A New Decision Making Algorithm for Airbag Control Syed Masud Mahmud, Member, JEEE, and Ansaf I. Alrabady Absraet— Recently, airbag systems have been introduced to supplement the seat bet system primarily to reduce head injuries. [Most of the present airbag systems use distributed mechanical sensors, which are costly, not easy to eallrate, and not effective {o trigaer the airbag on time for diferent types of crashes. An electronic sensor is more effective In the sen thatthe signal from the accelerometer can be dgidzed und analyzed to study the behavior ofthe signal for afferent types of crashes, Unfor tunately, most of the algorithms which are used for electronic ‘sensors still have some problems. They either fal to trigger the Airbag on time for several types of high speed crashes such as pole and angle crashes, or if they do trigger at these erases they Ako (rigger at low speed barrier crashes. “A new algorithm for an electronic sensor is presented inthis paper. This algorithm has been developed by analyzing a large Wolume of data obtained from actual erash eating experiments ‘The algoritym works for al types of crashes, The algorithm Ras ‘ben tested by applying it on real crash testing data A hardware clreut has also been proposed In the paper, which can be Used {o implement the algorithm for real-time operations. 1 ytRopucion IRBAG systems are used in many current vehicle re: straint systems to supplement seat belt systems primarily to reduce head injuries. Recent studies show thatthe stbag. reduces head and chest injuries significantly (1]-(3. ‘Twotypes of crashes are widely investigated in the literature [7-12]. One of these crashes is called the barrer-crash tand the other one is called the pole-crash. A crash is called 8 barriercrash when a car hits a very hatd stationary barrier such as a concrete wall. Ifthe cat hits the barrier perpendicularly, we say that itis a frontal barrier-crash oF 90° barrier-crask; otherwise, the erish is called an angular barrer-crash. A crash is called a pole-crash when a ca bis a pole or & small tee, Normally the aba is deployed if the collision force is equivalent to a frontal collision witha staionary barrier (90° barrer-rash) ata speed of 14 mh or higher [6]. The greater the collision impact the earlier the airbag should be deployed, Because, the atbag has to be deployed before an occupant ‘moves forward five inches relative tothe ear (4). Normally i takes 30 milliseconds for an airbag to be deployed after it gets 8 trigger signal from the airbag sensor. Thus, an airbag sensor is to be designed in such a way that it ean send a trigger pulse to the abag deployment circuit 30 ms before the time When Manas recive Joe 20,196: reve Dnsry 31,1998, Thee wa ply soppned by gt cn Ford Mor Company Tih ator at wih he Depurinent of Elta and Compt Enger ng. Wayne Site Users Deo MI 202 USA, EEE og Naber 911137 the occupants head moves forward five inches with expect to the car. Tis wigaering point is often called the 5-30 ms point [9]. For a barrercrash at 30 mi the 5-30 ms point comes within 26 me after the crash occurred ‘An airbag can be used only once. Since replacing a used airbag costs at least $500 (6), it's obviously critical forthe airbag to inflate only when absolutely necessary. Thus, the airbag should not be deployed when a car moves through a very rough or bumpy road, or when it hts a small trafic sign pole, eg., top or yield sign pole ‘Most of the present airbag systems use distributed me- chanical sensors. All the mechanical sensors work almost fon the same principle. In a typical mechanical sensor stainless steel ball is held by a magnet at one end of a closed cylinder. The erash impact overcomes the precisely calibrated ‘magnetic bas and the Ball moves through the eylinder to close electrical contacts and complete the airbag triggering circuit [415]. Mechanical sensors canbe designed to take the above mentioned specifications into consideration, but the necessary calibeation for the sensors i8 not that simple, Very litle work is available in the literature on electronic sensors for airbag systems and their algorithms for airbag AV, Within 35 ms after the erash occurred, then it is sure that the crash occurred in the deployment zane and therefore, the