Reliability Engineering Snapshot TM

INTRODUCTION

The challenge in performing the failure analysis was more than a matter of presenting the technical information. The technical information had to be presented to non-technical senior management in a way that management could understand. In this way, they could reach a decision regarding implementation of recommendations that were inherent to the analysis. The results in this paper are familiar to the reader. Intergranular attack, lack of weld fusion, cyclic loading, and maximum constraint, are all design issues understood by failure analysts. Factors that challenged the failure analysis were the time and cost to perform the analysis, the credibility and perceived value of the results, and the ability of the managers to understand the results. The managers had to believe that the results and recommendations were practical and would yield a suitable return on investment. If the recommendations based upon the failure analysis were impractical in terms of installation time and lost production, as well as future maintenance requirements and impacts on operations personnel, those recommendations were unlikely to be implemented.

Winning approval to perform the comprehensive failure analysis had to do with what people perceived to be important to them. They had to be comfortable with all of the facts and recommendations. If the maintenance manager didn't appreciate the value of taking samples from the rotary dryer in order to analyze them, those samples weren't going to be prepared. If the operations superintendent didn't understand or appreciate how a failure impacted production, the funding wasn't going to be approved.

Recommending the failure analysis was nothing more than a sales pitch for a service that few people at the decision making level comprehensively understood. To the production superintendent the issue was pounds out the door; to the maintenance superintendent the issue was manpower, plus safe and easy maintenance. With today's ever changing economic pressures and the future viability of the business, one had to convince management that the results would yield more and better product in a safe manner and for less expense. If that were not accomplished, then management would have lived comfortably in its current situation no matter how illogical it may have appeared.

This paper will illustrate some of the events that guided the course of the failure analysis, and note where previous failure analyses may have failed because the scope of the analysis was too limited.

BACKGROUND

The rotary dryer was installed in 1987 as part of a new process. It was over 16 meters (52.5 ft.) in length and 2.1 meters (7 ft.) in diameter. It rotated horizontally on top of 4 rollers and operated 24 hours a day, 345 days a year. Approximately 4 million load cycles were imposed per year. The rotary dryer was a counterflow design with liquid coming in at the cold feed-end while hot air came in at the hot discharge end. Paddles that were affixed to the inside wall tumbled the product into the air (Fig. 2 right). These paddles, known as "lifters," were distributed throughout the length of the dryer. The finished granular product exited the hot end of the unit. Within six months of initial installation and start-up, cracks were observed to have developed underneath the cold feed-end tire (Fig. 1 above). In addition, numerous cracks initiated at the ends of nearly every lifter and traveled along the attachment weld (Fig. 3 below). In addition, other critical components started failing. However, it was the problems with the cold feed-end tire and numerous lifter cracks that attracted the most attention during the early time frame.

The original equipment manufacturer (OEM) was asked to review the actual loading conditions of the dryer. The review indicated that the dryer was overloaded. It appeared that the actual loading conditions were different from the design conditions given the OEM. Since it was a unique process, this could not have been foreseen by either the OEM team or the in-house process design team.

In 1989 an in-house team of engineers configured a finite element (FE) model for the reinforcing ring underneath the feed-end tire to solve the cracking problem of the ring. The impact stresses induced by the jolters (Fig. 4 right) and the induced thermal stresses were reviewed separately. The jolters were a unique piece of equipment with hammers that hit the top of the shell to jar the product loose from the inside wall. Modifications recommended by the study were made to the reinforcing ring (i.e. install stiffener-rings) and to the lifters (i.e. cut relief grooves). No modifications were made to the jolters. These modifications were effective over the next few years. However, the lifter cracking problem gradually reappeared, and the stiffener-rings started to crack. Immediately after this, a large 120 degree shell crack opened along the circumferential weld on the downstream side of the feed-end tire. This was where the failed stiffener ring was located. Also, pieces of the lifters were breaking off and jamming the equipment further downstream. Worse yet, what were suspected as being shell buckles appeared in two separate locations, one of them being next to the same tire. Repairs began to require an inordinate amount of maintenance manpower and this demand impacted other parts of the plant that also required maintenance resources. By 1997 it became clear that the problems with the dryer were increasing in frequency and becoming serious. Steps needed to be taken to address the failures. It was feared that nothing would be done until the level of maintenance became so high that a new duplicate dryer would be purchased. If the causes for the failures were not identified, there was good reason to believe that the same problems would start again in the new dryer. As it turned out, identification of the design flaw regarding the problems at the feed-end was only one part of the ensuing failure analysis. The failure analysis was soon recognized as a legitimate part of an iterative design procedure.

GROUNDWORK - AUDITING THE CRACKS

In early 1997 a decision was made to audit all visible cracking. A digital camera was used to document the cracking and these photographs were distributed to various plant managers via the computer network. However, closer inspection of the pictures would cast doubt on the presumed failure causes of the original design.

A DILEMMA WITH THE EXISTING FAILURE MODELS

One of the dominant causes of failure was cracking of welds that attached the lifters to the shell. The established theory was that the lifters were failing due to high thermal stresses and jolter shock loads that were imposed upon the attachment welds. Grooves had been cut at the ends of the lifters to relieve both of these stresses, but the cracking continued. Early in the audit, it became evident that there was a distinct and repeatable pattern on the fracture surface of the lifters. The fracture path ran either along the top or bottom toe of the weld, but never through the throat (Fig. 5 below). Additionally, the topography of the fracture surface showed chevron markings that indicated the direction of crack propagation. This pattern indicated that the cracks were starting from both sides of the lifter surface and traveling inwards along the weld metal/base metal interface. This type of failure was not expected if the cause was due to thermal stresses.

