This action might not be possible to undo. Are you sure you want to continue?

# Engineering Failure Analysis 11 (2004) 515–535 www.elsevier.

com/locate/engfailanal

**The railway accident of Eschede – technical background§
**

Volker Esslinger, Rolf Kieselbach*, Roland Koller, Bernhard Weisse

EMPA, Uberlandstrasse 129, CH-8600 Dubensorf, Switzerland Received 27 October 2003; accepted 2 November 2003

Abstract In 1998 a very severe railroad accident occurred in Germany. The case went to court for negligent homicide after a preliminary investigation had been performed. The accident had been caused by fracture of a wheel and the manufacturer of the wheel and the railroad company were accused. The defendants engaged a number of experts to investigate the diﬀerent technical aspects of the accident for their defence. In spring 2003 the court decided to employ an unique procedure, to hear all experts consecutively to get the best possible overview of the diﬀerent opinions and possibly ﬁnd the real cause of the accident. After the court had heard the testimony of these 13 experts from 5 diﬀerent countries it decided to discontinue the case since the guilt of the accused was deemed to be very small, if there was any guilt at all and that further technical investigations and expert testimony would most probably bring the court no nearer to a conviction. A failure analysis proper was not the subject of the court procedure and therefore a complete investigation was not carried out. The result of the hearing was in the opinion of most experts, that the accident could not be explained by the results of the investigations performed. Rather a singular incident or technological material phenomenon could have initiated the fatigue crack, which then caused the accident. Since all realistic aspects of the accident had been thoroughly investigated by the experts, only speculations on such incidents or phenomena were possible. # 2004 Elsevier Ltd. All rights reserved.

Keywords: Accident investigation; Railway engineering; Fatigue failure; Strain analysis; Finite element analysis

1. Case history and accident 1.1. Development of the rubber-sprung wheel BA 064 In 1987 it was decided to use the high velocity train ICE in larger numbers. After grave problems with oscillations a manufacturer of wheels was commissioned to further develop the existing design of rubbersprung wheels for usage in ICE-trains. In 1991 these wheels with the designation BA 064 after extended test

§ The information compiled in this publication corresponds to the contents of the experts’ presentations at the trial in Hannover of February and March 2003. The conclusions represent the subjective choice and view-point of the authors although complete objectivity has been striven for. * Corresponding author. Tel.: +41-1-823-5511; fax: +41-1-823-4014. E-mail address: rolf.kieselbach@empa.ch (R. Kieselbach).

1350-6307/$ - see front matter # 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.engfailanal.2003.11.001

516

V. Esslinger et al. / Engineering Failure Analysis 11 (2004) 515–535

drives were approved by the railroad authorities for use at speeds up to 280 km/h. These wheels, which had already been successfully employed for commuter-trains and trams for disk quite a time consist of a wheel disk and a tyre made of steel and of rubber pads which are placed between around the circumference of the disk. Thus smaller irregularities of rail and wheel are damped. Fig. 1 shows the design of the wheel in and tyre section. The tread diameter of a new tire was 920 mm. When worn down in service it could be reproﬁled several times on a special lathe down to a permissible minimal diameter of 854 mm. The wheel was designed following the recommendations of the UIC. These are intended for monobloc wheels and wheels with steel tires but not for this design of a rubber-sprung wheel. Nevertheless the load assumptions and most general ideas could be applied to the rubber-sprung wheel as well. Since the unsprung masses are smaller one can assume that stresses are at least not higher than with conventional wheels. It has to be noted that the UIC-recommendations do not take account of any inﬂuence of velocity. This way of proceeding, to follow standards for similar problems or objects if speciﬁc standards for the problem at hand are lacking, is quite common in mechanical engineering and is used for many new developments. The material was an alloyed quenched and tempered steel B5 according to UIC 810-1 [5]. The data sheet of the steel showed the values of Table 1. The loads on axles and wheels of railways are determined according to UIC 510-3 [3] or prEN 13979-1 [8]. The ﬁgures in these standards are equivalent, no inﬂuence of velocity of the trains is taken into account. The loads are related to the mass Q of the carriage per wheel. In this case Q ¼ 8000 kg was the default value. The individual load cases have to represent the run straight ahead without lateral forces, through curves with a lateral force acting on the wheel ﬂange from the inside and over switches with a lateral force acting on the wheel ﬂange from the outside (Fig. 2).

Fig. 1. Design of the wheel BA 064.

