Fracture Analysis of Compressor Impellers in Olefin Units: Numerical and Metallurgical Approach

Department of Mechanical Engineering, Faculty of Engineering, Persian Gulf University, Bushehr 75169-13817, Iran Department of Chemical Engineering, Faculty of Petroleum, Gas, and Petrochemical Engineering, Persian Gulf University, Bushehr 75169-13817, Iran Department of Mechanical Engineering, Faculty of Engineering, Shiraz University, Shiraz, Iran R&D Process Engineer, Morvarid Petrochemical Company, Shiraz, Iran Department of Mechanical Engineering, University of Science and Culture, Tehran, Iran Department of Mechanical and Aeronautical Engineering, University of Pretoria, Pretoria 0002, South Africa Department of Medical Research, China Medical University Hospital, China Medical University, Taichung, Taiwan


Introduction
Compressor is one of the vital equipment in petroleum, chemical, and natural gas industries. On the other hand, the impeller is the most critical part of a compressor. So, the failure of the impeller can impose heavy losses on a system. Depending on the compressor type, the rotational speed of the impeller can reach 2000 to 100000 rpm. ere are some methods and tools to study the reasons for impellers failure and the examination of compressor performance. e combination of the numerical optimization technique with the computational fluid dynamics (CFD) is a practical tool to study compressor performance [1]. One of the most important phenomena which usually takes place in the compressors is the formation of liquid droplets. e liquids can have numerous impacts on the surface of impellers due to the velocity of the stream, and the liquid droplets can impose a huge impingement on the impeller of the compressor [2]. Some studies show the effect of droplet impingement on the performance of compressors. For example, Surendranand and Kim [3] carried out research that the purpose of the study was to investigate the effects of liquid droplets on the flow field of a centrifugal compressor stage numerically. Degendorfer et al. [4,5] carried out numerical and experimental research on the compressor blades; the results indicated that the unsteady forces throughout the operation of the compressor could result in destructive phenomena named resonance response, which can push forward compressor into fatigue cycle. Hong-Li et al. [6] investigated rotor dynamic behavior of turbo-expander involving droplet impact. Since the size and location of the droplets are not known precisely, a stochastic model with beta distribution and Bernoulli distribution has been developed, and the forces resulting from droplet formation have been formulated. e droplet that impacted the impeller blade can lead to a crack in the surface. After that, the stress caused by the rotational velocity in the vicinity of the cracked area can be altered. erefore, this can decline the performance of the system and also causes breaking down of blades in low temperatures. Consequently, they resulted that the number of channels and blades must be reduced to prevent impeller fatigue. Hundseid et al. [7] evaluated a hydrocarbon multiphase flow, performance parameters, and a wet gas performance analysis.
eir results clearly demonstrate that liquid properties greatly impact compressor performance. Since different loads such as centrifugal loads, exciting loads, and aerodynamic loads are applied, impeller work under the hard conditions that include cyclic and dynamic loads. Based on working conditions, impellers are made of different materials such as titanium alloys, stainless steel, and aluminum alloys. Load types, temperature, medium, material, structure, and processing technology are examples of effective factors in the fracture of impeller blades. Accordingly, different mechanisms have been identified as causes of failure of the impeller blades, with fatigue as one of the most common. Because of the high strength and low density of the 7xxx series of Aluminum alloys, these materials have been used to fabricate impellers. However, there are numerous reports of the failure of aluminum impellers fatigue, the dominant mechanism in these failures [8][9][10][11]. To evaluate the failure of impellers, the stress distribution should be obtained. Furthermore, the microstructure, machining and final geometry should be investigated carefully. Due to the operating condition of the impeller, it is practically impossible to measure the stress distribution of the impeller blades. However, the stress distribution of the impeller blades can be predicted by finite element analysis (FEA) [8]. Accordingly, the critical positions of the impeller will be identified clearly. e exploitation of this technique can be beneficial in fracture analysis. e combination of the numerical optimization technique with the computational fluid dynamics (CFD) calculation was used to study the correlations between the parameters of geometry and the compressor performance [1]. However, the role of the microstructure of the alloy should be considered in all cases. Azevedo et al. [9] have claimed that the fatigue cracks have initiated from small defects such as impurities. Also, inclusions are suitable stress concentration sites for nucleation of fatigue crack [10]. e inclusions can be the result of the introduction of impurities or various additives during the production process of impellers. Due to the substantial difference between inclusions and base metal, stress concentration can be formed in the interface. erefore, the microstructure of materials is the most effective factor in fatigue life [10][11][12][13][14][15]. Yu et al. [11] reported that defective casting process caused shrinkage cavities in the microstructure, which are the stress concentration during loading, and ultimately leads to preimpeller failure. Troshchenko and Prokopenko [12] studied the effects of manufacturing residual stresses, surface defects, and nonmetallic inclusions on fatigue behavior of steels and a titanium alloy. ey claimed that there is a complex relationship between the impact of various factors on fatigue strength. It has been verified by many researchers that poor machining can also cause stress concentration sites and the onset of fatigue cracking [11,13]. Also, the heterogeneous structure, such as welded joint, can be a site of stress concentration and reduce the fatigue life of the impellers [12,13]. A 3-dimensional (3D) point cloud is a new technique that has found lots of applications in reverse engineering recently. It has been used to prepare 3D drawings of engineering parts with high accuracy [13]. Studies have shown that 3D point cloud data has sufficient accuracy in extracted pipe spools [14] and cylinders and estimation of their parameters [15]. erefore, it is useful for comparing the geometrical changes that occurred before and after servicing the pieces. Wang et al. [13] reported that (3D) point cloud be used as a new technique for geometry quality inspection in a review article. However, the application of this technique is more in the field of reverse engineering than geometry quality inspection. e literature has shown that the systematic investigation of the failure mechanism of impellers has not been carried out. So, in this study, three fractured impellers have been analyzed to determine the cause and mechanism of failure using CFD, finite elements (FE), and metallurgy analysis.

