Paper1-hemant.doc

CONDITION BASED MONITORING OF FAN USING VIBRATION ANALYSIS Hemant Kumar, N. Tandon and S. Fatima Industrial tribology, machine dynamics and maintenance engineering center, IIT Delhi, Hauz Khas, New Delhi 110 016, India e-mail: [email protected]

In this paper, the detection of faults in fan at early stages is carried out using vibration signals so that proper corrective measures can be performed before final failure. The early detection of faults will reduce the downtime of critical equipment. The unbalance fault is seeded on the fan by putting m-seal masses in one of its fan blade and detection of this fault is carried out by using vibration signal. These detections are carried out at two different speeds. In the second phase broken blade fault is seeded on the fan by cutting fan blade along its width into small parts in incremental manner and then the detection is performed at two different speeds. Both the faults detection data of fan is compared with data of healthy fan and notice the variation in amplitude at their characteristic defect frequency. Signal processing like STFT and wavelet transform analysis is carried out in broken blade fault condition by using time domain data. A comparison of signal processing between FFT, STFT and wavelet transform analysis using time domain data is performed.

1. Introduction Condition monitoring (CM) is a process of monitoring the condition of machinery in order to identify the significant change which indicates the faulty behaviour of machinery and its components. Rise in noise level, temperature, and vibration, etc. are some of the parameters which are used to identify the faults in machinery. Vibration analysis used accelerometer to receive vibration signal from machines and these sensors are of piezoelectric type. Due to the piezoelectric effect, the sensors convert one form of energy into another to provide an equivalent electrical signal in response to the measured quantity. Electrical vibration problems occur due to unequal magnetic forces acting on the rotor or stator. The unequal magnetic force may be due to broken rotor bars, unequal air gap, open or short windings. Schoen et al. [1] tested two types of faults, hole drilled through the outer race and indentation produced in both the inner and outer surfaces and detection is done using vibration signal. Obaid et al. [2] analysed that act of disassembling, remounting, and realigning the test motor bearing can significantly alter the vibration signal. From these papers it can be said that defected vibration signals can be predicted and analysed at particular frequency which is also known as defect frequency. Xui and Marangoni [3] carried WESPAC 2018, New Delhi, India, November 11-15, 2018

1

out vibration analysis of motor flexible coupling rotor and they observed that there is a change in amplitude at 1xRPM in vibration spectra for motor unbalance and change in amplitude at 2xRPM in vibration spectra for misaligned coupling. Algule and Hujare [4] carried out an experiment on unbalancing of shaft rotor using vibration analysis and observed that there is a rise in amplitude in 1xRPM with speed for unbalancing. From the previous two paper it is concluded that unbalance defect is analysed at 1xRPM. Pradhan et al. [5] carried out vibration analysis of the pump by creating defect in impeller vanes by breaking one and /or multiple vanes. Frequency domain analysis of vibration shows that as the severity of defect increases, height of side bands (SB) increases and the difference the amplitude of vane pass frequency (VPF) and side bands decreases. The below is the relation to compute predictable frequency and their harmonics on which vibration amplitude is checked for cracked blade defects in fan f = mfd……………………(1) fd = nfs……………………(2) f – Predictable frequency, fs – Supply frequency, fd – characteristics defect frequency, n – number of blades, m- 1,2,3,…… (for computing harmonics) Kar and Mohanty [6] studied the defect in gearbox by creating gear no.2 as defected. By analysing the low frequency and high frequency range steady vibration signals they observed that FFT is very efficient for analysing low frequency range signals. Poursaeidi and Salavatian [7] carried out failure analysis of generator fan blade and notice that failure of blade occurs at the junction of blade and its base and at the curve parts of blade because here stress is maximum and both of them also uses short-time frequency transform (STFT) as a signal processing of timefrequency analysis to detect the fault in gears. In wavelet transform analysis time domain data is used to study the variation of defected signal. While transforming a signal into frequency domain view using FFT analysis, the time domain information is lost. This is the main drawback of Fourier analysis. To overcome this drawback wavelet transform is used. In STFT fixed window size is used, but in the wavelet transform longer window is used for smaller frequencies and smaller window for larger frequencies to achieve finer resolution of frequency. The DWT represents a signal into approximation component (A) and detailed component (D) which represent its low and high frequency components respectively [8]. The DWT tree decomposition is shown in the Fig.1.1 for up to level 3.

