Loading…

Flash Player 9 (or above) is needed to view presentations.
We have detected that you do not have it on your computer. To install it, go here.

Like this document? Why not share!

Limamali

on

  • 779 views

Passenger seat is main part of vehicle which has direct effect on her/his convenience. Seat suspension can remove unwanted and harmful vibration if right parameters were selected. Each of human body ...

Passenger seat is main part of vehicle which has direct effect on her/his convenience. Seat suspension can remove unwanted and harmful vibration if right parameters were selected. Each of human body organs has specific natural frequency. When vehicle vibration reaches to this natural frequency, resonance will occur, and this phenomenon is harmful in long term. Usually lumped models used to predict human body response to vibration. In this paper, via Kitazaki biodynamic model, the seat to head vibration transmissibility was minimized by artificial neural network method. By this method, the optimum spring constant, damper coefficient and mass values were found.

Statistics

Views

Total Views
779
Slideshare-icon Views on SlideShare
779
Embed Views
0

Actions

Likes
0
Downloads
3
Comments
0

0 Embeds 0

No embeds

Accessibility

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

    Limamali Limamali Document Transcript

    • Conference paper at University of Cairo, Egypt. (December 2011) Spinal Vibration Reduction in Ve rtical Vibration based on Artificial Neural Network Optimization by Using of Kitazaki Biodynamical Model Li mam Ali Faculty of Agricu lture (Machinery), University of Cairo, Egypt. ____________________________________________ ___________________________________________ ABSTRACT Passenger seat is main part of vehicle which has direct effect on her/his convenience. Seat suspension canremove unwanted and harmful v ibration if right parameters were selected. Each of human body organs has specificnatural frequency. When vehicle vibration reaches to this natural frequency, resonance will occur, and thisphenomenon is harmful in long term. Usually lu mped models used to predict human body response to vibration. Inthis paper, via Kitazaki biodynamic model, the seat to head vibration transmissibility was min imized by artificialneural network method. By this method, the optimu m spring constant, damper coefficient and mass values werefound.KEY WORDS: Art ificial Neural Network, Spinal Vib ration, Kitazaki Biodynamical Model, Vertical Vib ration. INTRODUCTION Nowadays, people have become more aware of v ibration and they are looking for more co mfortableenvironments. One of the vibration and shock sources is traveling inside vehicles such as cars, buses, trains andheavy construction machine such as tractors, loaders and bulldozers. Drivers are exposed to the whole bodyvibrations due to the interaction between uneven roads or tracks with wheels. This leads to some injuries like spinaldisorder, weakness in sight, pain in internal tissues and heart problems in a long term. Many researchers studiedabout human body responses to vibration, and how they can minimize the negative effects of shock to human body(Paddan and Griffin , 1998, Boileau and Rakheja, 1998,Wang et al. , 2008) . The responses have been assessed interms of the apparent mass, driving point impedance and transmissibility. The forces and motion of pelvis and spineare studied by the first two, and the third looks into the motion of chest, head and spine. The results oftransmissibility of seat-to-head from previous researchers indicate that the first principal response was foundbetween 4 and 6 Hz, and the second natural frequency appears at 8 to 12 Hz (ISO 5982). The human body is a co mplex structure, and responses of that to the dynamic excitation are more comp lex. Fro mthe results of many studies (Muksian and Nash, 1974, Suggs et al., 1969 , Wan and Schimmels, 1995, Liu et al.,1998), several of biodynamical models are availab le to explain human body behavior exposed to the oscillat ion.These models consist of lumped, mult i body and finite element models. In the lumped model, human bodies areconsidered as a set of masses, springs and dampers. Multi body models have a series of segments which areinterconnected by some kinematics jo ints. These joints are usually spherical, revolute or universal jo ints. In this typeof model, spine behavior is modeled by flexible elements. Finite elements methods consisting of several elementswere used to model the various parts of human organs, by meshing and stiffness matrix defin ition. This method canalso reproduce behavior of flexib le and rigid parts with rheologic properties. Although lumped models are theearliest generation in modeling, those are already applicable because of its simp licity. Mechanical impedance, seat to head vibration transition, and apparent mass are important functions which areused to driving model fro m experimental test, and also these are useful for optimizat ion