January 2017

Heat Transfer

Failure analysis of olefin furnace radiant tubes

A detailed failure investigation into the cracking of ethylene furnace radiant tubes included visual, fractographic, metallographic, scanning electron microscopy/x-ray spectroscopy, and hardness and chemical examinations. These analyses showed that the tube crack was initiated by thermal shock, accompanied by restrained thermal expansion due to internal coke layer formation. Details of the failure’s contributing factors, and recommendations to avoid similar failures, are shared here.

Al-Meshari, A., van Zyl, G., SABIC Technology Centre

A detailed failure investigation into the cracking of ethylene furnace radiant tubes included visual, fractographic, metallographic, scanning electron microscopy/x-ray spectroscopy, and hardness and chemical examinations. These analyses showed that the tube crack was initiated by thermal shock, accompanied by restrained thermal expansion due to internal coke layer formation. Details of the failure’s contributing factors, and recommendations to avoid similar failures, are shared here.


A fire in the radiant section of an olefin furnace necessitated an emergency plant shutdown. An inspection of the furnace internals showed the presence of a circumferential crack at one of the weld joints of the sixth pass, which led to a hydrocarbon leak.

The furnace radiant coils, which had been in service for 3 yr, contained six passes. The sixth pass tubes (i.e. hottest pass) were made of centrifugally cast, heat-resistant alloy 35Cr-45Ni-Nb Micro. The tube metal temperatures of the sixth pass during the normal operation and decoking process were 1,030°C and 1,100°C, respectively.

About 10 months before the failure, a power dip occurred in the middle of run, and the induced-draft fans of all furnaces tripped due to the failure of the emergency generator. After the power failure, the furnace operation was reinstated using ethane feed.


Visual examination

Neither the external nor internal surfaces of the tube sample submitted for analysis showed signs of corrosion, overheating or thinning (FIG. 1). A straight crack running through the weld joint appeared to have initiated from the tube internal surface at the weld root pass. Dimensional measurements performed for ring samples cut from the tube showed no appreciable deformation or bulging. The tube minimum outer diameter (OD) and wall thickness were both greater than the original specification.

Fig. 1. Photos of the tube sample showing the shape and location of the crack.

A macro-fractographic examination clearly showed a single crack initiating from the tube inner diameter (ID), with a relatively flat fracture surface that became rough and showed signs of creep rupture after approximately 1 mm (FIG. 2). A tube ring was cut for a carburization test in accordance with National Association of Corrosion Engineers (NACE) macroetching standard test method 0498-2006. The tube ring was immersed in a mixture of 670 ml distilled water, 200 ml nitric acid (HNO3) and 70 ml hydrofluoric acid (HF). No carburization was visible.

Fig. 2. Photos showing the fracture surfaces.

Chemical analyses

The tube and weld materials were identified using x-ray fluorescence spectrometry (XRF) and a carbon/sulfur analyzer (TABLE 1). The analysis confirmed that the chemical compositions of the materials are comparable to those of 35Cr-45Ni-Nb Micro and filler UTP 3545 Nb. The slightly higher C concentration detected in both materials could be attributed to in-service carburization effect.

Cross-sections of the failed weld were cut, mounted and polished for metallographic examination. FIG. 3 reveals a cross-section with a single, unbranched, oxide (and carbide)-filled crack running perpendicular to the tube surface. The crack originated from the tube internal surface at the weld root pass. Creep microvoids and internal fissures were noticed around the crack, with a higher concentration at the crack tip. The crack appeared to have propagated through oriented microvoids formed at the carbide/matrix interface.

Fig. 3. Crack at the weld root surrounded with numerous creep microvoids (as polished).

Interestingly, cracks were also observed initiating from the tube (ID) base metal near the weld zone (FIG. 4). The cracks appeared to have propagated through metal carbides precipitated within the microstructure.

Fig. 4. Cracks at the tube ID near the weld (A) and inside the weld (B, as polished) propagating through metal carbides, etched by glyceregia.

A weld microstructure examination showed different distribution of chromium and niobium carbides at different sites (i.e., root pass, middle passes and cap passes). Decarburized ID and OD layers caused by depletion of chromium as a result of high-temperature oxidation measured approximate thicknesses of 400 µm and 500 µm, respectively. Internal dark oxide islands were observed scattering at the decarburized layers.

Other than creep cavities and fissures, no signs of other abnormalities were observed in the microstructures of the heat-affected zone.