A second observation on the majority of lifters was that the cracks initiating in the stress relief grooves progressed inwards at 45 degrees from the groove down to the weld (Fig. 6 right). This observation did not seem to fit the existing failure theory either. The goal of the failure analysis was to find the common denominator that tied all of the failures together. Each proposed model would have to be tested against each failure. If the jolters were the cause for all of the lifter failures, why were similar failures appearing in areas where there were no jolters? If thermal stress was the cause for all of the lifter failures, then why were similar failures appearing in regions that were essentially cold, and more importantly, why were some of the cracks moving away from the welds and into the shell? Had something of importance been missed in the first failure analysis? Significantly, no samples of failed weldments were examined during the first analysis. It was suspected that somebody did not want to cut holes in the dryer shell.

QUESTIONING THE JOLTER THEORY

Experience with three similar dryers within the corporation revealed that cracking always initiated at the end of the lifter and propagated into the shell (Fig. 7 below). The inspection reports went on to indicate that there was poor welding technique in all of the failed welds. There was additional ultrasonic test information that indicated those welds without visible cracks did in fact contain cracks. Still, the early assumption was that the jolters were the cause for lifter cracking underneath the jolter area. However, during this failure analysis the actual service exposure of the lifters in the region of the jolters was calculated. Results of the calculation indicated that there was some 90 minutes between impacts on the same lifter, or 16 impacts per day. The jolters were blamed for lifter cracking 12 months after initial start-up, which corresponded to 5,840 impacts. This number was considered too low to initiate fatigue cracking (Ref. 1). This suggested that vibration waveform analysis be utilized to determine more information regarding the details of the load caused by the hammer blows.

QUESTIONING THE THERMAL STRESS THEORY

The initial investigators assumed that all lifters were cracking due to thermal stresses. However, no infrared imaging had been done to support this assumption. Subsequent inspection of the dryer's discharge-end during this analysis revealed that the lifters had creep yielded (Fig. 8 right). Further investigation lead to questions. Normal design values for alternating thermal stress was on the order of 40,000 cycles for a 20-year design life (Ref. 2). If the rotary dryer was operated in the worst possible manner having an inconceivable four thermal shocks per unit start-up, the dryer would have experienced only 3,300 cycles by 1999. The thermal stress failure model was initially proposed in 1989, only 12 months after lifter cracking was first detected. This corresponded to 350 thermal stress cycles; this too, was suspiciously low. If thermal stress was the cause for failure then the two hottest rows of lifters should have been cracked, and they were not cracked. Were the relief grooves effective in only this region? Was there a significant temperature difference between the lifters and the shell to induce a stress? Infrared thermography did not show a large thermal differential along the weld profile (Fig. 10); the worst case was 400 C (104 F). The result placed the thermal stress model in question.

HURDLE No. 1 - TO STUDY OR NOT TO STUDY

Beginning a comprehensive failure analysis was a large undertaking. Besides the failures discussed in this paper, 29 additional design weaknesses had been identified. People were so committed to the existing models of failure that it was difficult to accept the implications of the initial inspections in 1998. It was believed that something must have been missed in either the previous evaluation or the current inspection, or both. Therefore, it was decided to seek new outside help so that fresh ideas could be developed. However, several concerns had to be considered before approaching management for failure analysis funds.

The first concern was that the six inspection reports issued between 1997 and 1998 convinced management that there was a problem. The first reaction was apprehension. Two other sister plants had already replaced their rotary dryers for similar reasons. The inspection reports convinced management that the dryer was likely to fail catastrophically and that installation of a new dryer seemed the only available course of action. To them, there was little need for any further investigation. The proposed new failure analysis was almost rejected out of hand.

The second concern was that advocating a failure analysis in order to assist in developing specifications for a new rotary dryer would cause the proposal to again be rejected out of hand. Years of experience indicated that unless any piece of equipment was on the verge of a catastrophic failure, a large capital outlay was not going to be approved. The dryer continued to break down and consume maintenance resources. To some, a large capital expenditure was easier to approve when there was no remaining useful life. Some retired managers referred to this style of operation as "harvesting." Emergency maintenance had gone on for eight years, and there was no way to determine how much longer the dryer would continue to operate. The rotary dryer didn't have the prerequisite "look" to it yet, but with numerous shell buckles and a 1200 circumferential weld crack next to the feed-end tire, the signs were emerging.

The third and most important concern was that business profit margins were not as high as they had been, and allowing equipment to run to failure no longer made economic sense to the new managers. With the recent advent of corporate reliability programs such as "Responsible Care" and "Mechanical Integrity", it became imperative to the "new-age" managers that the term "harvest" be buried. Based upon these three very important aspects, the presentation to management suggested that if the dryer could be repaired for $200K, then the cost of a new $2MM dryer could be avoided, and capital could be saved. The new analysis would first argue why the dryer couldn't be repaired before recommending replacement. Repairing the dryer was more attractive to management than replacement. However, everyone agreed that $200K, although small in comparison to $2MM, was a lot of money, especially if spent on the wrong repairs. In the end, approval was obtained for the study described in this paper.