airbag must be iggered. But itis not possible to select a threshold value of AV” which ean be used to wigge the aia for 19 mifhipole crash (see Fig 2). Because, if such a threshold vale is selected to trigger the altbag within 60 ms after the ‘occurrence of a 19 mifhpole crash, them that threshold value will aso tigger the atbag for an'8 midh!90" barrier crash, Thus, the parameter AV” can not be used to discriminate a 19 rishypote crash from an 8 mivBv0® barrie crash until itis 0 Tate to trigger the airbag for a 19 mi/hipole crash. Therefore, ‘only AV’ alone is not enough to make a decision whether ‘or not the sirbag is to be triggered forall types of crashes However, the parameter AV’ can be used in the airbag control algorithm as Follows: Te AV>AVy, within 35 ms after the crash ‘Then crash occurred in the develop- type of the crash is; further paraneter © necessary to determi: angular barrier cra Ele In onder to select a good parameter for discriminating pole crashes against frontal barir crashes (90° barier crashes) we reed to clearly understand the differences between barrier and Pole cashes, During a frontal barrier crash the entire bumper ofthe vehicle is affected. But during a pole crash only a par of the bumper is affected. Thus. the vehicle stars lsing energy ata higher rate fora batier erash than fora poe crash, AS a result, for a frontal barrier crash the vehicle velocity decreases rapidly than fora pole crash of equivalent collision force. A pole crash, which requires airbag deployment, normally breaks 4 par of the bumper. The pole then cus through the bumper and hits other distributed solid objects (motor, radiator, ec) in the vehicle, When the pole hits a solid object the magnitude of the vehicle deceleration increases, ie, the vehicle slows. down ropidly. After tha ifthe pole breaks the solid object, then the vehicle deceleration will decrease, ic, the vehicle will rot slow down as rapidly as it was. Ifthe pole its another solid object. then the magnitude of the vebicle deceleration will increase again. From the above discussion itis clear that the accelerometer signal of a pole crash (which requites airbag ‘deployment wil have more swings than a frontal barter crash which does not require airbag deployment. We kept this fact in mind in order to determine a parameter which will allow us to trigger the airbag atthe appropriate time for a pole crash The above fact led us to investigate the accelerometer signal in det, ‘THE ACCELEROMETER SIGNAL AND I1S COMPONENTS ‘There is no doubt that during a crash the accelerometer signal, s(¢), contains vital information about the crash It ea bbe assumed that the signal s(¢) contains. two components: the vehicle acceleration component and the nose component. Thos, we can write a(t) = alt) +t) @ where a(?) is the vehicle acceleration component and n(#) is the noise component, The vehicle acceleration component, f(t) isa negative number during a crash. The component a(t) contains the important information about the rash. The noise ‘component depends on several factors, such as: the location of the accelerometer, the vibration and deformation of the vehicle during a crash, For example, ifthe accelerometer is placed at the front of the vehicle, then the noise component is higher than i iis placed inthe passenger component. Normally the accelerometer signal contains t00 much noise iT itis located in a soft part of the vehicle, or in that part of the vehicle which is vulnerable 10 a crash, In onder to design = good electronic eras sensor itis desirable to place the accelerometer atthe least noisy location in the vehicle, so thatthe vehicle acceleration component, f), can easily be separated from the accelerometer signal, #1) Some authors modeled a(t) by a haversine pulse which is a low frequency component (0-10 Hz) (7). [10]. Thos. they filered the signal s(t) using a low-pass filter of cutoff frequency 10 Hz to investigate several parameters such asthe first derivative of o() (ek) [8], the energy in af) and the energy in the noise component m(t) [10]a ie a SS a ee eh ee may © Fg.3._() Acceromaer signal) colt fom an aslo laced lhe outer near he ders doe () Pegs reponse of) We believe that fitering the signal down to 10 Hz will not ‘only eliminate or