Similar to the weld cross-section, the tube base metal microstructure showed decarburized ID and OD layers of approximately 400 µm and 500 µm, respectively. The tube material possessed an austenitic microstructure containing a complex network of coarsened and coalesced chromium and niobium-containing carbides, which outlined the boundaries of the original dendrites and also precipitated within the matrix. Such a microstructure is typical of thermally aged, centrifugally cast, heat-resistant alloys. The tube internal surface showed a relatively higher concentration of carbides, which is indicative of a carburization effect.

Micro-hardness testing

Micro-hardness profiling using a Vickers hardness tester with 500-kgf load showed no significant variations between the weld and the base metal (FIG. 5). The average hardness of the base metal was approximately 250 HV, whereas the average hardness of the weld was approximately 240 HV. It is worth noting that the hardness at the crack region was significantly higher than the hardness inside the intact weld. This may indicate the hardening effect associated with the cracking mechanism.

Fig. 5. Locations and results of micro-hardness testing.


Elements constituting scales/layers inside the crack and on the tube internal surface were identified using an energy-dispersive x-ray spectrometer (EDS) attached to a scanning electron microscope (SEM), shown in FIG. 6. The crack contained oxides of mainly silicon and chromium, as well as chromium carbide particles. Different compositions, shapes and morphologies of oxides were detected on the tube and weld internal surfaces. Significant reduction in chromium concentration was noted at the tube substrate. This was caused by chromium diffusion toward the surface and resulted in the formation of chromium oxides. All of these attributes were expected for the tube material as a result of exposure to the high-temperature environment for a relatively long time.

Fig. 6. SEM/EDX analysis of scale formed inside the crack.

Two backscattered electron images of the weld microstructure clearly indicated the variation in carbide distribution at different areas on the weld. Although such distribution can result in variations in the weld properties, this would be extremely difficult to avoid.

As indicated in FIG. 7, in addition to chromium carbide precipitates, a silicon, niobium and nickel-rich phase was detected in the weld and base metal microstructures. This phase is sometimes referred to as “G phase.”

Fig. 7. Variation in carbide distribution at different areas on the weld. The left micrograph was taken from a root pass, while the right was taken from the center weld.


Examination of the failed tube sample revealed a straight circumferential crack that initiated from the tube internal surface at the weld root pass. The crack had a flat fracture surface that, after approximately 1 mm of propagation, turned rough, suggesting the involvement of more than one cracking mechanism:

  • The appearance of the crack initiation region resembles a brittle/thermal shock fracture
  • The dominant fracture mechanism, following the initial brittle fracture, appeared to be creep deformation.

The brittle fracture caused by thermal shock can manifest itself as a long longitudinal crack, circumferential crack and/or a “window” that falls out of the tube. The orientation of the crack depends on the direction of complex stresses generated and applied during the thermal shock. The microstructural analysis indicated crack propagation through carbide phases (i.e., carbide split), which is deemed a sign of brittle fracture.

Brittle fracture in the furnace tube can be produced by stresses generated from significant and rapid temperature drop, similar to that caused by the power dip encountered in this case. During the furnace trip, a temperature drop of more than 500°C was reported. The resulting strain during such a trip is estimated to be within 0.75%–1.5%, which is equal to the rupture ductility of a thermally aged and/or carburized alloy between room temperature and 600°C. This explains why the furnace tube cracked in a brittle manner by splitting the carbides. The probability of brittle cracking is determined by several factors:

  • Temperature drop
  • Thickness of the coke layer inside the tube
  • Degree of the alloy thermal aging and carburization.

Another factor that may have contributed to the crack initiation was the considerable difference between thermal expansions of the alloy and the coke layer, assuming that some coke formation had been present before the power failure. This difference becomes important during normal operation/decoking mode switch and/or during power failure, as the coke buildup inside the tube represents a “coke tube” that is adherently joined to the internal surface of the radiant tube. The thickness of the coke layer formed inside the tube during the period from the decoking process until the power dip was unknown. The thermal expansion coefficient of the alloy is much higher than that of the coke. This causes a larger contraction in the tube material compared to the coke layer, resulting in the application of high-compressive stresses on the coke layer. Due to its high-compressive strength, the coke layer does not crush.

Consequently, high-tensile stresses are induced in the tube alloy and result in a strain that is directly proportional to the difference in thermal expansion coefficient and the temperature drop. The high-tensile stresses relax during the decoking process. For example, during the initial stage of a normal decoking process, a temperature drop of 100°C–200°C can occur, producing a strain range of 0.15%–0.30%, which corresponds to high stress levels (approximately 300MPa). During the subsequent decoking procedure, the high stresses relax by creep (i.e., cyclic creep relaxation). During each cycle, the tube creeps a small amount. At the end of life, the material reaches its creep ductility.