HURDLE No. 2 - CHOOSING THE TECHNICAL RESOURCES

Should the OEM be brought in, or a consultant? Should we use a sister plant's new dryer design and clone it to fit our needs? How much of our proprietary operating knowledge should be shared outside the plant? Utilizing outside consultants required a level of trust in the knowledge of the consultant. How did one know what expertise to use in order to solve a problem? Who would know what to ask for? What requisite backgrounds and appropriate qualifications should the consultant have? Some managers were unaware of the information to be gained from any type of failure analysis technique. Who would know whether the consultant had addressed the proper issues? Experience with some outside consultants indicated potential difficulties with making the report understandable. In some cases, the technical conclusions were correct but the recommendations fell short of the company's needs. This usually occurred because the consultant was not always fully aware of the company's needs. In defense of the consultants, managers didn't always think logically. In defense of the managers, life and making a profit wasn't always logical, timely, or predictable. Some managers weren't fully aware that there might be more than one way to obtain the same answer, and thereby solve the problem. If a recommendation wasn't favorably received an alternative solution was usually not requested, and the recommendation was not implemented. The successful failure analysis would involve breaking up the problem into well defined objectives within manageable fields of expertise.

A second administrative problem had to be resolved involving the trust level. After the request for funding was submitted, but before action was taken, the lead author attended an ASM seminar on fractography taught by the third author of this paper. Dr. Becker was able to answer several questions the lead author had regarding crack initiation. At this point the decision was made to include Dr. Becker in the investigation. Dr. Becker agreed to examine 24 fracture samples removed from the dryer. The second step in putting the team together was to have a finite element analysis performed. It was important that this person have knowledge of root-cause failure analysis together with practical experience, so that an adequate trust level would support computational results. Mr. Jan Smith of Applied Reliability Inc., the second author, and a prior professional supervisor of the lead author, agreed to join the team.

HURDLE No. 3 - OBTAINING THE SAMPLES

Another problem was peer pressure. Failure analysis, microstructural interpretation and fractographic analysis were unfamiliar to the workforce and to frontline supervision. Their opinions influenced the day-to-day operation and maintenance of the plant. If they didn't understand something, they wouldn't do it. If they were forced to do something, ridicule would follow. It was intimidating to the uninitiated. Thus, it was important to have their "buy-in." The lead author had been working with the general plant population over the past 10 years illustrating where failure analysis had directly influenced the outcome of various maintenance repairs. He also pointed out where specific modifications were the result of such analytical findings. Building this trust level allowed for easier acceptance of taking the fracture samples by cutting holes in the dryer. There was still a risk; this trust level could be easily lost if taking the samples created new failures. This would cast a dark shadow on failure analysis in general. Faith persevered, and after convincing a cautious management that it was necessary, 24 samples were removed (Fig. 11). Usually five samples were taken from each region of interest. Hopefully, this would minimize the potential problem of having a fracture damaged to the point that a fractographic analysis could not be obtained. It also provided for consistency in the evaluation of the results. There were four regions of interest: (a) the feed end, (b and c) two regions in the mid-section, one directly underneath the jolters and the second away from the jolters, and (d) the hot discharge end.

FINDING THE UNEXPECTED

Everyone assumed that the welding procedures used to attach the lifters provided adequate strength of the joint, as had been assumed in the initial analysis in 1988. However, when the samples were cut from the dryer many of them fell apart and revealed what appeared to be a very large amount of lack of fusion and lack of penetration (Fig.'s 12 & 13 respectively, below). Could such a degree of poor welding found in the samples be representative of the rest of the weldments? If this were indeed the case the jolter and thermal stress failure theories would be compromised and would have to be revisited. There was a good possibility that additional aggravating failure causes existed and the search for these new causes would have to begin in earnest. To challenge this finding of bad welding technique, a certified NDE inspection company was brought in to ultrasonically check the welds using shear wave technology. The technique can differentiate between weld porosity, lack of fusion and lack of penetration based upon the returning shape of the waveform. The results showed a severe degree of lack of fusion and lack of penetration in nearly 60% of the total linear length of welds not yet visibly cracked.

THE ANALYSIS BEGINS

Samples were sent to Dr. Becker with a well defined request. Using a combination of microstructural and fractographic information, characterize the fracture surface in terms of crack initiation site, crack propagation mechanism (cyclic, monotonic), and crack growth direction. Secondly, to determine any role of the microstructure in influencing any of the fractographic information.

To alter the validity of the initial assumption that cracking was initiated in the lifters due to thermal stresses, there would have to be fractographic evidence that cracking was initiated due to bending loading. The first sets of samples to be analyzed were taken midway between the feed-end and feed-end tire location (Fig. 14 right). At this location they were far enough away from the tire and therefore could not be influenced by any conditions at the tire.

All samples were usually cut at two locations, being either cut at the crack tip, or cut ahead of the crack in what usually appeared to be good material. Macroscopic examination showed the presence of beach marks (Fig. 15right) and microscopic examination showed the presence of fatigue striations (Fig. 16 below). Initial findings indicated the presence of bending fatigue in this region (Fig. 15 & 16). Crack initiation always occurred at the toe of the weld, sometimes in a single location at the end of the lifter, and sometimes in multiple locations (Fig 15). In the latter case, cracks initiated not only at the end of the lifter, but also along the sides. Cracks at the end of the lifter propagated radially towards the outer diameter of the shell (Fig. 15).

