ENGINEERING RESEARCH INSTITUTE THE UNIVERSITY OF MICHIGCAN ANN ARBOR Semiannual Progress Report No. 5 SHOCK ON ELECTRICAL COMPONENTS IN TRACK-LAYING AND WHEELED VEHICLES December 2, 1955 to May 31, 1956 H. S Bull Project 2145 DETROIT ARSENAL, DEPARTMENT OF THE ARMY CONTRACT NO. DA-20-089-ORD-.5654 CENTER LINE, MICHIGAN June 1956

-pa, j.).I -4 (lof, E4! I Scr # 1.9, 0 L 5 A-eD

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN FOREWORD This is the fifth semiannual progress report on a research program being carded on in the Department of Electrical Engineering of The University of Michigan, under the supervision of the author. Most of the work reported here represents the labor of Mr. Harris Olson, who has devoted all his time to this research. In addition, the project has benefited from the expert counsel of Professor Jesse Ormandroyd of the Department of Engineering Mechanics, Mr. John Riordan of the Data Reduction and Computation Service, and from the generous cooperation of the Chrysler Corp., the Electric Auto-Lite Co., the General Electric Co., the Tung-Sol Electric Co. the Westinghouse Electric Corp., and members of the SAE Impact-Tester Panel. ii

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN' TABLE CF COfEMINTS Page LIST OF TABLES iv LIST OF FIGURES v ABSTRACT viii OBJECT viii CONCLUSIONS AND RECOMMENDATIONS ix II ARSENAL-TYPE INPACT-TESTER CORRELATION STUDIES 1 A. Single-Filament Lamp Tests 1 B. Double-Filament Tests 11 C. Chi-Square Test 17 D. Testing for Socket Bias 19 E. Observation of Failures 21 F. Conclusions Drawn from Correlation Studies 21 II. -CCOTINT ED STUDY OF "T.E ROTAR_-DRUM TVACT ESTER 25 A.. Lamp-Holder Alterations 25 B. Theoretical Study of the Rotary-Drum TEester 28 III* STUDIES OF EXPERIMEN TAL IAMP DESI(GN 34 A. Experimental Lamp Tests 34 B. Conclusions of Experimental Iamp Tests 47 APPENDIX 49 BIBLIOGRAPHY 54 iii

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN LIS.. CJ' r TAS;3ES No. Page IT Computing Form for',eree-Way Classification Analysis of Variance 9 II. Summary of the Analysis of Varalnice for the Arsenalt., Auto-Lite, Cbhysler, General Electriz, and T<'ug-So l iMachi$es 10 III. Summary of the Analysis of Va!riance for the Arsernal, Cnzysler, General Electric, axnd cng-Sol Mac:ines, Usirg the Iat;a from the Single-Filament Tests 10 IV. Summary of Analysis of LVariance'Using the Second 1251 AutoLite Test and the Arsenals CLjrysle, General Electric, and Tung-Sol 1251 Data 11 V. SummarYy of Analysis of Variance for the Double-Filament 1034 Lamp (Six Machines) 15 VI. Summa;ry of Analysis of Variance for the DI)oble-Filament 1034 armp, Excluding the Wvestil rjghouze Machine 16 VII. Summary of Aal.ysis of Vsariarce for Five Arsena-.i-Type:mpact Testers (Excluding the We-st inghous e Macine) of the 1.034 Lamps for 10 Time Periods 17 VIII. Summarized Resul.ts of X2 Test for Tr'iaL vs -Mach.ine Average for the Five:1251 and i034 Tests of 4J5the Arsenal, Auto-Lite, Chrysler, General El-c tricm, ung-Sol, and Westingh-ouse Impact Testers (For 15 Deg:rees of Freedom;.5 = 24e996; o01 30.578) 18 IX. Summarized Results of X2 Test Comparing I;tae iMackine Totals with the Average of AUli. Six Macb.ines for the 1251 and 1034 Tests (For 15 Degrees of Freedom;.05 24.996; o01 = 30.578) 19 X. Summary of Analysis of Variance for the 1251 Lamp BEased on Classification of Defective Iamps by Machine and Trial (Arsenal, Auto-Lite, Cbuysler, General Electric, and TungSol Machines) 22 XI. Summary of Analysis of Variance for the 1034 Lamp Based on Classification of Defective Lamps by Machine and Trial (Arsenal, Auto-Lite, ChrysLer, General Electric, and TungSol Machines) 23 iv'

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN LIST OF FIG~URES Page No. 1. Mortality curves of type 1251 single-filament lamps tested on the Arsenal modified impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 2 2. Mortality curves of type 1251 single-filament lamps tested on the Auto-Lite modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 2 3. Mortality curves of type:1251 single-filament lamps tested on the Chrysler modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 3 4. Mortality curves of type 1251 single-filament lamps tested on the General Electric modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 3 5. Mortality curves of type 1251 single-filament lamps tested on the Tung-Sol modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 4 6. Mortality curves of type 1251 single-filament lamps tested on the Westinghouse modified impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 4 7. Comparative composite mortality curves of type 1251 singlefilament lamps tested on six different modified Arsenal-type impact testers. 5 8. Mortality curves of the type 1034 lamp (minor filament) tested on the Arsenal modified impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 12 9. Mortality curves of the type 1034 lamp (minor filament) tested on the Auto-Lite modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 12 10. Mortality curves of the type 1034 lamp (minor filament) tested on the Chrysler modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 13 v

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN LiSTU OF FiGUES (continued) No. Page 11. Mortality curves of the type 1034 lamp (minor filament) tested on the General Electric modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 13 12. Mortality curves of the type 1034 lamp (minor filament) tested on the Tung-Sol modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 14 13. Mortality curves of the type 1034 lamp (minor filament) tested on the Westinghouse modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 14 14. Comparative composite mortality curves of the type 1034 lamp (minor filament) tested on six different modified Arsenal-type impact testers. 15 15. The performance of two mortality tests using flat spring clips for lamp holders as compared to previous tests using similar lamps soldered to the lamp holders of the rotary-drum tester. 26 16. Modified spring clip holder. 26 11.7. Mortal'ity results of tests using revised fLat spring clips fora lamp holders as compared to previous tests using simila:r lamps,soldered to the lamp holders of thie rotary-drum tester. 28 18. A three candle power lamp with a tVee-segment filament mounted on rigid leads. 34 19. Comparative mortality results of special three candle power lamps and regular 1251 lamps from the same manufacturer. 36 20. Comparative mortality results of special three-candle-power and regular 1251 lamps evaluated on an Arsenal-type impact tester by the lamp manufacturer. 36 21. A typical cold failure near the support of a three candle power special lamp. 37 22. Modified 1251 lamp with double-anchored filaments. 38 vi

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN, LIST OF FIGT~JES (concluded) No. Page 23. Comparative mortality results of double-anchored 1251 lamps and regular 1251 lamps from the same source tested equally on the rotary-drum impact tester. 39 24. Typical failure of the special double-anchored 1251 lamps. 40 25. Modified 1251A lamp with an unsupported filament. 41 26. Modified 1251B lamp featuring a rigid support clamped to a twosegment gapped filament. 41 27. A lamp with four-wire, flange-stem construction mounted in a B-6 bulb with unsupported 1251 filaments. 41 28. Typical filament distortion of the modified 1251A lamp. 43 29. Comparative mortality results of modified 1251A and regular 1251 lamps of the same manufacturer, equally tested on the rotarydrum impact tester. 45 30. A typical faiiure of the modified 1251A lamps. 44 31. Comparative mortality results of modified 1251B and regular 1251 lamps from the same maafacturer, eqal ly tested on the rotarydrum impact tester. 45 32. A typical hot failure of a 12513 lamp after sexrvice on the rotary-drum. impact tester. 46 33. A typical cold failure of a 1251B lamp after service on t1he rotary-drzm impact tester. 46 34. Comparative mortality results of the 1251C and regular 1251 lamps from the same manufacturer, equally tested on the rotarydrum impact tester. 47 35. A typical 1251C lamp failure as the result of impact tests on the rotary-drum tester. 48..... vii

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN ABSTRACT This is the fifth semiannual progress report covering the period from December 2, 1955, to May 31, 1956. The accomplishments may be summarized as follows: 1. Studies of comparative performance have been completed for the six modified Arsenal-type impact testers. 2. Additional theoretical studies of lamp impact testing on the rotarydrum tester have been completed. 3. A new method of lamp mounting for the rotary-drum tester has been devised. Preliminary tests indicate fairly satisfactory results are to be expected from this spring-clip mounting. 4. A number of modified groups of type 1251 lamps have been impact tested with the results indicating evidence of life improvement, OCBJECT The objects of the research project are: 1. to determine practical means of increasing the operating life of incandescent lamps used on military vehicles, with particular reference to their resistance to mechanical shock and vibration; 2. to study the presently accepted method of impact testing vehicular lamps and to determine practicable means of improving the tester and correlating the results obtained by several testers now in service; and 3. to design, study, and evaluate a new and smaller impact tester which may possibly supplant the impact testers now being used. viii