average some of the important information in the signal s(t), but also will incur along time delay in the fered signal which could be longer than the fring time, It is important to study the whole slowing components which contain pulses at moderate frequency due to several Aistributed solid objects in the car (bumper, motor, radiator, te.) These pulses are very important to measure the intensity fof a crash, A higher speed crash causes a higher frequency and amplitude pulses than a lower speed crash To analyze the slowing component a(t) we took several accelerometer readings forthe same crash a different locations in the vehicle. We found that different accelerometer loations yield to different signals. There are three reasons for tht Firstly, the accelerometer measures the slowing ofthat part of| the vehicle where tis located. For example, an accelerometer placed in the front of the vehicle will give higher values ‘than an accelerometer placed in the passenger compartment oF at the end ofthe vehicle. Because, asthe front pat of the vehicle breaks down by absorbing most of the vehicle energy, the other pars of the vehicle still keeps moving. Secondly, each part of the vehicle vibrates ata different frequency, and there are several locations which oscillate at low frequencies, and this oscillation component may overlap with the slow ing component a(t). Thirdly, the way the accelerometer is ‘mounted in the vehicle: this means tht if the accelerometer is loosely fixed in the vehicle, it will shake and vibrate at 4 frequency which could overlap withthe slowing component {a(t}, After analyzing a larg set of crash testing data we found that the best location ofan accelerometer is atthe rocker near the driver's or passenger's door. Fig. a) shows the signal s(¢) collected from an accelerometer which was located at the rocker near the driver's door, and Fig. 3(6) shows is Frequency response. From the frequency response itis clear thatthe noise an bre "hems © Fg. the vee acoleton component a) of eet crates () 8 [NO barr rss (I, C2). 8 1 WSO" ares eer (3, Cy, nd) 19 tht crashes (C3. ‘component is widely separated from the signal component Most of the signal components are within 0-500 Hz, and the noise is very dominant within the frequency range 500-1800 Hz, Thus the veicle acceleration component a(t) can easily be separated from the noise component n() using a lowpass filter of appropriate cutoff frequency. Easier in this paper we analyzed six crashes (see Fig. 2) to determine the change in vehicle velocity during the crash. We are going wo further analyze the same six crashes. Fg, 4 shows the component a(t) ofthese six crashes, obtained by filtering them using a low pass filter of eutot frequency S00 Hz, As mentioned earle, it \vas difficult for us to discriminate » 19 mi/hipole crash from an 8 miv90° barrier crash by monitoring only the change in vehicle velocity AV’ (oe Fig. 2). But Fig. 4 shows that the a(¢) component of @ 19 miftvpoe crash has more swings than an 8 mi/tv90° barter crash, as it was expected from our previous discussion in Section IIL. The fact that a pole crash ‘hich requires an airbag deployment has more swings in the signal a(t) than a 90° bari erash which does not requ airbag deployment can be used to determine whether or not a pole crash occurred in the deployment zone. IV, THE Ammeag ConTROL PARAMETERS AND THE ALGORITHM We have already mentioned that AV, the change in vehicle velocity, is a good parameter fr frontal barrier erases. A second parameter is introduced in this section which caneet Fp $. ieemi! eagh AL of cane ot bbe used 10 discriminate a deployment pole crash from a rnondeployment barier crash. An airbag contol algoritom is then developed in this section based on the two parameters siscussed in the paper. ‘The second parameter which is introduced in this section ives a quantitative measure of the swinging effect of the fihered accelerometer signal. Fig. 4 shows that for a 19 rmiftpote crash, the amplitude of the swings of a() is higher than that of an ®-miNV9O? barrier crash. Thus, we wanted {0 ‘determine the second parameter in such a way that it will be a ‘monotonously increasing function of the number of swings and the total amplitude of the swings. I is obvious that the value of such a parameter will be higher for & 19 mistipole crash than for an 8 mifly90" barrier erash, We chose the length of the a(f) curve as our second parameter, because this quantity Increases with the numberof swings and the swing amplitudes. ‘The differential length A. of the a(?) curve between times and * + AT, as shown in Fig, 5, can be determined as [Batt + (ary ((@y~)» Hence. the length of the curve a(!