Você está na página 1de 7

IET Science, Measurement & Technology

Research Article

Improved condition monitoring technique for ISSN 1751-8822


Received on 14th August 2016
Revised 4th November 2016
wind turbine gearbox and shaft stress Accepted on 12th January 2017
E-First on 28th March 2017
detection doi: 10.1049/iet-smt.2016.0338
www.ietdl.org

Abdulwahed Salem1 , Ahmed Abu-Siada1, Syed Islam1


1ElectricalEngineering Department, Curtin University, Kent Street, Bentley, Australia
E-mail: a.salem@postgrad.curtin.edu.au

Abstract: Condition monitoring has been widely used to detect mechanical and electrical faults of wind turbine in order to avoid
any potential catastrophic failures, reduce operational and maintenance cost, and enhance the reliability and availability of the
equipment. Although several papers about wind turbine condition monitoring can be found in the literatures, adopting a reliable
and cost-effective technique for wind turbines that usually experience severe mechanical stress due to the harsh weather
conditions they are exposed to is still challenging. Although statistical studies show that gearbox failure rate is low, the resulted
downtime and the replacement or repairing cost is substantial. This study introduces an improved technique to monitor the
condition of the wind turbine gearbox based on gearbox vibration and shaft torque signatures analyses. In this context, a test rig
that emulates real wind turbine operation has been developed to analyse the behaviour of wind turbine under various
mechanical fault levels. Shaft torque and mechanical vibration signals are detected using high-resolution sensors and are
analysed using two signal processing techniques: wavelet and order analyses to examine the impact of investigated fault levels
on the wind turbine drive train. Both signal processing techniques are compared based on their sensitivity to detect incipient
fault levels.

1 Introduction through analysing the vibration signal sideband frequency is


introduced. Molins et al. [20] illustrate that the electromagnetic
Condition monitoring of wind energy conversion system (WECS) torque of the induction generator shaft can be controlled through
is essential to decrease maintenance cost and avoid any potential of controlling the terminal voltage of the generator using static
catastrophic failures [1]. Various wind turbine (WT) gears failure synchronous compensator in order to reduce the gearbox fatigue
modes such as surface disturbance, scuffing, deformation, surface and stresses, and extending the lifespan of the WT gearbox. Tang
fatigue, cracks and gear box tooth breakage have been reported in and Luo [9] demonstrated that vibration-based condition
the literature [2]. Failure of a WT key component such as blades, monitoring is an effective technique in recognising gearbox faults
gearbox, breaking system and tower can cause an extreme particularly in the high-speed stage which is attributed to the high
downtime to the entire system [1, 2]. Statistical studies show that accuracy of accelerometers to measure high-frequency vibration.
although the gear box failure rate is less than the failure rate of Asmine et al. [21] introduced a method for WT mechanical stress
other WT components, the downtime it causes to the entire system reduction based on model validation for WT generator. Yuan and
along with the cost of replacement or repairing is of the highest [3]. Lilong [22] compared the vibration signals of a healthy and
Few studies in the literatures have investigated different signal damaged gearbox using variable amplitude Fourier series. Zhang et
processing techniques to detect various WT fatigue and gear box al. [23] combined both, data mining algorithms with a statistical
faults based on vibration signature analysis [4, 5]. quality control concept in order to improve a monitoring model to
Discrete wavelet transform (DWT) is known by its ability to recognise gearbox technical problems. Wang et al. [24] utilised
extract most of the signal features in both time and frequency wavelet method for fault diagnosis of a gearbox under fluctuating
domains [6]. The exceptional advantage of the wavelet application loads. Gearbox bearings along with oil quality and lubrication
is that its frequency bandwidth varies in time, which makes it an system have been reported to have a significant correlation with
appropriate method to study non-stationary signals [7, 8]. gearbox failures [25]. Thiringer and Linders [26] have examined a
Harmonic wavelet-based data filtering has been used for enhanced WT with fixed pitch angle through controlling rotor progress.
time frequency features to analyse multiple sensors signals for WT Asmine et al. [21] presented various condition monitoring
gearbox faults identification and quantification [9]. Authors of [10, approaches for WT blades which include visual inspection,
11] reveal that 59% of WT failures are attributed to the gearbox. ultrasonic and thermography techniques [27]. Wu and Lai [28]
Therefore, industrial gear fault diagnosis has been given a great estimated the fatigue life for a small composite sandwich WT
concern by researchers [12, 13]. The gearbox is a key element in blades through calculating the stress limits under fatigue loading.
WT s condition monitoring due to its challenging accessibility, Combet and Gelman [29] investigated the vibration signal sideband
complexity design and dynamic operating conditions. Different ratio to examine the gearbox health condition under load variation
issues confronting WT gearbox including rotor blades experiencing which may result in a passive effect on the gearbox and produce
wind gust that may cause an impermanent misalignment between non-linear transmission path and disturbance to the vibration signal
the two rotor shaft sections connected through the gearbox which [30].
may harm the bearing loads and decrease the WT lifespan [14, 15]. While several studies can be found in the literatures as detailed
Gould and Burris [16] have investigated the impact of wind shear above, most of the investigated techniques for gear fault detection
on WT rotor blades by predicting the pitch moments from wind neglected the dynamic behaviour of the turbine blades and assumed
shear. Authors of [17, 18] categorised the causes of gearbox it to be of constant speed. As the measured vibration signals are
failures into two types of stresses: contact stress and root bending very sensitive to load and turbine blades speed variations,
stress. The former type causes gear pitting and the later causes reliability of the proposed techniques is questionable when applied
tooth crack or tooth damage. Both types can lead to premature gear to practical WT. In this paper, the dynamic behaviour of the load
wear and failure. In [19], a method for gearbox fault detection and blades speed is realised and its impact on the measured