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN CONCLUSIONS AND RECOMMENDATIONS 1. Studies of comparative performance have been completed for the six modified Arsenal-type impact testers and it is concluded that: a. Each machine tests lamps consistently, That is, each may be used to evaluate a particular group of lamps in comparison with a standard group already tested on the same machine, b. Each machine may have its data evaluated either on the basis of comparison with an arbitrary mortality curve or on the basis of the number of defective lamps in a sample group. c, Only the Arsenal, General Electric, and Tung-Sol machines are sufficiently alike in performance to permit complete interchange of test data. 2, Additional theoretical study has been given to the relationships of cam speed, cam offset, and velocity of impact on the rotary-drum tester. 3. A new spring-clip method for lamp mounting on the rotary-drum tester has been completed and found to offer some promise. Further tests will be undertaken to determine the exact potentiality of this type of mounting. 4. Several groups of modified 1251 lamps have been impact tested, some of which offer a definite physical advantage over the original design. Additional samples of the more promising modifications are to be tested in the laboratory and the field. ix

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN I. ARSENAL-TYPE IMPACT-TESTER CORRELATION STUDIES The SAE Impact-Tester Panel, in a meeting held at The University of Michigan, Novemiber, 1955, agreed that a series of correlation tests of the modified Arsenal-type impact testers would be desirable. Plans were made for Auto-Lite, Chrysler, General Electric, TungSol, Westinghouse, and the project to test an agreed number of single- and double-filament lamps on their respective tester. Each series of tests will be discussed separately and conclusions stated for the entire study. A. SINGLE-FILAMENT LAMP TESTS The single-filament 1251 lamps, made by TungSol, were distributed to the six interested laboratories after thoroughly mixing the homogeneous lot. The conditions specified for these tests were: 1. The 100 lamps will be divided into 5 groups of 20 lamps. 2. The testing cycle will be 25 minutes "on"' and 5 minutes "off" for 16 cycles (8 hours) or 75% failure, whichever occurs first. 3. The voltage supplied to the lamps will be 28 volts. 4. Minimum observation of these tests will be during the cold cycle and preferably once just before the cold cycle occurs. The results obtained are graphically illustrated for each tester by Figs. 1 to 6, and composite curves for all machines are shown in Fig. 7. The individual failure schedules are located in the Appendix. An analysis of variance was made of the data of these 1251 tests. The Westinghouse data were not available until after the analysis was completed, so consequently they are not included. The general handling of the data conformed to the description which follows. The machines were ordered alphabetically and numbered in series (i.e., Arsenal, l, Auto-Lite, 2,..., Westinghouse, 6). The 8-hour test period was divided into 16 equal periods numbered consecutively. Let Xijk be the number of lamps of the ith machine, in the jth test, which burn out in the 1

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN 10oo Act s. —- FIRST TEST',, < C 4- - — SECOND TEST -----— THIRD TEST a.h~~~~~~~~ 80 -- --— FOURTH TEST,.4 C \,,FIFTH TEST w,a,60 0 LI. 040O n20 O 2 O0 4 00 600 B8 0 96 O TIME IN MINUTES Fig. 1. Mrtality curves of type 1251 single-filament lamps tested on the Arsenal modified impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 100 I00K - --- FIRST TEST ----- - SECOND TEST -C —-— THIRD TEST <80 - -- — FOURTH TEST T IMFIFTH TEST'60 0.. O40 z Fig. 2. Mortality curves of type 1251 single-filament lamps tested on the Auto-Lite modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact;Tester Panel. 2

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN I00 ~...... -— FIRST TEST - M —SECOND TEST 100 > >- a —— I — THIRD TEST 0n80 -----— FOURTH TEST >60 0L2X'4 200 400 600 800 960 -- 0 cc r20 200 400 600 800 960 TIME IN MINUTES Fig. 3. Mortality curves of type 1251 single-filament lamps tested on the agreed procedure of the Impact-Tester Panel. I00 lO - -—.IRST TEST... -— SECOND TEST CO — ~~T....... — ---—.THIRD TEST a.80 -.. --- N-'- -—.FOURTH TEST 4 ~ FIFTH TEST 0 0.20 00 200 400 600 800 960 TIME IN MINUTES Fig. 4. Mortality curves of type 1251 single-filament lamps tested on the General Electric modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel......,,,,.....3

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN I00 FIRST TEST. \ - SECOND TEST -co THIRD TEST 280 --— FOURTH TEST j 80 | -... >'-"_4 __ FIFTH TE ST 60 0. 0 0 200 400 600 800 960 TIME IN MINUTES Fig. 5. Mortality curves of type 1251 single-filament lamps tested on the Tung-Sol modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Paneli 10O ~....~ - -> - - t —— FFIRST TEST - ---— ________. _- _ --- SECOND TEST ------ - THIRD TEST o0 - -— _ FOURTH TEST (NO FAILURES) _______ FIFTH TEST w >60....... 20 640......._ z ho 2(40 2~60 8. 96o TIME IN MINUTES I

I00 z:~~' —- ARSENAL ___-..-AUTO-LITE ---— CHRYSLER 0) 80 __~~~~~~~~~~ —GENERAL EETI _______TUNG-SOL. —-WESTINGHOUSE ~60Z w 0-C 040 I-~~~~~~~~~~~~~~~'' z wz 0 a-2C 0~~~~~~~~~~~~~~~~~ 0 200 400 600 8 TIME IN MINUTES Fig. 7. Comparative composite mortality curve of type 1251 single-filament lamps tested o modified Arsenal-type impact testers. ~ — M.......~~~~~~~~~~~~

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN kth time interval. (Note: In those tests where less than 20 lamps were used, each lamp was weighted accordingly; e.g,.if 18 lamps were used, each one was counted as 20= 1.1 lamps. Where a machine was not run for the full 8 hours, fictitious a8ta were supplied. Although such modifications were not extensive, when they were made, additional tests described in Section E indicated that they did not significantly affect the results.) If the Xijk were summed over one or more indices, the sum was denoted by Xijk with the index (or indices) replaced by a dot: e.g., Xijk Xik, XiJk = X.k i,j IXijk X i,j,k More specific examples are X1.2 = X112 + X122 + X132 + X142 + X152 - and XS.. = X511 + X512 +..o + X51(16) + X521 + X522 +.e + X52(16+) + X551 + X551 + X552 + + X55(16) = ZXSjk j,k We denoted the number of machines by Ni, of tests by Nj, and of time periods by Nk,-and tested by analysis of variance the following hypotheses: a. The machines are consistent; i.o,, repeated tests on the same machine give similar results. b. The machines are uniform; i.e., similar tests on different machines give similar results, c5 The time periods are uniform; i.eu, no time period favors the failure of lamps more than another. 6 6

ENGINEERING- RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN If these hypotheses are true, we can get several estimates for the variance of the population from which the data are drawn. We do this as follows: Let Li = Ni X. - X2 L = -X2 - Lk = 7 X2k - X2 Lij = Ni NL, Xj - x2 Li.k Ni Nk X2k - X.LJ Nj.. ijk kL Xijk - 0 and. Ii. = Li = Lj., Ik = ~ oL Iij = LiJ - L* Lj., Iik = Li.k Li.. L I.jk = L.Jk - 7OJk Iijk = LiJk - *, j. -,k - Iij. - Ii.k - Ijk It canl then be s1own that estima.tes for the variance are ~Ii.. Ij. ____I * _i - 1) - - ) Iij. Ii.k =1jk (N - )(N 1) (Nji -l)(Nk - 1) ( zijk The denominators of these fractions are ca:sled the degrees of freedoma of the est imat;esA

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN In addition it can be shown that the ratio of two estimates of a variance has a distribution depending only on the degrees of freedom of the estimates. That is, if a,, and a2 are two estimates of a variance, the probability of the ratio F = a2/a being greater than any value, t say, can be calculated if we know the degrees of freedom of *2 and add In fact, for certain combinations of degrees of freedom, the values t for w'ich thle probability that F = t is 5% or 1%?have been tabulated. The proofs off the above s4tatements are omitted because of their length and may be found in most statistical texts. The computations are put in a standard form, which is shown in Table i following that given in Dwyer's Linear Computations, pages 309-514. Table II summarizes the computations involved in this analysis of variance for the single-filament lamp tests of the Arsenal, Auto-Lite, Chrysler, General Electric, and Tung-Sol machines. A study of this analysis of variance reveals the truth of the hypothesis that each machine was consistent. That is, each tester repeatedly gave similar results in testing the 1251 lamps. T.e hypothesis that the failure rate is uniform for each time period is also true. For the data from the tests of the 1251 lamp, our analysis shows that we should reject the hypothesis that the machines are uxiiform, for the variance which,_ arises because of' the variatioo n between machines is significantly greater than the variance of ahl. t he data0 The ratio of these estimates for the variance is 5,7 (with. 4 and 240 degrees of freedom), which is significant at the 1% level. An inspection of t.he data leads us to believe that Machine 2 (Auto-Lite) is diffezrent fron the other.s. The summary of the analysis of variance performed on tLhe data from the other four machines as shown in Table III confirms thls suspicion. Auto-rLite was requested to repeat their 125. tests upon the completion of this analysis. Tlhese tests were operated the same as previously prescribed, except that each was terminated at the end of 360 minutes. These test results were then used for a second analysis of variance for all machines except the Westinghouse. The summary of this analysis, as shown in Table IV, still indicates that the machines are not uniform (at the 5% level). As was pointed out earlier, the Westinghouse data were not included in this analysis. It is quite evident from an inspection of the data that the results of the analysis would not hinave been influenced by their inclusion. That is, the significance of the analysis would not have been altered. 8