Extensive pitting attack in the weld metal provided microscale stress concentrations that were superimposed on the stress concentration associated with the weld joint (Fig 17 right).

Metallographic examination of the majority of samples in other regions of the dryer revealed the same thing. The welding problems did not appear to be the cause for the cracks but they appeared to assist in crack propagation. Cracks initiating along the sides propagated towards a lack of fusion defect located at the juncture of the lifter and the shell, along the centerline of the lifter (Fig. 18 right). Lack of fusion was identified in every sample submitted. None of the samples between the feed-end and the midsection of the dryer gave any indication that crack initiation was due to net section yielding in the shell that might be caused by an overload condition.

In addition, microstructural examination of the lifters showed no changes in microstructure that would encourage crack initiation or propagation (Fig. 19 right).

ANOMALY OR THE SMOKING GUN

The search was still on to find a sample that would support a thermal stress theory. The finite element thermal model did show that the last four rows were "thermally stressed." There was such a sample (Fig. 20 below), and it was ideally located within this region. It was located more than eight feet away from the jolters. The crack had some of the classic macroscopic features that were typical of nearly every other fracture sample. The crack ran through the full width of the end of the lifter for several inches, and ran along the top toe of the weld, on the lifter side. In addition, the weld was the original weld and not a repair weld; the relief groove had been put in prior to any crack initiation. If any sample was a good candidate, it was this one.

If the lifter were growing more than the shell then the fracture surface should have had an oblong open-ended dimple configuration when viewed under SEM. In addition, the closed ends of the dimples would have pointed in the direction of travel of the lifter with respect to the shell (Ref. 3). The orientation would have clearly showed the lifter growing, in shear, toward its free end. That was not to be the case. Fractographic analysis showed ductile transgranular microvoid coalescence on the fracture surfaces that indicated monotonic overload (Fig. 21 right). This usually happens in axial or bending loading and not in shear. What caused the tension at the leading edge of the crack? If bad welding, thermal stresses, and jolters didn't propagate the crack then what did?

STRESS CORROSION CRACKING OR STRESS CORROSION FATIGUE

The hot discharge end of the dryer was fabricated from a micro-alloyed austenitic heat resistant material. Field inspection revealed that all cracks at this end propagated outward toward the outer diameter of the shell (Fig. 22 right).

An important question to be answered was whether the cause for failure was stress corrosion cracking (SCC) or stress corrosion fatigue (SCF). Chemical analysis of the product stream had revealed the presence of 650 to 700 ppm of chloride. However, three factors were important in discarding the model for SCC. According to the material manufacturer, SCC should not occur in this material at the elevated temperatures associated with the normal operation of the dryer. Secondly, SCC is always transgranular in this micro-alloy; the normal cause of failure was intergranular.

The third factor, metallographic examination showed that intergranular cracking occurred in the shell with essentially no concurrent transgranular plastic deformation (Figs. 23 - 26). The only evidence of plastic deformation was slightly bent annealing twins in a few grains adjacent to the crack (Fig. 23 right). This implied that the material was not loaded above the net-section yield strength. Microhardness traverses from the shell through the weld metal and into the lifter indicated a slightly higher hardness of the shell material, but the grain boundary precipitate produced a material with little inherent ductility under the operating conditions (Fig. 25 lower right). Top toe weld cracks in the 309 stainless steel lifter material tended not to propagate as far as the bottom toe weld cracks, and in no specimens did the cracks propagate to failure.

Of importance here, as in the rest of the samples, it was observed that the crack path wasn't necessarily associated with any distinct changes in microstructure. Instead, it appeared that the crack path was controlled more so by geometric and loading variables.

The role of chlorides in an SCF model is currently under study.

FINITE ELEMENT ANALYSIS - ESTABLISHING THE FINITE ELEMENT (FE) MODEL CREDIBILITY

It was important that the FE model be a valid representation of the stresses occurring within the rotary dryer. The model would be valuable in evaluating whether or not the dryer could be modified in the field to remedy the chronic failures, or whether replacement was justified. If replacement were required, then the model would prove valuable in assessing the capabilities of potential manufacturers by specifically querying them on design weaknesses found in the model. The lead author felt that in order for the model to be valid the dryer needed to have the lifters represented in the model. The industry normally omits the lifters to simplify the model and they accept the assumption that the lifters contribute very little in terms of structural interaction with the shell. As was to be seen from the model, this was not the case. However seemingly small in magnitude, the lifters did in fact carry a structural load (Fig.27 right). They acted like structural angle iron supports.

The size and complexity of the dryer, the unknown mechanical and thermal loadings, and the numerous cracks whose existence, location and direction had to be predicted by the model should have made it obvious that the task was impossible. However, the lead author's optimism apparently was contagious. Most frequently, a large structure has only one or a few areas of interest and the FE model mesh is coarse in all but these areas of interest. However, the dryer was cracking in numerous locations so that the large structure had to be modeled in some detail in its entirety (Fig. 28 lower right). The second author's approach was to use a moderately refined full model, identify the most appropriate locations for high detail, and then insert those sections of high detail into the full model. The finite element analysis consisted of a family of 34 different FE models (e.g. Fig. 27), many of which had numerous submodels. The largest single model contained over 50,000 elements and required over 250,000 equations to be solved simultaneously.