IET Sci. Meas. Technol., 2017, Vol. 11 Iss. 4, pp. 431-437 431
© The Institution of Engineering and Technology 2017
fluctuating load, and unstable actions [35]. Order tracking
spectrum is a method to analyse the vibration signal based on
related spectral components rising with the instantaneous rotating
speed. In case of the collected vibration data comprise numerous
spectral components with a significant rotational speed variations
during measurement time, the processed signal will be unclear and
challenging to analyse which may lead to a misinterpretation [36].
Comprehensive discussion and more details including strengths
and limitations of order analysis technique are presented in [37,
Fig. 1  Frequency range includes the details and approximation 38].
coefficients
3 Experimental setup
vibration signature is investigated. In this context, a test rig that
Fig. 2a shows a real photograph of the setup test model that has
emulating real WT operation has been developed. High-resolution
been implemented to carry out the experiments, while Fig. 2b
sensors are used to accurately measure the shaft mechanical torque
shows a schematic diagram of the test rig setup. The developed test
and the gearbox vibration signals. Wavelet and order analysis
rig consists of an induction motor connected to a generator through
signal processing techniques are employed to analyse the measured
gearbox, shaft and coupling system. Numbers of accelerometers of
signals. Both techniques are compared based on their sensitivity to
low (0.1 Hz) and high (1000 Hz) frequency with sensitivity of
detect incipient fault levels.
±10% were installed on the drive train at different locations to
collect the vibration signals. ISOTRON Accelerometer
2 Signal processing techniques (ENDEVECO model 2258A-10), with output sensitivity of 10 
In this section, a brief elaboration on wavelet and order analysis is mV/g was mounted on the gearbox towards the low-speed stage of
presented. the shaft. The other (AC133-1D high-performance low-frequency
accelerometer with a sensitivity of 500 mV/g was mounted on the
top side of the gearbox casing. The two-stage helical gearbox
2.1 Wavelet technique
which is the most sensitive and expensive component of the setup
Wavelet technique has been utilised to detect shaft misalignment has tooth ratio of 1: 4.34 and is lubricated with mineral oil SOVG
and bearing troubles by analysing the output power signal of 220. Torque transducer (FUTEK model TRS605) has been fitted
variable speed WTs [31]. Fig. 1 shows the approximation and between the prime mover and the low-speed shaft of the gearbox to
detail coefficients frequency range of a wavelet function that can measure the shaft torque; rated torque is 100 Nm. A comprehensive
be expressed as [30] study on test rig model for WT generator can be found in [21].

j−1
4 Experimental results
Xn = ∑ ai 0, j
φi
0, j
n + ∑ ∑ di jψ i j n
, ,
j i = i0 j 4.1 Torque signature