TABLE I COMPUTING FORM FOR THREE-WAY CLASSIFICATION ANALYSIS OF VARIANCE rT1 Z X111 X112 XL13........ X11(18) Xll. X121 X122 X123............. X12(18) X12. X131 X132 X............... X13(18) X13. X141 X142 X143.........(... X14(18s) X14. X151 X152 X153............. Xm5(1e) X15. X1.1 X1.2 X1.3............... Xi.(is) Xi.. X211 X212 X213 (21(18) (21. X221 X222 X223... X22(18) X22. X231 X232 X233...2(.. X23(18) 2(23. m X241 X22 X243..( X......24( 24. xi (24 X251 X252 X253 ~ "' ~... X25 (18) X25. X2.1 X2.2 X2.3.o ~~... X2.(18) X2.. X511 X512 X513....5.......... X1(16) 51. (X521 2(522 X523.2(O 0.0... X52(18) X52. Xs531 X532 X533.2 (........ X53(18) X(53 X541 X534 X534.......... X54(18) X54( 2(551 X535 X535 2(ss.e)0 (5 5s 2(5.1 (X5.2 X5.3........2(s.(ie) X5.. 2.11 X.12 X(13.20. X.i(i) X.i. X.21 X22 X23...2.......... X.2(16) X.2. X(31 X(.32 X(.33.2(... X-3(16) 3. m 2.41 2.42 X.43....... X.4(18) X.2*4 x.si X2.2 X.53............. X.(ie) X2.. X..j X..2 X. 3 ~,3~~. ~.0~ X.. (18 X.. 0 Source N X 12 L or I DF LorI/DF F Ni X... EX? L. Ni X? - 2.. Ni - a- i../Ni -1 a/ajk j Nj X... EX(j. L~ = Nj X4J. - X,.. Nj -1 = L.j./Nj - 1 k Nk X C... 2E.k L..k = Nk2.k 2-X..- = LN L..k/Nk.1 a-jk ij NNj X. 2 Iix. = NiNj Xj. - X.. - L... (N - 1)(Nj - 1) j = I./(N - )(N - ik NiNk s21 Exs1. Ii.k = NiNk2.k -.. - Li.. - L..k (N - 1)(Nk - 1) k = I.k(Ni - 1)(Nk - 1) k/jk jk NjNk X... Exk I.jk = NjNk k -... - L... - L..k (N - 1)(Nk - 1) Ok I.jk/(Nj - 1)(Nk - 1) k/Jk Z NNN x.2 I NxNN.....2..N.. 1)(N... 1)(Nk 1) x..1)N X. - X k. - X.s............... X. j(ki) Xk k. - - - X Lj.k X- Xi.s -i..k. - X.k

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN TABLE II SUMMARY OF THE ANALYSIS OF VARIANCE FOR T]E ARSENAL, AUTO-LITE, CHRYSLER, GENERAL ELECTRIC, AND TUNG-SOL MACHINES Source N X L or I DF L or I/D F Significance Level i 5 247.7 13574 6516 4 1629 5.7 1% j 5 247.7 12372 505 4 126 k 16 247.7 4159 258_ 15 346 1.2 _ ij 25 247*7 2879 3606 16 225 -- ik 80 247*7 1079 13301 60 222 jk 80 247.7 993 12432 60 _ 207 - - ijk 400 247.7 427 67854 240 282 (Figures rounded to integers) TABLE III SUMMARY OF THE ANALYSIS OF VARI.AN'CE FOR TEIE ARSENAL, CHRYSLER, (GENERAL ELECTRIC AND TUNG-SOL MACHINES, USING THE DATA FROM THE SI1NGLE-FILAENTT TESTS Source N X Z 2 L or I DF L or I/DF F Significance Level i 4 170.7 7645 1443 3 481 2.4 j 5 170-7 5922 472 4 118 - k 16 1170-7 2060 35826 15 255 1. - ij 20 170.7 1650 1953 12 163 - Jk 80 170.7 547 1036 6o 172 ik 64 170.7 634 6200 45 138 - _ _ ijk 320 170.7 280 36170 180 201 (Figures rounded to integers) 10

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN TABLE:N SUIMMARY OF ANALYIS OF VARXA~E E U7S'iLoT rTE SEC')M) 1251 ATO2 -LITE TEST AND THE ARSENALj, CIRkLER, S. GERAI E~CTREI'1 AN7uD TUONG-SOL 1251 DATA Source N X X' L or I DF L or I/DF F Significance Level i 5 163 5733 2227 4 557 2.98 5% j 5 163 5302 73 4 8 - k 12.L65 _20.374`L 325 1.74.ij 25 163 1248 2450.6 153 ik 60 163 674 8193 44 186 - - k 60 163. 636 8055 44 - -83 ijk 300 163 280 32893 176 187 (Figures rounded to integers) B. DOUWiLE-FILAMT' TESTS Thje double-filament tests were initiated upon the completion of the first series. A homogenous group of double-filament lamps, supplied by GeneraL Electric, was'niscu>lJ. xixed and disr,ib.t.ed t o thce participating laboratories, Threse laps were to e estested under the following conditions:.A Thle tester ih-ould conform as nearly as possible to the recommended modifications and pbyrsiCaL setting;ls wf!)ch.ic we-e dCesc-ilbed in a -letter to the panel members dated March. 16, 1955. 2. The 100 lamps will be divided in-to 5 groups of 20 lamps. 3- %T,.e lamps will be ori.enti;ed in their ho.lders with the axes of the filaments in a horizorti;al plane and 4 2Ae minor fii.smen-t shall be located in the upper position. 4. Thre testing cycle will be 20 minltes "on" for the minor filament, 5 minutes both filaments lighted, and 5 minutes both filaments "off." This 30-minute cycle will be repeated 15 times (8 hours). 5. Tlhe voltage applied to the lamps will be 14 volts. 6. These lamps shall be observed closely during the conbined cold cycle, and at regular intervals dting the hot cycles. ThEe data received from thet cooperating laboratories are illustrated by the mortality curves shown in Figs0 8 to 13, with all the data from each L1

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN tester combined in Fig. 14. The individual failure schedules for each test are included in the Appendix. 10O... --- FFIRST TEST't I - - --- - SECOND TEST a~\\8~N. ->-~~- -----— THIRD TEST cn~8C-' ----— FOURTH TEST FIFTH TEST >40 02 n,. 0 Iz, 2 60 20960 TIME IN MINUTES Fig. 8. Mortality curves of the type 1034 lamp (minor filament) tested on the Arsenauto-Lte modified impase-tpe icnt tester in accordance with the agreed procedure of the mImpact-Tester Panel. KX),-~~~~~~'''~'' "TEST ---— THIRD TEST u08 --— FOURrH TEST _______FIFTH TEST 0, TIME IN MINUTES Fig. 9. Mortality curves of the type 1034 lamp (minor filament) tested

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN 100 - -- FIRST TEST ------, SECOND TEST | —-l|| - THIRD TEST u)80. _' iF —-— OURTH TEST ___ FIFTH TEST >60 20 4C 0 TIME IN MINUTES Fig. 10. Mortality curves of the type 1034 lamp (minor filament) tested on the Chrysler modified Arsenal-type impact tester in accordance with the agreed procedure of the Trpact-Tester Panel. 100 --- FIRST TEST 10S -— SECOND TEST -----— THIRD TEST C0O60 _\ - _ ---- FOURTH TEST FIFTH TEST >6o0 X....... a 0 0 TIME IN MINUTES Fig. 11. Mortality curves of the type 1034 lamp (minor filament) tested on the General Electric muodified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel. 13

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN KO-FIRST TEST —'SECOND TEST \Nk A\. —~...~ ~~THIRD TEST co)80 FO - U ---- FOURTH TEST __ __ __ FIFTH TEST >60' 4 0. 0 6(t 0 40 0`) 600 o8o 960 TIME IN MINUTES Fig. 12. Mortality curves of the type 1034 lamp (minor filament) tested on the Tung-Sol modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-mester Panel. --- FIRST TEST "- -- SECOND TEST ------- — THIRD TEST Ma8t —. —FOURTH TEST >60 \\ ----------— FIFTH TEST a. Ok 2 (6) 4(0 60"_ 8 0 0 9t: TIME IN MINUTES Fig. 13. Mortality curves of the type 1034 lamp (minor filament) tested on the Westinghouse modified Arsenal-type impact tester in accordance with the agreed procedure of the Impact-Tester Panel.......... 14.

ENGINEERING. RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN -— AR-ARSENAL ~\ -- - -— A AUTO-LITE --------- CHRYSLER cm80 "' " -., = —---— GENERAL ELECTRIC ----- TUNG- SOL |_ - W ESTINGHOUSE,_60 --— _.I o0. 0 " ".. 4tes0.... z C, TIME IN MINUTES Fig. 14. Comparative composite mortality curves of the type 1034 lamp testers. A similar analysis of variance was made of the collected doublefilament data. It should be noted tbhat only the minor-filament data were treated, since only a very nominal number of major filaments failed during any of the tests. Table V summarizes the analysis of variance for the 1034 lamp. TABI;E V SUMMARY OF ANALYSIS OF VARIANCE FOR.THE DOCUL3LBE-F LAM7NT 1034 LAMP (SIX MACHINES) Source N X X2 L or I DF L or/F F Significance Level i 6 274 13939 8558 5 1712 5.04 1% j 5 274 15314 1494 4 374 - k 16 274 5612 14716 15 981 2089 _ ij 30 274 2946 3252 20 163 - ---- ik 96 274 1366 32786 75 437 1.29 ak 80 274 1389 19834 60 1 - iJk 480 274 537 102044 300 340 (Figures rounded to integers) 15.