2. By this the switch was toggled to the neighbouring track and the hind part of the train redirected there. to UIC 510-3. The collapsing bridge buried . when near the village Eschede the tire of a rubber-sprung wheel. jammed under the ﬂoor of the carriage and then got stuck in the tongue of a switch. / Engineering Failure Analysis 11 (2004) 515–535 Table 1 Material properties of the tire Re (MPa) Nominal value Measurement None. 22 16. 29. broke at a speed of approximately 250 km/h. the centrifugal stress in the tire then is approximately 38 MPa. 1. Course of the accident The train ICE 884 with 12 carriages and 2 locomotives (Fig. 2 Load case Case 1: run straight ahead (sinusoidal motion) Case 2: run through curves Case 3: run over switches Fz (kN) 1:25 Á Q ¼ 98 1:25 Á Q ¼ 98 1:25 Á Q ¼ 98 Fy2 (kN) – 0:6 Á Q ¼ 47 – Fy3 (kN) – – 0:6 Á 0:6 Á Q ¼ 28 In addition to the loads shown in Table 2 the tire of the rubber-sprung wheel is stressed by the pre-stress of the rubber pads which is applied when the wheel is assembled and the retaining ring is attached and also by centrifugal forces during the run. only for information 462 490 Rm (MPa) 800–920 828 853 A5 (%) 5 14 17. 15 Fig.V. 3) travelled on June 5. The tire detached from the wheel. 2. Esslinger et al. 1998 on the route Munich–Hamburg. to UIC 515-3 for railway wheel forces as shown in Fig. At a velocity of 250 km/h and a tread diameter of 860 mm the wheel rotates with a frequency of approximately 26 Hz.7 15 517 KU at RT (J) 5 15 22. 24. was dragged along. This led to derailment and collision of the derailed train part with the pylon of a road bridge leading over the tracks. type BA 064. Table 2 Load cases acc. Formal load cases for railway wheels acc.

In the train there were mounted conventional monobloc-wheels as well as rubber-sprung ones. a part of the train. Esslinger et al. which uses its Oﬃce for Research and Experiments (ORE) for this purpose [2].1997 3.5 39 .1998 Distance run (km) New 157.6] valid at that time.1995 7. The so-called history of the wheel tire could be reconstructed as shown in Table 3. The assumptions for external loads are normally independent of the design details and most experts found it justiﬁed to take them from the UIC-guidance [4.789.5 40. which are a kind of technical guidance in the ﬁeld of railway engineering and thus correspond to the engineering rules in this ﬁeld.000 Tread diameter (mm) 920 910 893 875 865 862 Thickness of tire (middle) (mm) 68 63 54. of totally 56 wheel-sets 38 had rubber-sprung wheels of type BA 064 and 18 had monobloc wheels.518 V. EN-standards at present are either still drafts or already in revision again [7].8.8 million km corresponds to approximately 640 million rotations. The UIC publishes recommendations based on the discussions of its commissions of experts.3.1. / Engineering Failure Analysis 11 (2004) 515–535 Fig.1994 23.3. 3. The wheel with the broken tire was the last but one on the right side of the ﬁrst carriage.000 1.6. The distance travelled of 1. Table 3 History of the ICE-884 wheel tire Reproﬁlation 0 1 2 3 4 Date 14.536. For the type of rubbersprung wheel there was no special standard or guidance at the time of its development nor is there one today. Research contracts are given by the Union Internationale des Chemins de fer (UIC). The stress analysis was carried out experimentally on a prototype (measurement with strain gauges). 1.2. The consequences of this catastrophe were besides with enormous material damage and interruption of train service 100 dead and more than 100 injured passengers. Design principles The design of railway components usually is based on experience and on recommendations of the international association of railways (UIC).000 853. The European Rail research Institute (ERRI) is concerned with a certain focussing of projects. Research and development of railroads in Europe is mainly done by manufacturers and railway companies.5 45.000 1. After the accident the parts of the broken wheel and other comparable wheels were seized and all necessary information on them obtained. Load tests were carried out on a test stand on the prototype and on wheels in bogies during test runs on the track.270. At the limiting dimension of the tread diameter deemed permissible of 854 mm the thickness of the tire would have been still 35 mm in the middle.1996 17.000 1. Composition of the train of the accident.1997 27. The material used corresponded to a common railroad tire-material [5]. the wheel was designed and developed as in common engineering practice. Therefore.6.

V. Fig. For the ICE-wheel the crack was able to consume a cross-section of approximately 80% before rupture occurred. The investigations of the broken ICE-tire have shown. that it was not yet possible to say anything about the origin of this crack and that at the point of the crack start no ﬂaws had been found which could have promoted the crack. which in very short time gave as their ﬁrst statement. see Fig. / Engineering Failure Analysis 11 (2004) 515–535 519 2. whether the accident had been a singular event or whether similar failure cases on wheels of this type were to be expected and which measures had to be taken in that case. 2. 2.2. On the fracture surface colours and structures can be distinguished which imply a very discontinuous crack-growth.3.1. One of the most urgent questions was. The crack grew at ﬁrst more into the depth of the tire and spread later in a semi-elliptic way. In all. 4. At the same time the public prosecutor had started his investigations and found in his report suﬃcient reason to ﬁle suit for involuntary manslaughter and bodily injure caused by negligence. Experts of diﬀerent faculties were consulted. so that a quite long during phase of crack initiation can be assumed. that the direct cause of failure had been a fatigue crack in the wheel tire. (see Fig. Based on an approximative design calculation a reuptake of service for wheels having a tread diameter of at least 880 mm was deemed permissible. 4) of the broken wheel tire beach. This fatigue crack had already cut through a great part of the section (approximately 80%) when the ﬁnal fracture occurred. this suggests a rather extended stable crack-growth phase. that there were no surface or material ﬂaws on the inner side of the wheel tire. All material of the train relevant for the accident including all wheel-sets was taken and stored for further investigations. Analysis of the fracture surface On the fracture surface (Fig. 5). 4. Expertises to take immediate measures After the accident all carriages having bogies with rubber-sprung wheels had been taken out of service. The fracture appearance clearly shows that the material of the tire is very crack-tolerant.marks as the clear indicators of a fatigue crack can be detected by visual inspection already. Fracture surface of the broken wheel tire. Only when the section had been damaged to about 80% by crack-growth ﬁnal fracture occurred. Investigations after the accident 2. Phases of very slow crack growth or of crack arrest and phases of faster growth alternate. The crack grew faster laterally than into depth. Esslinger et al. Discussion of the crack-growth phase The fatigue crack started near the point (‘‘roof-top’’) where in the load case of straight course the maximum stresses occur at the inside of the wheel tire. .

