An analysis of failure in high speed blower blade is presented, which took place a few weeks of operations after its installation in a flare system. Evidences obtained from the fractographic studies and dynamic audit of the moto-blower show that the failure was produced due to the combination of two causes: flaws in the material generated during the blade smelting process and resonance operation of the first lateral vibration mode. The combination of causes lead to the fracture by high cycle fatigue in one of the blades, and this loose portion later impacted and fractured the other blades. These findings allowed, with the participation of the blowe´s manufacturer, the auditing of the design and manufacturing of the blades to establish necessary corrections.
The blower is mounted on a 36 inch. diameter air admission pipe (Figure 1). It has a total of 6 rotational blades manufactured in aluminum smelting and 9 static blades placed at a certain distance from the rotational, which appropriately guide the air. This is run by a 100 HP electric motor capable of operating at 900 and 1800 r.p.m., depending on the volume of gas circulating by the flare.
With only a few weeks of operation in apparent intermittent fatigue regime; the blower blades failed catastrophically. Then, a formal process of failure analysis was initiated to determine its causes and define the preventive/corrective actions.
Visual inspection of damages on site
During the inspection, the following was found:
• Fracture at root level in six rotational blades
• Damage due to breaking of protection grid
• Damage due to impact in static blades
• Fragments of the rotational blades were found at a distance of 60 meters around the flare.
As part of the failure analysis, inspections and tests were performed, which are described in detail in the following sections.
Metallurgic Analysis through Electronic Sweep Microcopy (MEB)
aces of the fractures from the six samples at the base of the blades were assessed through electronic sweep microscopy.
In Figure 3, a microfractographic register of the simple from the blade is observed showing the fracture surface with the lowest irregularity (major flatness). An area with microstructural flaws associated with an apparent fatigue zone is shown, which was located in the external border of the fracture´s surface. The microstructural defects are shown as porosities resulting from the contraction or shrinking of the microstructure (shrinkage porosities), which contain dendritic packages and secondary micro-cracks, characteristic of an incomplete fusion by sudden cooling during the manufacture of the blades. This kind of flaws can work as concentrating stress points which originate micro-cracking in the structure, and could represent the starting point of the failure.
In Figure 4, an amplified image of the fatigue zone is shown in which high cycle fatigue striations initiated in the external border of the surface are observed, precisely where the cracks originate due to the microstructural flaws. The striations are seen regularly spaced with similar characteristics to those produced by load service variations. According to these evidences and the results from the vibration studies, it is confirmed that the failure occurred due to cracking resulting by a high cycle fatigue mechanism.
Audition the Blower´s Dynamic Response
The fracture pattern and striations found in the fractograpic analysis suggest the possibility that the failure might have been produced by fatigue induced by mechanical resonance. The resonance is produced when the alternative forces acting on a system do so at a frequency that coincides with some of the natural frequencies within a range. To determine if the blower operated in resonance it’s necessary to build an Interference Diagram or Campbell Diagram, with the natural frequencies, the excitation forces or sources and the bandwidth for coincidence. Moto-blower natural frequencies were requested from the manufacturer who notified not having them.
Natural Frequencies of moto-blower
Among their practices, the execution of impact tests to determine the static natural frequencies of the blades is inexistent, even less the dynamic. The natural static frequencies were obtained through impact tests on site, taking advantage of the existence of dynamic blades in another blower operating at a lower speed. Measuring of the dynamic natural frequencies was out of reach from this analysis due to a lack of sophisticated instrumentation (strain gages, data transmission system by telemetry and independent air source to produce a high speed jet disturbing the blades as they rotate). However, it is known that the dynamic natural frequencies are slightly higher than the static, due to the stiﬃng eﬀect produced by the centrifugal forces during rotation. Figure 5 shows the arrangement of equipment used for the test, among them:
- Blower blade.
- Instrumented Hammer.
