Fatigue in the shell of a conveyor drum

The shell of a belt conveyor broke in operation due to fatigue in the area of the weld seam between the axle disk and the ation. The conveyor delivered broken rock from a tunnel under construction. The belt was consequently cut The expected service life was calculated for different combinations of belt tension: continuous normal oper- ation with a belt tension of 250 kN, pure start-up operation with a tension of 400 kN, and mixed mode oper- ation. The calculations were performed according to the FKM-guideline [1]. * Corresponding author. Tel.: +41 44 823 45 26; fax: +41 44 823 40 11. E-mail address: christian.affolter@empa.ch (Ch. Affolter). Engineering Failure Analysis 14 (2007) 1038–1052 1350-6307/$ - see front matter � 2006 Elsevier Ltd. All rights reserved. by the sharp edges of two circular fractures and had been partially wound into the inside of the drum (cf. Fig. 1). The failure caused a long production down-time and high costs for the repair. Additionally, the pos- sibility of a recurring failure for a similar installation was cause for concern. The main scope of the investigation was to answer the question of whether the design or the manufacture of the drum was responsible for the damage. Extensive finite element models were built for two design options: the specified and the executed geometries. In order to obtain a realistic loading for the drum, the belt was mod- elled as a reinforced hyper-elastic part and the belt forces were transmitted over contact definitions (geometric and material nonlinearities). cylindrical shell. The investigation had to provide evidence whether an overloading of the drum, or a deficient design or fabrication lead to the failure. After the first damage hypothesis had been ruled out, two finite element models were built to compare the original design with the actually manufactured drum by means of nonlinear calculations. Based on the stress components of the ‘hot spot’ on the circumference, the service life of the drum was calculated for the given loading conditions. For the design according to the drawing, a sufficient service life was verified. The estimated ser- vice life of the failed drum corresponded with the effective operation time of the conveyer. High shear stresses contributed significantly to the deterioration. Since the realisation of the design proved to be problematic regarding the welding tech- nique in the area of the seam root and at the shoulder of the axle disk, an optimised design was developed and proposed. � 2006 Elsevier Ltd. All rights reserved. Keywords: Weld seam; Fatigue; Conveyor drum; Nonlinear finite element analysis 1. Introduction The shell of a discharging drum of a 2.6 km long conveyer in a mining site collapsed after 4400 h of oper- Fatigue in the shell of a conveyor drum Ch. Affolter *, G. Piskoty, R. Koller, M. Zgraggen, T.F. Ru¨tti EMPA, Swiss Federal Laboratories for Materials Testing and Research, Ueberlandstrasse 129, CH-8600 Duebendorf, Switzerland Received 31 August 2006; accepted 30 November 2006 Available online 12 February 2007 Abstract www.elsevier.com/locate/engfailanal doi:10.1016/j.engfailanal.2006.11.071 Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 1039 Conveyor drums and pulleys often prove to be critical components in large or heavily loaded installations. Many publications deal with the difficulty of fatigue and/or manufacturing technique, specifically for welded components (design of weld seam, welding procedure). Critical areas include the axle disc, the shell of the drum or the connection of different components. Jones [2] described fatigue cracks on the bottom of different drums (axle disk or web). Kim et al. [3] performed durability analysis on pulleys with additional holes in the web, which can reduce the stresses in the axle disk. Chen et al. [4] examined the flexural stiffness of cylindrical shells and the influence of longitudinal stiffeners. But the focus was mainly on the global bending load of drums mounted on two symmetrical external bearings. For consideration of local bending effects and buckling in the area of the belt margin, a detailed simulation of the drum with the finite element (FE) method is considered more reliable than a global analytical approach. Ravikumar and Chattopadhyay [5] developed a special FE code for calculation of conveyor pulleys. For the presented study, the commercial finite element code ABAQUS [6] was chosen due to availability and its broad functional extent. Non-linear material behaviour and the contact between shell and belt could both be consid- Fig. 1. Broken conveyer drum with cut belt roped into the shell, one shell segment cut off. ered in the model. 