The tube life is, therefore, affected by:

  • Number of decoking cycles
  • Temperature drop during decoking (or power dip)
  • Creep parameters (temperature, material, creep rate and ductility).

In short, the combination of carburization and creep ductility exhaustion is a “normal” failure mechanism for pyrolysis tubes in ethylene plants.1,2 However, this failure mechanism is accelerated by furnace trips (e.g., power dips).

To demonstrate the above effect, a tube made of 25Cr-35Ni alloy internally covered with a coke layer is considered here. The tube and coke thermal expansion coefficients at 1,000°C are estimated as 17.8 × 10–6 °C·–1 and 2.0 × 10–6 °C·–1, respectively. Therefore, the temperature change required to produce offset strain of 0.2% is calculated using Eq. 1:

ε = [α (tube) – α (coke)].  ΔT = 126°    (1)

The development of centrifugally cast, heat-resistant alloys, such as 35Cr-45Ni, has motivated numerous studies concerning the role of different alloying elements, such as niobium (Nb), silicon (Si), titanium (Ti), etc., on the alloys’ performance. It has been documented that niobium carbide (NbC) is unstable, particularly at temperatures above 1,000°C. Consequently, a partial transformation of NbC to a nickel-niobium, silicide phase (G phase) has occurred.

An increased amount of silicon (> 1.6 wt%) promotes the formation of G phase. Also, high Nb and Ni fractions are necessary for G phase formation. G phase in 35Cr-45Ni alloys is observed in the 982°C–1,093°C temperature range after exposure for 500 < T < 10,000 hr. Research has suggested that steels containing G phase exhibit good performance during long-term, high-temperature service, with indications of greater time to rupture. However, other works have concluded that the presence of G phase adversely affects the material creep life. It is obvious that the effect of G phase on the material properties at high temperatures is not yet well understood.

More importantly, the formation of G phase results in extreme brittleness at ambient temperature. Accordingly, in the case of welding, solution annealing must be carried out to improve the weldability of in-service tubes. Care should be taken during startup and shutdown to avoid relatively quick cooling or any sudden impact on the tubes.

The formation of G phase can be considered a normal consequence of in-service aging. Nonetheless, it has been suggested that the formation of G phase can be controlled by ensuring that Ni and Si levels are kept low.3,4

This investigation showed no major deficiencies in the weld. However, due to their geometry and different microstructures, welds usually represent points of high stresses and act as “metallurgical notches” that concentrate stresses. Therefore, the likelihood of failures at the welds is significantly higher than the probability of failure at the base metal.

The failed weld microstructure contained different distributions of chromium (Cr) and NbC at different regions. It is understood that concentration, shape, size and orientation of carbides could affect the weld cracking resistance and other properties. Long chains of carbides significantly contribute to the degradation of the material mechanical properties. Consequently, brittle or thermal fatigue fractures may occur through carbide layer and through grain boundaries. However, because they are manufactured through centrifugal casting, heat-resistant alloys such as 35Cr-45Ni possess relatively non-homogeneous microstructures. The welds, due to their nature, also have non-homogeneous microstructures that are extremely difficult, if not impossible, to optimize through the welding process.

Study results

The radiant tube crack was initiated by thermal shock that was caused by a rapid temperature drop in the coils during the power failure. It is possible that a coke layer of some thickness was present in the tube before the power dip. Therefore, the contribution of restrained thermal expansion to the crack initiation during the power dip is possible.

Creep (and perhaps coke formation inside the crack) contributed significantly to the crack propagation. The creep process was probably accelerated by the crack initiation that led to increased tensile stresses in the joint.

It is advised that the operation department launch a study into possible ways to alleviate the effect of rapid temperature drop during power dip/furnace trip on the furnace tubes. HP


  1. Jakobi, D. and R. Gommans, “Typical failures in pyrolysis coils for ethylene cracking,” Materials and Corrosion, Vol. 54, No. 11, 2003.
  2. Diehl, C. W. W., P. C. Zeltzer and C. W. Klein, Troubleshooting in a modern cracker plant–case study.
  3. Ibanez, R. A. P., G. D. de Almeida Soares, L. H. de Almeida and L. Le May, “Effects of Si content on the microstructure of modified–HP austenitic steels,” Materials Characterization, Vol. 30, Iss. 4, pp. 243–249, June 1993.
  4. Berghof-Hasselbächer, E., P. Gawenda, M. Schorr, M. Schütze and J. J. Hoffinan, Atlas of Microstructures, DECHEMA e.V., Materials Technology Institute.

The Authors

From the Archive