3.1. that the load on the wheel tire at fracture and during the preceding crack-growth phase cannot have been high. then with ﬁnite elements. Since the wheel load itself was unknown. Since some of these investigations relied on simpliﬁed assumptions or methods which were not easily comprehensible for the other experts and the results of which often considerably diﬀered from those of the other experts controverses were unavoidable.04 h . 5. Expertises and their results 3.2. Origin of the fatigue crack as seen in the scanning microscope. of the metallurgical and fractographic investigations of the fracture. also outside of Germany. Expertises for the court and for the accused The results of the inspection on the accident site. / Engineering Failure Analysis 11 (2004) 515–535 Fig. The points investigated by them are compiled in Table 4. this factor could only give a relative value. The accused let execute investigations and expertises of the quintessential questions by various institutes and experts. For example the experimental strain analysis on a roller type test stand (see Fig. So it can be stated again that the fracture of the ICE-tire occurred only after extended crack-growth at a very small residual cross-section. It has to be emphasized that the purpose of the investigations performed by these experts was not primarily the analysis of the failure case itself but rather the provision of suﬃcient arguments for the defence of the accused. Determination of service loads The smoothness of travel of the ICE was tested during passage of so-called Q-measuring-points. Esslinger et al. Approximate calculations showed wheel loads of less than 80 kN at rupture.520 V. This also indicates. an experimental stress analysis using a roller type test stand and ﬁnally a calculation of structural durability [1] were presented at court by the individual experts. Here a kind of dynamic factor was determined. 1998 at 10. At the Q-measuring-point in Edesheim before the accident on June 3. an approximate stress calculation for the tire. of tests of static and fatigue strength of the broken tire. 3. 11) of one expert gave stresses which were smaller at higher speed which might be explainable by shortcomings of the equipment or of the methods they employed.

. When the development of the rubber-sprung wheel started. 3. Esslinger et al. Some experts used for their calculations of structural durability load collectives which came from measurements under diﬀerent conditions than the ICE train service. this collective nevertheless at least qualitatively shows the order of magnitudes and relations. It was loaded stepwise up to 150 kN with a point load using a short section of a rail. Example of a load collective for train service.47. the lowest values being around 1.V.13 has been measured. The measuring chain was adjusted to zero at load zero when the wheel was already assembled. 7 shows as an example. Although the numbers do not correspond to the ICE train service. 6.2. on the left side of the same axle it was 1. An example of this shows Fig.3. 6. / Engineering Failure Analysis 11 (2004) 515–535 Table 4 Expertises and the topics covered in them Experimental stress analysis Germany 1 Germany 2 Germany 3 Germany 4 Japan Sweden Switzerland South Africa X Computed stress analysis X X X X X X X X X Stress analysis ﬁnite elements Conventional design calculation X Design calculation of structural durability X X X X X X Material tests Dynamic load tests on wheels Dynamic load tests on tire segments X 521 Statistical analyses X X X X X X X X X X X X X X X X X X X X X at 247 km/h on the right wheel at wheel-set 7 a dynamic factor of 2. there existed no results of measured service loads for this kind of wheel using the intended bogies and tracks. Experimental stress analysis 3. Even today no results of such measurements have been published. On a wheel tire of 854 mm tread diameter strain gauges were attached as Fig.1. Measurements with strain gauges Such measurements were performed by most of the experts.3. This wheel had been machined to the minimal tread diameter to be able to measure the highest possible strains. Thus the stresses occurring Fig.