Metallurgical Analysis.
ree compressor impellers were selected for this study to evaluate fracture analysis. All working conditions were similar, but the service life was varied. e code and the lifetime of each sample have been listed in Table 1. e impellers were designed and fabricated by various companies. In fact, to choose the best product from different companies, random samples from each company have been selected for investigation. e standard composition of 7175 Al alloy was extracted from ASM handbook volume 2. However, all impellers failed in less than the designed time. Figure 1 shows three fractured impellers, samples 1, 2 and 3, respectively. e diameter of all impellers was 16 cm. All fracture points and then the critical points of the fracture zone were cut, and microscopic evaluation was performed on them. Table 2 represents the chemical composition of samples. is test was performed with a spectroscopic device (SOLARIS ccd plus model).
e base metal of impellers was evaluated by microstructure, phase characterization, and hardness tests. Metallographic specimens were prepared by cutting, grinding, mounting, and sanded using 180-3000 mesh sandpapers, and polished by special felt, followed by etching in a solution that contained 95 ml distilled water, 2.5 ml HNO 3 , 1.5 ml HCl, and 1.0 ml HF, and the duration of etching was 6-9 seconds. Optical microscopy (OM) and scanning electron microscopy (SEM TESCAN-Vega 3) equipped with energy-dispersive X-ray spectroscopy (EDS) analysis system were applied to characterize microstructure. e phases of microstructure were identified by X-ray diffraction (XRD) method. e prepared specimens were inserted into the Philips X'Pert PW 3040 device to accomplish this test. e XRD test was performed at a scan rate of 0.05 degrees.
Furthermore, the Brinell hardness (HB) was measured on all specimens at least three times as well as before and after service at a load of 500 kg. Fractured surfaces were examined by an ordinary camera and above mentioned SEM to identify the location of the cracking initiation and the   Advances in Materials Science and Engineering 3 failure mechanism. In addition, 3D scanner equipment was used to obtain point cloud data from specimens before and after service. Evaluation of the deformation behavior of the blades of the impeller was the primary purpose had been the results of this technique. For this experiment, Steinbichler 3D scanner set-up was used.

Numerical
Analysis. e procedure was divided into four steps for numerical simulation of the flow inside the compressor blade. In the first step, the CAD model of the mentioned blade was generated. Figure 2 and Table 3 indicate the CAD model of the blade with boundary conditions and the dimensions of the model, respectively. In the second step, the computational domain to discretize equations has been prepared that is shown in Figure 3. All the discretized equations have been solved by using a CFD code at the third step. FVM numerical approach has been employed to solve governing equations. A second-order upwind scheme was applied for discretizing the momentum equation, turbulence equation, and energy equation on the provided grid. e SIMPLE algorithm has been considered for Pressure-Velocity coupling. Also, a steady state rotating reference frame (RRF) handles the rotation of the wheel of the compressor. e final step of the modeling was postprocessing of the outcomes from CFD computations which have been analyzed and compared with the experimental results to evaluate the blade fatigue. e information of both computational grids and models has been provided in Tables 3 and 4, respectively.
For FEM analysis, the boundary and loading conditions were considered as follows: e effect of the centrifugal field was modeled as a volumetric force, which extends from the center of the rotating body to the outside. e rotating field was designed for 34500 rpm angular velocity in static modeling. e loading due to fluid flow was modeled as the pressure difference between the two sides of the blades. is loading was applied to the blade surfaces as compression.
Since the blade was moving at a constant angular velocity, the effect of centrifugal force was applied to it in the form of a field, and the problem was static requiring boundary conditions of fixed support on the part connected to the shaft.