Fig 1.1 Wavelet tree decomposition up to 3 level According to DWT, frequency band for approximation ACL and detail DCL at level L are given by ACL = [0, DCL = [

]…………………………….. (3) ,

]……………………………. (4)

Where fs is the sampling frequency. After reading many literate some of them are mentioned above it was observed that there is a lot of research work done on bearing, gears and from that it was decided to do condition monitoring on the fan by using basics of vibration signal detection. In industries and power plant, very heavy and large size of fan used for cooling and power generation purpose.When there is break down of fan then whole power plant will shut down if there is no alternate option and recovery or WESPAC 2018, New Delhi, India, November 11-15, 2018

2

maintenance of plant will take time. So this study will help to prevent the fan before final failure and save the losses. In the coming section details of experimental setup for detection of fan vibration are described

2. Experimental setup and details A fan of axial type having 5 blades and of 33 watt is used in our test. Fan is mounted on a wooden frame so that proper suction of is maintained and a speed controller is also connected with fan in series for speed variation. An accelerometer (Type 4368 B&K) is used for measuring the vibration of the fan due to unbalancing and blade defects. Accelerometer have capable capacitance of 114 pF and charge sensitivity of 4.64 pC/ms-2 . The output of accelerometer is connected to charge amplifier because it is used to convert high input impedance to low output impedance. Its main purpose is impedance matching. Direct connection of accelerometer output to the measuring system is difficult because of high impendence output. To eliminate this, accelerometer output is fed through pre-amplifier which has high input impedance and low output impedance suitable for connecting to low input impedance of measuring and analysing instrumentation. It has integrated circuit so they can give velocity and acceleration also from displacement. Charge amplifier is then connected to ono sokki FFT analyser where frequency domain and time domain data are recorded. The whole experimental setup and fan used are shown in Fig 2.1. Frequency domain data is further used to draw frequency spectra, spectra were monitored with frequency span of 200Hz. The FFT analyzer used has two input channels both of which are BNC type, which can bear maximum absolute input voltage of 100Vrms AC for 1minute, full scale accuracy of ±0.1dB at 1 kHz. The data received by the FFT analyzer at the sampling rate of 4096 and sampling frequency of 640 Hz was saved for further analysis. The time domain data is used for analysis in STFT and wavelet transform signal processing analysis. Two types of faults is simulate on the fan (a) unbalance and (b) cracked blade type fault and detection is performed using FFT, STFT and wavelet transform as a signal processing of time and frequency domain data.

(a) (b) Fig 2.1. Experimental setup for vibration analysis (a) setup and (b)fan a) Unbalance type of fault In this category, mentioned above (a), fan is simulated with unbalance defect by putting various m-seal masses in one of its fan blade from 1.12, 2.007, 2.50 to 3.00 g. and a healthy fan is used to check the variation. The whole is carried out with two RPM 1500 and 2300 rpm. b) Cracked blade type fault

WESPAC 2018, New Delhi, India, November 11-15, 2018

3

In this category, cracked blade defect, fan is simulated with cracked blade defect by Chipping Fan blade along the width from 5mm,10mm,15mm,20mm to 25mm by using a hacksaw and a healthy fan is used to check the variation. Detection is done at two different

3. Results and discussions The fault detection of fan is done by checking of spectrum magnitude and characteristics defect frequencies. By comparing with the healthy fan spectrum, if some large magnitude deviation is observed around specific characteristic frequencies, then it can be say that variation in amplitude is due to that specific defect. The monitoring of fan is done at two different faults condition (a) unbalance and (b) cracked blade and at different speed as well. (a) Unbalance The result shown below in the Fig.3.1 are of healthy and unbalance vibration spectra over the span of 160 Hz frequency. The frequency highlighted in the spectra is 1xRPM used for unbalance fault detection. The variation at that frequency is checked for increase severity in unbalance.

(a) (b) Fig 3.1 Vibration spectrum at 1500 rpm(a)healthy and (b)1.12g unbalance mass

Fig 3.2 Amplitude variation in vibration of unbalance fan The variation of vibration amplitude at different unbalance masses and different speed is shown in Fig 3.2. It is concluded that on increasing the add-on mass quantity on fan blade the amplitude of vibration increases i.e. severity increase. These vibration amplitude also increases with increase in RPM (b) Cracked blade The results presented shown in Fig. 3.3 are of the vibration spectrum of cracked blade type of fault over the span of 250 Hz. First the data for healthy condition of fan is taken which is used to WESPAC 2018, New Delhi, India, November 11-15, 2018

4

compare it with the faulted condition in vibration spectrum at blade pass frequency calculated by using eq (2) . The amplitude at blade pass frequency is highlighted in the fig. and if the increase in amplitude is observed at blade pass frequency then the defect is due to the cracked blade.