Fig. Point 3 in this example was where the highest strains were expected. For a wheel diameter of 854 mm at 250 km/h the latter is "centr ¼ 175 mm/m. Location of strain gauges on the wheel tire. As expected. Since the grid of the strain gauges used is several millimeters long (typically 6 mm). . Several measurements were performed in which the wheel was rotated for a certain angle.522 V. The same spot at rotation suﬀers a maximal compressive stress of approximately À70 MPa. example. 9 one can see that at strain gauge 3 (DMS 3). the point in the middle shows the highest stress with approximately 300 MPa (at 150 kN. 8. By measuring at ﬁve diﬀerent points distributed over the width of the tire the local diﬀerences were to be covered. where the crack started. Esslinger et al. From Fig. 8). To the stresses in Table 5 caused by the wheel load the stresses due to assembly (pre-tension of rubber pads) and to centrifugal forces have to be added. Because the wheel is stressed on the rail in diﬀerent distances from the wheel ﬂange this inﬂuence was also investigated. 7. Fig. the measured values for tangential strains are a mean over the grid length. by assembly itself were not included in these measurements. the stress is practically independent of the position of the loading plunger. By this procedure the diﬀerent local strains occurring when the wheel rotates could be measured. / Engineering Failure Analysis 11 (2004) 515–535 Fig. Stresses measured at the tire inside at 150 kN.

Esslinger et al.854 mm. 1 À 2 ð2Þ ð1Þ Since according to ﬁnite-element calculation T % À3z . and for a 2-axial state of stress T ¼ E Á ð"T À "z Þ. Table 5 Stresses in MPa calculated from measured strains wheel . 10 schematically shows. / Engineering Failure Analysis 11 (2004) 515–535 523 Fig. 9. Stresses on resting and on rotating wheels By loading a resting wheel the point directly under load suﬀers the highest stress.3.2. The stress from measured strain " is ¼ E Á D".V. When the wheel rotates under a load Q a point on the circumference consecutively experiences the stresses 1–5. loaded ð4Þ ¼ 0:3. test load F ¼ 150 kN Position 0 – maxima 50 – minima Stress range DMS 1 185 À47 232 DMS 2 142 À36 178 DMS 3 297 À67 364 DMS 4 187 À57 244 DMS 5 228 À69 298 3.3. Determination of circumferential stress from measured strain The strains were measured by most experts only uni-axially.3. one gets ð"T À "z Þ ¼ 3 Á ð"z À "T Þ with the adjusted stress becomes T ¼ 1:14 Á E Á "T : The stress T is underestimated by 14% if only the uniaxial measurement is considered. This decreases over the circumference and reaches a minimum at Æ45 as shown in Fig. 8. Inﬂuence on maximal stress of the position of load initiation on the tire. To be able to compare better the stresses calculated from those measurements with those calculated with ﬁnite elements considering multi-axiality a correction can be applied. A resting wheel. the stress range accordingly is A þ B as Fig. 3. ð3Þ .

From measurements and ﬁnite-element calculation one knows that the positive amplitude is approximately 82% of the total stress range. / Engineering Failure Analysis 11 (2004) 515–535 Fig. This also means that these simple analytical models do not yield useful results without a priori information. 11. that it would have been possible to determine the stresses in the wheel tire suﬃciently exact by a computational stress analysis using simple models. By one expert the strains were measured on a rotating wheel using a roller type test rig at velocities up to 250 km/h at a load of 80 kN using strain gauges. Thus the conditions of a rotating wheel can be experimentally determined in a quasistatic test on a resting wheel. loaded by a point load were suggested (Fig. in contrast to directly measuring the stresses with strain gauges in a load test. compressed diametrally and the beam on an elastic foundation. Approximations which use the diametrally compressed ring as a model assume.4.g.3. these approximations lead to unrealistically high stresses. Computational stress analysis using simple models One charge of the prosecution was. By these it would then have been apparent that the wheel tire was too thin. 12). On these no useful design calculation could be based. e. that the tire is an independent load bearing structure in the rubber-sprung wheel. 3. the measured maximal stresses and the stress range decreased with increasing velocity which could not be satisfactorily explained neither physically nor mechanically and possibly was an eﬀect of the test rig or of the measuring methods.524 V. by determination of the proper bedding coeﬃcient. Fig. Esslinger et al. . Thus it would neither have been reasonable in 1992 to use such approximations. Stresses measured by one expert on a roller type test rig. Since this assumption is wrong. 11 shows. Diﬀerence of stresses comparing the resting and the rotating wheel. by F is stressed only with an amplitude of A. As Fig. At F % 1:2 Q the stress range would be equivalent. 10. As possibilities for such simple assessments the ring. One expert got satisfactory results with the model ‘‘point load on an elastically bedded beam’’ which were in good accordance to ﬁnite-element calculations after he ﬁrst had adjusted the bedding coeﬃcient accordingly. For an assessment of structural durability the eﬀect of mean stress level has also to be accounted for.