∞ Fig. 3a shows the torque signal measured by the torque transducer
ai n = x[n]φi n dn (1)
0, j
−∞
0, j during normal operating condition. The torque amplitude is
measured at a rotor speed of 200 rpm when there is no mechanical


stress applied on the drive train.
di, j n = x[n]ψ i, j dn
−∞
Fig. 3b shows the approximation coefficients of the healthy
torque signal using wavelet db14 (Daubechies 14), while Fig. 3c
i /2 i shows the detail coefficients of the torque signal. Fig. 3d shows the
where ψ [n] is the mother wavelet, and φi, j[n] = 2 0 φ(2 0n − j) is a signal in frequency domain using order analysis. The signals of
i
scaling function at a scale s = 2 0 shifted by j. Similarly, Fig. 3 are stored as a healthy reference signature to be compared
i with the future measured signatures in order to detect any variation
ψ i, j[n] = 2i /2ψ(2in − j) is the mother wavelet at a scale s = 2 0
due to possible faults.
shifted by j, ai , j represents the approximation coefficients at a A stressed shaft is implemented by injecting dc current of 85%
0
i of the generator rated current (9.39 A) through the generator stator
scale s = 2 0, and di, j are the detail coefficients at a scale s = 2i.
terminals to emulate real fault conditions on the WT blades such as
A high-order wavelet (Daubechies 14) has been chosen in order icing accumulation, wind shear and gust wind. The injected current
to minimise the overlapping between the frequency bands. DWT is will produce an opposite electromagnetic force that tends to retard
able to localise correctly the frequencies with accurate time the shaft speed which can emulate the above-mentioned real
locations and hence unforeseen transient variations that could take situations.
place within any steady-state regime can be detected by this As can be seen in Fig. 4a, there is a significant increase in the
application [32]. The use of orthogonal wavelets families results in torque amplitude when compared with the case of healthy
no redundancy which means N data points produce N wavelet condition that is shown in Fig. 3a. Most of rolling elements such as
amplitudes [33]. According to [34], the decomposition levels can bearings are designed to operate with a relatively smooth transition
be tested based on the following formula: between loaded and unloaded modes. The rapid and significant
amplitude change shown in the torque signal of Fig. 4a is attributed
Log( f s / f ) to the fact that, the machine is delivering more mechanical power
nd > +1 (2)
Log(2) to overcome the stress on its shaft which result in significant
impact on rolling elements surface durability.
where f s and f are the sampling and supply frequencies, The wavelet approximation coefficient shown in Fig. 4b
respectively. exhibits a significant drop in the amplitude when compared with
the normal operating condition signal shown in Fig. 3b,
2.2 Order analysis consequently the detail coefficients of the wavelet signal has
significantly decreased as can be seen in Fig. 4c. Power spectrum
Fluctuating load conditions and deformity in rotating machines can of the order analysis technique could not be captured for this case
cause constrained vibrations. According to the signal properties of because the high stator injected current caused high stress on the
the machine rotating parts, a vibration signal will be formed. By rotor of the generator and forced the whole drive train to stop when
involving spectra and time histories to the speed of the moving the shaft speed reaches the critical limit.
parts, order analysis has the ability to detect the vibration of

432 IET Sci. Meas. Technol., 2017, Vol. 11 Iss. 4, pp. 431-437
© The Institution of Engineering and Technology 2017
Fig. 2  Real photograph of the setup test model
(a) Test rig setup for WT, (b) Schematic diagram of the test rig setup for WT

Fig. 3  Torque signal under normal operating conditions (healthy conditions with rotor speed 200 rpm)

IET Sci. Meas. Technol., 2017, Vol. 11 Iss. 4, pp. 431-437 433
© The Institution of Engineering and Technology 2017
Fig. 4  Torque signal under stressed shaft (with rotor speed 200 rpm)

Fig. 5  Vibration signal under normal operating conditions (with rotor speed 200 rpm)

