifDX195937_1_1013.tifDX195937_1_1015.tifDX195937_1_1017.tifDX195937_1_1019.tifDX195937_1_1021.tifDX195937_1_1023.tifDX195937_1_1025.tifDX195937_1_1027.tifDX195937_1_1029.tifDX195937_1_1031.tifDX195937_1_1033.tifDX195937_1_1035.tifDX195937_1_1037.tifDX195937_1_1039.tifDX195937_1_1041.tifDX195937_1_1043.tifDX195937_1_1045.tifDX195937_1_1047.tifDX195937_1_1049.tifDX195937_1_1051.tifDX195937_1_1053.tifDX195937_1_1055.tifDX195937_1_1057.tifDX195937_1_1059.tifDX195937_1_1061.tifDX195937_1_1063.tifDX195937_1_1065.tifDX195937_1_1067.tifDX195937_1_1069.tifDX195937_1_1071.tifDX195937_1_1073.tifDX195937_1_1075.tifDX195937_1_1077.tifDX195937_1_1079.tifDX195937_1_1081.tifDX195937_1_1083.tifDX195937_1_1085.tifDX195937_1_1087.tifDX195937_1_1089.tifDX195937_1_1091.tifDX195937_1_1093.tif