by which those ﬁnite-element calculations without validation by measurements bring only approximate results. There the rubber properties were adjusted until the measured values were attained in the calculation. The material behaviour of the rubber pads is strongly non-linear.3 Rubber pads FHFN 85 (H85) 2. The present case is a profoundly non-linear problem. Simple models for computational stress analysis.100. Table 6 shows the values used by one of the experts.000 MPa 0. Because of the large deformations for the rubber pads also a geometrical nonlinearity has to be taken into account (conventional mechanics always assumes small deformations). Esslinger et al.45 0.3 1. Stress analysis using ﬁnite elements Today the method of ﬁnite elements is the standard method for determination of stresses in mechanical engineering design. Finally the components of the wheel are not connected rigidly but loads are transferred by contact and friction. In most ﬁnite elements calculations used by the experts the solution according to Mooney–Rivlin was used for the properties of the rubber pads.00 MPa 0. Table 6 Material properties used in some ﬁnite-element models Wheel tire Material Modulus of elasticity (E) Poissons ratio () Mooney–Rivlin parameter C10 (for a hardness of the rubber pad of 85 shore) Mooney–Rivlin parameter C01 (for a hardness of the rubber pad of 85 shore) Steel B5 210. / Engineering Failure Analysis 11 (2004) 515–535 525 Fig. by the fact that most software packages contain not only one material law and one solution for parameters of contact and friction but rather a whole set of them from which the user has to choose the one which is in his opinion the most appropriate. 3.V. One expert took Mooney–Rivlin parameters just from a table of the software-house. 12.000 MPa 0. The latter even today is a very challenging problem with ﬁnite-element calculations. This is on one hand done by experience so that a certain subjectivity comes into play.4.363 .3 Wheel disk Stahl C2 210. When assembled in the wheel the rubber pads are pre-stressed for a certain amount such that the deformation range in service is reduced. Before 1990 only linear problems could be solved by this method by industry. others carried out load tests on rubber specimens which then were modelled using ﬁnite elements. This becomes apparent among others. on the other hand by validation with experiments and measurements.

with exception of one who used a too simple ﬁnite-element model. Geometry. The calculated stress values correspond to the measured ones. / Engineering Failure Analysis 11 (2004) 515–535 Although only one half of the wheel has to be modelled because of symmetry as shown in Fig. Most experts did also calculate the load cases ‘‘curve’’ and ‘‘switch’’. Fig. the resulting stresses can be displayed in function of the wheel load Q as in Fig. Then the quasi-static stress can be determined for any wheel load. Regarding the maximal stresses along the inner circumference one can clearly see the inﬂuence of the support by the rubber pads as shown in Fig.526 V. 14. tire alone (middle) and hot spot in tire (right). . 15. but additional stress peaks occur at both sides below the circumferential ledges. 14 one can see that the assembly stress is considerably smaller than service stresses at 98 kN load and that the maximal mean stress m and the stress amplitude a have their maximum in the middle. Results of a Finite-element calculation of the rubber-sprung wheel BA 64. an example is shown in Table 7. Since the calculations were performed in the elastic range. 16. From Fig. that is at the crack origin. which means a high expense of calculation time even for modern computing equipment. Esslinger et al. The calculated results of the experts did not diﬀer signiﬁcantly. possible service loads and the rubber pads necessitate the solution of diﬃcult non-linear stress and contact problems using three-dimensional ﬁnite-element analyses which today can be carried out only by a few experts in this ﬁeld. Fig. Finite-element model of a rubber-sprung wheel type BA 064.000 elements resulted. 13. The stresses were determined in all cases for the ‘‘roof’’ at the inside. nevertheless large models having more than 100. assembled with point load (left). 13.

Complementing calculations. 3.5. example for numerical results from a ﬁnite-element calculation of a tire for various loads and tire diameters Tire-. Fig. that the loads of the tire can only be transferred up to a certain wheel load by the rubber pads. 17. . 15. / Engineering Failure Analysis 11 (2004) 515–535 527 Fig. Maximal stresses along the inner circumference of the wheel tire. Esslinger et al. (mm) 920 866 862 Load (kN) 150 150 98 141 227 98 141 227 129 150 Tassembly (MPa) 32 48 50 Tcentrifugal (MPa) Stress due to external load (less Tassembly and Tcentrifugal ) Tmax (MPa) 126 275 182 261 414 191 271 430 305 334 Tmin (MPa) À46 À71 À49 À69 À106 À50 À71 À104 À81 À81 DT (MPa) 172 346 231 330 520 241 342 534 386 415 zmax (MPa) 26 90 58 85 140 62 90 149 109 116 Misesmax (MPa) 117 251 38 860 51 38 854 54 277 304 At the time of the development of the rubber-sprung wheel such calculations were possible neither seen from the programs nor from computer capacity. Stresses in the wheel tire as a function of the wheel load Q. interpretation of results One phenomenon. At higher wheel loads or possibly after heavy wear of the rubber pads the latter can be compressed so far that the tire sits directly on the ledges of the wheel disk as schematically shown in Fig. Table 7 Load case ‘‘straight-on’’.V. 16. which has not been duly recognized by the prosecution has the eﬀect.