A steady state material balance, and visual observations of product flow through the unit during the typical start-up, allowed a transient material balance to be estimated. The product weight distribution along the axis and the total product weight were derived from this transient material balance. The distribution of product weight around the circumference at each axial location was estimated from the visual observations during start-up. With this data, 1,450 vertical forces of appropriate magnitude and location were applied to the FE models to simulate product loads.

VALIDATION

If any location was well documented with failures, it was the feed end tire. The tire had been a continual maintenance problem since initial installation in 1987. Axial cracking occurred underneath the tire in the thicker reinforcing ring. The first finite element model results obtained in 1989 addressed this cracking, but the global model was never configured to obtain other results. The lead author gave a general description of the problems experienced on the dryer to the engineers at Applied Reliability who were to do the finite element analysis. They were not provided information regarding observed cracking. If the model were correct, it would have to show that there was a problem on the downstream side of the feed end tire. The model of the shell underneath the tire did in fact show that there was a highly stressed region located along the circumferential weld opposite the feed end side of the tire (Fig. 29 circled). This weld had failed 120 degrees around the circumference.

In addition, it showed that the lifters were actually loaded in the spill region (Fig. 27). The stress pattern on the lifters was such that an overload would tend to cause these lifters to bow in a manner similar to what had been observed in the field (Fig. 30 below). The model was valid.

FE MODEL OF FEED END LIFTER STRESS

It was important to verify that there was an interrelationship between the lifters and the shell. The best place to verify this was at the feed end lifters. It was at the intersection between two rows that fractographic and SEM examination revealed a bending fatigue failure. Would the FE model also show such a relationship? Did the lifters play an integral role in the failure? The high-resolution model revealed high shell stress at the intersections between the two rows (Fig. 31 right). A typical fracture is shown in Fig. 32 below. If the lifters didn't have a space between them there would be no problem. It also showed that although the maximum principle tensile stress between the lifters was relatively small at the top of the dryer's rotation, the alternating principle stress was five times greater. In terms of fatigue, this high alternating stress and related strain, was severe.

PROVING THE FE THERMAL MODEL

It was crucial that the FE thermal model be as accurate as possible. Therefore, it was important to obtain accurate and representative field data. Fortunately, the rotary dryer's temperature service life was obtained by reviewing the process computer database at four hot locations. It was easy to see exactly what the most severe temperature swings looked like. From 70 individual product runs, a representative time-temperature profile was selected and utilized as input data for the FE thermal calculations.

A conventional heat transfer analysis was conducted on the shell wall at each row of the lifters (not including the lifters). This was done in order to obtain the convection coefficients that would be used in the two-dimensional FE thermal models of the walls with the lifters. The output of each 2-D model was the temperature profile of the shell and the lifter at that particular location along the axis of the dryer. Inside and outside shell temperatures, derived by conventional heat transfer methods on the wall, and the FE method, matched at the shell, remote from any lifter. This validation allowed the FE models to be used to determine the temperature of the lifters. The convection coefficients obtained and used in the FE models were effective convection coefficients that included the effect of product insulating the inside wall. This was because for any given axial location along the dryer, there was some inside surface around the circumference that was in direct contact with the hot gas and some that was completely covered with up to a few feet of product. This complication was readily resolved by recognizing that the rotating dryer allowed averaging the inside surface conditions and the inside convection coefficients derived in the thermal loading procedure.

Prior to running the model of the entire unit, a model was configured for an individual row. This row was in what was believed to be the field of influence for any thermal strains. The model showed a very low temperature difference at the lifter's weld interface (Fig. 33 above). Infrared images of the shell at the same location (Fig. 10) closely paralleled the thermal profiles of the 2-D FE thermal models. With this validation, the temperatures from the 2-D models were then extracted and applied to the 3-D models.

CHALLENGING THE THERMAL STRESS THEORY

The FE thermal model showed thermal stresses were present, although the stress pattern was different from that initially assumed. Instead of being along the entire length of the lifter, the highest stresses were located at the ends of the lifters (Figs. 34 right & 35 lower right). This pattern was consistent throughout the entire length of the rotary dryer. When the model's deflection mode was turned on, it revealed that the first three rows of lifters closest to the hot discharge end deformed. The tips of the lifters were hotter than the shell and were expanding more than at the lifter/shell junction.

This difference caused the lifters to arch (Fig. 36 below). Subsequently, the ends of the lifters were digging into the shell in a high compressive three-dimensional state of stress (Fig. 35 above right). The shell distortion was verified out in the field (Fig. 37). The troubling part was that of the first four rows of lifters that were thermally stressed, only one end of the same row of lifters was severely cracking, and those cracks lead into the shell. In terms of numbers, that meant only 10 locations out of 80 were cracking. Those weren't good odds for a thermal stress theory. What was so different about this end of lifters?

A higher resolution FE model was developed that included just the ends of two lifters (Fig. 38 right). The end of one lifter was from row "A" (the problem row) while the other lifter was from the next row down, row "B" (no cracking). This model showed that there was a difference in the stress level between the two rows. Results indicated that the stresses at row "B" were almost half as much as those at row "A". How could there be a difference in "thermal" stress levels when the ends of each row were right next to each other and on the same circumferential line? The only difference, and a big one, was that row "A" was welded to a thick reinforcing ring underneath the hot discharge end tire. The lifter jutted out past the reinforcing ring by 18". The lifter couldn't bow uniformly. Therefore, the strain went in the path of least resistance, or out into the considerably thinner shell region. This reaction effected only this row of lifters. Row "B" lifters were not tied to the reinforcing ring; they were simply welded to the thinner, and comparatively weaker, shell.