4.2 Vibration signature exhibits a crest value of 80 which is 20 times the amplitude of the
normal signal shown in Fig. 7b.
The vibration signal during normal operating conditions is shown It is also observed that the detail coefficients of the measured
in Fig. 5a which reveals that the vibration signal oscillates between signal are increased from 0.65 for normal operating condition to 12
maximum and minimum limits with some spikes. When analysed for stressed shaft condition as can be seen from Figs. 7c and 8c,
using wavelet technique, a better feature of the healthy vibration respectively. On the other hand, power spectrum for the faulty
signal can be gained and stored for future comparison as can be signal (Fig. 8d) exhibits more harmonics with less amplitude when
shown in the approximate and detail coefficients in Figs. 5b and c. compared with the healthy signature power spectrum (Fig. 7d).
Fig. 5d shows the harmonic contents in the order power spectrum Based on these results, a code to quantify the level of drivetrain
of the vibration signal during healthy operating conditions. stress along with the recommended asset management decision can
When 85% of the full load current is injected to the generator be developed as listed in Table 1. The table is developed based on
stator terminals, while maintaining the shaft speed (200 rpm) and the maximum amplitude of the vibration signal and the
the sensors locations, the vibration signal was recorded as shown in corresponding duration. The recommended asset management
Fig. 6. To show the impact of rotor speed variation on the decisions range from continuing normal operation to taking the WT
measured vibration signature, Fig. 7 shows the vibration signature out of service for serious levels of drivetrain stresses.
for a normal operating condition with a rotor speed of 650 rpm
while Fig. 8 shows the vibration signal with stressed shaft at this
higher speed. Results of Figs. 7 and 8 show that rotor speed has a 5 Comparison analysis
great influence on the measured vibration signature. Higher Studies in [37] show that order analysis can effectively identify and
rotational speed is corresponding to higher vibration amplitude. detect faults within the pinion gear in the gearbox. However, this
Also, it can be observed that the faulty signal shown in Fig. 8b digital signal-based technique requires the collected vibration

434 IET Sci. Meas. Technol., 2017, Vol. 11 Iss. 4, pp. 431-437
© The Institution of Engineering and Technology 2017
Fig. 6  Vibration signal under stressed shaft (with rotor speed 200 rpm)

Fig. 7  Vibration signal under normal operating conditions (with rotor speed 650 rpm)

signal from the machine to be sampled at a fixed angular increment applied to the test including changing in rotor speed which
which means the shaft speed should be kept proportional to the represents the WT blades as affected by the fluctuating loads. The
sampling rate. Without extra specified equipment, such as ratio shaft torque and gearbox vibration signals are measured using high
synthesiser and anti-aliasing tracking filter, the measured signal sensitive sensors for normal operating conditions and for a stressed
will not be accurate. Comprehensive discussions and more details drivetrain shaft which is implemented by injecting a current to the
including strengths and limitations of order analysis technique are generator stator winding to create an opposite torsional torque on
presented in [38, 39]. On the other hand, wavelet analysis the main shaft. The healthy and faulty signals are processed and
technique can identify the coefficients accurately in the low- and analysed using wavelet and order analyses techniques.
high-frequency range which makes it a powerful technique to Experimental results show that wavelet analysis technique has the
monitor and diagnose the health condition of rotating machines. ability to localise the frequencies with accurate time locations, as
Based on the above results, the wavelet technique is found to be well as detecting unforeseen transient variations based on the
superior and more powerful than the order analysis technique that different faulty and healthy scenarios investigated on the developed
may fail to analyse the shaft torque signal under high mechanical test rig. While order analysis shows more harmonics in the
stresses as explained in the case study shown in Fig. 4. analysed faulty signal, it is hard to be diagnosed to detect the root
failure causes. A code to quantify the mechanical stress level on
6 Conclusions the drivetrain is developed based on the measured vibration
signatures and asset management decision is recommended.
Gearbox and generator shaft are vital mechanical components in
WECSs and hence a reliable technique is required to maintain the
reliability of the system. In this paper, a test rig is developed to
emulate real operation of a WT. Different scenarios have been

IET Sci. Meas. Technol., 2017, Vol. 11 Iss. 4, pp. 431-437 435
© The Institution of Engineering and Technology 2017
Fig. 8  Vibration signal under stressed shaft (with rotor speed 650 rpm)

