Life assessment of steam reformer catalyst tubes - MCCm-c-c.nl/wp-content/uploads/2006/05/publ-16.pdf · Life assessment of steam reformer catalyst tubes - diameter decrease is not

Paper no. 5F

Life assessment of steam reformer catalyst tubes

- diameter decrease is not negative creep ! -

by :

R. Gommans G. Kamphuis J. M. Brear JA. Schelling P. Bakker J. M. Church DSM Stamicarbon Methanor ERA Technology Geleen, NL Delfzijl, NL Leatherhead, GB

prepared for presentation at

the 46th Annual Safety in Ammonia Plants and Related Facilities Symposium

in Montreal, Quebec, Canada on September 17-20, 2001

copyright J.A.Schelling, DSM Stamicarbon

UNPUBLISHED

AIChE shall not be responsible for statements or opinions contained in papers or printed in its publications

- 1 -

Life assessment of steam reformer catalyst tubes

- diameter decrease is not negative creep ! -

ABSTRACT Steam reformers provide a primary source for hydrogen in syngas production. The

highly endothermic reaction takes place in vertical, catalyst-filled tubes that are

directly fired. Because of the severity of the operating conditions, these tubes are

fabricated from centrifugally cast, thick section material, typically made of HK40,

HP-Nb or HP-MA.

The dominant loading on these catalyst tubes is the through-wall thermal stress and

life consumption is by cyclic creep relaxation, on a time-scale controlled by the

operational pattern of the unit. Strain and damage accumulate through life and may

respectively be monitored by diametral measurements and non-destructive

techniques based on eddy current or ultra-sonic methods.

Detailed investigation shows that the evolution of diametral strain is complex. Early

in life, the outside diameter may decrease whilst the internal diameter increases.

Later, both internal and external diameters increase with time.

A rigorous understanding of this behaviour is necessary in order to reconcile tube

measurements and to allow accurate predictive modelling of catalyst tube life.

This paper elaborates the theoretical basis for this behaviour and provides

validation from a reformer serving a methanol plant. The implications for analytical

life prediction and practical tube monitoring are discussed.

- 2 -

1. Introduction Steam reformer units are critical to many processes in refining and chemical plants. They are used in the production of hydrogen for oil refining, ammonia production and direct iron reduction, hydrogen and carbon monoxide for nickel reduction and purification, and syngas as a basis for producing methanol, acetic acid and various other chemicals. In this specific case the issue is production of methanol. The life management of reformer units is ordinarily dominated by the service capability of the radiant catalyst tubes. Because of the severity of the operating conditions, these tubes are fabricated from centrifugally cast, thick section material, typically to the generic specifications HK40, HP-Nb and HP-micro or their proprietary derivatives. Tube life is primarily limited by creep, driven by a combination of internal pressure and through-wall thermal stresses that are generated during start-up cycles and operating transients. Creep life exhaustion is evidenced by progressive grain boundary cavitation which, due to the significant influence of thermal stresses generated during operating transients, initiates within the tube wall towards the bore. The design of tubular components is ordinarily based on pressure stresses, outside wall temperatures and factored lower-bound materials rupture data. However, this does not provide a realistic basis for remaining life assessment of steam reformer catalyst tubes. Indeed, life prediction by an inverse design procedure using actual materials properties and service conditions can result in highly optimistic estimates of future operational capability which are not borne out by service experience. This is because the influence of start/stop-cycles (which have a significant and negative influence on tube life) is not taken into account in this procedure. In the past some attempts were made to incorporate the influence of start/stop-cycles [ref.1,2], but they were not completely successful. ERA Technology has developed a software program, called REFORM, that takes both loading mechanisms (steady-state creep and thermal cycling) into account. The REFORM analysis is performed probabilistically. Process conditions and metal skin temperatures vary with both space and time throughout a reformer unit. A probabilistic approach allows variations in process conditions and material properties to be accounted for. Further it allows the interactive effects of forward creep and stress redistribution to be reconciled for any position along a selected reformer tube. Further information about the scientific background of REFORM can be found in: [ref.3,4,5]. The REFORM analysis was applied to the methanol plant of Methanor-1 in Delfzijl (NL). As a result of the analysis it appeared that a decrease in external diameter had ocurred, a prediction that was later confirmed by diameter measurements. This paper describes the process of the REFORM analysis for the Methanor-1 reformer tubes and the (scientific) background of the diameter decrease, which is not to be seen as negative creep! 2. Specific features of the REFORM model The model has many specific features relating to the treatment of operational, mechanical and materials behaviour in an overall probabilistic framework. These features are briefly described here, but more extensive background can be found elsewhere [ref.3,4,5].