2. Situation and hypothesis A schematic overview of the installation is shown in Fig. 2. The main parameters of the belt conveyor were as follows: � Length: 2640 m. � Conveyor height: 342 m. � Output: 205 m3/h. � Belt width: 0.8 m. � Nominal drive power: 600 kW (3 · 200 kW). � Belt speed: 3.0 m/s. The modelled main part consisted of two turned disks welded to a cylindrical shell. Two short rings were welded to the disks to extend the total drum width. The outer shell surface was machined to a concave shape after the welding process. A section of the drawing is shown in Fig. 3. After a first visual inspection of the failed installation on-site, an initial rupture of the belt with the subse- quent failure of the drum could be excluded (postulate). Thus there remained the following hypotheses for the failure: Part A The load capacity of the weld seams was too low due to: A.1 incorrect dimensioning or layout; A.2 material flaws, corrosion; A.3 manufacturing faults (wrong dimensions); A.4 poor welding (inclusions, incomplete root penetration, lack of fusion, porosity). Part B The loading of the weld seams was higher than estimated at dimensioning due to: B.1 incorrect assumptions for calculations (friction, amount of conveyed material); B.2 the overall installation diverged from the plan (built longer or steeper); B.3 higher tension on the belt (incorrect control parameters); B.4 additional dynamic loads (vibrations, rolling stones). Fig. 2. Schematic of the conveyer system in the upper discharging area. 1040 Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 Fig. 3. Drawing of the conveyor drum with turned axle disks. As no indications could be found which would explain an overloading of the drum (protocols of measured drive power and pre-tension load cells, re-measurement of the complete installation), the investigations mainly focused on part A of the damage hypothesis to verify their plausibility. 3. Investigations 3.1. Macroscopic inspection of the failed drum Based on the macroscopic appearance of the broken area, a fatigue failure in the area of the circumferential weld seams was considered possible. In the peripheral zone of the fracture system, no macroscopic plastic deformation was found that would have been an indication for fatigue. Almost all fracture surfaces were pla- nar and perpendicular to the axis of the drum. Only approximately 10% of the entire fracture surface could be attributed to a forced rupture. This area could easily be interpreted as remaining fracture surface of a fatigue failure. At the bottom of the drum, in the rupture area on the inner side, a longer burr of almost square cross-sec- tion was found, see Fig. 4. This burr was still partly connected to the axle disk. Similar parts were found inside the drum together with rubber parts of the belt. The dimensions of the burr were 4–5 mm by 3–5 mm, which almost matches the geometry of the shoulder inside the drum disks, cf. Fig. 5. Due to this appearance, it was suspected that the shoulder of the axle disk had not been properly penetrated and connected to the shell of the drum. Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 1041 Fig. 4. Traces on the bottom of the drum (axle disk A). Fig. 5. Detail of a burr segment; denomination of surfaces. The measured shell thickness in the area of the crack proved to be between 18.9 and 21.7 mm (nominal size according to the drawing: 21.2 mm). 3.2. Metallographic investigation Due to consequential damages, the fracture surface could not be analysed. The metallographic examination of cross-sections through the weld seam showed that the crack had propagated through the weld metal (see Fig. 6). This was an indication for an increased notch effect in the root of the weld seam . Apart from that, no significant welding imperfections could be found that would indicate a precipitate failure (e.g. inclusions), only smaller pores in the root of the seam could be observed. 3.3. Stress analysis 3.3.1. Finite element models: mesh and boundary conditions (BC) In order to examine the influence of differing dimensions in the fabricated drum and to verify the impact of the executed of the weld seam compared to the designed case, two finite element models were built. Symmetry was utilised to reduce the size of the model and the number of solid elements. The models had the following properties (see Table 1). The drum was built of 3D-hexahedron and wedge elements, the belt was modelled with shell elements, cf. Fig. 8. For the stress analysis, linear elastic properties were assumed for all materials. � Drum: Steel with E = 210 GPa, m = 0.3. � Belt: Reinforced rubber with E0 @ 5 GPa, m = 0.45. The orthotropic properties of the reinforced belt were neglected, as previous calculations had shown 1042 Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 that the influence on the stress distribution is of a minor order. The coupling between axle disks and shaft Fig. 6. Cross-section micrograph of the weld. Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 1043 Table 1 Two model options regarding geometry of the shell and execution of the weld seam was simplified with a rigid connection, which tends to be a stiff reproduction. In option 2, the shoulder which was presumably not carrying any load was removed. The meshes for both options are shown in Fig. 7(c) and (d). Model Geometry of the drum Execution of the weld seam Option 1 Fig. 7a According to the drawing of the manufacturer ‘Ideal’: full penetration, load-carrying shoulder; as intended by the designer Option 2 Fig. 7b Actual values according to drawing, but shell thickness of the drum t = 19 mm Supposed real execution) partial penetration, shoulder therefore suppressed Fig. 7. Modelled cross-sections in the weld area according to drawing: ‘ideal’ (a) and executed (b); and FE-meshes in the seam area: option 1 with full root penetration (c), and no load-bearing shoulder in option 2 (d). Fig. 8. Finite element mesh of option 1 (quarter model using symmetry) with loads and BCs. The boundary conditions applied to the model were the following: � pivoted shaft as a simplified representation of the bearings, � symmetry conditions for YZ- and ZX-plane, � Co 1044 Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 The belt tension for start-up operation results to: F Belt;max ¼ 3 �MSU;max 0:5 � D þ F Pt ¼ 3 � 37125 0:5 � 0:63þ 60 ¼ 414 kN: ð2Þ For the finite element calculations, a constant value for start-up of 400 kN was assumed. 3.3.2. Results for option 1 The radial displacements in the drum of option 1 at continuous operation are shown in Fig. 9. Two main deformations can be observed. A global bending including shaft and drum, and local buckling and bending of the shell on the axle disk. Looking at the distribution of Mises stress, two potential locations with a high notch effect (so-called ‘hot-spots’) can be observed, cf. Fig. 10. These hot-spots are subject to high local stresses and generally also high stress amplitudes over an operation cycle, which is critical in the evaluation of the service life. Table 2 Belt tension at the discharge drum for different drive powers Drive power P (kW) Belt tension FBelt (kN) 600 260 445 208 Neglecting the losses in the drive, the belt force results in: F Belt ¼ F Pt þ P=v: ð1Þ For the total power of the drive motors, two extreme values were assumed: Pn = 600 kW ) Nominal power according to operating company. Pmeas,max = 445 kW ) Max. achieved drive power as per measurement report. Table 2 summarizes the resulting belt tensions. With these two values, a mean belt tension of FBelt = 250 kN for continuous operation was a conservative assumption. The maximum belt tension at start-up could be estimated with the maximum starting torque of the drive motors: � max. starting torque per motor: MSU,max = 37125 Nm (motor rating plate). � Belt tension of pretensioning system: FPt = 60 kN. � Drum diameter D = 0.63 m. nveying speed, v = 3 m/s. � distributed load of total FBelt/2 in the belt, � contact conditions between drum shell and belt. The belt tension had been estimated by means of measured data of the drive power and the pre-tension FPt applied over the pretension carriage (measured with a load cell used in the control loop). The values were cal- culated for the two load cases ‘start-up’ and ‘continuous operation’. During continuous operation, the belt tension at the discharge drum (FBelt) was estimated with the following known parameters. � Belt tension at the pre-tension carriage, FPt = 60 kN (measured ± 5%). � Total power of the drive motors, P [kW]. Fig. 9. Radial displacements in the complete conveyor and definition of cylindrical co-ordinates (deformations are scaled up for visualisation; model includes a quarter of the shaft; interior view). Fig. 10. Mises stress in the conveyor drum (shaft suppressed). Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 1045 A detailed study of the stress components for each hot-spot is necessary to determine whether a proof of service life requires the consideration of multi-axial stress states. In Fig. 11, the bending stresses parallel to the shaft axis are plotted. It is obvious that the hot-spot #2 is dominated by bending stresses. The maximum and minimum values are directly apparent from the fringe plot, i.e., the amplitude can be calculated directly. If one looks at the shear stresses in the u–Z direction (Fig. 12), the maximum is located at hot spot #1. Here, the peak-to-peak values of stress can not be read directly from the plot. If the hot-spot sees a negative shear stress at position u = 0� (Fig. 9), the sign will change to positive after half a rotation of the drum to the position u = 180�. Thus, even if due to symmetry only one half of the drum in u-direction was modelled, the stresses have to be evaluated over the full circumference of the drum. For the detailed evaluation of the stresses which are required to deliver the proof of a certain service life, the stresses were plotted as a function of angular position for hot-spots to be considered on their path during one cycle, i.e., over the full circumference. This graph is shown in Fig. 13. Because both hot-spots lie approxi- mately on the same circumference, this evaluation only had to be done once for each geometry and load case respectively. 