However, there was still a problem. The FE thermal model showed that the lifter tips were in a high state of compression, a condition in which cracks do not propagate. Something other than thermal stresses had to be opening the cracks up. There was an explanation. It had to do with what was known in pressure vessel design as a "self-limiting" load. In its simplest terms, such a load when first applied will cause a minute degree of plastic deformation, after which, when the load is removed there are residual stresses that remain. As long as subsequent loads are equal to, or less than, the original applied load, the material response will be elastic (Ref. 4). In other words, the thermal load caused the lifters to bow, which in turn distorted the shell in compression (Figs. 37 & 39). This in turn created residual tensile loads in the lifters. Depending upon whether the lifter had a relief groove cut in it or not, the crack either started at the root of the relief groove, or at the end of the lifter (Fig. 39 below). The thermal expansion stresses in the lifters were neutralized as evidenced by the yield creep that could be seen (Figs. 8, & 37). However, the residual stresses had already been established prior to cutting the stress relief grooves.

The thermal model was turned off with just the mechanical and gravitational loads applied. A closer look at lifter "B" in the 3 o'clock position (loaded) revealed that the lifter was under a tensile stress that was perpendicular to the longitudinal axis of the weld (Fig. 40 right). The highest stress was at the tip. This tensile load opened the cracks. Microstructural examination (Fig. 21), field inspections, and the FE thermal and mechanical models supported this failure model.

THE COMMON DENOMINATOR

Additional evidence of shell-lifter interaction was found in the mid-section of the dryer underneath the jolter area. The shell had deformed in compression and subsequently imposed tensile loads on the lifters (Figs. 41 - 43 right). This was in an area far away from any thermal stress affects, but it was located directly underneath the jolters. Obviously, the jolters were not the cause of the shell deformation. The deformation was probably due to an intermittent or one-time overload condition. Similar types of shell compression failures that were based upon other contributing factors had been documented in the kiln industry (Ref. 5). It may have coincided with the severe deformation of some lifters that were located near the feed-end tire (Fig. 30); this was unknown. However, the deformation allowed the lead author to see how the lifters interacted with the shell on a global scale. The FE models were showing an interaction. Both microstructural evaluation and FE analysis determined that pitting attack at the feed-end of the dryer, coupled with the shell stresses between the lifters, caused the lifter cracks to initiate and propagate. Microstructural analysis also showed where there was poor welding throughout the dryer. This too would promote crack propagation. It was obvious from the analytical work that the imposed stress concentrations would initiate cracks.

There was a common denominator throughout the dryer. The lifters created a significant proportional discontinuity with respect to the shell at the edge of the lifter. Because of this discontinuity, were the cracks a result of "maximum constraint?" The theory of "maximum-constraint" stipulated that strain must always accompany stress; otherwise, a crack would result in order to accommodate the required strain. If a three dimensional state of stress existed then it might be difficult to accommodate that strain, especially with such a large proportional discontinuity as a lifter, and a crack would result.

Was this theory applicable? Inspection of every lifter revealed that cracks always started at the ends that did not have relief grooves cut into them. Also noted was that wherever there was a relief groove, the crack always started at the root of the groove (Fig. 44 right). Wherever a relief groove wasn't cracked, it was due to a crack running along a bad weld from the end of the lifter. Moreover, when these weld cracks were repaired, a crack would eventually start at the root of the relief groove. Most cracks eventually turned into the shell. All of this was occurring regardless of lifter location, temperature differential, and jolter location. The maximum-constraint failure model was applicable.

CHALLENGING THE DESIGN ITSELF

All of the evidence was pointing toward a problem with the lifter welds and the way the lifters interacted with the shell. Although all of the evidence was carefully scrutinized and cross-checked, the final acid test was to compare this dryer's performance with other dryers in the plant. The plant managers would certainly ask this question after the failure analysis report was presented. There were more than two dozen rotary units operating throughout the plant. Most of them had their lifters welded full length. A review of inspection reports over the last several years on six other slower and much older rotating units that were in similar service showed that they were indeed failing in a similar manner when it came to their lifters. The lifters were failing at the ends, and never in the middle. Cracks progressed through to the shell's outer surface. Although they were older in terms of years, they were the same in age when comparing load cycles. Most of the other units operated at 4 to 5 RPM. It took them 15 to 20 years to attain the same number of load cycles that our rotary dryer attained in just 8 years.

HURDLE No. 4 - CHOOSING THE RECOMMENDATIONS

It was clear that the space between the feed end lifters acted like a discontinuity. Stresses in the shell exceeded the allowable values for fatigue. It was clear that the FE model correctly explained everything that was happening, having been corroborated by much of the work of W. Becker. With the model, the bending fatigue failures could be solved, the feed end tire cracking could be solved, and the stress cracking minimized at the hot discharge end. All of this could be accomplished as a repair. It was a known fact that making recommendations was the easy part; it was yet another thing to get them implemented. It was important to pick the right recommendation and to answer the question "Will management perceive the recommendation as interfering with the desired performance of the dryer?" Making the dryer last longer was inconsequential to the ramifications that would be encountered if the repairs ruined the product in some fashion, or created a new maintenance headache. A positive perception of the recommendation was the essential element.