- 3 -

Materials creep behaviour. The creep rupture life is generally given as a parametric curve (such as the Larson-Miller curve). However, this only describes the moment of failure, and not the complete strain-time behaviour that is necessary to predict ductility exhaustion. Also, creep damage (cavitation) is not described by such a parametric curve. The simplest established materials model that simultaneously predicts strain and damage with time is the continuum damage mechanics model developed by Kachanov and Rabotnov [ref.6]. At higher stresses, flow processes dominate and the stress dependencies of creep rate and rupture life are identical. At lower stresses, damage processes (such as cavitation) predominate, leading to premature rupture, lower ductility and a drop in the stress dependency of rupture life compared with that for creep rate. Thus, it reflects the high stress start-up situation experienced by reformer catalyst tubes as well as the lower stress steady-state regime. Figure 1 shows a data set for an HP type alloy to which are fitted several typical materials data descriptors: a power law, an exponential law, a Larson-Miller parameter and the Kachanov model [ref.4]. Examination of these four diagrams shows that it is only the Kachanov model that is capable of adequately describing both the high stress and the low stress behaviour. The exponential and power law fits are not particularly good at the extremes of the stress range and the parametric fit cannot be reliably extrapolated. Figure 1 Comparison of stress-rupture descriptors for HP-Nb type alloy

- 4 -

Creep life consumption. A strain based life fraction rule is employed, since a critical program of creep and stress relaxation testing has demonstrated that this is more realistic, over the stress ranges involved, than a time based approach. A concentric ring model allows the time and position at which local failure first occurs, corresponding to crack initiation, to be recorded. Ordinarily this is at, or close to, the inner tube surface, thus matching service experience. However this does not represent the end of life, since additional service during the crack propagation stage is possible. Indeed a significant proportion of life can be spent with the tube in this condition. The mechanistic creep model employed in the life consumption algorithm can be extended to predict through-wall crack growth, using a damage front propagation method. Examination of samples taken from service, see figure 2, shows that multiple, parallel cracks form, all of similar length.

Figure 2 Cross-section showing damage front propagation

It is thus appropriate that a damage front propagation model is used rather than creep fracture mechanics [ref.6,7]. This method is implemented by recalculating the wall loads as each successive defined ring fails, whilst maintaining the original temperature distribution. The failure times for each ring are noted, enabling the crack growth to be followed through to final failure. It is also possible to calculate, at each time, the corresponding diametral strains and to record the levels of damage predicted in each uncracked ring, thus enabling a complete prediction of crack, damage and strain behaviour.