methods. Severalresearchers have investigated on parameters which effect on the seat-to-head transmissibility (STH)( Paddan, andGriffin, 2000). The variation of STH functions like as apparent mass and mechanical impedance is large. Theyfound that the posture, feet support and backrest can influence the STH. Bo ileu and Rakheja (1998) employed a fourdegree of freedo m (4-DOF) model to predict hu man responses to vertical vibration. They obtained first resonanceof STH at 4.8 Hz although experimental results showed at 5 Hz. Wan and Schimmle established a 4-DOF lu mpedmodel that the total mass was 60.67 kg, and the first peak in STH appears at 4 Hz. The goodness -of-fit was 91% forthis model. Wang et al. (2008) studied the relationships between the apparent mass and seat to head transmissibility.They conducted test with 12 males as subject who were exposed to whole body vibration in vertical d irection.L. Ali
    • Conference paper at University of Cairo, Egypt. (December 2011) The frequency was in the range of 0.5-15 Hz, and the magnitudes of acceleration were 0.25, 0.5 and 1.0 m/s 2 inrms. The measurements were done in three postures: without backrest, vertical backrest and inclined position. Also,two driver of hands position is considered such as hands in lap and hands on the steering wheel. The results showedthat both response functions (seat to head transfer and apparent mass) in primary resonances showed goodagreements between two postures. In addition, in secondary resonance there are considerable differences betweenseat to head transmissibility and apparent mass for the two backrest positions. Seat to head transfer function is littlemore sensitive than apparent mass in responding to the hand positions. As a result, seat to head vibration transfer function shows wide variation in resonance first and second peaks(fro m 3 Hz to 8 Hz) which are depended on posture, within or without backrest and feet supporter. The summationof most important studies was represented as standard which is called ISO 5982. The seat to head transmissibility ofthe seated human body is shown in Fig.1. In this standard, the first resonance in seat to head transmissibility isbetween 3 to 5 Hz. Fig.1. The seat to head transmissibility of the seated human body exposed to the vertical oscillat ion. A modal analysis based on experimental data was made by Kitazaki (1992). This model is a b io mechanicallu mped model which is able to exp lain biodynamical behavior of torso, head, and spine. In spite of simple base, thismodel can reproduce first five modes of vibration in acceptable accuracy. Fig. 2 shows the parts of Kitazaki modelwhich has 15 degrees of freedom. The most impo rtant components of this model are the five vertebrae (T1, T6, T11,L3 and S2) considered in the model as separate parts, and it is possible to monitor the modes shapes of them in theanalyses. Fig.2. Hu man body biodynamical model and its degrees of freedom [10]. In vehicles, suspension systems are used to reduce unwanted vibration, and isolate the passengers. Most of heavyduty vehicles remove the transition vibration by passive or active seat suspensions. In passive suspension systems,finding the mass of seat, damping ratio and spring constant are essential for design. Lu mped parameter models are frequently used to find optimized values of damping and spring constant invehicle seat or suspension design. One of the methods which are useful in optimizing and mathemat ical modeling isartificial neural network (ANN). This procedure was based on artificial intelligent method and machine learningwith some samples. These s amples consist of sets of input and output values. Artificial neural network can trainedwith examp les which finally produce desired output values. This step is called training. After training, the networkwill be able to simulate and predict the outputs for each entered values. ANN models can solve and simu latenonlinear systems with acceptable accuracy. In this paper, artificial neural network was used to determine the seatL. Ali
    • Conference paper at University of Cairo, Egypt. (December 2011)suspension properties (mass, spring constant and damping ratio) that are able to min imize the vertical unwantedvibration based on Kitazaki hu man body model. Material and Methods As mentioned in the literature review, Kitazki biodynamical model shows good results with regard to the realconditions [10]. This model was considered as a benchmark for an establishment of an ANN model. Proposed modelincludes human body parts (head, spinal column, visceral colu mn and pelvis model) in sitting posture and seatpassive suspension system components. Optimizing suspension coefficients (mass, spring and d amper) was done inthree steps:1- Data preparation : The human body modeled by using Working Model 2D software as shown in Fig. 3. The input function was aharmonic sinusoidal acceleration with equation: Z0  0.05 sin t  and   1 rad s ( Liang and Chiang, 2006).The input acceleration (seat) and output values (head) in the time do main were obtained in this software for 20seconds (Fig. 4). The seat to head (STH) function frequency peaks were calculated by using FFT (Fast FourierTransform) function in Matlab software. This step was repeated 11 times for 11 different values of mass, spring anddamper coefficients.2- Optimization with neural network The first two peaks of STH values were entered into neural network as input set (for 11 example). Also, the values of mass, spring and damper were entered as output of network. The network was trained by these eleven examples.3- Validation of optimized system The seat suspension system coefficients were set by the values obtained. Then, peak values of STH were recalculated and were co mpared to the proposed values. Fig.3. Hu man body and eccentric oscillator circle which was modeled in Working Model 2DL. Ali
    • Conference paper at University of Cairo, Egypt. (December 2011) Fig. 4. Vertical accelerations (m/s 2 ) of a) head and b) seat versus time (t). Data Preparation: Human body mechanical modeling in Working Model 2D Working Model 2D has some features like as mechanism modeling and dynamic analyses. Parts of human bodywere modeled with rigid co mponents, torsion spring and damper co mponents as illustrate in Fig.3. The displacementof masses was constrained in the vertical axes. The components properties were set according to Table 1 (Kitazaki,1992). An eccentric circle and a rotary engine were located under the model to produce harmonic movement according to Z0  0.05sin t . The neural network model needs some examp les for t rain ing and optimization, thusthe values of seat mass, spring and damper were set to eleven values. These values are listed in Table 2. It isobserved that the amplitudes of the first 2 peaks in STH are low for mass in the range of 14 to 15 kg, dampingcoefficient 125 to 130 Ns/m and spring stiffness from 8100 to 8150 N/ m. The accelerat ion of input point (seat) andoutput point (head) were achieved in separate diagrams which these values are time dependents. In the next step,these two acceleration values were keying into the program which was written in Matlab environ ment. This programdivided the accelerations values of head to seat. The output values were converted to frequency by using FastFourier transformat ion function to determine the peak values of seat to head transmissibility (STH). Table.1. Stiffness and damping values of the lumped model (Kitazaki, 1992). Articulations K (Nm/rad) C (Nms/rad) C0 18 0.1 T1 20 0.8 T6 625 0.8 T12 92 0.2 L3 224 0.9 S2 643 0.1 Table. 2. Seat mass, spring, damper coefficients, first and second peaks of STH in each of 11 examp les. Amplitude of Amplitude of Example No. Mass (kg) C(N.s/m) K(N/ m) first peak in second peak in STH STH 1 15 125 8100 0.756 0.174 2 13 120 8150 1.316 1.004 3 15 130 8150 0.743 0.177 4 14 128 8100 0.751 0.174 5 16 120 8200 0.754 0.181 6 13 130 8000 1.31 0.99 7 13 100 8100 0.8 0.178 8 14 125 8130 0.753 0.176 9 11 120 8200 0.754 0.181 10 7 80 7500 1.084 0.935 11 17 120 8200 1.325 1.009L. Ali
    • Conference paper at University of Cairo, Egypt. (December 2011) Neural Network Modeling and Seat Suspension O pti mization Neural networks are robust tools which can solve non-linear equation in optimizat ion problem. In this research,the problem was stated in mathemat ic relat ionship as multi input-output function which has a polynomial nature. Infact, seat to head transmissibility peak values ( p1 and p2 ) is a function of mass, spring constant and dampercoefficient.STH( p1 , p2 )  f (m, C, K) A feed forward back propagation architecture network was chos en for this modeling because this case isoptimization problem. Two matrices were made for input data and output. The first matrix was 2* 11 wh ich includesfirst and second peaks of STH values with eleven examples. The output matrix was 3*11 which three rows weremass, spring and damper coefficients. Various training functions, adaptation learning function, threshold functionand error reduction function were tested to obtain best model fitting with high ac curacy. Due to this aim theTRAINLM function (Levenberg-Marquardt backpropagation algorith m) was considered as training function, andadaptation learning function was LEA RNGDM (grad ient descent with mo mentum weight and bias learningfunction). This architecture has 4 layers in total which it has 3 hidden layers with ten neurons (Fig. 5). Also,TANSIG (Hyperbolic tangent sigmoid transfer function) was selected in Matlab software for threshold function andminimu m squired error (MSE) was function to error reduction and weight adjusting loop. Fig.5. The architecture of neural network fo r optimization. After network training, the performance graph and gradient were obtained. Furthermore, the regression graphbetween the target and output was plotted. In addition, R2 for all regressions was calculated. Finally, the networkwas simulated to find the output with first and second STH peaks. The outputs of the network will be the optimu mvalues of suspension coefficients (m, C, and K) which produce minimu m value of seat to head transmissibility. RES ULTS AND DISCUSS ION After the network was entered with examples, six epochs were done to complete the train ing. The R2 for training,validation, test and in overall were 0.99994, 1, 1 and 0.99995, respectively (Fig. 6). To find the best values of mass,spring and damper constants, the amplitudes of first and second peak of STH were considered 0.5 and 0.09. Thesevalues were entered to the trained network to find outputs of that. These value were obtained as 15.88, 8 000 and128.748 for seat mass, spring and damper coefficient, respectively.L. Ali