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN Tihne hLypothesis that the mac sines are consistent still holds true for the 1034 tests. The tests show: however, that the time intervals were not uniform, for reasons that are not clear. Th.e most probable cause was inaccurate reporting of data. The hypothesis that the six machintes are uniform should be rejected. Here again the variance which arises because of variation between machines, is significantly greater than the variance of aLl the data. The ratio of these two estimates for the variance is 5C04 (with 5 and 300 degrees of freedom), which is significant at the 1% level. If the Westinghouse machine (No. 6) is excluded, we cannot reject the hypothLesis that the machines are uniform as is shown in Table VI. This indica-e that tt he Westinghouse machine is at variance with thLe other machines. However, it should be noted that the variance due to between-machine differences is very high, i.e., F = 2.34 with 4 and 240 degrees of freedom. TABLE VISUMMARY OF ANALYSIS OF VARIANCE FOR THE DOUB-LE-FTAMIINT 1034 LAMP, EXCLUDING THE WEST1INGHOTYSE MACHINE Source N X X2 L orI DF L or I/DF F Significance Level i 5 255 13572 2804 4 701 2.34 J 5 255 13243.161 4 290 go k 16 255 4952 14L 5 316 - 2 ij 25 255 2858 2443 16 1.53 ik 80 255 1304 22283 60 371 1.24 -l jk 80 255 1278 21877 6_ l_.,22 | ik 400 255 o504 71842 240 299 (Cfigures rounded to integers) Additional analyses of variaace were performed to cornf'i.m the fact that the supplied fictitious data did not affect the result. (Note: One Chrysler test was discontinued af-ter 300 minat-es.) Tabole VII summarizes the result of this analysis where only ten time periods were used for all machines except the Westinghouse, which was not included. The F value due to lbetween-nmachinet" -variance is significant at the 5% level when taken at 10 time periods. T. his might be expected since four of the machines were quite closely related at the end of ten time periods and the previous F value was so high. The statements relative to the other two hypotheses still remain the same......... 16.......

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN TABLE VII SUMMARY OF ANALYSIS. OF VARIANCE FOR FIVE ARSENAL-TYPE IMPACT TESTERS (EXCLUDING THE WESTINGHOUSE MACHINE) OF TLHE 1034 LAMPS FOR 10 TIME PERIODS Source N EX X2 L or I DF L or I/DF F Significance Level i 5 166 5966 2252 4 563 2.85 5% j 5 166 5643 638 4 159 k 10 166 3595 8372 9 930 4,.72 ij 25 166 1309 2268 16 142 - ik 50 166 908 7224 36 201 jk 50 1166 86L 6583 36 1835 - ijk 250 166 333 28412 144 197 - (Figures rounded to integers) C. X2 (CHI-SQUARE) TEST To test further the hypotheses about these machines, an alternate test was used. The Chi-square test was used to determine whether frequencies in the sample differed significantly from frequencies predicted by a hypothesis about the parent population. If to each observed frequency O there corresponds an expected frequency E, we calculate x2 - 2' the sum being taken over all observed frequencies. It can be shown that this sum has a distribution depending only on the nuniber N of (independent) observations. In fact, calling N-1 the number of degrees of freedom of X2, we can find from tables, for this number of degrees of freedom, the probability that X2 will have a value as large as any given value. If we calculate X2 from an actual sample and with a hypothesized frequency distribution, and tile probability is small that X2 would take on as large a value as we calculate (e.g.,.05 or.01), we reject the hypothesized distribution. To test the consistency of each lamp tester, we compared (by using x2 test) the frequency distribution of lamp failures for each test with the average frequency distribution for the five tests on that machine. That is, 17

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN for each. j we computed (2 = I F(Xijk - /5Xi k?1 k L 1/5 Xik J This follows a (2 distribution with 15 degrees of freedonm The results of the test are srzmi:,ized in Tabi e VIII. The machines proved to be consistent in bothl;' the silgi.e- and doubl.e-filalent tests. TABLE VIIr SUMARIZED RESULTS OF X2 TEST FGR. 1ERI;AL VS MACH:rNE AViEERAGE FOR TKE FIVE 1251 AND 1034 TiESTS OF ri11:E ARSENAL., AU'IT1-,ITE~E, CHERYSLEIR, GENERAL ELECTRIC, r7ANG-S ND, D WTEWSGTINCHGOUE IMPACT T.ESTERS (FOR 15 DEGREES OF FREEDOM;.05 = 24996; 01 = 30.578)'- Trial vs Miiachine. Ave-rage ITrial vs Machine Average 125-' Tes i.Lf34 Test MIachine 1 2 3 4 5 1 2 3 4 5 1.7 1-4 2.L4 k14 7 12 13 12 2 11 10 1.7 11. L4! z 13 16 6 3 11 7 12 22 9 9 4 14 9 10 4 12 23 25(24.7) 9.6 14 13 17 19 14 5 10 11- 12 15 5 12 ii 8 8 7 6 7 i 3 6 4 6 3 2 8 To tes't the uniformity of macLine.i we compared t1he frequency distribution of lamp failures for each macaine (all t-ests on. that machine) with the average of these for the six mac-hines Th.~at is, we computed aX == 2 [ $ which again has aX2 distribution with i5 degrees of freedom, In testing the cantly from the average In testing t~he double-fi:ament lamp, Machines 3 and 6 deviated significantly from the average. These results confirm our suspicion that Machine 2 gave significantly different results with the 1251. lamp, as did Machine 6 with the 1034 lamp. In addition, the~2 test shows that Machine 3 is significantly different with the 1034 lamp. This might be expected since our analysis of variance had a very high F number.

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN TABIL IX SUMMARIZED RESULTS OF X2 TEST COIPARING THE MACHiNE TOTALS WIET THE AVERAGE OF ALL SIX MACHIES FOR T'HE 1251 AND 1034 TESTS (FOR 15 DEGREES OF FREEDOM; oo5 = 24.996;.ol = 30.578) Machine Totals vs Average Level of Level of Machine 1251 Significance 1034 Si ficance 1 8 -- 9 2 51 1 13 - 3 17 44 4 11 22 -- 5 12 17 6 35 18 52 D. TESTING FOR SOCKET BIAS There has always been some question as to whether or not the Arsenal-type impact tester has treated every socket equally. Inspection of some scattered test data seems to indicate taat some sockets were biased, but there were never enough data available to confirm thiis observation fully. Socket bias was tested from two different points of view. The hypothesis that each socket treated. lamps snimilarly was tested first and then a test was made to determine if each row had proportional numbers of failure. Thne hypothesis that the failures were randomly distributed over the sockets was tested by the X test. bThe test is given by Z ot! e - in which Ot = number of times lamp t failed and Et = expected number of failures in a given socket =. 20 1 Z19

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN Machine 1 2 3 4 56 X2 (1251 lamp) 13 7 7 9 23 I (L034 lamp) 1L 11 15 17 14 22 For 19 degrees of freedom.05 = 30.144,.02 = 33.687, and.01 = 36.191. From the X2 test on individual sockets we are unable to conclude that there is socket bias. However, due to lack of sufficient data we cannot conclude that the failures were random. The 1251 data for the No. 6 machine were unavailable when this calculation was performed, but probably would have given an inconclusive result with so few failures. The equality of failure by rows was tested by performing X2 tests on the hypothesis that 7/20 of the lamps which fail are in the first row (i.e., sockets 1-7), 6/20 in the second row (i.e., sockets 8-13) and 7/20 in the third row (i.e., sockets 14-20). Again, ) =2 7=(0s - ES )2 where Os number of lamps in sth row that failed, Es 7/20 (Rows 1 and 3), and Es = 6/20 (Row 2), and with 2 degrees of freedom. Machine 1 2 4 5 6 all machines 2 (1251 lamp) 1.45 1.8.05 1.1 2.6.98 A. (10O4 1ar) 2,.82,.44 5.57 2,23.97 - 11.33 For 2 degrees of freedom.05 = 5.991,.02 = 7.824, and.01 = 9.210. The hypothesis may be accepted that the failures were uniform in each row for the 1251 tests. This same hypothesis must be rejected, however, for the 1054 lamps since one value is significant at the 2% level. --. 20

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN When the data from all the machines are combined, there is a definite bias which leads to a rejection of the hypothesis that the rows are uniform in testing the 1034 lamp. Thus, of the 267 lamps which failed, 119 were in the first row, 63 in the second, and 85 in the third. This gives a x2 value of 11.33, which is significant at the 1% level. E. OBSERVATION OF FAILURES It would be convenient in making impact tests to observe only the number of failures at the end of a predetermined time interval instead of making regular observations during the test. A hypothesis to this effect was tested with the available 1251 and 1034 data. A lamp was called defective if it failed in 12 or fewer cycles (i.e, 6 hours), and the data were then classified according to defective and nondefective lamps. An analysis of variance was made of the data and summnarized for the 1251 lamp in Table X and for the 1034 lazmp in Table XI. The 1251 analysis indicates that the hypothesis that the machines are uniform must be rejected. The analysis of the 1034 tests also leads us to reject the hypothesis of machine uniformity. If the 1034 data from the No. 6 machine are included in the analysis, the significance is at the 1% level. The 1034 results are variant from the first analysis of the data, but it must be recalled that the F number was very close to.05 significance. These results are substantially in agreement with the other analysis of variance and the X2 test. This would indicate that only one observation for failures would be necessary for some types of impact tests. F CONCLUSIONS DRAWN FROM CORREIATION STUDIES It has been found that each machine tests lamps consistently. This indicates each machine may be used to test lamps against a standard lamp that has been tested on that machine* Fairly frequent checks should be made to determine if this consistency still holds. These tests may be made either on the basis of comparison with a mortality curve or by observing the number of "defective" lamps in the sample. (The decision must be made on the basis of what qualities are desired in the lamp.) 21