Results and Discussion
3.1. Fractography. In Figure 1 schematic of the test sample is shown. In continue, its macrostructure can be seen for sample 3 in Figure 4. is fatigue fracture surface illustrates different regions that include: S (Stage I) A and B (stage II), and C (ductile fracture). e red arrows indicate crack directions. e crack initiated in S and branched to A and B regions. e plastic accumulation method [16] is dominant in the B region, and this area has undergone shear deformation. In contrast, the dislocation blockage mechanism is predominant for region A and a different fracture surface is created than in region B. erefore, region A shows the mussel-like surface or so-called fatigue-induced beach marks and region B illustrates shear dimples. e region C indicates final fatigue fracture with brittle fracture surface. Fatigue crack in zones A and B have reached its critical size. Finally, the C region has formed abruptly and rapidly in the last step and completed the fatigue failure.
In all specimens, fatigue crack initiated at the surface and propagated with different paths. Figure 5 illustrates high magnification of the initiation crack site (S region in Figure 4) for samples 1 and 3. In addition, it can be seen different cracks in this subsurface area. Figure 6(a) shows the high magnification of Zone A with beach marks. In zone A, the direction of crack propagation has been perpendicular to the direction of load and beach line marks. is pattern is quite common for fatigue failures. In this zone, the crack tip undergoes plastic deformation and progresses slowly. In general, the cracks propagate in the perpendicular direction of the highest tensile stress [17]. A comparison of Figures 6(a) and 6(b) indicate that the crack progressive was different in the two regions (A&B). In fact, the crack direction has angularity with applied tension mode in the B region. For this reason, deformed dimples are noticeable in region B. In addition, the EDS results confirm the presence of coarse Al 2 MgCu precipitations in this area.
Given the geometry of blades and hub connection, it seems that the critical crack length was about 2 mm. is area is shown in Figure 6(a) by a yellow dashed line. Figure 7 shows OM images of all specimens. Sample 2 has more recrystallization microstructure than others, but the average size of the precipitations of sample 2 is higher than both samples 1 and 3. Also, it can be seen that all microstructures have been overaged. e more magnification of microstructure of sample 2 is illustrated in Figure 7(d). is phenomenon is confirmed clearly by the presence of precipitations in the grain boundaries. Arrows in Figure 7 indicate some of them. It was mentioned that this alloy is susceptible to overageing [18]. e EDS results of the grain boundary precipitation have been illustrated in Figure 8. It can be seen that Al 2 MgCu (S phase) precipitation has formed in the grain boundary. is phase has a centered-orthorhombic crystal structure [19]. Also, this precipitation on a large scale can be the site of crack initiation under loading.