} from time 0 t0 # ean be determined as "( [aay sf (V( ay ») at Fie 6 shows the lng of the o() curve vers ime for fot crashes: two 19 poe crashes (curves C3 and Co) and two 8 0° hier rashes ures Cand C2), om this gure ii een that tthe Sing point ofthe 19 maple crash there alae gap betwecn te curves of 19 inple races athe curves of # mim/9O" bare rashes. Ts i it very eany to discriminate 19 mile cashes om the 8 Ins" bar cases using ou second purer. whichis the lng ofthe scseation eure (). The pial mang Of tht Ht gives quantiatine measure of the swinging effet of the filtered accelerometer Sina. The value of Le ncresses with te number and mp te of swings inthe accelerometer signal. The accelerometer nal of deployment pole cash has more high ample Sings than nondeployment baer cash This fat male the parameter sf lor developing ur sap: cond algorithm, @ ” Fig The eng of (cae or four crates C1LC:8 m0" Mair A. Development of the Airbag Control Algortion [Now we are ready to develop an algorithm for the airbag control. Both the above mentioned parameters must be used the algorithm. One parameter without the other will not be a good choice. Because, we have already seen that only AV. the change in vehicle velocity, can't be used to tigger the airbag for a pole crash. Similarly only the length of the f(t) curve, L. may not be a good choice. Because, the value of L may be large when driving through a rough or bumpy road. However, the change in velocity over 8 period of (ihe firing time for the minimum speed fontal barrier crash Which requires an airbag deployment) willbe much less when driving through a rough or bumpy road. Hence, even though T may be large, the very small value of AV” will indicate tht it is not a crash. We need to select a threshold value of AV, say AV, which will discriminate crash (deployment ot nondeployment crash) from deving through rough or bumpy ‘oad and from other types of slowing conditions ofthe vehicle, such as braking the vehicle. Such a threshold value can easly be determined for a parcular vehile by knowing. how fast, the vehicle can be braked. Assume that a vehicle moving at {60 tifa can be completely stopped within 3 s (which is a very optimistic assumption) by applying brakes. The average change in vehicle velocity over a period of 7 =35 ms for such a braking condition will be @ni/r 3s 1661/0 nr ajs 1760 x 3f1/n ToApms/s av x 5m Which is a very small number. Note that for an 8 mistv90® barrier erash the change in vehicle velocity over a perio’ of 35 ms is between 2-4 fs. Thus, even forthe nondeployment barrier crash, the change in vehicle velocity overa period of 35. rms is 2-4 times higher than that forthe very optimistic braking condition, explained before. Hence, a value of AV; (threshold value of AV) can easily be selected for a given type of vehicle to discriminate a crash (deployment or nondeployment crash) from other types of slowing conditions of a vehicle. Thus, if the value of AV’ over a period of 7; is less than V4, then the airbag must not be fred no matter what the value of [ the length of a(t) curve) is. IF AV’ (over a period of 7\) is greater than V3, then itis understood that some kind of ‘rash hs been detected, But whether or not the airbag i to be fired will depend on the results of further testing. Once a crashhas been detected, another threshold value of AV, say AV, is necessary to determine whether or not itis a deployment barrier crash. For the crashes shown in Fig. 2, a value of 65 {us ean be used for AV» to detect a deployment frontal bari nash. Similarly a threshold value of 1, say Ly i necessary to determine whether or not its a deployment pole crash. Let a be the fring time for the minimum speed deployment pole crash. The value of L must then be computed over a period of Ta, For the crashes shown in Fig. 6, value of 1200 can be