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN TABLE X SUMMARY OF ANALYSIS OF VARIANCE FOR T.3E 12512 M BPEASED ON CLASSIFICATION COF DFECTIVE;LAMPS BY:MACRE AND -TRIAL (ARSENAL, AUTO-LsTE, CHRYSLER, GEERAL ELECTRICS At) TUING-SOL MACHIINES) Machine Trial 1.2 43. 1 X1 X12 X14 X15 Xi. 2 X21 X22 x2s X24 X25 - X2, 3 X31 X32 X33 X34 X35 Xs 4 X41 X42 X43 X44 X45 X4. 5 X51 2 X X54 X55 X5 X.1 X.2 X.3 X.4 X.5 X L Ic ~~j Lij Li. Iij Lij Machine Trial 1 2 5 4. 1 6 15 9 4 7 47 2 5 15 5 8 3 36 3 6.6 14 9 4 4 37.6 4 4 9 11 1 7 32 5 8 12 7 5 3 35 29.6 65 41 22 24 181.6 7842 6232 66O2 1586 395.t __..3..,6660._~urce.....L r I Level of Source N 1X X2 L L.or.I:DF DF F Signiflicance Machine 5 181L6 782 62-32 6232 14 i558 63 1% Trial 5 181.6 6602 33 33 4 8 Interaction 25 181.6 1586 6660 395 16 25 (Machine x Trial.) _ 22 522

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN TABLE XI SUMMARY OF ANALYSIS OF VARIANCE FOR TCB 1034 LAP BASED ON CLASSIFICATION OF DMEECTIVE LAMPS BY MACEINE AND,) TRIAL (ARSENAL, AUTO-LITE, CHRYSLER, GCEFIAL ELECTRIC, AND TUNG-SOL ACHINES) Machine Trial 1 2 3 4 1 8 2 12 4 8 4 2 12.1 6 12 5 2 37. -3 8 12 9 lo 5 44 4 12 1 1 8 6 48 5 10 8 9.6 4 2,4 34 50.1 39 53.6 31 23.4 197.1 8413 3214 7928 1825 2777. _ 94 2777 6785 L or I Level of Source N j X E x2 L L or I DF DF F Significance Machine 5 197 8413 3214 3214 4 804 4.6 5% Trial 5 197 7928 794 794 4 198 Interaction 25 197 1825 6785 2777 16 174 _- There is disagreement between the 1251 and 1034 tests as to the uniformity of failures for time periods. No definite reason can be assigned to this nonuniformity. Further mortality tests, if of a favorable nature, could eliminate this nonuniformity. The machines as a group are not uniform, i.e., similar tests on alternate machines do not give similar results. ThEis is demonstrated by the fact that a different machine was at odds with the group in both the 1251 and the 1034 tests. It is demonstrated that the Arsenal, General Electric, and Tung-Sol testers are uniform and can be used interchangeably with one another. It is also shown that the Auto-Lite, Chrysler, and Westinghouse testers can not be used interchangeably. It is possible that after further tests the Chrysler and Westinghouse machines could be used interchangeably with the others if some handicap or weighting were given to their results. The AutoLite machine, on the basis of a second 1251 test, is still significantly di.fferent from the group of five machines. The sockets are not biased relative to the location of failures, but the rows of sockets do have a definite bias. These results do not appear to 25

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN have much significance since tae total nuiber of lamps involved per socket is not great. These results lead to a recommendation that the modified Arsenaltype impact testers be used only for testing lamps against a standard lamp that has been previously tested on tie same maceine.

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN II. CONTINUED STUDY OF TIE ROTARY-DRUM IMPACT TESTER A. LAMP-HOLDER ALTERATIONS In the fourth semiannual report for this project, a method was described for holding test lamps on the rotary-drum tester. Briefly, this method employed a soldered connection between the lamp base and a supporting strip attached to a bearing. Electrical connections were established by flexible lead wires attached to their respective points. This method gave a positive and reliable mounting assembly, but it was felt that it would be too time consuming and involved and would not be practicable with aluminum bases. With these defects in mind, a search was undertaken for a new holding device. Most mechanical fastening methods (e.g., clamps and sockets) had been previously ruled out either because of the mass involved or their inability to hold the lamp firmly. Flat spring clips seemed to be the most likely type of fastening to be investigated. To determine the feasibility of a spring clip fastening, a Tinnerman spring clip (Part No. C3278-017) was modified and soldered onto the regular pivoted leaf. The lamp base was then inserted just far enough into the clip to place it in the same position it would have if soldered to the leaf. Figure 15 shows the results of two mortality tests, using these holders, as compared to a previous range established with soldered holders. The results seemed encouraging enough to continue the exploration of this method. Consequently, a holder was designed that would incorporate both the arm and the spring clip, and a sample is shown in Fig. 16. Pendulum-swing tests of this assembly produced values comparable to those previously reported for the original type of holder, as shown in the following table (page 27). Two mortality tests were run using these spring clip holders with the results given in Fig. 17. It will be evident that their performance compares favorably with the earlier tests of Fig. 15, and there appears to be good justification for continued study of this type of holder. 25

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN I00 TEST NO. 105 TEST NO. 106 u)80 >o2 0 a. 20 0 200 3 400 600 800 960 TIME IN MINUTES Fig. 15. The performance of two mortality tests using flat spring clips for lamp holders as compared to previous tests using similar lamps soldered to the lamp holders of the rotary-drum tester. Fig. 16. Modified spring clip holder.'' 26

Im z W F Cycles/ X K I lk 4X Velocity t As sembly grams grams sec inches inches inches in./sec sec m Soldered G-6 9.62 4.08 2.32 0.848 1.242 2 1.100 1.047 7.27 0.017 brass-base lamp Spring clip, m brass-base lamp 11.21 5.48 2.43 0.981 1.277 2 1.207 1.1 7.64 0.015 m Spring clip, S aluminum-base lamp 9.8 4.6 2,26 0.939 1.335 2 1.052 1.025 7.17 0.018 ro C m I =.distance from pivot to impact contact X = distance from pivot to center' of gravity W = weight of assembly F = force to hold assembly in horizontal position K = radius of gyration of assembly around pivot z

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN I00 TEST NO. 126 TIME___________ --- TEST NO. 127 080. B",,,/IPREVIOUS TEST RANGE ~60 0. 040 0 /C: Previous eeren data (pag 19 fourth sem6 report )60 TIME IN MINUTES Fig. 17. Mortality results of tests using revised flat spring clips for to the lamp holders of the rotary-drum tester. B. THEORETICAL STIJDY OF THE ROTARY-DRUM TESTER indicated a definite relationship between cam offset arl/or speed and lamp mortality. Further theoretical analysis hai now been completed which investigated the factors affecting the total relative velocity of impact. AX= ZA, Y28-Y= 28

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN x2 + y2 = r2 But x = X y =Y + — 2 X2 h2 2 h2 2 - 2 X2 - hX + h —+ + hY + h r2 x2+ y2 _h (X -Y) r2 h 2 Let @ be measured from the Y axis; then X = p sin Q Y = p cos 9 p2 sin2 + p2 cos2 - h (sin 0 - cos ) p = r2 h 2 p2 _ h (sin - cos ) p 2 - = p = r -(sin @ + cos @)2 (cos -sin) 8r 2 h p i r - (cos - sin @) p2 +h (cos 9 - sin @) p - (2 2) = 2p. ~+ h (-sin - cos ) p + h (cos 9 - sin 9) d.L h (s F+,, h (-sin h +cos ) 29

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN sin G + cos h do~3e~h h di 2 1 + J r (cos @ - sin G)J Compared with 1, 00 J [2- (cos - sin 0 1r p 2 (sin @ + cos G) dG 2 Slope at G = dp _- 1 h (sin @ + cos 9) rd@ 2 r dt dO dt dG 2 A silmmary of the results is: p 5 r - 2 (cos @ - sin @) 2'do h (sin G + cos O) Slope at @ is.p 1 h (sin G + cos O) rdQ 2 r dp _d dQ = ap. dt da dt dO hz (sin G + cos O) Let y be measured from center of rotation, x be measured from drop off:

- ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN. x. rG Yo = _ 2 io = n) y = -g Y - gt +2 1 k2 ~ h Y = gt +-t + r + - 2 2 2 X 0= I x =rwit whent t1 x=x1 t l = l.at tmne of contact rw = + 2 hw _Xkr h 2 r2cW2 2 rc+ 2 But x = rG and xl = r=G 1 gG h Y1 = + + r +- 2 W 2 2 But Y1 -P (GI).+2 h9z_ h h J2 W2 2 + 2 - 2 (cos %1 - sin 01) 2,2 2 2 2 This position of impact, I,. is calculated from cos Q1 - sin @ = 2 -01 - 1