(a) (b) Fig 3.3 Vibration spectrum at 1500rpm (a) healthy and (b) 15mm chipping from 1st blade

Fig 3.4 Vibration amplitude variation of cracked blade fan From the above Fig 3.4, it can be said that there is in increase in amplitude on increasing chipping size. There is rapid increase in amplitude while chipping blade from 10mm to 15mm and from 20mm to 25mm (Full blade chipping). The amplitude is also increase with increase in speed. STFT STFT is a time domain signal processing tool which plot the spectrogram between frequency and time. Spectrograms for vibration signal of defected fan are shown in Fig 3.5.

(a) (b) Fig 3.5 STFT of fan at 1500RPM (a)1 blade broken (b) 2 blade broken The defected frequencies lines become more yellowish in two blade broken fan as compared to 1 blade broken fan as seen from the above Fig 3.5. The more yellowish colour indicate more severity level like higher amplitude in FFT. Thus it can be said that STFT is an effective tool in signal processing to detect the fault. WESPAC 2018, New Delhi, India, November 11-15, 2018

5

Wavelet Transform In my case sampling frequency is 640 Hz and level select for wavelet is 5 and characteristic frequency lie in level 2 and 3 frequency band i.e., 160-320 and 80-160.

0.05531

(a) (b) Fig 3.6 Decomposition signal (a) and statistic (b) of d3 level of 1 blade broken fan at 1500 rpm (vibration) It is cleared from the above wavelet decomposition Fig 3.6 that wavelet show maximum variation in d2 and d3 detail component because characteristics frequency lie in d2 and d3 frequency band so statistics of that frequency band is analysed and maximum value is notice which is increasing while increasing speed and severity level. So wavelet transform show result at particular frequency band.

4. Conclusions Condition monitoring became an effective tool in maintenance and prevention of machines in industries now a days. In this paper vibration analysis of defected fan is performed. The unbalance type of fault is checked at 1xRPM frequency and cracked blade defect is checked at blade pass frequency of fan in FFT spectra and the amplitude at that frequency increases as increase in severity level like speed, unbalance mass cracked blade size. In STFT more yellowish colour at characteristics frequency shows more severity. The maximum value in wavelet transform statics increase in a particular frequency band in which characteristic frequency lies. So FFT is effective to than STFT and wavelet transform but if FFT is used with wavelet than result will be more accurate.

References [1] R. R. Schoen, T. G. Habetler, F. Kamran, and R. G. Bartheld, “Motor bearing damage detection using stator current monitoring,” IEEE Trans. Ind. Appl., vol. 31, no. 6, pp. 1274–1279, 1995. [2] R. R. Obaid, T. G. Habetler, and J. R. Stack, “Stator current analysis for bearing damage detection in induction motors,” in Proc. SDEMPED, Atlanta, GA ,vol. 3, pp. 182–187, 2003 [3] M. Xui and R.D. Marangoni, “Vibration analysis of motor flexible coupling rotar system subjected to misalignment and unbalance,”Journal of sound and vibration,vol. 2, no. 13 pp. 681-691, 1994 [4] S. R. Algule and D. P. Hujare, “Experimental study of unbalance in shaft rotor system using vibration signature analysis,” International Journal of Emerging Engineering Research and Technology,vol. 3, no. 17, pp. 124– 130, 2015. [5] K. Pradhan, R. Mohanty, P. Nitaigour, and G. Dastidar, “Fault detection in a centrifugal pump using vibration and motor current signature analysis,” Int. J. Automation and Control, vol. 6, no.2, pp. 261-276, 2012. [6] C. Kar and Ã.R. Mohanty, “Monitoring gear vibrations through motor current signature analysis and wavelet transform,” Mechanical Systems and Signal Processing ,vol. 20, no. 8,pp. 158–187, 2006. [7] E. Poursaeidi and M. Salavatian, “Failure analysis of generator rotor fan blades,” Eng. Fail. Anal., vol. 14, no. 5, pp. 851–860, 2007.

WESPAC 2018, New Delhi, India, November 11-15, 2018

6

[8] K. C. D. Kompella, M. V. G. Rao, R. Srinivasa, and R. N. Sreenivasu, “Estimation of bearing faults in induction motor by MCSA using Daubechies Wavelet Analysis,” IEEE Trans. Ind. Appl., pp. 1–6, 2014.

WESPAC 2018, New Delhi, India, November 11-15, 2018

7