Esslinger et al. 18 shows. 19 can be seen that the tire sits on the disc at a load of approximately 200 kN if its diameter is 862 mm (state at accident). that no calculation of structural integrity corresponding to the state of the art had been performed before putting the wheels into service. Load shifting in the system tire-rubber-disc at higher wheel loads. This phenomenon can be simulated using ﬁnite elements as Fig. Nevertheless the calculated values were up to 10% higher than the measured ones. Calculation of structural durability Since a further charge of the prosecution had been. From Fig. two experts Fig.528 V. .6. Wheel load at which the tire sits on the disc. 18. point-wise load introduction in ﬁnite-element model. i. 19 is satisfactory. that is the diameter it had at the accident and thus higher loads are transferred by direct contact steel on steel which prevents increasing stresses at the ‘‘roof’’ at higher wheel loads.. / Engineering Failure Analysis 11 (2004) 515–535 Fig. diﬀerent position of rubber pads in measurements to give space for the strain gauges. The correspondence of measured and calculated values for circumferential stresses in Fig.e. In reality the results of course also depend on the state of wear of the rubber pads. 3. which inspires conﬁdence in the ﬁnite-element calculations. The diﬀerence can be ascribed to several circumstances: mean-value over length of strain gauge. 17. also the loss of stiﬀness was usually so small that the rubber pads could be reused several times. When the wheels were disassembled for revisions it showed that the wear. depending on wheel diameter (tire thickness). tread machined cylindrically for measurement in contrast to the slightly conical shape of the real structure which was also used in the ﬁnite elements model. not exact material parameters in the calculation.

collective 2. a very conservative assumption. / Engineering Failure Analysis 11 (2004) 515–535 529 Fig. the founder of fatigue tests was commemorated. The experts in the railway sector are of the opinion that traditionally all the railway equipment has to be designed fatigue proof (fail safe) (Fig. Esslinger et al. 21 schematically shows the procedure. The material was the railway steel B5 according to UIC.V. Indeed there was also a question whether such a calculation of structural durability was necessary at all. 19. 20). Fig. Nevertheless the calculations were also performed according to this FKM-Guideline. 21). Comparison of measured and calculated circumferential stresses in the tire as a function of the angle of rotation. carried out such calculations based on the newest edition of the guideline in use for this in Germany [9]. Variants for dimensioning: fatigue proof. In this connection also Woehler. collective 1 or ﬁnite fatigue life. 20. The application of such a guideline has the great advantage that the calculations are based on a widely accepted. . The stress collective assumed here was a constant load amplitude. who had developed his experiments exactly to the purpose of proving the fatigue limit for railway axles. at least the safety relevant ones. material and load parameters the structure can withstand higher stresses although for less load cycles if calculated for structural durability using load-collective 2 instead of load-collective 1 where the number of endurable cycles theoretically is unlimited but at a lower stress. Rm ¼ 828 MPa and bW ¼ 372 Fig. As an example the calculation was carried out for the assembled wheel using the local concept with stresses taken from the ﬁnite-element calculation and for a diameter of 860 mm according to Table 8. corresponding to a steel C60 with material parameters according to the shop certiﬁcate from 1992 of Rp0:2 ¼ 462 MPa. For a Woehler-line of the structure which has been determined using information from geometry. validated and systematic procedure (Fig.

the assessment of the tire can be made (Table 10).0 280.sR BK. Using a factor accounting for residual stresses and one for the mean stress sensitivity the components’ fatigue limit can be calculated. / Engineering Failure Analysis 11 (2004) 515–535 Fig.14 Wheel load 141 kN 18.0 342.35 for grave consequences of failure but with regular inspections and a second one of 1. for bending and for surface roughness (3.8 319.2 mm). The structural durability of the tire as a component in this case was identical to the component fatigue limit because one had calculated conservatively using the components fatigue-limit (Woehler-line) (Table 9). According to this assessment the tire would never have fractured.sZ 301.8 .0 360.4 320. loaded with 98 and 141 kN respectively.6 459.. 21.0 189. With the safety factor 1.sT BK.0 159. Procedure for an assessment of structural durability according to the FKM-Guideline.45 Á Rm).530 V. Table 9 Structural durability of the tire Permissible amplitude (MPa) Wheel load (kN) 98 BK.50 for no regular inspection.0 0.5 241.05 MPa (=0.6 453.0 0. The design values for the calculation account for supporting eﬀects.7 141 kN 305. Table 8 Stresses in the tire of 860 mm . Esslinger et al. load case straight run. including assembly stress of 51 MPa and centrifugal stress of 38 MPa Stress (MPa) uT oT mT DsT RsT Wheel load 98 kN 39.