- 5 -

Modelling of process transients. A summary of shut-down/start-up cycles and significant process upsets is also needed. It is normally sufficient to classify transients into two groups. Minor transients, caused by smaller process upsets or irregularities in firing, primarily affect the through-wall temperature gradient rather than the absolute temperature level. Major transients also affect the absolute temperature of the tube and can lead to the re-imposition of elastic stress levels comparable to those generated at the initial start-up. For conservatism, therefore, major transients are assigned the same severity level as complete shut-down/start-up cycles. Modelling of thermal stress. During start-up, shutdown and operational transients, through-wall thermal stresses are generated. Once the transient has passed, these stresses relax by creep towards a steady-state equilibrium. It is important to model this correctly as the high stresses involved lead to enhanced life consumption during this period of stress relaxation and redistribution. An analytical thermal stress analysis routine is used, which has been confirmed by finite-element analysis methods [ref.5]. Probabilistic procedure. In order to obtain a realistic life prediction for a particular unit, a probabilistic treatment is employed. Statistical distributions of all input variables are determined and sampled using a Monte-Carlo method. Typically, sample sizes of order 10,000 are used, depending on the operator's required critical probability level and the relative ranges of the input distributions. The routines employed can provide an estimate of the confidence of the prediction. Sensitivity analyses may also be performed to investigate the effects of any particularly critical or uncertain input parameter. For each sample set, the crack initiation time, crack growth behaviour and final failure time (tube leakage) are recorded, together with times to pre-selected diametral strain and damage levels. Cumulative probability curves for times to crack initiation, failure and each chosen strain or damage level are produced. Other statistical functions can be derived from these as required. Simulation. A great advantage of a predictive model is that consideration can be given to hypothetical changes in future operation. Frequently it is necessary to investigate the effect of an improvement in catalyst efficiency, feedstock or firing pattern on tube life, or to evaluate the cost benefit of different choices of replacement tube material or geometry. Provided a reasonable estimate of the potential new parameters can be made, then the model can be run several times using the actual history with each candidate future scenario. From an operational point of view, it is often instructive to consider the sensitivity of tube life to the severity or frequency of operational transients. The greatest advantage of the REFORM model is that it allows simulation of changes in operation before the change is actually effected in the plant. This can save much tube life and thus availability of the furnace. Also, furnace capacity may increase without significant loss of reformer tube life.

- 6 -

3. The Methanor-1 reformer The reformer studied here serves a methanol plant. It is of Foster-Wheeler design, terrace fired with 576 tubes in four cells. The majority of the catalyst tubes are of HK40 material, though a few replacements in HP-Nb were made following an operational incident. For assessment purposes, the reformer has been considered to comprise four sections (in this case, the four cells). These were analysed separately. In this paper, for simplicity, only the North-West cell, containing HK40 tubes, is described. 4. Input for the REFORM model The REFORM model needs design and operating input data; these are listed in table 1 for the subject unit. The operating data need to be available as a function of time. From experience it is known that some parameters have a large influence on the calculated life. Specially tube skin temperatures and initial dimensions (diameter and wall thickness) have a large influence.

Table 1a: Input Parameters - Constant Parameters common for all parts of the unit

Fired Tube Length, [m] 11.35 Number of Tubes, [-] 576

Tube Spacing (No. Diameters) 2.1 Heat Duty Absorbed, [kW] 2.0E+08

Current Operational Hrs (HK40 Tubes) 205375

Table 1b: Input Distributions - Parameters common for all cells

Mean St Dev Min Max Dwell Time, [Hrs] 2265.0 2394.0 72.0 12000.0

Process Inlet Temperature, [°C] 519.2 10.7 External Diameter, [m] 0.1087 0.0003 0.1080 0.1099 Internal Diameter, [m] 0.0840 0.0000 0.0840 0.0840

Table 1c: Input Distributions - Parameters specific to each cell (showing NW-cell only)

Section of unit Mean St Dev Min Max Inlet Pressure, [MPa] NW 1.87 0.07 1.55 2.00

Outlet Pressure, [MPa] NW 1.61 0.04 1.45 1.69 Process Inlet Temperature, [°C] NW 494.7 14.0 415.9 542.3

Process Outlet Temperature, [°C] NW 825.7 8.4 780.7 843.5Maximum Metal Skin Temperature, [°C] NW 950.1 13.5 910.0 986.0

- 7 -

Tube skin temperatures were recorded only as maximum temperature readings during each shift. At first, the average temperatures were not recorded. Since the average temperature would be much lower than the maximum temperatures (see figure 3), it was decided that the average temperatures were to be recorded during a limited period of time, to improve the realism of the prediction. Using these average temperatures extended the expected life of the reformer tubes compared to that obtained using only the maximum temperatures.

Remark : in figure 3, the Gauss-curve for the average skin temperature contains only a limited amount of data (sampling during a few months). The Gauss-curve for the maximum skin temperature contains many data (sampling during years), and is therefore higher.