- SKF Microlog Vibration Data Collector (Channel 1).
- Computer with Excel and MathCad to process information.
- Information processed.
The accelerometer was attached to the tip of the blades with bee wax since they are made of aluminum smelting and they are not magnetic. Its output was saved in the data collector and the time-based response was recorded. Preliminary tests helped determine the optimal adjustments of simple frequency, activation of pre-trigger, and saturation scale of the collector so that the response (vibration) could be completely captured without distortion, as shown in Figure 6. The axis of abscises is the time in seconds and the one for the ordinates is the response of the accelerometer in volts.
With the instrumented hammer – the tip of the blade was impacted, near the accelerometer. The hammer is provided with a charge cell measuring the impact force, but it couldn’t be connected to the SKF collector due to having just one channel. It must be said that, with the simultaneous register of the accelerometer and the hammer, not only the natural frequencies or eigenvalues but also the vibration form or eigenvectors and the dynamic parameters K (stiﬀness), M (mass) and C (buﬀering) associated to each present mode in the frequency range of interest were obtained. However, with only one cannel, the natural frequencies (response peaks) were obtained, and the decay rate σ σσ σ of the response, buﬀering rate ζ ζζ ζ (eq. 1), the logarithmic decrease δ δδ δ (eq. 2) and the amplification Q of the peak (eq.3), as well.
An Excel program connected to MathCad processed the data from the collector. The time-based response from the accelerometer was read from the Excel sheet and transferred to a MathCad sheet, in which the Fourier (FFT) and Hilbert Transforms were applied to generate the spectrum and surround function, respectively. Results returned to the Excel sheet helped identify and calculate the fn σ σσ σ, ζ ζζ ζ, δ δδ δ and Q. The taking and the data processing described before were repeated several times on three of the six blades of the blower.
Figure 7 shows the spectrum of response to an impact up to a range of 1,000 Hz. The axis of the abscissae is the frequency in Hz, and the one from the ordinates is the response of the accelerometer in volts. In it, the peaks corresponding to the natural frequencies are observed, with one outstanding at 57 Hz. Another spectrum up to 100 Hz, Figure 8, reveals that there are really two natural frequencies close to 56.5 and 58.0 Hz.
Although eigenvectors are required to identify the mode (lateral, torsional, angular, mixed, etc.) associated to each peak, it is believed that one of them is the first lateral mode and the other is the first angular, as seen on Figure 9.
The predominant height, in comparison with the string and thickness of the blade, make it very flexible in lateral and angular directions, and Little energy is required to excite both modes with a blow in the tip. The peak at 60 Hz seen on Figure 8 is of electromagnetic origin and should not be associated to the mechanic characteristic of the system. As we mentioned before, with the time-based response, σ σσ σ, ζ ζζ ζ, δ δδ δ and Q were obtained. Calculating σ σσ σ, single when it comes to one natural frequency, got complicated due to the existence of surrounding natural frequencies (56.5 and 58.0 Hz) and the peak at 60 Hz. Figure 6 shows the modulation source that produces this fact on the transitory response and the surrounding. Despite of this, little modulation was observed between 1.5 and 2.25 sec and the adjustment of σ σσ σ by minimum squares was limited to this segment of the surrounding, Figure 10.
Tables 1 and 2 summarize the static natural frequencies and the results from the impact tests. The σ σσ σ, ζ ζζ ζ and δ δδ δ values are expected to increase and the Q values to decrease in the rotating blower due to the increase of the drag forces in the blade-cube joint and the addition of buﬀering from the air circulating between the blades.