Since approximately the year 2000 a new test stand shown in Fig. Alternating load tests on wheels The most realistic test next to a provisional service would be a test with a wheel-set on a roller type test stand using service load spectra.6 0. that at the point of crack start no signs for a pre-existing ﬂaw of material or manufacture could be detected at all.500 0. Since the test results were obtained with diﬀerent specimen types not all of them are directly comparable. Thus the specimens were taken from a greater number of tires which in Table 11 gives a good average. Fatigue strength The materials fatigue strength also was tested by some of the institutes involved in the process using diﬀerent procedures and specimens taken from tires.1 0.02 800–920 A5 (%) 17. Material tests 3.1.7. To apply this to components several reduction factors have to be used which could be debated. 3. Table 11 Results of tensile tests on wheel tire steel Rp0:2 (MPa) Mean Standard deviation Coeﬃcient of variation Nominal value 492 23. One expert for example used for the tire the technological factor for forged components. At the time of development and commissioning only one test stand of this kind was available in Germany.694 0. Load tests on wheels and on tires 3. The metallographical tests have shown a normal structure for this kind of material. 3.9 1.7. corresponding to steel B5 according to UIC 810-1 [5] was tested by a number of institutes. The nominal values were obviously attained. 3.9 0.V.2. which though had not been used then nor now to run tests for the investigation of the accident.1. / Engineering Failure Analysis 11 (2004) 515–535 Table 10 Utilisation ratio as indicator of the assessment Utilisation ratio Wheel load 98 kN Safety factor1.556 531 Wheel load 141 kN 0. 22 of DB AG is available for tests of wheel-sets. In reality the forged parts were machined on a lathe to a quite ﬁne ﬁnish of all surfaces.3.771 3. Esslinger et al. Tensile tests The static strength of the material of the tire.50 0.05 – Rm (MPa) 851 15.35 Safety factor 1.7.8. Also the chemical analysis and the measured hardness proﬁle over the tire section had no negative results. The fatigue strength measured reached from 326 to 392 MPa.8. If the fatigue limit is calculated from the ultimate static stress according to [9] one gets 383 MPa.06 5 14 . Metallurgical investigations The investigations of the fracture surfaces in the scanning electron microscope have already been described above and have given as a result.7. meaning that he assumed a rough forged surface.

5 175.5 203. in some cases even 854 mm.5 Æ 188.0 Á 106 + 2.0 Á 106 + 2.0 Á 106 + 2.34 Á 106 C + 1. . multiple ﬁgures for load-cycles indicate repeated tests on the same wheel.5 154 Æ 153. Table 12 and Fig.0 Á 107 I 1.81 Á 106 C 0. Since these tests were carried out on resting wheels.5 Æ 160.5 Æ 146.0 Á 106 I + 0. C: crack. At a number of institutes sinusoidal tests were carried out on complete wheels.5 80 Æ 70 85 Æ 80 140.5 80. several institutes also carried out sinusoidal tests on segments taken from wheel tires.5 189. 3.0 Á 107 + 1.0 Á 106 I 2. But the available bogies would permit such loads only after some reinforcements.5 Æ 139.5 80 Æ 70 Cycles 2.5 154 Æ 153.0 Á 107 I I: intact.5 Æ 87. The tests were performed with wheels having tread diameters of 860 mm.5 Æ 174.0 Á 107 I 2. Table 12 Results of sinusoidal tests on rubber-sprung wheels type BA 064 Institute 1 1 2 2 3 3 3 3 3 3 3 4 4 5 Diameter (mm) 860 860 860 860 860 860 860 860 860 860 860 854 854 860 Load (kN) 140 Æ 139.0 Á 106 I 2.2.0 Á 106 I 2.94 Á 106 C 2.54 Á 106 C 2. / Engineering Failure Analysis 11 (2004) 515–535 Fig.55 Á 106 C 2. On this test stand velocities up to 300 km/h and loads up to 400 kN are possible.0 Á 107 I 1.5 161.5 Æ 79. 22.0 Á 106 + 2. Because the tire sits on the disc at loads above 170 to 200 kN.0 Á 106 I + 0.0 Á 106 + 5.0 Á 106 I 1.0 Á 106 I 2. Roller type test stand for wheel-sets of DB AG (from brochure of DB AG). Sinusoidal tests on segments taken from wheel tires Because the sinusoidal tests using complete wheels had shown that cracks in the tire could only be produced under unrealistically high loads and these tests were very costly.0 Á 106 + 2. the stress range produced by the test loads corresponded only to about 80% of that which would have been achieved with a rotating wheel. the ledges on the inner face of the tire and on the disc were partially removed before the tests. Esslinger et al.0 Á 106 I + 0.532 V.04 Á 106 C + 1.5 147.8. 23 show a summary of the results. which allowed higher test frequencies and thus shorter testing times.98 Á 106 C + 1.5 88.5 Á 106 C 1.0 Á 107 + 1.98 Á 106 C + 0.5 Æ 202.

25. 24. 24. To reproduce the real conditions in the tire of a rotating wheel the stress at the critical point (‘‘roof’’) has to be adjusted accordingly. Results of sinusoidal tests on complete rubber-sprung wheels. / Engineering Failure Analysis 11 (2004) 515–535 533 Fig. Results of sinusoidal tests on segments of wheel. Fig. One test stand is shown in Fig. Segment of tire in 3-point bend test on a testing machine. Esslinger et al. by comparison with strain gauge measurements. Fig.V.g. 23. .. These segments then were tested in 3-point bend tests. e.