Tsevp Il 5 Tcc

Figure 3

Two Gauss curves with average and maximum TMT's

0.00

0.01

0.02

0.03

0.04

0.05

800 850 900 950 1000

Temperature, °C

Freq

uenc

y

Maximum Metal Skin Temperature, [°C]

Average Metal Skin Temperature, [°C]

he initial outer diameter was only available from the drawings with a relatively broad catter (-0, +1.5mm). To refine this estimate, the outer diameters were measured at the cold nds during a turnaround and these measurements were taken as equivalent to the initial alue. Because of the improved knowledge of the initial diameters, the accuracy of the redicted life improved.

n general, if operating data are available from digital control systems or advanced data oggers, this greatly improves the speed at which the REFORM-analysis can be performed.

. Results of the REFORM-analysis for Methanor-1

he results can be presented as CP-curves (Cumulative Probability curves) for time-to-rack initiation, time-to-failure, time to reach a certain strain level, and time to reach a ertain damage level.

- 8 -

Such CP-curves are presented for the NW-cell of the Methanor-1 reformer (see figure 4). The operating time at the moment of assessment was 215,000 hours. This time is indicated on the figure, together with a series of small lines which each represent one additional year of operation. In figure 4a (crack initiation and failure), it can be observed that crack initiation was predicted to have occurred already, but that tube failure was not to be expected within a short period of time. Some results are :

- the first tube is expected to fail after another 50,000 hours (~6 years) ; - 10% of the tubes are expected to fail after another 200,000 hours (~25 years) ; and - 50% of the tubes are expected to fail after another 400,000 hours (~50 years).

From these results it is clear that –with similar future operating conditions– replacement of large amounts of reformer tubes is not to be expected for a long time. Furthermore, with proper inspection techniques the damaged reformer tubes can be replaced on a planned schedule [ref. 8].

Figure 4 Cumulative probability curves for Methanor NW-cell a. time to crack initiation and failure

0.01%

0.10%

1.00%

10.00%

100.00%

10000 100000 1000000

Total Service Time (Hours)

Cum

ulat

ive

Prob

abili

ty

Initiation Time

95% Confidence Interval

Failure Time

95% Confidence Interval

Current Operational Hours

Subsequent Years Of Operation

The predicted damage levels (represented by failure of the consecutive rings in the model) are given in figure 4b. When a damage level of 30-50% is reached, non-destructive techniques such as US-attenuation will start to register damage. In this case the model predicted that ~25% of the tubes already had a damage level of 50%. Furthermore, the predicted strain levels diagram (Fig 4c) is of interest to observe. The model predicts negative strains, meaning a decreasing outer diameter! This is very strange and was not expected beforehand; one could think it means negative creep ! This strange phenomenon is further explained in detail in section 7.

- 9 -

Figure 4 Cumulative probability curves for Methanor NW-cell

b. time to x% through wall damage c. time to x% strain

0.01%

0.10%

1.00%

10.00%

100.00%

10000 100000 1000000

Total Service Time (Hours)

Cum

ulat

ive

Prob

abili

ty

10% Through Wall Damage

20% Through Wall Damage

30% Through Wall Damage

40% Through Wall Damage

50% Through Wall Damage

Current Operational Hours

Subsequent Years Of Operation

0.01%

0.10%

1.00%

10.00%

100.00%

10000 100000 1000000

Total Service Time (Hours)