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN /400 300 200 100.1.2.3.4.5.6.7.8.9 e, (RADIANS) Position of impact, 91 (radians), for different can offsets and speeds. S'. = -gr- + 2 jYl =.- e+ _ downward,~i 2 Magnitude of Sri downward = g/w (@1) - hm/2; the radial velocity of the point of impact upward is hw - (sin g1 + cos G1) upward the relative radial velocity is the sum of the two. Vrel. impact = o - + h- (sin 91 + cos O1) Li) 2 2 This is not all of the relative velocity of impact since the lamp envelope strikes the roller tangentially and the speed of the roller adds to the relative velocity of impact. 32

,ENGINEERING RESEARCH INSTI:TUTE U:NIVERSITY;'OF MICHIGAN DIRECTION OF TOTAL IMPACT yRELATIVE RADIAL VELOCITY \ahh /V lNO,-tCOS5 -jJ \fRELATIVC TANGENTIAL VELOCITY r.'fANGENT TO DRUb' SURFACE DRUM SURFACE Velocities here are velocities of lamp envelope relative to the roller surface. Total relative velocity of impact: (g + [sin 1 + cos 1 - 1 cos + rau sin r p = 2 [sin 91 + cos Q1] r dQ 2 r cos,; 1 sin h [sin G1 + cos O1] 2 r hw Vimpact'= a 1 +hzo (sin @1 + cos 91) - In this equation Q1 is calculated from the relationship cos Q1 - sin =1 = k - 1 -12 33

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN T'he relationship of cam offset, cam speed, and relative velocity of impact is graphically shown by the;Iccmlrpanying figure. It can be noted that the relative velocity of impact is significantly ihanged when the cam offset and/or cam speed is altered. These curves support previous experimental findJ.:;gJ of this relationship which were reported in the last semiannual report. It is quite evident that the velocity of impact of the lamp can be increased to the point where the filament will fail by creep instead of fatigue. The borderline where this phenomenon takes place is apparently at or near 8 inches/ second and probably only can be determined experimentally. 11.0 10.0 IO.O. -J34a. > 9.0 00 z 0o a. 8.0. z E 7.0 h =0-.46 -J 6.0 5.0_ 150 200 250 300 350 R.P.M. Relative velocity of impact of lamp envelope for different cam off-sets and speeds. 33a

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN III. STUDIES OF EXPERIMENTAL LAMP DESIGN A. EXPERIMENTAL LAMP TESTS The 1251 lamp is one of the weaker lamps used in large quantities by Army Ordnance. Since this lamp is expected to operate under severe conditions, considerable effort has been expended in recent months on ways and means of strengthening it. The more frequently observed types of failure for this lamp were listed on page 23 of the fourth semiannual report for this project, where it was noted that entanglement of the filament segments was perhaps the most serious fault. These observations were transmitted to representatives of the cooperating lamp companies during several conferences, leading ultimately to several different attacks on the problem of strengthening this lamp. The first modified design is shown in Fig. 18, the important characteristics of which are listed below: Base - single contact bayonet Bulb - S-8 Filament - three-segment gapped filament clamped to four nickel lead wires Fig. 18. A three candle power lamp with a three-segment filament mounted on rigid leads. 34

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN The following was noted of the filament structure: Lead-wire diameter 0.022" Lead length to point of filament attachment 21/64" Support length to point of filament attachment 9/16"? Lead and support tip spacing 11/64 ~o0 —i/4" —1 0 Filament wire diameter (approximately) 0.001" Coil diameter (approximately) 0.010" Number of turns per segment (approximately) 69 Total number of turns (approximately) 207 Typical filament arrangement Current (at 28.0 v).236 amp Natural frequency of a filament segment 480-560 cps Natural frequency of support (average) 1400 cps Natural frequency of lead (average) 2100 cps The initial inspection showed no evidence of damage from shipment. All lamps appeared very uniform in filament structure. The lot was divided into two equal groups for impact testing on the rotary-drum tester and the filament were oriented such that the two supports were beneath the current-carrying leads. The results of these two tests can be observed in Fig. 19 where the mortality curves are superimposed upon a comparative range of regular type 1251 lamps that were tested under similar conditions. It can be noted that there is no overall life improvement of these sample lamps as compared to the original design. The manufacturer evaluated a group of these special lamps on their own impact tester and confirmed these results, as clearly noted in Fig. 20. The failure schedules from both series of tests indicated that better than 90% of the failures occurred during the cold cycle. About 2/3 of these failures occurred at or near the longer supports. Evidently the longer supports with their lower natural frequencies were a critical feature of the design. 35

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN FIRST TEST c - — SECOND TEST cL~o~ ~~~ ~~~PREV IOUS RANGE _60..... a.0. 0 I0 C 2 640 8Oo 960 TIME IN MINUTES Fig. 19. Comparative mortality results of special three candle power lamps and regular 1251 lamps from the same manufacturer. K)O SPECIAL FIRST TEST — _ \ \lSECOND TEST uO REGULAR cr80. REGULARFIRST TEST ~k _ _____SECOND TEST 60 C. 4o I I........ I. 20'v_ 0( 2' ~'2 4 0 6 0 8(' 90O TIME IN MINUTES Fig. 20. Comparative mortality results of special three candle power and regular 1251 lamps evaluated on an Arsenal-type impact tester by the lamp manufacturer.

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN A large number of failures occurred in the coiled section, as is shown in Fig. 21. Also, it was not unusual to find fractures midway in the coiled section. These fractures at random points indicate the possibility that slippage occurred between grains that may have grown to occupy the entire cross-sectional area of the wire. Figure 21 also illustrates a typical manner in which the filament segments survived the tests without appreciable distortion or sagging and consequent entanglement. Fig. 21. A typical cold failure near the support of a three candle power special lamp. The conclusions to be drawn from these tests are. 1. The overall life of these lamps when subjected to shock and vibration has not been increased when compared to regular lamps. 2. Some types of filament failure can be either reduced or eliminated by this filament arrangement. 3. A number of unexplained failures have occurred at random locations in the filament. This modified design does seem less prone to distortion and shorting of the filament segments. This gain alone seems important enough to warrant further experimentation with the arrangement. If further work is to be done, it is suggested that 1. the current-carrying leads and the supports be interchanged, and 2. all supporting members reduced to a minimum length.

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN The manufacturer has indicated that the project will be supplied with a number of additional samples incorporating these suggestions. A second group of type 1251 lamps with an altered construction was received early in April from a cooperating lamp maker. A double-anchored filament was the only deviation of this sample lot from the regular lamp. It was felt these two anchors (i.e., two supports) would strengthen the lamp by 1. dividing the filament into three segments, thereby raising the resonent frequency of each segment, 2. preventing the entanglement of the filament segments by more positive separation methods, and 3. eliminating possible escape of the filament from the anchor by better control of the entire coil. The lamp illustrated in Fig. 22 is typical of this lot of lamps. The characteristics of these lamps were: Base - single contact bayonet Bulb - G-6 Filament - 2C - 2F.j.?::.:;:i:.i::::.:i:::':;:~,::.::::::!:~!':::::?::............. Fig. 22. Modified 1251 lamp with double-anchored filaments. It was noted from an inspection of these lamps that they appeared similar to a regular 1251, except as previously noted, and that the two mounts were quite widely separated. This separation was enough (about 1/8 inch) to isolate completely each mount from the other. It was also noted that the an-: g p~''':'i''0g'jt':g:00't f i: i::4;' t;0''i3'"i 8i; igi S:: i

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN to excite a coil segment at its resonant frequentcy. The lot of lamps was divided into three equal groups of 19 lamps each for evaluation on the rotary-drum tester, and the test results are shown in Fig. 23. -— TFIRST TEST -- — SECOND TEST ----— THIRD TEST 80.. aC Fi. AVERAGE..J3 ~/////~~PPREV IOUS RANGE I>I 4o z C j~200 42 0 60 8 ~0060 TIME IN MINUTES Fig. 23. Comparative mortality results of double-anchored 1251 lamps and regular 1251 lamps from the same source tested equally on the rotary-drum impact tester. The mortality curves are compared to a previous range established by a series of tests on the rotary-drum tester using 12-51 lamps supplied by the same manufacturer, and it is quite evident that the life expectancy of these lamps has been increased by a factor of from 2 to 3. The failure schedules indicate almost a complete absence of hot failures during these tests. Undoubtedly the two anchors controlled the filaments so effectively that no shorting of the filaments could occur. Although no major entanglements occurred, in many cases the coil was stretched and distorted near the supports so that some individual turns did short. About 751 of the failures occurred at or very near the anchor, as shown in Fig. 24, which suggests a sawing action of the anchor on the filament. The remaining fractures were located centrally on one of the three segments. The manufacturer evaluated a number of these lamps on his Arsenaltype impact tester, using a procedure similar to that outlined for the SAE single-filament tests. The results of their mortality tests apparently confirm our findings. There was also agreement in the two series of tests as to the character and location of the observed fractures. 39

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN Fig. 24. Typical failure of the special double-anchored 1251 lamps. It might be mentioned that a number of the test-lamp envelopes blackened to a considerable degree. Undoubtedly this defect would be completely eliminated if this lamp were produced in commercial quantities. This special lot of 1251 lamps has exhibited marked improvement over the regular lamp produced by the same manufacturer when evallated in terms of shock and vibration, and the supplier has been urged to explore further the possibilities of this double-anchored construction. A shipment of a similar lot is expected soon. A conference with representatives of a second cooperating lamp company produced another series of modified 1251 lamps, which, it was hoped, would increase the resistance of the lamp to shock. Modified samples of this lamp were supplied to the project for evaluation and were as follows: 1251A-a lamp like the present 1251, except without support wires and with widest possible lead tip spacing (illustrated by Fig. 25). 1251B-a lamp like the present 1251, except with a rigid copper support clamped to a two-segment gapped filament (sample illustrated by Fig, 26). 1251C-a lamp with a four-wire, flange-stem construction with unsupported 1251 filaments, mounted in a B-6 bulb (sample illustrated by Fig. 27). Sample lamps 1251A and 1251C were received late in February and the 1251B lamps were received several weeks later. Initially these sample lots were inspected and found to be in good condition with no apparent damage from l-40

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN Fig. 25. Modified 1251A lamp with an unsupported filament. Fig. 26. Modified 1251B lamp featuring a rigid support clamped to a two-segment gapped. filament. Fig. 27. A lamp with four-wire, flange-stem construction mounted in a B-6 bulb with u~nsupported 1251 filaments. 4...............1...........

