Vibration Analysis of the Windflow Turbine Gearbox

Epicyclic Gearbox with Load Sharing Planet Pins

The gearbox must convert 48 rpm at the rotor blades into 1500 rpm for the synchronous generator. It must also be as light and compact as possible. This was achieved using a 4 stage design, with three of the stages using epicyclic (planetary) gears. These were chosen for their high torque rating in a compact housing.

The highest torque acts on the slow turning first stage. By using more planets, the load can be shared by more gear teeth simultaneously. Four planets is the common upper limit, as above that number it becomes very difficult to machine the gear with such precision that all share equal load.

Windflow overcame this by using flexible planet pins, thus allowing them to use eight planets in the first stage; but these are not simple straight pins, as that would cause twisting of the planet’s rotation axis, thus distorting the gear mesh. Instead they use a double cantilevered design which allows for a massive 0.5 mm of axial deflection, without twisting.

Commtest’s Initial Involvement

We were interested in learning about the issues involved in condition monitoring on wind turbines, and had contacted Windflow about a mutual field trial arrangement. The first opportunity occurred during their pre-commissioning testing of the gearbox + generator assembly in a factory in Christchurch. The assembly was being driven by a test rig motor and gearbox as shown.

They were seeing very high vibration levels at the running speed of the generator. They knew that the test rig motor was running at a similar speed, but could not determine which was the actual culprit.

By taking a few high resolution spectra it was clear that the 1480 rpm test rig motor was totally dominating the vibration from the generator at 1500 rpm. Windflow were very happy!

GMF of Epicyclic Gearboxes

After that initial success things quickly got trickier. Three epicyclic stages meant a lot of shafts, gear meshes and bearings, all in one compact housing. The spectra were very confusing!

I created a spreadsheet to calculate all the various fault frequencies (planet pass, carrier, sun, ring, and all the usual bearing ones at whatever speed they happen to be turning).

Of particular concern was the huge peak at 311 Hz, which was frustratingly close, but not equal, to the second stage GMF of 313.36 Hz. No number of recalculations could bring the two together!

I suspected a modulation effect, caused by the planets passing the location of the accelerometer. The “Frequency based Waveform Analysis” function in the Ascent software enabled me to look at the waveform in a narrow band around 311 Hz.

This showed clear modulation! So the actual peak must be just a sideband of the GMF. The difference between the two cursors on the waveform shows modulation at about 10 Hz, which does correspond with the Planet Pass frequency (Carrier fc of 2.51 Hz x 4 planets).

BUT the difference between 311 Hz and the actual GMF is about 2.5 Hz, not the 10 Hz…. so that’s not quite it…

Later in the proceedings Windflow engaged the services of many acoustics and vibration experts. One of whom was Lan Le-Ngoc from IRL. He uncovered a paper by J. Mc Names titled “Fourier Series Analysis of Epicyclic Gearbox Vibration”. The complex math he used predicted that the phased signal from each of the planets would cancel out the others, except at multiples of the number of planets. As an example he used a gearbox with 119 teeth on the Ring gear, for which you would expect to measure a GMF at 119 times the Carrier rpm. He predicted that the vibrations would cancel at all frequencies except for multiples of the 8 planets, ie at 112, 120, 128 (x Carrier rpm).

Measurements on a Cobra AH-1S helicopter gearbox proved this to be exactly the case. The GMF appeared at 120 cpr, not 119!

The Windflow gearbox second stage has 4 planets and 125 ring teeth. So the closest #planet multiple is 124 cpr, and:

124/125 * 313.5 Hz (actual GMF) = 311 Hz (observed GMF). Eureka!

So we were now sure that large vibration peak was coming from the gearbox second stage. But why was it so much larger than that from the (almost identical) first stage?

Note that the first stage has the same number of ring teeth (125) as the second stage, but has 8 planets instead of 4, so the closest #planet multiples are a lot further away from the actual GMF, at 120 and 128…

Windflow were already behind schedule for installing the turbine on the hill, so I concluded my (unofficial) testing report with the disclaimer:

“Using ISO 10816-3 as a reference (rpm<=600, rigid, group2), most locations have ‘Acceptable’ overall vibration levels in the 10-1000 Hz band…

“However I suggest that the 311 Hz frequency region should be closely monitored when the Wind turbine is operating. It seems significant that by comparison the 1st stage GMF at 102 Hz is almost non-existent in the spectra.”

Online Condition Monitoring

During all of the above, Windflow had several bearing failures in the gearbox. These were mainly around the complex concentric shafts of the torque limiting 4^th stage. Meanwhile we at Commtest had developed an online monitoring system, the vbOnline™. We agreed on a mutually beneficial field trial partnership, in which they would gain surveillance and some measure of shutdown protection, and we would gain insight into the challenges of online monitoring on wind turbines.

The turbine is at a remote site on the Port Hills, lacking even a telephone land line. We decided to install a PC on-site to control the frequent protection recordings, and use a wireless data modem to allow access from the head office.

The on-site PC was intended to be a simple, low cost solid state PC with no moving parts. These are sold under names like WinTerm and ThinClient. They use flash memory, have processors around 1 GHz and cost only US $500. This PC runs the OnlineManager program which schedules the recordings on the vbOnline module, stores them in a database, checks alarms and controls the module’s output relays.

We established a machine structure in which daily gearbox surveillance recordings were kept separate from the very frequent protection recordings.

For the protection readings, alarm checking was complicated by several factors:

• the significant changes in vibration levels at different power output levels
• the gusty nature of the wind, causing rapid power level changes
• the desire to have a “second opinion” reading, before causing a turbine shutdown

We addressed these issues by reading the generators’ power output level from the turbine’s PLC control system. We established Recording Criteria to separate the recordings into three generation power levels, each with their own alarm thresholds. Recording Intervals alternated the Check1 and Check2 recordings, each of which controlled an output relay.

We wired the two relays in series so both would need to be activated to cause a shutdown.

This solution proved to be adequate, but is rather complex and the separation into the generation power levels is not perfect. Development is underway to streamline this, which should yield benefits not just for wind turbines but for all users of variable speed machinery.