used for L (see Fig. 6) t0 detect a deployment pole crash B. Computation of AV and L Since the fring ime forthe lowest speed barrie rash which requires an airbag deployment is T;, the value of AV’ must bbe computed over a period of T). AC any given time the value of AV can be computed using the accelerometer signal ofthe ‘most recent T; time. A sliding window of length T; can be tused for computing AV. At any given time, whatever part of a(t is available within the window, only that part can be used to determine AV. Every time a new sample comes in, the window is moved by one sample. Ths means thatthe least fecent sample is moved out ofthe window when a new sample ‘moves into the window. Every time the window is moved, the value of AV is recomputed. Similarly another sliding window ‘of length T; can be used to determine the value of. Based on the above discussion we present the following algorithm for atbag contol ‘ALGORITHM Iniviatzarion 1. Use a window, say Window 1, of length 7s. Where Ti is the firing time of the minimum speed frontal barrier crash which requires an airhag deployment 2. Use another window, say Window 2, of length 72, Where Ts is the firing time ofthe minimum speed pole crash which requires an airbag deployment. 3. Initialize the contents of both windows to 0. Normal Mode 4. Get a new sample from the accelerometer and slide Window 1 and Window 2 by one sample, 5. Determine AV within Window 6. Determine F within Window 2. 7. WAV > av ‘Then go to step 8 (STAND BY MODE). This means that a crash has been detected, Otherwise g0 to step 4 Stand By Mode 8. Get a new sample from the accelerometer and slide Window 1 and Window 2 9. Determine AV within Window 1 10, Determine Z within Window 2, ILIA < ay, Then go to sep 4 (roturn to NORMAL MODE). ‘This means that te crash does not require an airbag deployment. IMAV > AV; of L > Lin ‘Then go to step 14 (FIRE), 13. Go to step 8 (sty in STAND BY MODE) P00 eee =a) Ese eyed aan eae Teme Fig. 7. A popored hrdnace timp he icons, Fire Mode 14, Fire the airbag, Fg. 7 shows a proposed hardware which can be used to control the airbag, The accelerometer signal a(t) is filtered by a low-pass ter to get the vehicle acceleration component 4(t). The a(t) component is then converted to digital ouput using an A/D converter. ‘The digital outputs are then stored into & first in fst out (FIFO) butter, The size of the FIFO bulfer is equal 10 T>. “This means that this bufler will keep the a(t) signals of the most recent Tp time. As anew sample comes into the bulfer, the least recent sample is moved out of the buffer. The a(t) signals of the most recent T; (T, < Tz) time are then fed into a cicuit which computes the change in vebicle velocity, AV. Note that the change in vehicle velocity, AV, over a petiod of T; can be computed by integrating either s(t) oF 4(¢) signal. Because, the integral over the noise component will be approximately zero. This means that av = if Moat = [mwas [moa [own o ‘The a(t) signals of the entire butler are fed into another circuit which computes the length of the a(t) curve, The circuits for computing AV and LL can be designed using systolic architecture to achieve the real-time speed. The out- puts of these wo circuits are fed into a third circuit which will make a decision whether or not the airbag is 1 be fied, This decision making circuit can be implemented as 4 synchronous sequential machine_with only three states: NORMAL, STAND-BY and FIRE. The state diagram of this machine is shown in Fig. 8Come) GaP) Fg. 8 Ste diagram of the decison makings, V. TWST RESULTS AND Discussions ‘The algorithm is ested by applying it on many crash testing data, and the results are presented in this section. The test results ate very close to the requirement, and in no case the airbag failed t0 fire if it was required to fie. The test results ae also discussed in this section. Finally some suggestions ae presented which explains what changes can be made in the structure of a vehicle in order to have more accurate fring time from the algorithm, A. Test Results We have tested our algorithm by spplying it on many. sets of crash testing data from two types of vehicles. Since wwe didn't have suficient crash testing data from other types fof vehicles and