    • Conference paper at University of Cairo, Egypt. (December 2011) Fig.6. Correlation coefficients in neural netwo rk model in training, validation, test, and all steps, respectively (T is target values, Y is predicted output by model, and R correlation coefficient between output of model and target values). For evaluation of result, seat suspension was adjusted again in Working Model and acceleration of head and seatwere reached. STH diagram was plotted with new values of suspension parameters. Fig. 7 illustrates the STH inoptimized seat properties, and this ratio decreased to below 1 which means that purposed results were obtained asdesired. Fig.7. First and second STH peak in optimized seat suspension. In Fig.7, amp litude of STH in optimized seat suspension system were0.76 and 0.18, and it appears that the firstand second natural frequencies are in the desired range according to ISO 5989. Alizadeh et al. (2008) found out thatthe seat mass, damper and spring coefficients were 14 Kg, 125 Ns/m and 8130 N/ m, respectively by Quasi- Newtonmethod. The STH after optimization in their test was near 1.8 while optimized seat to head vibration by KitazakiL. Ali
    • Conference paper at University of Cairo, Egypt. (December 2011)(1992) model have better reduction in vertical vibrat ion transmissibility, and it looks this model can be considered asreference model for simulation and optimization. CONCLUS ION Optimizing the seat properties (spring, damper and mass) by neural network is very useful in reducing thevertical v ibration transmissibility. The result of th is study shows that it is possible to consider spinal column motionexposed to vibration for optimization. In addition, if a suitable seat suspension is used, the vibration which istransmitted to the head can be reduced by 94%. ACKNOWLEDGEMENTS The author would like to thanks from Un iversity of Cairo. REFRENCES1. G.S. Paddan, M.J. Griffin, A review of the transmission of translational seat vibration to head , Journal of Soundand Vibration 215 (1998) 863-882.2. P.E. Boileau, S. Rakheja , Whole-body vertical biodynamic response characteristics of the seated vehicle driverMeasurement and model develop ment, International Journal of Industrial Ergonomics 22 (1998) 449- 472.3. W. Wang., S. Rakheja, P.E. Bo ileau, Relationship between measured apparent mass and seat-to-headtransmissibility responses of seated occupants exposed to vertical vibrat ion , Journal of Sound and Vib ration 314(2008) 907–922.4. ―Mechanical Vibration and Shock—Range of Idealized Values to Characterize SeatedBody BiodynamicResponse Under Vert ical Vib ration,‖ ISO 5982, International Organization for Standardization, Geneva, 2001.5. R. Muksian, C.D. Nash, A model fo r the response of seated humans to sinusoidal displacements of the seat ,Journal of Bio mechanics 7 (1974) 209–215.6. C.W. Suggs, C.F. Abrams, L.F. St ikeleather, Application of a damped spring-mass human vibration simulator invibration testing of vehicle seats , Ergonomics 12 (1969) 79–90.7. Y. Wan, J.M. Sch immels, A simple model that captures the essential dynamics of a seated hu man exposed towhole body vibration. Advances in Bioengineering, ASM E, BED 31 (1995) 333–334.8. X.X. Liu, J. Shi, G.H. Li, Biodynamic response and injury estimation of ship personnel to ship shock motioninduced by underwater explosion. Proceeding of 69th Shock and Vibrat ion Sy mposium, vol. 18 (1998) St. Pau l 1–18.9. G.S. Paddan, M.J. Griffin, Evaluation of whole-body vibration in vehicles, Journal of Sound and Vibrat ion 253(1) (2000) 195-213.10. S. Kitazaki, Application of experimental modal analys is to the human whole-body vibration, Proceedings of theUnited Kingdom Informal Group Meeting on Human Response to Vibration, The University of Southampton,Southampton, Hampshire(1992) 17–39. 11. Cho-Chung Liang, Ch i-Feng Chiang, A study on biodynamic models of seated human subjects exposed tovertical v ibration. International Journal of Industrial Ergonomics 36 (2006) 869–890 12. H.A. Alizadeh, A. Sedaghat, M. Sadeghi Mehr, D. Naderi, Determining and Optimization of Mass, Stiffnessand Damping Coeffic ients of Tractor Seat by Quasi-Newton Method Using Coupled Human-Seat Model, Journal ofAgricultural machinery Science 4(1) (2008) 51-55.L. Ali