There weren't too many ways that the stress could be reduced. The technically correct solution would have been to relocate the feed-end support pier. This would significantly reduce the stresses in this region to below those required for overload. It would eliminate most of the problems that were occurring in addition to the lifter problem. However, there would be major modifications to the shell and to the building. Both would be very expensive, and would severely impact production. Another course of action would be to leave the feed-end support pier where it was and stiffen the rotary dryer. This could be accomplished one of two ways, by either increasing the shell thickness or by adding stiffening rings to the dryer. Again, increasing the shell thickness would be a good idea for a new dryer. However, an OEM would be requested to design and fabricate an "off-spec" dryer, an even more expensive alternative. Everything would have to be upgraded to handle the increased shell weight.

The last alternative, adding stiffening rings, seemed to be the recommendation of choice. There were many risks. They might not work. The welding could be poor and they could break loose. Worst of all, there was the perception that they would interfere with the performance of the jolters. Just the mere perception of downgrading the performance of the jolters was enough to squelch the idea of using stiffening rings. Nevertheless, the idea was tempting. Therefore, an FE model was developed with stiffening rings incorporated into the model (Fig. 45). If the model showed significant reductions in the overall stress levels then a study would begin on the effect of the rings on the jolters. As it turned out, the stiffening rings significantly reduced the stress levels throughout the dryer. They were a viable repair alternative. Three of them would have to be placed where the jolters were located so it was time now to evaluate how the jolters worked in order to determine whether the stiffening rings would adversely affect them.

HURDLE No. 5 - DEBUNKING THE JOLTERS

There was one very-important piece of equipment that proved itself invaluable over the years in helping to make the rotary dryer perform up to expectations. It was known as "the jolter." There were several of them. When they weren't performing well, the rotary dryer wasn't performing well. Production personnel were very concerned with the term "stiffening ring." To them that meant the rings would stiffen the shell to the point where the jolters would not be able to deflect the shell and hence, dislodge the product. A recommendation that could ruin jolter performance would not be favorably received.

It was important to understand how the jolters worked, so it was time to learn. Vibration waveform analysis was utilized to capture the dynamic energy imparted by the jolter hammerhead. The jolters were positioned closely together. If one jolter failed while in service, the product would build up immediately. Therefore, it was safe to assume that the effective field of influence of a jolter wasn't large. The questions to answer were what made the jolter effective, and what was the distance before they became ineffective? The vibration study showed that the energy imparted by the hammerhead blow was equivalent to a short 5,900 kg (13,000 lb.) burst of energy lasting less than 10 milliseconds (Fig. 46 right). This amounted to a unit stress on the shell itself of only 3.2 Mpa (460 psi).

As the pulse traveled away from the hammerhead, the pulse RMS energy level decreased significantly when compared with the average RMS energy level (Fig. 47 right). Therefore, it was the pulse of energy that did all of the work, and not the average energy. This wasn't a surprise. The surprising thing was that the energy dissipated so quickly. It was noted that there was a dead spot in-between each jolter where there was very little pulse energy. From this, it was determined that the location of the stiffening rings would be out of the field of influence of each jolter. Now it was just a matter of convincing the Production personnel. This was accomplished in the presentation. An appropriate analogy was used, something everyone understood.

THE LAST HURDLE - PEOPLE

This failure analysis came with emotional "baggage." The plant was over 50 years old, and many people had well over 25 years of experience working there. To some people a failure analysis was equivalent to an investigation that would purposely point blame on someone. Like it or not, this was the perception taken by the workforce. This type of managerial approach was never tolerated by the company. However, the perception was engrained into the aging work force. Fear and suspicion were the operative words of the past. Maintenance personnel blamed the operators for "torching the unit" while production personnel blamed the mechanics for not knowing how to weld or fix components. Between the two of them it was a "fact" that if the operators knew how to run the unit and if the mechanics knew how to fix stuff none of this would be happening in the first place. To make matters worse, it seemed that the innuendoes and blaming were always preceded by the word "they." Who were "they?" It was never possible to follow up and verify. any innuendoes because "they" couldn't be found." The analysis required asking everyone questions and watching how they did their jobs, this in turn created a level of uneasiness. Normally ten years ago, there would not have been any interaction, the hourly workforce would not have participated. However, the only way that the lead author got cooperation was due to a trust level that took eight years to establish. It was important to recognize this fear and acknowledge its power. Throughout the ongoing failure analysis, it was important to continually explain the investigation and the findings to everyone. As it turned out, it encouraged further cooperation from the hourly workforce. The failure analysis gave the appearance that many people would be "exonerated," and the real problems finally brought to light.

It seemed that with the cooperation, and the facts fitting nicely into place, it would be easy to make some recommendations. It wasn't that easy. The time for "second guessing" began. Just what did everyone want anyway? Well Maintenance surely didn't want to work on it anymore. They'd rather see a brand new rotary dryer instead. On the other hand, traditional Production values meant that they wanted it fixed but they didn't want to give the unit up for any extended length of time in order to fix it, an oxymoron in itself. Then there was the conservative approach; everyone else in the corporation had replaced their rotary dryers, so maybe we should.