Table 1 Correlation between drivetrain stress level and [12] Wang, W., McFadden, P.: ‘Early detection of gear failure by vibration analysis
i. Calculation of the time–frequency distribution’, Mech. Syst. Signal
vibration parameters along with recommended asset Process., 1993, 7, pp. 193–203
management decision [13] McFadden, P.: ‘Examination of a technique for the early detection of failure in
Fault type Vibration Time, s Level Asset gears by signal processing of the time domain average of the meshing
amplitude management vibration’, Mech. Syst. Signal Process., 1987, 1, pp. 173–183
[14] Ciang, C.C., Lee, J.-R., Bang, H.-J.: ‘Structural health monitoring for a wind
decision turbine system: a review of damage detection methods’, Meas. Sci. Technol.,
drivetrain 0–0.2 0.5–0.6 lowest
continue normal 2008, 19, p. 122001
stresses operation [15] Vries, E.D.: ‘Wind turbine gearboxes and the effort to improve their
reliability’ (Wind Power, Windcomms, UK, 2010)
0.252–0.265 0.61– low visual inspection is [16] Gould, B., Burris, D.: ‘Effects of wind shear on wind turbine rotor loads and
0.65 recommended planetary bearing reliability’, Wind Energy, 2015, 19 (6), pp. 1011–1021
[17] Li, S.: ‘Effects of centrifugal load on tooth contact stresses and bending
0.269–0.55 0.66–0.7 average scheduled for
stresses of thin-rimmed spur gears with inclined webs’, Mech. Mach. Theory,
maintenance 2013, 59, pp. 34–47
0.56–1.00 0.71– high must stop [18] Tiwari, S.K., Joshi, U.K.: ‘Stress analysis of mating involute spur gear teeth’,
0.73 Int. J. Eng. Res. Technol., 2012, 1 (9), pp. 1–12
[19] Mark, W.D., Lee, H., Patrick, R., et al.: ‘A simple frequency-domain
1.01–1.25 0.74– highest take out of service algorithm for early detection of damaged gear teeth’, Mech. Syst. Signal
0.76 Process., 2010, 24, pp. 2807–2823
[20] Molinas, M., Suul, J.A., Undeland, T.: ‘Extending the life of gear box in wind
generators by smoothing transient torque with STATCOM’, IEEE Trans. Ind.
Electron., 2010, 57, pp. 476–484
7 References [21] Asmine, M., Brochu, J., Fortmann, J., et al.: ‘Model validation for wind
turbine generator models’, IEEE Trans. Power Syst., 2011, 26, pp. 1769–1782
[1] Salem, A., Abu-Siada, A., Islam, S.: ‘A review of condition monitoring [22] Yuan, X., Cai, L.: ‘Variable amplitude Fourier series with its application in
techniques of the wind turbines gearbox and rotor’, Int. J. Electr. Energy, gearbox diagnosis – part II: experiment and application’, ScienceDirect, 2005,
2013, 2, (1), pp. 53–56. doi: 10.12720/ijoee 19, pp. 1067–1081
[2] Kar, C., Mohanty, A.: ‘Monitoring gear vibrations through motor current [23] Zhang, Z., Verma, A., Kusiak, A.: ‘Fault analysis and condition monitoring of
signature analysis and wavelet transform’, Mech. Syst. Signal Process., 2006, the wind turbine gearbox’, IEEE Trans. Energy Convers., 2012, 27, pp. 526–
20, pp. 158–187 535
[3] Sheng, S.: ‘Report on wind turbine subsystem reliability – a survey of various [24] Wang, X., Makis, V., Yang, M.: ‘A wavelet approach to fault diagnosis of a
databases’. National Renewable Energy Laboratory, 2013 gearbox under varying load conditions’, J. Sound Vib., 2010, 329, pp. 1570–
[4] McFadden, P., Smith, J.: ‘A signal processing technique for detecting local 1585
defects in a gear from the signal average of the vibration’, Proc. of the Inst. [25] Fischer, K., Besnard, F., Bertling, L.: ‘Reliability-centered maintenance for
Mech. Eng. C, J. Mech. Eng. Sci., 1985, 199, pp. 287–292 wind turbines based on statistical analysis and practical experience’, IEEE
[5] Zakrajsek, J.J., Townsend, D.P., Decker, H.J.: ‘An analysis of gear fault Trans. Energy Convers., 2012, 27, pp. 184–195
detection methods as applied to pitting fatigue failure data’, DTIC Document, [26] Thiringer, T., Linders, J.: ‘Control by variable rotor speed of a fixed-pitch
1993 wind turbine operating in a wide speed range’, IEEE Trans. Energy Convers.,
[6] Gaeid, K.S., Ping, H.W.: ‘Wavelet fault diagnosis of induction motor’, cdn. 1993, 8, pp. 520–526
intechopen. com, 2009 [27] Besnard, F., Bertling, L.: ‘An approach for condition-based maintenance
[7] Haghjoo, F., Shahrtash, S.M.: ‘Wavelet transform based decomposition and optimization applied to wind turbine blades’, IEEE Trans. Sustain. Energy,
reconstruction for on-line PD detection and measurement. Part I: narrow band 2010, 1, pp. 77–83
components decomposition’, Eur. Trans. Electr. Power, 2010, 20, pp. 901– [28] Wu, J., Lai, F.: ‘Fatigue life analysis of small composite sandwich wind
914. doi: 10.1002/etep.373 turbine blades’, Procedia Eng., 2011, 14, pp. 2014–2020
[8] Kia, S.H., Henao, H., Capolino, S.G.A.: ‘Diagnosis of broken bar fault in [29] Combet, F., Gelman, L.: ‘Novel adaptation of the demodulation technology
induction machines using discrete wavelet transform without slip estimation’. for gear damage detection to variable amplitudes of mesh harmonics’, Mech.
Industry Applications Conf., 2007. 42nd IAS Annual Meeting. Conf. Record Syst. Signal Process., 2011, 25, pp. 839–845
of the 2007 IEEE, 2007, pp. 1917–1922 [30] Lei, Y., Kong, D., Lin, J., et al.: ‘Fault detection of planetary gearboxes using
[9] Tang, J., Luo, H.: ‘Wind turbine gearbox fault detection using multiple new diagnostic parameters’, Meas. Sci. Technol., 2012, 23, p. 055605. doi:
sensors with features level data fusion’, J. Eng. Gas Turbines Power, 2012, 10.1088/0957-0233/23/5/055605
134, pp. 042501–042501 [31] Wiggelinkhuizen, E., Verbruggen, T., Braam, H., et al.: ‘Assessment of
[10] Link, H., LaCava, W., Van Dam, J., et al.: ‘Gearbox reliability collaborative condition monitoring techniques for offshore wind farms’, J. Sol. Energy
project report: findings from phase 1 and phase 2 testing’, Contract, 2011, Eng., 2008, 130, pp. 030301.030301–031020.030312
303, pp. 275–3000 [32] Antonino-Daviu, J.A., Riera-Guasp, M., Folch, J.R., et al.: ‘Validation of a
[11] Astridge, D.G.: ‘Helicopter transmissions – —design for safety and new method for the diagnosis of rotor bar failures via wavelet transform in
reliability’, Proc. of the Inst. Mech. Eng. G, J. Aerosp. Eng., 1989, 203, pp. industrial induction machines’, IEEE Trans. Ind. Appl., 2006, 42, pp. 990–996
123–138 [33] Newland, D.E.: ‘An introduction to random vibrations, spectral & wavelet
analysis’ (Courier Dover Publications, 2012)