Cum

ulat

ive

Prob

abili

ty

-0.2% Strain Accumulation

-0.5% Strain Accumulation

0.2% Strain Accumulation

0.5% Strain Accumulation

1.0% Strain Accumulation

Current Operational Hours

Subsequent Years Of Operation

- 10 -

6. Results of the inspections of the Methanor-1 tubes US-attenuation measurements and (outer) diameter measurements were performed during the turnaround of October 1998 by DSM Stamicarbon. More information about DSM Stamicarbon’s inspection techniques can be found elsewhere [ref. 9,10,11]. From the inspections it became clear that the diameter at the upper terrace was larger than at the lower terrace of the Foster-Wheeler designed reformer furnace. At first instance it was thought that diameter increase (and creep) was progressing faster at the upper terrace. However, when the diameters measured along the tube length were compared to the original specification and the (extra) “zero” measurements at the cold ends, it became clear that the tubes were decreasing in outer diameter. This is represented in figure 5. Thus, the predicted negative strains by the REFORM-analysis correlated well with the measured strains. The US-attenuation measurements showed that more than 10% of the tubes showed significant sound attenuation, which indicates a damage level of 30-50%. This is in agreement with the predicted damage levels by the REFORM-analysis 7. Diameter decrease is not negative creep ! For reformer catalyst tubes, internal pressure stresses during operation are low, such that thermal through-wall stresses dominate tube response. Initially, the thermal stresses (both hoop and axial) are compressive at the outer surface, whilst tensile at the tube bore. This is because during start-up of the reformer the outer part of the tube wall becomes hotter because of the radiation from the burners. Therefore, the outer part of the tube wall wants to increase more in diameter compared to the inner part of the tube wall. However, the material at the outer part will be held by the material at the inner part. This results in compressive stresses at the OD and tensile stresses at the ID. With time at temperature the through wall stresses redistribute to become wholly tensile through wall, reflecting the steady state pressure stress distribution. A situation can be envisaged therefore whereby the outer part of the tube initially strains compressively and subsequently in a tensile manner, whilst towards the bore of a tube response is tensile throughout the same period. Constancy of material volume during stress redistribution implies that the tube will increase in length during this transition to steady state. Significantly, these observations suggest that creep damage, associated with tensile stresses, can occur towards the bore of a reformer tube whilst the outer surface is straining compressively. Ultimately, failure may occur at very low tube dilation levels, even for modern alloys (such as HP-Nb and HP micro -alloy) with reported creep ductilities up to 3-8%. Figure 5 illustrates average external diameter measurements taken for the NE-cell of the Methanor unit over a series of inspections. Included on the figure are one standard deviation confidence intervals and the design tolerances for tube manufacture. In the absence of initial diameter measurements, these have been inferred on the basis of cold end measurements taken during the latest inspection. The data indicate a trend of reducing diameter with time, with the average tube diameter for the last inspection falling below the minimum design specification.

- 11 -

Figure 5 Tube diameter as a function of time, NE-cell A strict cothe probabdeterminisstrain alonextent of cconsistentof the tubediameter. towards thexample gtube dilati Since idensimilar beinspectiontechniquein the refo 8. Sensit Because oinvestigatinput). Wiwill thereftemperatu ERA has temperatutube skin methane smakes the

107

108

109

110

1970 1980 1990 2000

Year

Tube

dia

met

er, m

m

Fired portion

Cold end, 1998

Specification

mparison between observed and predicted tube response is not possible in view of ilistic nature of the REFORM methodology. However, on the basis of mean tic response, figure 6 illustrates predicted time contours of external diameter hoop g the length of a catalyst tube. Initially all positions strain compressively. The ompressive strain is position dependent, being greater towards the top of the unit, with the maximum through wall thermal gradient. Eventually, the upper portion reaches a turning point after which tube dilation occurs, but from a reduced

This upturn in response reflects the contribution of creep damage accumulation e bore of the tube and associated load shedding on to undamaged material. For the iven, tube failure is predicted to occur towards the top of the unit at negligible on levels.

tifying the phenomenon at the Methanor-1 reformer unit, ERA has observed haviour on other similar reformer units. On the basis of the findings, subsequent s at the Methanor plant will further use Stamicarbon’s US-attenuation inspection , in preference to tube dilation monitoring, as a means of monitoring creep damage rmer tube.

ivity study

f the predicted long remaining lives of the catalyst tubes, Methanor decided to e the possibility of increasing the tube skin temperatures (by increasing the heat th an increase in outlet temperature, the methane slip will be reduced and this ore save large amounts of money. With an increase of 30°C on tube skin re, an extra 50 tons of methanol per day is realised.

performed a sensitivity study in order to quantify the effect of increased tube skin res on tube life. Given the results of this, Methanor has decided to increase the temperatures and the outlet temperature in order to profit from the decreased lip. The decrease in tube life was easily compensated. This simulation facility REFORM analysis a valuable tool in decision making.

- 12 -

Figure 6 Modelled Strain Accumulation Along a Tube

-0.014

-0.012

-0.010

-0.008

-0.006

-0.004

-0.002

0.000

0 2 4 6 8 10 12 14

Height [m]

Dia

met

ral S

trai

n, %

99660

201585

405435

423555

475650

Time, hr

9. Concluding remarks The REFORM analysis can predict : - crack initiation and tube failure ;

- strain accumulation with time ; and - damage development with time,

in the form of cumulative probability curves. With the above mentioned results, recommendations can be made about:

- advised future inspections (inspection timing, frequency scope, and inspection method) ;

- advised availability of spare parts - availability of the catalyst tubes and the reformer furnace as a whole.