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN shipment. It was also noted that the lamps in the A and C lots appeared to be very uniform in structure while the lamps in the B lot had notable differences in their mount structures. The following was noted of the filament structure: l251A L251B 1251C 0.016" and Lead-wire diameter 0.0101" 0.010" 0.0181 Lead length to point of filament attachment 0.14" 0. 09" 0.2511 Lead tip spacing 0.20" 0.15" 0.371" Filament-wire diameter (approximately) 0.001" 0.001" 0.0011 Filament-coil diameter 0.0081 0.008". 008" Number of turns per filament segment (approximately) 116 69 115 Total number of turns per lamp (approximately) 232 276 230 Barrel length (approximately) 0.27" 0.18" 0.291" Natural frequency of filament segments (cps) 480-530 490o-700 550-600 Current (at 28 v) 0.243 amp 0.222 amp 0.243 amp The luminous intensity of these three modified lamps was compared to that of a regular 1251 lamp of the same manufacturer. It was noted that the luminous intensity of the regular lamp was greater than that of the 1251A, the 1251B, and the 1251C, in that respective order. The impact-test procedure for each test followed that outlined in the Appendix unless otherwise noted. The 1251A lamps were the first to be subjected to impact tests. A1most immediately the hot filaments became distorted as shown in Fig. 28, but after the first few moments little or no further distortion took place. The results of these four tests are shown in Fig. 29. The four mortality curves are compared to a previous range established by a series of tests 42

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN,0W v Fig. 28. Typical filament distortion of the modified 1251A lamp. -- FIRST TEST -\ --— SECOND TEST \. — THIRD TEST oi80 --— FOURTH TEST AVERAGE X<-L —-_ PREVIOUS RANGE >_60 L40 06 0 8.... TIME IN MINUTES Fig. 29. Comparative mortality results of modified 1251A and regular 1251 lamps of the same manufacturer, equally tested on the rotary-drum impact tester. 433

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN on the rotary-drum tester, using type 1251 lamps supplied by the same manufacturer. It is readily noted that the mortality rate of these lamps has been decreased by a marked degree. By scrutinizing the failure schedulles, several interesting things become evident. The hot failure has been reduced to a very small number; of the 47 recorded failures, only 5 occurred during the hot cycle and only one of these was the result of a filament entanglement and shorting. All the remaining failures (42) were during the cold cycle and usually at random points on the coiled section of the filament. A typical failure is illustrated by Fig. 30. Fig. 30. A typical failure of the modified 1251A lamps. From these tests it seems evident that the elimination of the support dramatically decreases the mortality rate and changes the character of filament failure to predominantly cold failures. It also almost completely eliminates the type of failures previously noted in the type 1251 lamp. After short, severe service the filament does become distorted from the original shape, but this seems to have no harmful effect on its impact resistance. The impact evaluation of the 1251B lamps was carried on in a similar manner, with the results shown in Fig. 31, where the tests are compared to a previously established range by testing a number of regular 1251 lamps. It is quite evident that the 1251B is stronger than the regular lamp. Only during the early portion of the test does the mortality rate approach or equal that previously established. After an initial heavy mortality, the rate of failures is reduced to -44

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN ----— FIRST TEST -- --- SECOND TEST \ -' —-. THIRD TEST C4O 80"- -- FOURTH TEST a,80~'d~~~~~~~~~ \ ar _AVERAGE PREVIOUS RANGE >40 0 TIME IN MINUTES Fig. 31. Comparative mortality results of modified 1251B and regular 1251 lamps from the same manufacturer, equally tested on the rotary-drum impact tester. a very nominal figure. The total percentage of failures is the lowest yet experienced for this particular type of lamp. The failure schedules of these four tests reveal some rather inconsistent facts. In the first test, seven of the nine failures occurred during the hot phase of the cycle while in the third test only one of the eigh.t failures occurred during this period. The filaments of these.hot fa.lures had few if any of the characteristics of previous hot failures of the other tests (i.e. filament entanglement and shorts). A typical hot failure is illustrated in Fig. 32, which shows the break to be at a point between the lead wire and the coil Figure 35 depicts an average cold falwe with the break in the coiled section of the filament,. rThese tests indicate that there were a number of weak Lamps in each group which failed quite early. This is shown by the fact that 50% of the failures occurred in the first quarter of the test. The most plausible reason for the numerous early failures is that the mounts were manually assembled and probably were subject to more manufacturing differences. Except for this shor coming, the lamps seemed mezchanically strong and showed a definite improvement over the regular 1251 lamp. Thxe test procedure for the 1251C lamps was similar to the 1251A and 1251B tests, except for the following variations: 45

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN Fig. 32. A typical hot failure of a 1251B lamp after service on the rotary-drum impact tester. --::- -..- w:::I:.:::! Fig. 33. A typical cold failure of a 1251B lamp after service on the rotary-drum impact tester. 1. Lamps were soldered to a holder designed for a S-8 bulb. 2. Eighteen lamps constituted a test. The test results are presented in Fig. 34, where they are compared with regular 1251 lamps from the same manufacturer. It will be noted that there was a marked reduction in lamp mortality. For instance, at the end of

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN --- FIRST TEST - - SECOND TEST o.. —%:,THIRD TEST C0.80..-. | —FOURTH TEST > - - AVERAGE 0,.,-PREV IOUS RANGE -~ 20'46 600 b 8o 90 TIME IN MINUTES Fig. 34. Comparative mortality results of the 1251C and regular 1251 lamps from the same manufacturer, equally tested on the rotary-dr.im impact tester. four hours testing only about one-third the usual number of failures had oc - curred. The failure schedules reveal that only 3 of the total (33) failures occurred during the hot cycle. The rezmcainder of these faillures all occurred during a cold period. and were located at intermediate localtions on the filament coil, as illustrated in Fig. 35. The lamp shown in Fig. 35 is also typical of the manner in which the filaments survived the tests with little or no distortion. The following conclusions may be i,~rawr from the results of these tests: 1. The 1251C lamp modification is superior to the 1251 in its resistance to impact tests and shows no signs of distortion or entanglement of the filament segments. 2. The character of the observed failures has been changed to a predominant cold failure at a random location. B. CONCLUSIONS OF EXPRIMENTA LAMP TE1STS All of these modified 1251 lamps, with one exception, exhibited a notable increase in resistance to shock and vibration. This improvement has

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN Fig. 35. A typical 1251C lamp failure as the result of impact tests on the rotary-drum tester. been achieved by modifying the lamp with either a double anchor, a rigid support, an unsupported filament, or an unsupported filament mounted on rigid leads. The double-anchored filament and the unsupported filament lamps seem to be the most practical modifications, since both are physically interchangeable with regular production lamps and offer few manufacturing problems. The other methods of modification should be explored further unless they appear to be impracticable from either a manufacturing or service viewpoint...~~~~~~

ENGINEERING RESEARCH INSTITUTE * UNIVERSITY OF MICHIGAN APPENNDIX TEST PROCEDURE USED FOR THE ROTARY-DRUM IMPACT MORTALITY TESTS 1. The cam offset was set at 0,o63" (+ 0.001") and the drum speed was adjusted to 175 rpm (+ 2 rpm). 2. A random sample selected from the test lot was soldered to the appropriate lamp holder. 35 The lamps were operated 25 minutes "on" and 5 minutes "off" for a predetermined number of cycles. 4. The lamps were operated at 28 volts a-c, unless otherwise specified. 5. Observation during the cold cycle was made by means of neon indi cators switched in series with the cold filament. 6. Obbservation of the hot cycle was made by periodic visual inspection of the lamps.