436 IET Sci. Meas. Technol., 2017, Vol. 11 Iss. 4, pp. 431-437
© The Institution of Engineering and Technology 2017
[34] Khalaf, S.G., Hew, W.P.: ‘Wavelet fault diagnosis and tolerant of induction [38] Nikias, C.L., Mendel, J.M.: ‘Signal processing with higher-order spectra’,
motor: a review’, Int. J. Phys. Sci., 2011, 6, pp. 358–376 IEEE Signal Process. Mag., 1993, 10, pp. 10–37
[35] Fyfe, K., Munck, E.: ‘Analysis of computed order tracking’, Mech. Syst. [39] Saavedra, P., Rodriguez, C.: ‘Accurate assessment of computed order
Signal Process., 1997, 11, pp. 187–205 tracking’, Shock Vib., 2006, 13, pp. 13–32
[36] Bossley, K., Mckendrick, R., Harris, C., et al.: ‘Hybrid computed order
tracking’, Mech. Syst. Signal Process., 1999, 13, pp. 627–641
[37] Eggers, B., Heyns, P., Stander, C.: ‘Using computed order tracking to detect
gear condition aboard a dragline’, J. South African Inst. Min. Metall., 2007,
107, p. 115

IET Sci. Meas. Technol., 2017, Vol. 11 Iss. 4, pp. 431-437 437
© The Institution of Engineering and Technology 2017

Você também pode gostar