every cash testing is very expensive, we ‘ould not apply our algorithm on any other types of vehicles, However, we believe thatthe findings of our research wil help other researchers to develop better airbag contol algorithm, From now on we will indicate one type of vehicles as Type-1 vehicles, and the other type of vehicles as Type-II vehicles ‘The test resulis for the Typed and Type-II vehicles are presented in Tables I and IL, respectively. The crash testing ata foreach type of vehicles were divided into two groups One group of data was used to detenmine the threshold values AV), AV, and Ly, and the other group of data was used {o test the algorithm, The different threshold values for the ‘Type vehicles ae: AV = 1.08 fs, AV = 6 fs, and Ley = ind those forthe Type-II vehicles are: AV}, = 1.03 fs, AVs = 9.5 fs, and Lay = 1200, B. Discussion of the Results From Table 1 it is seen that the algorithm satisfied the required firing condition for all the crash modes presented inthis table. The required fring condition was determined by monitoring the movement of the dummy during a crash. For to ofthe three 19 mifhypole crashes the algorithm fied the airbag significantly carer than the requcement, Early fring is acceptable as long as the algorithm does not fre in the rnondeployment zone ofa crash. If we don't use the parameter © determine whether oF not the aitbag isto be fied, then the fring doesn't occur in time for the 19 milhipole crashes. For these crashes the firing is delayed by at least 20 ms. ‘Test results of a number of crashes of Type-II vehicles are presented in Table Il In addition to frontal barrier crashes and Pole crashes, the algorithm was als tested on thee other types of erases: an angular (30°) barter crash, a bumper overide trash and a ear to car frontal crash with 50% overlap on the Jef side. Ifthe frontal body of a vehicle hits an object before its bumper, then that crash is called a bumper override crash, =e ra ‘eer apa wer [ae ry Tami rser [a Ey Sinpuseh wren [28 = Sinpraewrrar ae z Repwsehwrern | sp | ss For example, if car crashes with a lage truck, then itis ‘bumper override crash. Table It shows thatthe algorithm ‘Satisfied the requited firing condition for almost all the crash modes of Type-Il vehicles presented in this table. However, ‘only for one crash mode the fring time of the algorithm is, ‘oft by only 1 ms from the required timing. Even for the Type I vehicles if we don't use the parameter L in the algorithm then firings forthe pole crashes are also delayed. The delays for the 17 mifm and 31 mifh pole erashes are 25 and 6 ms, respectively From both Tables I and I it is observed that forthe 90° burier crashes the fring time fom the algorithm matches very closely to that of the requirement. However, for other types of crashes sometimes the algorithm fre significantly earlier than the requiement. Early firing is acceptable if the erash occurs in the deployment zone. But ifthe early fring condition is present in an algorithm, then most ofthe time tht algorithm right trigger the sithag for gray zone crashes. AS it was ‘mentioned earlier, a deployment or a nondeployment of the airbag is equally acceptable for gray zone crashes. However, since an airbag ean be used only once and the replacement of an airbag is expensive it would be desirable to minimize the umber of firings For gray 2one crashes . Fromal Barrier Crashes versus other Crashes ‘The frontal (90°) barier crashes are very easy to detec, because during such a crash the entire bumper is. pushed toward the vehicle, and asa result the erash force is uniformlydlistrbuted over the entire bumper. When the bumper hits the body of the vehicle, the crash fore is uniformly distributed ‘over the entire frontal part of the vehicle. Thus, the vehicle decelerate very rapidly, and thereby tis relatively easier for the algorithm to detect this crash and determine whether or ‘not the crash occurred in the deployment zone. For other types of erashes only a part of the bumper is hit. Thus, if the bumper is not strong enough then a part of it the part which is hit by the object) breaks down and the Vehicle Keeps on moving until it hits the object. Hence, for this type of crash the vehicle docs not decelerate as fast a8 it does for