Metallurgical Characterization.
is phenomenon is displayed in Figure 9 for one precipitation in sample 3. In other words, the clear discontinuities (yellow arrows) are visible between precipitation and Al alloy matrix. It should be noted that this image was created without the etching process. Comparison of the different samples showed two points: the number of precipitations varies for different specimens and severe precipitations discontinuity with the Al matrix in sample 3. Although low deposition of this precipitate can improve the mechanical and corrosion properties [20], increasing the amount of this precipitate has the opposite effect on the mechanical properties. e existence of the latter precipitations is one of the significant weaknesses of this Al alloy microstructure. Moreover, Al grain boundaries may have been embrittled by large S phase precipitation. It can be seen large cracks between precipitation and Al matrix in Figure 9. e main reason of this phenomenon is related to the incoherency of precipitation. According to Chen et al. [21], the coherency degree of the precipitates with the 7055 Al alloy is the main factor in the fatigue crack progressive. In the present work, the crack growth rate was very high due to     Advances in Materials Science and Engineering low coherency between precipitations and Al matrix (Figures 8 and 9). According to Table 2, specimens have a close chemical composition close to 7175 Al standard composition, such as ASTM B227.
is alloy should have at least 1.2%wt Cu according to this standard. Specimen 3 has a lower Cu in its composition. Due to the role of Cu element in the strength of this alloy, it was expected that this sample would exhibit weaker mechanical properties. Lim et al. [22] claimed that the amounts of Zn, Mg, and Cu have a decisive role in the formation of reinforcing phases and affect the strength of 7175 Al alloy. e XRD result of the samples is shown in Figure 10. It should be noted that excessive amounts of Cu can also lead to the formation of the destructive S phase (Al 2 MgCu). erefore, sample 3 was expected to have the least amount of destructive S phase. But this is not an accurate conclusion. Because the phase formation depends on the heat treatment applied in addition to the chemical composition.
e results of the hardness test are shown in Table 5. e hardness of all samples is lower than the standard limit of 7175 Al alloy (152 HB). It can be seen that samples 1 and 3 have the highest and lowest hardness, respectively. e coarsening of S phase precipitations was the main cause of this result. A comparison of the number and size of S phase precipitations between different specimens has shown that sample 1 has better conditions due to the microstructure of sample 1 has a uniform distribution of precipitations. In other words, the uniform distribution of fine precipitations resulted in higher Brinell hardness. However, it should be reminded that all samples have been over-aged. erefore, the distribution of precipitations is another main factor in fatigue strength.
e same results were reported by Kobayashi et al. [23] for 20-11P steel alloy.  Advances in Materials Science and Engineering 7 Figure 11 shows the result of the CFD simulation for HP and LP Compressor. e results revealed that the maximum pressure of the gas stream for the rotor speed of 34500 rpm has taken place in the fractured area of the blade. e shear stress (WSS) contours are shown in Figure 12, which also indicates the maximum in the breakdown zone. e temperature gradient for HP and LP Compressor is shown in Figure 13.

Finite Element and Numerical Analysis.
In addition, fluid velocity vectors in a row of the compressor from the inlet to the outlet are shown in Figure 14. e results showed that the pressure, stress gradients on the HP blade were significantly higher than the LP compressor, providing the reason for the blade fracture in the HP compressor. e parametric geometry model of the impeller was made according to technical drawing using ANSYS software.    Advances in Materials Science and Engineering e final model of the impeller consisted of 114600 nodes and 125101 elements for one blade. As already mentioned, the impeller of the compressor endures a combination of centrifugal and surface loads [24,25].

Validation.
To evaluate the results of CFD simulation, the industrial data was provided by the R&D Unit of Morvarid Petrochemical Company (MPC) has been used, and for both LP and HP the same impeller has been used. According to Table 6 the boundary condition which was set up for both inlet and outlet were kind of constant pressure named pressure-inlet and pressure-outlet, respectively. erefore, after model convergence to validate the model results, the calculated mass flow rate from the CFD simulation was compared to the actual mass flow. Table 7 shows the comparison of the simulation results and the industrial results of HP and LP compressors. As can be seen from the obtained results, there is a good agreement between the results of modeling and the operational results of the process, which indicates the accuracy of the modeling. e surface load is caused by gas fluid pressure. Figure 15 illustrates the contour of the total deformation of the impeller at 0.1 bar pressure difference in 34500 rpm. e value of pressure difference was extracted from process conditions. Since the fluid analysis results show that the pressure difference between the two sides of the blade is less than 1 bar, there is no need to transfer the pressure distribution to the structural analysis. Instead, the pressure is applied uniformly to the lower surface of the blade. Figure 15(a) illustrates that the maximum deformation is about 0.1 mm. e extracted contour of elastic strain and von-Mises stress for mentioned conditions are shown in Figures 15(b) and 15(c), respectively. e extracted contours illustrate the maximum stress has accrued in the root of the blades of the impeller. is is due to the interference of the torque due to the centrifugal force and the fluid pressure, which is maximized at this point. Moreover, Von-Mises stress has a maximum value of 231 MPa in the root of the blades of the impeller.
is phenomenon is evident in Figure 16.
In the following, to estimate the effect of the maximum fluid pressure and make the more critical situation with respect to the actual situation, the pressure difference was increased to 0.5 bar, and the calculations were repeated. e results of the latter state have illustrated in Figure 17. In this state, the maximum total deformation was 0.45 mm on the tip of the blades. It seems the fluid pressure has predominated to centrifugal load in this location. However, the root of the blades has maximum stress of 249 MPa in Figure 18. erefore, due to the high proportion of centrifugal loads, the maximum stress does not increase as the pressure difference increases in the root of blades. e compared results between pressure differences of 0.1 and 0.5 are shown in Table 8. In summation, based on Figures 15 and 17 the location of maximum stress occurs at the root of the compressor blade. For example, for pressure difference 0.5, in Figure 17, the value of Von-Mises stress in the root is in the interval 166-250 MPa, while this value for other parts are below 110 MPa, this difference is considerable. Based on this analysis the probability of failure in the blade's root is much higher than in the other parts. e fatigue strength on the titanium alloy studied according to the literature [26], for 10 7 cycles is about 200-250 MPa. e present study's maximum stress is between 230 and 250 MPa.
e results show an acceptable agreement of the present work with the previous literature.   Figure 19 shows traditional camera photos of samples before and after service. Since the original sample was the basis for preservice parts verification, it has been avoided to reproduce the preservice photos of samples 2 and 3 in Figure 19. It can be seen that all specimens undergo local deformation near the blade to hub junction. is site is near the fractured surface on the ruptured blade. It can be concluded that discontinuities accumulated at this position. is means that deformation bands are formed by applying a large external load at the concentration stress sites due to asymmetric slippage of the crystal planes. With the formation of deformation bands, crystalline defects such as voids move in the same direction to accumulate at the point of crack initiation. erefore,  microcracks develop easily in stress concentration sites. On the other hand, the size of these microcracks depends on the microstructure of the material and the concentration of stress due to the external load. e same result has presented by Affonso [17]. In this stage, the crack is tiny and unobservable with an unaided eye. e 3D points cloud technique was used for exact investigation of deformation of all sections of compressor impellers after service. Figure 20 illustrates 3D images of sample 3 before and after service with high magnification.