3.3.3. Results for option 2 The FE-model for the design as manufactured (option 2) showed a stress distribution which was similar to the results of option 1. Due to the lower shell thickness, the stress level was higher, and the notch effect at the hot-spot was increased. This followed from the removal of the shoulder in the FE-model of the axle disk, since it was not carrying any load due to the applied welding process. The evaluated stress components for the hot-spot of option 2 is shown in Fig. 14. The position was in the groove on the inner surface of the shell (Fig. 7d). The summary of the results for both options is given in following Table 3. 1046 Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 Fig. 11. Bending stresses rZ in the drum of option 1 (shaft suppressed). Fig. 12. Shear stresses suZ in the drum (shaft suppressed). Fig. 13. Stresses versus angular position for hot-spot 1 in option 1 (continuous duty, FBelt = 250 kN). Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 1047 1048 Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 Fig. 14. Stresses versus angular position for the hot-spot in option 2 (continuous duty, FBelt = 250 kN). 3.3.4. Influence of mesh refinement The FKM code describes different calculation runs for a proof with nominal stresses, structural stresses or notch stresses. For a proof with notch stresses, the weld seam and all fillets have to be modelled accurately, which results in a fine mesh with very small elements. The matrix system would get very large and calculation costs tremendous. In the present study, a proof with structural stresses for the weld seam transition was applicable. Structural stresses are calculated by extrapolating the stress values in the vicinity of the notch in the direction of the high- est stress gradient towards the hot-spot [1]. The exact notch stresses need not be known, and the results do not depend very strongly on the mesh refinement. 4. Proof for fatigue strength The calculations for the proof of the broken weld joint were performed according to the FKM-guidelines [1]. All used symbols and abbreviations will be in compliance with this code. The butt weld between axle disk and drum shell was considered. The proof was based on the FE calculations described in the previous chapter. For both load cases, start-up and continuous duty, a constant amplitude load spectrum was assumed. Thus, pure start-up and continuous loading were considered. The consequence of this simplification will be discussed later. The calculational run with local stresses for planar, welded components was considered. For proofs with local stresses, the FKM-guideline prescribes separate verifications for the root of the weld seam and for the weld toe (transition from seam to structure). In the present case, both geometrical options have a notch in the weld toe which causes higher stresses in the transition compared to the weld root. There- fore the proof for the seam root can be neglected for both options. Table 3 Main components of stress for the three calculated cases Maximum Stress components (MPa) Option 1 (‘Ideal’ case) Option 2 (realised case) Normal operation FBelt = 250 kN Start-up FBelt = 400 kN Normal operation FBelt = 250 kN First principal stress 25.0 38.8 31.0 Third principal stress �29.4 �45.6 �37.3 Max. shear stress ±25.3 ±39.2 ±34.0 Max. Mises stress 46.6 72.0 59.0 The material for the drum according to the manufacturer was steel grade EN 10025 – S235 (formerly St37- 2), which has the following mechanical properties for sheet metal: � Yield strength: Re = 235 MPa. � Tensile strength: Rm = 340–470 MPa. rW sW thus mr De comp Fo KE,r mean Ch. Affolter et al. / Engineering Failure Analysis 14 (2007) 1038–1052 1049 KAK;r ¼ 1þMr � rmra for �1 6 Rr 6 0; ð7Þ KAK;s ¼ 1 1þM s � jsmjsa for � 1 6 Rs 6 0; ð8Þ and includes the effect of mean stresses unequal zero. The mean stress sensitivity M in the above formula is given by Mr ¼ aM � 10�3 � RmMPaþ bM; ð9Þ M s ¼ fW;s �Mr: ð10Þ For steel, the used constants are as follows: aM = 0.35, bM = �0.1 and the shear fatigue limit under com- pletely reversed stress fW,s = 0.577. With these values, the mean stress factor KAK results in Table 4. The component fatigue strength rAK and sAK results in Table 5. Table 4 Mean stress factor at the transition of the weld seam Designation Continuous operation Start-up operation Option 1 Option 2 Option 1 Option 2 KAK,r 1.0074 1.0074 1.0072 1.0076 KAK,s sAK ¼ KAK;s � KE;s � sWK: ð6Þ and KE,s consider the influence of residual stresses and both were set to 1.0 (high residual stresses). The stress factor KAK is given by 1 rAK ¼ KAK;r � KE;r � rWK; ð5Þ sWK ¼ KWK;sK ¼ sW ¼ 37 MPa: ð4Þ r planar welded components the component