Nobody was staying in the same supervisory position for more than two to three years. As one group of people became familiar with the issues, they would then be promoted, reassigned, or leave the company. Then a new group of supervisors and superintendents would come in, and it would start all over again. Once again, trust levels would have to be nurtured before convincing anyone that work needed to begin on the rotary dryer. The value of failure analysis would have to be re-learned. What did this have to do with making the recommendations? If the recommendations weren't favorably received by everyone, there was a good chance that nothing would be done. As in the past, people would leave, a new set of management people would come in, and the problem would continue. We would continue repairing the dryer in the same old-fashioned way until it finally broke into two pieces.

THE FAILURE ANALYSIS FINDINGS AND EXONERATION

Before things could move forward with presenting the recommendations, there had to be closure on a lot of emotional collateral damage. How was it presented? Simply like this: The operators didn't torch the unit; the hot discharge was failing because the material had reached its useful life. The mechanics weren't bad welders because the repair welds were on a dryer that was structurally too weak. Production didn't intentionally overload the dryer. The process operating parameters of the dryer were the same then as they were today. The OEM wasn't at fault, the design was based upon theoretical process assumptions given to them. The process design team wasn't at fault because it was an unknown process at the time. If the dryer had rotated at 5 RPM instead of 10 RPM the propensity of problems would not have been as great. Above all else, none of the reasons had anything to do with how people did their jobs.

THE RECOMMENDATION

The dryer could be repaired by installing stiffening rings. This modification would reduce the working stresses by 37% without modifying any other component. This would allow the business to recapture the remaining 10 years of life. The repair would cost around $250K. A new "in-kind" dryer had 29 documented design weaknesses that would begin failing within six months of installation. A new re-designed dryer would cost significantly more than $2MM. The majority consensus was to repair the rotary dryer.

Was the failure analysis successful? That all depends. The plant was in the process of downsizing. New economic and personnel pressures would be brought to bear on everyone. There would be a shift in management and the new managers would be involved in deciding the rotary dryer's future. These managers would have to become acquainted with the dryer's problems. Would the rotary dryer problem be able to compete with the other important challenges facing the new managers? The credibility and perceived value of the failure analysis would be scrutinized again, and rightfully so. Of course the recommendations from the failure analysis would be championed by those closest to the project, but it was frustrating to be so close and yet apparently, fall short again. A chilling thought came to mind; would those who had championed the cause still be there in a year to see it through?

SUMMARY

The rotary dryer could be repaired instead of replaced. Replacing the dryer in-kind would simply repeat the same failures within the same time frame. In the course of this failure analysis it became evident that an effective combination of troubleshooting tools proved to be the use of microstructural interpretation, fractographic analysis, and finite element analysis. Working together, all three disciplines clearly showed there were other issues that plagued the rotary dryer. Peoples' perception of the problems had turned into fact and it took a lot of intensive analysis to debunk those engrained perceptions. The primary causes for failure were an overloaded dryer, bad lifter configuration, and poor weld quality of the lifters. These problems were hidden behind a myriad of circumstantial evidence, innuendo, and false accusations.

CONCLUSION

In an industrial environment, there are good failure analyses and there are successful failure analyses. The difference being that a successful failure analysis can be measured in terms of whether the key recommendations are implemented. It is not enough to just discover the failure cause and categorize it for the client. It is not enough to present technically correct recommendations. Beyond all of that, it is imperative that the failure analyst makes certain that the client understands the circumstances and events that lead up to each failure cause. Otherwise, that client isn't going to believe in the recommendations no matter how technically correct. Industrial failure analysis is a never-ending commitment toward educating the general plant population and management in regards to various failure analysis techniques and their benefits. To discount their perception of a problem is a mistake. Perception is based largely on experience, right or wrong, and it is a cornerstone of a person's value system. The entire course of this failure analysis had to do with listening to people. From the operator to the mechanic, from the welder to the foremen, from the engineers to the superintendents, they all had their own perception of the problems with the dryer, maintenance and production alike. The challenge was in finding the common thread that would pull all of it together. A common thread that everyone would believe in and support. As with previous failure analyses on this dryer, the recommendations were an amalgamation of compromise and integrity designed to be effective in the real world of both physics and economics. One thing is certain, reliability is a never-ending journey toward a continuing process of improvement, one failure at a time.

ACKNOWLEDGEMENTS

Finite element modeling and heat transfer analysis contributions - David Mele PE, and Brad Bourgeois PE, respectively, Applied Reliability, Inc., Baton Rouge, LA.

Metallography, micro-hardness testing, macro-photography, and SEM specimen preparation contributions - Jason Bartlett and Faris Ohan, under the direction of W. Becker, Dept. of Material Science and Engineering, Univ. of Tennessee.

References

1. American Welding Society, ANSI/AWS D1.1:1998 Structural Welding Code - Steel, pp. 13-17, AWS (1998)

2. Specialty Steel Industry of North America, Designer Handbook Design Guidelines for the Selection and use of Stainless Steels, p. 14, SSIA (1995)

3. American Society of Metals, ASM Handbook Volume 12 Fractography, pp. 12-15, ASM International (1992)

4. H. Bednar, Pressure Vessel Design Handbook 2nd Edition, pp. 24-27, Van Nostrand Reinhold Company, New York (1986)

5. R. Chapman, Recommended Procedures for Mechanical Analysis of Rotary Kilns, pp. 12-13, Fuller Company, Bethlehem, PA. (1985)