3D Points Cloud Technique.
e 3D image of sample 3 is analyzed before and after service for better comparison. It can be seen that deformation is the highest near the junction of the blade and hub of the impeller. So that the radius of the connection of the blade to the hub has changed from 6 mm to 5.9 mm for sample 3. is change was lower for samples 1 and 2. As the radius becomes smaller, the stress concentration at this point will increase. erefore, this technique can well show small geometric changes, especially for critical points of the impeller. According to the cyclic load system and fatigue condition for impeller blades, accumulation of elastic strain lead to local plastic deformation in this site. Due to the presence of coarse precipitations in the grain boundaries of crystalline structure, elastic strains have accumulated around them. Recently, Malitckii et al. [27] reported the strain accumulation during microstructurally small fatigue cracks by the digital image correlation technique. erefore, given the over-aged microstructure (coarse precipitations), sufficient conditions have been provided for a new impeller fracture surrounding the blade-to-hub connection area.
On the other hand, the presence of stress concentration can accelerate the crack initiation step. Accordingly, large       Advances in Materials Science and Engineering 13  Figure 19: Camera image of sample 3 (a) before service (b) after service. 14 Advances in Materials Science and Engineering precipitations have accelerated the fatigue crack formation in all three samples. e same result has reported for AA2524-T3 alloy [28].
In general, some design features may act as stress concentration sites such as fillets between blades and hubs on impellers. Also, all investigated impellers have inappropriate microstructure causing intergranular precipitations. All these factors led to the fatigue failure of the junction of the blade to the hub of impellers. Depending on the number and distribution of precipitations, the service life varies from sample 1 to 3, maximum to minimum, respectively. e same results have been reported by Moreira [29] for AA 2618 T652 alloy compressors wheels. However, the impellers were in a corrosive environment in their samples, and this degradation was more severe.

Conclusion
e failure impellers were operated at 34500 rpm. e initial evaluation indicated a similar failure mode for all specimens. e simulation of impeller service showed stress concentration in some regions of the impeller, such as the root of blades, and 3D points cloud technique showed local deformation in impellers.
(1) e phenomenon of fatigue has been the most important factor in the failure of all impeller blades. In addition, inadequate microstructure due to the overageing of 7175 Al alloy has led to premature failure of the blades (2) e stress due to centrifugal motion has been much higher than fluid stress (3) It was found that the root of the blades has maximum stress of 249 MPa, which leads to a deformation of 0.45 mm (4) e radius of curvature of the impeller root has changed by 0.1 mm, and this amount has intensified the stress concentration conditions in this area (5) 3-D points cloud technique can detect local deformations of the impeller with accuracy and prevent sudden failures with greater degradation in overhaul times According to the results, it seems that the precise control of the alloy composition and its microstructure can lead to an increase in the service life of the impellers. However, the authors have put another proposal on the agenda to consider other alloys such as Ti-6Al-4V titanium alloy to evaluate its performance in this situation.
Data Availability e data will be available upon request.

Conflicts of Interest
e authors declare that they have no conflicts of interest.