a frontal barrier crash of an equivalent collision force, until the object hts a very hard part of the vebicte ‘This isthe reason why the change in vehicle velocity during the first few tens of milliseconds for a minimum speed pole crash which requires an airbag deployment is almost similar {o that of a maximum speed frontal (90°) barier crash which does not require the deployment of the airbag. Fora pole and other types of nonfrontal barier crashes, the vebicle stars decelerating ata high rate once a hard part ofits body is hit by the object. But that time is t00 late to trigger the airbag, This is the reason why the change in vehicle velocity during the frst few tens of milliseconds can not be used forall types Of crashes to determine whether or not the crash occurred in the deployment zone. D. Suggestions for Changing the Structure of a Vehicle From the previous discussion itis clear thatthe breakage of the front bumper makes it difficult (if not impossible) to determine whether or not a pole or nonfrontal barrier crash ‘occured inthe deployment zone. The crash sensing technique can be improved if hard bumper instead of a soft bumper is used in a vehicle, so thatthe bumper is not easily breakable. a hard bumper is used in a vehicle chen the shock absorber Which is used between the bumper and the body ofthe vehicle will absorb most of te energy and will break dawn before the ‘bumper breaks down. I the bumper does not break down then the collision force will be distributed over the entre bumper, and the vehicle will decelerate at a higher rate for this case than forthe case where the bumper breaks down very easily “Thus, the use of a hard front bumper wil let us have an early indication (from the knowledge of AV), whether an aibag is to be fred fora pole or other nonfrontal barrier crashes VL Conctustons A new airbag contol algorithm is presented in this paper, This algorithm uses two different parameters to predict the Severity of the crash. The first parameter is the change in vehicle velocity, and the second parameter is the length ofthe filtered accelerometer signal, a(t), curve. The fist parameter Js used to determine whether or not a frontal-barrer crash ‘occurred in the deployment zone, The second parameter is used to determine whether or not the airbag i to be fred for pole ‘or other nonfrontal barrier crashes. We tested our algorithm by applying it on a number of crash testing data, Our algorithm fired the airhag for all the cates where the deployment of the airbag is necessary. In most of the eases our algorithm fred the airbag earlier than the requirement, except in one case where the firing occurred only 1 ms after the requirement Since every crash testing is very expensive, we could not apply ‘ur algorithm to many types of vehicles. However, our study will belp other researchers in designing their airbag contol algorithm by incorporating the facts which we presented inthis paper. We have also given some suggestions regarding how to design the front bumper, so that even a beter algorithm can be developed for the airbag contol [REFERENCES (0 G.. Nish nd R. Mota “Face deftones a, tesoneExerml Tce p38 cage cay SEP 1 and ak angry ig, ene of by min iprinentl Meco (3) Aan Rata. oak sion ch Sp tron: Enger pp. 2023, (a) Bn ng ln el ay. tate New, 28, Sem 38 nh 1s) HPSkin. Re Opt, F. Tala and M. Flas. “Sinsation and {sing fa abrir sie Antomate Bees. Frat lone oe 1a) B Coat, "Rituee” Poplar Mecha. p8, ne 191 TT] ten “Pomerat cath seas forse stati,” SAE (8) T. Gina, "A prediive bed alg for action of ag: 1p] DVS. Brod av Csi, “Teds in sensing foal impacts” SAE 50750 9p. 3-8, (10) RW Bile, "leone sensing of alomedie cash for ag dep Ine SAE S036 gp oot a1) BS Bred and ¥ Cla, “Probie in design and engineering of bag syste” SAE SATS, pe 130. (iz) DiS V Cont ed F Sloat Are ar rer alice for evauang a Sear pestormance”SAE 9084 Fy. 165-17 Syed Manu Mama (5'52-M'¥,forphtorph and biography, eth Ansa L Aleabady eeted 8S (Hom) epee {nents ad Computer engicerng or uy Unies of Since and Tecnology. Jot. in 1930 an MS. degre ie computer einen ‘fom Wayne Ste Unversy, eto: Min 1998 tas presen a PRD, candi, i 902 wa esearch air at Wayne Sse sient) dana wach te beware on pect ‘rom Ford Mate Company. Deut. Ml Since