Excitation Sources and Bandwidth
Table 3 summarizes the excitation sources, bandwidth and vibration modes normally considered by manufacturers, standards and independent consultants. In this case, the four first orders of the rotation speed of the rotor (1X, 2X, 3X y 4X) were considered, and the two first orders of vane pass frequency (1VPF y 2VPF) and blade pass frequency (1BPF y 2BPF) as well. Not all the sources cited consider the blade pass frequency relevant in the eﬀort of the rotor blades. The fan from this analysis has six (6) blades in its rotor. And nine (9) static blades that straighten the flow. The vane pass frequency and blades correspond to 6X, 9X, 12X and 18X the rotation speed of the rotor. The interference band, tipically selected between ±10% and ±20% around the operating speed range of the rotor was fixed in ±10%. In this particular case, the fan rotates at 900 or 1,800 RPM, depending on the gas flow going through the flare.
Diagram of Campbell or Interference
Shown in Figure 11. The axis of the abscises show the speed of the motor rotation in Hz, the ordinates, natural frequency of the blades in Hz; horizontal lines corresponding to the natural frequencies of the blades identified in the impact tests; the diagonals from the origin represent the excitation sources and the zones between the vertical intermittent lines correspond to the interference bands surrounding the operation speeds from the fan-motor set.
It can be seen that, for both operation speeds, there are potential resonance points (inside circles and triangles). Those that involve natural frequencies and low excitation orders require special attention since they are easy to excite and their vibration modes amplify eﬀorts considerably. According to this, none of the interferences with triangle was considered of high risk, except the one in red circle, which involves the first lateral vibration mode and the 2X harmonic from the rotor speed rotation. This mode, similar to a rod in a cantiliver (Figure 9), provokes major eﬀorts in the base of the blade, right where the fracture was produced. The amplification factors from Table 2 also tell us that this mode, under resonance, increases the eﬀorts from 150 to 350 times.
- According to the evidences and results presented, it can be concluded that the failure on the first blade was produced by the high cycle fatigue mechanism. The high eﬀorts originated from the operation in resonance of the first lateral vibration mode, together with smelting flaws that worked as eﬀort concentrators. The first blade, fractured by fatigue, impacted and finally fractured the rest of the blades.
- Blower blades showed micro-structural flaws as porosities with dendritic packages and secondary micro-cracks (shrinkage porosities), produced by the contraction or shrinking of the micro-structure due to incomplete fusion and sudden cooling during its manufacture. This kind of flaws act as concentrating points of eﬀorts, somehow representing the failure’s starting point.
- Notify the manufacturer about the results of the analysis so the process of smelting can be improved. Also, request the information necessary to audit the current blade design and determine if any modification is necessary. The information typically required is included to perform a design audit.
- If any change is required, audit the new design proposed by the manufacturer in order to guarantee that it will be free of new resonances.
- Perform impact tests to the new blades after their manufacturing to guarantee that the tolerances in their making and assembling haven’t changed the natural frequencies of design significantly.
Annex – Information design audit
The information tipically requested from the manufacturers for auditing the blades design includes:
- Natural Frequencies (theory and measures) of blades for the first lateral, axial, torsional and combined modes. The values measured should be with the blades ensembled on the fan, to consider the eﬀects on structure, support cube-blade joint, etc.
- If the natural frequencies from the last point were measured with the fan without rotation, then the increase in these frequencies by the stiﬀening eﬀect due to rotation speed should be indicated.
- Diagram of Campbell or Interference. It must show the natural frequencies curves and the lines of posible excitation. As excitation lines, the following should be included: 1X, 2X, 3X and 4X from the rotation speed of the fan; 1X, 2X, 3X and 4X from the rotation speed of the motor, 1X and 2X from the directional static vane pass frequency 1X and 2X of obstructions or structural supports blocking the free air flow through the fan.
- In case of interference points within the band of +/- 5% around the operation speed range of the fan, including the eﬀort analysis and diagrams of Soderberg for the estimation of fatigue.
- Natural frequencies values and eﬀort under resonance obtained through finite element (FEA) are acceptable only if predictions from the FEA model have been calibrated with impact tests (modal analysis) for the static or “non-rotational” condition.
Luis Barreto Acuña
Enrique J. González.