SINGLE-FILAMENT MORTALITY SCHELULE FOR THE ARSENAL TESTER Percent Operative Test No. 1 Test No. 2 Test No.-3* Test No. 4 Test No. lamps lamp Time Lamp Time la Time lam Time Lamp Time 95 1 17 9 86 18 118 19 205 16 25 90 9 73 13 148 8 178 8 207 12 145 85 6 235 10 175 1 205 1/2 2 236 9 145 fl 80 15 296 17 265 15 235 11 357 2 146 75 16 322 18 299 10 344 13 385 14 175 70 20 357 1 425 13 357 20 387 18 235 65 8 388 8 478 3 476 7 415 19 35 60 13 416 5 415 13 326 55 5 416 15 417 6 415 50 7 477 45 40 35 m 30 25m *18 lamps SINGLE-FILAMENT MORTALITY SCHEDULE FOR THE AUTO-LITE TESTER rb Percent Operative Test No. 1 Test No. 2 Test No. 3 Test No. 4 Test No. lap lw Time LaPTime Time l Time LRUP L 95 1 56 13 56 7 3 4 83 8 22 90 8 87 3 85 17 55 6 100 9 116 85 4 145 7 95 20 115 11 100 19 116 80 10 145 16 113 1 130 18 142 20 127 75 11 175 9 113 4 155 19 145 U 203 0 70 13 175 4 145 5 165 2 265 4 2. 65 14 175 11 145 2 180 7 265 1 235 60 6 206 14 205 3 205 16 265 5 265 55 16 206 15 205 19 260 17 343 7 265 50 17 235 17 210 9 295 20 361 14 265 45 18 265 5 233 16 315 17 325 40 5 295 2 270 6 325 13 355 35 20 295 1 293 11 325 12 360 30 3 356 6 355 14 325 18 415 25 15 356 18 359 8 430 3445 SINGLE-FILAMENT MORTALITY SCHEUJLE FOR TEE CHRYSLER TESTER Percent Operative Test No. 1 Test No. 2 Test No. 3 Test No. 4 Test No Lamps lamp Time lmp Time lam Time lamp Tim lam 95 7 115 16 240 19 25 5 90 19 25 90 3 180 6 270 12 60 2 150 355 85 1 210 13 300 8 85 1 205 2 210 Q 80 12 225 20 330 2 90 8 205 9 270 75 20 240 7 360 9 120 9 210 1 300 70 10 240 1 385 11 180 13 210 13 300 65 2 240 17 390 3 240 16 240 17 330 60 15 270 15 450 15 240 300 8 390 55 11 330 19 480 20 355 10 300 11 420 50 9 420 17 390 15 355 45 16 445 14 420 7 360 40 5 450 18 420 35 6 480 19 420 30 12 450 25 6 480 14 480 z

SINGLE-FILAMENT MORTAITIY SCHEDIULE FOR THE GENERAL ELECTRIC TESTER Percentm Operative Test No. 1 Test No. 2 Test No.__3 Test No. 4 Test No. Lamps tamp Time tamp Time Lamp Time Lamp Time lamp Time 95 9 245 4 12 18 34 11 250 5 12 90 15 260 5 15 15 145 3 385 2 85 Z 85 6 350 1 200 3 145 17 415 1 116 80 7 555 14 205 14 525 14 142 75 10 385 16 205 11 360 15 295 70 8 420 2 270 20 385 65 5 475 6 270 6 418 m 60 15 270 4 445 55 11 415 9 445 rn 50 19 415 15 445 45 12 475 7 475 40 355 30 25 SINGLE-FILAMENT MORTALITY SCHEDULE FOR THE TUNG-SOL TESTES Percent ". Operative Test No. 1 Test No. 2 Test No. 3 Test No. 4 Test No T Lamps Laimp Time lapTime Lamp Time IapTime Lamp Tm H 95 2 176 15 116 8 40 11 2 1 116 90 8 208 14 256 6 206 14 116 15 526 85 9 208 7 237 3 297 1 117 15 357 IT 80 14 236 19 387 14 528 2 176 20 476 75 3 357 16 446 1 587 20 266 70 15 357 19 356 65 16 357 9 556 60 1 366 16 388 55 20 476 3 416 50 7 476 45 40 35 30 m 25 SINGLE-FILAMENT MORTALITY SCHEWULE FOR THE WESTINGHOUSE TESTER Percent Operative Test No. 1 Test No. 2 Test No. 3 Test No. 4 Test No. 5 Ianops LanEm Time Lamp Time Lamp Time lamp Time Lamp Time 95 15 150 17 25 (hot) 7 25 (hot) 2 295 (hot) 90 14 360 15 25(hot) 80 75 ( 670 I 60 55 I 50 45 40Z 35 30 25

DOIBLE-FILAMENT MORTALITY SCEIXULE (MINOR FILAMENT) FOR THE ARSENAL TESTER Percent Operative Test No. 1 Test No. 2* Test No. 3 Test No. 4 Test No. 5 Lasps Lamp Time Lawp Time Time Lamp Time Iamp Time z 95 7 145 2 87 9 86 7 116 10 87 90 2 207 1 176 5 115 1 176 7 116 85 5 267 5 179 1 146 12 236 1 145 7 80 18 297 4 206 14 146 11 266 8 75 6 325 7 208 15 149 17 266 6 205 m 70 16 355 13 237 17 175 4 267 20 263 65 3 355 14 267 7 208 8 269 11 264 60 1 356 12 296 6 267 2 296 14 297 55 11 415 18 298 8 446 6 297 16 326 50 10 356 13 447 16 326 9 355 45 17 356 4 476 3 328 4 40 6 418 20 359 2 386 35 15 446 13 447 13 386 30 14 476 17 416 25 18 476 m 20 *18 amps DOUBLE-FILAMET MORTALITY SCHEDULE (MINOR FILAMENT) FOR THE AUTO-LITE TESTER Percent Operative Test No. 1 Test No. 2 Test No. 3 Test No. 4 Test No. 5 Lamps Lamp TimLe L Time Time La Time L Tie 95 7 205 7 175 15 92 7 5 8 146 90 8 235 15 175 17 145 2 118 6 149 I 85 6 415 8 265 1 205 14 145 13 176 80 20 416 1 295 16 205 4 147 7 206 75 4 445 2 295 4 265 5 175 17 207 70 9 445 20 355 13 266 20 235 4 269 65 13 445 12 417 10 266 10 265 2 60 10 475 13 418 11 267 17 265 1 325 m 55 18 475 7 268 11 326 16 448 50 14 325 16 327 45 18 328 3 355 40 19 356 15 387 35 1 387 30 19 416 25 13 477 20 DOUBLE-FILAMENT MORTALITY SCHEDULE (MINOR FILAMENT) FOR THE CHRYSLER TESTER Percent Operative Test No. 1 Test No. 2* Test No. 3 Test No. 4 Test No 5* las TmLap TmTiest Ties Lo _I' --- -- - lP 95 13 90 11 55 2 90 1 20 1 20 90 4 120 1 120 6 90 2 90 6 90 85 14 120 2 120 8 90 16 90 150 80 1 145 3 20 9 180 4 120 7 210 75 2 150 14 120 14 180 7 180 2 240 70 7 180 4 150 11 180 3 210 14 270 65 6 210 7 180 4 240 14 210 13330 60 9 210 15 180 20 240 20 210 5 360 55 15 210 10 210 3 330 6 270 19 390 () 50 19 210 19 210 1 420 19 270 17420 45 18 300 8 300 5 360 40 20 330 17 30 35 20 0 *Stopped at 300 min. *16 lms Z _~~~~~~~~~~~2 0 1 1 ~

DOUBLE-FIIAMENT MORTALITY SCHEDULE (MINOR FILAMENT) FOR THE GENERAL ELECTRIC TESTER m z Percent Operative Test No. 1 Test No. 2* Test No. 3 Test No. 4 Test No. 5 Lamps Lamp Time LTamp Time Lap Time Lamp Time Lamp Time Z 95 7 145 7 29 17 86 1 150 1 115 m 90 4 250 1 205 6 117 7 180 7 115 85 8 295 6 295 10 119 5 205 11 117 80 2 330 2 298 3 185 5 205 5 25 6 75 9 415 10 560 13 185 11 235 6 375 Z 70 16 445 9 415 1 212 16 235 65 16 477 2 265 10 295 60 18 478 4 265 19 295 55 7 265 9 470 50 16 265 13 470 45 ll 475 14 470 40 17 470 tF 35 18 470 m 30 20 470 25 20 *19.aps I DOUBLE-FIIAMENT MORTALITY SCHEIILE (MINOR FILAMENT) FOR TEE TUNO-SOL TESTER Z Percent Operative Test No. 1 Test No. 2 Test No. 3 Test No. 4 Test No. 5* la _s Lamp Time lmp Time lmp Time Lap Time Lanp Time 95 1 145 13 86 14 206 6 176 11 176 ~3]u~~~ ~ 90 3 206 3 206 7 256 5 257 17209 4'J491~~~ ~85 14 206 10 386 5 236 14 236 16386 80 20 236 15 417 2 326 1 266 1 386 75 11 257 20 446 13 326 7 266 70 13 266 4 476 12 326 65 10 266 18 416 60 18 356 20 477 55 7 386 50 4 386 45 40 Z 35 30 25 20 17L DOUBLE-FIIAM MORTALITY SCEEDULE (MINOR FIIAMENT) FOR TEE WESTINGHOUSE TESTER Percent Operative Test No. 1* Test No. 2 Test No. 3 Test No. 4 Test No. 5 Laps lamP Time Lamp Time Lamp Time Lap Time Lap Time 95 5 205 3 420 1 340 3 390 9290 90 3 420 19 450 3 340 20 420 5475 85 7 480 4 350 19 445 20480 _ 80 2 390 15 480 75 13 390 70 19 390 65 10 450 60 55 50 _ 45 Z 40 35 30 25 20 *19 Lamps

ENGINEERING RESEARCH INSTITUTE ~ UNIVERSITY OF MICHIGAN BIBLIOGRAPHY Dixon, Wilfred J., and Massey, Frank J., Jr. Introduction to Statistical Analysis. New York: The McGraw-Hill Book Company, 1951. Dwyer, Paul S. Linear Computations. New York: John Wiley and Sons, Inc., 1951. Goulden, Cyril H. Methods of Statistical Analysis New York: John Wiley and Sons, Inc., 1952, 54

UNIVERSITY OF MICHIGAN 3 9015 02086 6425