Also, the inspection strategy can be optimised. When the model has not predicted crack initiation yet (or when only low damage levels are predicted), it is not yet necessary to perform an inspection based on the US-attenuation technique. For optimisation of the prediction it is advisable to compare measured diameters with predicted diameters. In many cases, this implies that inspection costs can be lowered. In summary, it has been shown that using a validated model approach and probabilistic techniques provides quantitative life assessment data in a risk of failure format. This, together with the consequence analysis, enables a risk based life management strategy to be adopted. This type of assessment also allows operational changes to maximise production in response to market demand to be planned and balanced against the price to be paid in terms of the life of the component. Alternatively it allows operational changes to be made to optimise both product yield and component/plant damage accumulation rates.

- 13 -

- 14 -

Acknowledgements The authors would like to thank their colleagues who were involved in this project: ERA Technology: Duncan. Humphrey, John Williamson Methanor: Ab Steenbergen, Jaspert Grootendorst DSM Stamicarbon: Jan Keltjens, Peter Molenaar, Hans Janssen, and Theo Huurdeman The authors would also like to thank the respective management of DSM Stamicarbon, Methanor vof. and ERA Technology Ltd. for permission to publish this paper. Martin Church is now working for the Nuclear Research Group (NRG) in Petten, NL. Rob Gommans is now working as an independent consultant under the name of Gommans Metallurgical Services (GMS) in Stevensweert, NL. References [1]. F.Simonen, “User’s manual for the computer programme TUBE for creep analysis of thick wall tubes”, Battelle project “Materials for steam reformers – II”, Battelle OH (15jan1976) [2]. T.Kawai, T.Mohri, K.Takemura, T.Shibasaki, “Stress analysis for prolonging tube life”, AIChE symposium Denver “Ammonia Plant Safety”, 24 (1983) p.131-139 [3]. J.Brear, P.Aplin, R.Timmins, “The effect of primary creep on the Kachanov -

Rabotnov model - results on ½CrMoV, 1CrMo and Type 316 steels”, ESIS/SIRIUS Int Conf ‘Behaviour of Defects at High Temperatures’ Sheffield, April 1992. ESIS Publication 15, ed. R. A. Ainsworth and R. P. Skelton, Mechanical Engineering Press, London, 1993, pp 401-422

[4]. J.Williamson, J.Brear,"Risk based life management of catalyst tubes and pigtails", Fourth Annual Ammonia & Urea Conference 'Asia 2000', Singapore, June 2000

[5]. J.Brear, J.M.Church, D.Humphrey, M.Zanjani, “Life Assessment of Steam Reformer Radiant Catalyst Tubes - the use of damage front propagation methods", Second HIDA Conf 'Advances in Defect Assessment in High Temperature Plant' MPA, Stuttgart, Germany, October 2000. Paper S6-3

[6]. L.Kachanov, “Introduction to Continuum Damage Mechanics”, Kluwer Academic, 1990. ISBN: 90-247-3319-7

[7]. G.Webster, R.Ainsworth, “High Temperature Component Life Assessment”, Chapman and Hall, London, 1994. ISBN: 0-412-58520-0

[8]. R.Gommans, D.Jakobi, J.L.Jiménez, "State-of-the-art of materials and inspection strategies for reformer tubes and outlet components", to be published at the Ammonia Safety Symposium [9]. J.Boogaard, “Detection of creep damage in reformer tubes”, DSM Stamicarbon symposium (nov.1996) [10]. A.Scheerder, Th.Huurdeman, “Experiences with non-destructive testing of static equipment”, AIChE symposium “Ammonia Plant Safety”, 38 (1998) p.241-252 [11]. A.Scheerder, “Experiences at DSM with NDT and evaluation of creep damages in reformer tubes”, 7th European NDT Conference (Copenhagen, 26-29. May 1998)