25. Before the wheel was commissioned there were performed strain measurements on a prototype of the wheel. Seen from todays view-point facilities for design would be better. i. Service loads had never been determined for this type of wheel. corresponding to a wheel load of 1:25 Á Q ¼ 98 kN. The material corresponded to all speciﬁcations.534 V.e. without notch only at about 7 times the nominal load. further test were performed using specimen with notches up to 2 mm depth and also tests on specimens without notches under loads up to 265 kN wheel load. fatigue tests on complete wheels and extended service tests on rail tracks. Conclusion from these tests was. For this type of rubber-sprung wheel no individual UIC-guideline existed. / Engineering Failure Analysis 11 (2004) 515–535 Fig. the fatigue crack must have started from other reasons. After the accident fatigue tests have been carried out on whole wheels and on segments of tires which gave as result that a fatigue load of nearly four times the nominal wheel load would have been necessary to produce a fatigue crack without pre-existing ﬂaw. Fracture surface of a segment specimen after a test.7 mm deep to produce a fatigue crack at a stress range of 200 MPa. After it became clear. that even with high loads no cracks could be produced if no notch had been applied at the critical point. 26.5-times the nominal wheel load. that only with extremely high loads a crack can be produced in a tire having no ﬂaw (notch). a fatigue crack could be produced only with an initial ﬂaw at least nearly 1 mm deep and a load 2. When the segments were tested. 4. The ﬁnal fracture only occurred after the remaining cross-section had dropped to about 20%. Notches had to be at least 0. . 98 kN. Results are shown in Fig. 26) corresponded quite well to that of the accident but the portion of the fatigue crack area is smaller. The tests were each started with un-notched specimens at a load corresponding to a wheel load of 1:25 Á Q. train and track. Esslinger et al. Since no pre-existing ﬂaw at all had been found at the crack origin. Summary and conclusions The fracture of a wheel tire which had brought about the accident of Eschede clearly was caused by a fatigue crack at a relatively low mean stress level. The impact factors which had been measured at a so-called comfort measuring point can only serve as approximate indicator. This procedure corresponded to the UIC-recommendations in force at that time. The recommendations for monobloc wheels nevertheless could correspondingly also be applied here.5-times the nominal load. no traces or pre-existing ﬂaws which could have initiated a fatigue crack could be detected. which is practically impossible because the tire comes into contact with the wheel disc already at wheel loads of 2–2. The crack-growth (Fig.

that realistic measurement of service loads and impact factors should be carried out previous to design work. which had been done in this case before commissioning of the wheels. References [1] [2] [3] [4] [5] [6] [7] [8] [9] Berger C. 2000. ¨ UIC_Merkblatt 510-5 (14. Major projects of joint European railway research. that the weak spots need identiﬁcation by tests on roller type test stands until fracture and nondestructive test of these weak spots ought to be done in the regular service checks. et al. Bahnanwendungen-Radsatze und Drehgestelle Laufradsatzwellen-Konstruktionsverfahren. On the approximately 5000 remaining wheels removed after the accident several cracks were found in tires but no fractures. Technische Lieferbedingungen fur Rohradreifen aus gewalztem. 1981. Jpn Railway Transport Rev 1996:16–21. Entwurf 2001). The case has shown as topics for improvement that modernisation of design rules or international standards is necessary. 1998. The nature of such a rare event can only be speculated about. Bahnanwendungen-Radsatze Drehgestelle-Rader Technische Zulassungsverfahren Teil 1: geschmiedete und ¨ ¨ gewalzte Rader. ¨ EN 13103.V. ¨ Rechnerischer Festigkeitsnachweis fur Maschinenbauteile VDMA-Verlag Frankfurt. A calculation of in-service integrity based on the stresses calculated with loads according to UIC has proven a suﬃcient safety of fatigue cracking. MTEL P 980016 Technische Zulassung von Vollradern. ERRI B196. Technische Zulassung von Vollradern. UIC_Merkblatt 510-3. Because such calculations would have been useful primarily for optimisation of design details and a validation of their results would have been necessary anyway by measuring strains. 4. This supports the presumption of a rare or singular event which could have initiated the crack. unlegiertem Stahl fur Triebfahrzeuge und ¨ ¨ Wagen. 2002. Korpanec I. et al. Int J Fatigue 2002.24:603–25. ¨ UIC-Kodex 810-1. ¨ . in the oﬃcial investigation of the case such topics were not brought up and because this wheel type is no longer used for high-speed trains no further research was done in this direction.1 8. had grown only very slowly or not at all and never reached a critical depth. The cracks had occurred also in wheels with larger diameters than that of the accident and without pre-existing ﬂaws. 2000. the design process cannot be criticized in this point. Eisenbahnfahrzeuge Drehgestelle-Laufwerke. which were detected later. Esslinger et al. On the approximately 100 wheels which had been taken out of service before the accident no cracks or ﬂaws have been reported when they were disassembled and scrapped. 1994. It also supports the assumption of a generally quite low mean loading level because these cracks. Verfahren fur die Berechnung von Radsatzwellen. Entwurf zum UIC-Merkblatt. Betriebsfestigkeit in Germany – an overview. Nevertheless it was disputed whether such a calculation would have been relevant at all because the railway companies demand a fatigue proof design. erweiterte Auﬂage. ¨ prEN 13979-1. / Engineering Failure Analysis 11 (2004) 515–535 535 A calculation of stresses in the tire using ﬁnite elements showed the stresses to be not signiﬁcantly high at nominal loading.