Time series data and predictive maintenance - addendum

A 30 year lifetime for a wind turbine is of course more attractive than just 10 years of operation. However, the plots for 30 years shown below are less illustrative than the plots for 10 years shown in the main article. With the system identification procedure applied, 259200 time steps (one hour samples for 30 years) constitute no problem at all - as documented below.

Putting it all together into a 30 years lifetime simulation

Combining

• the computational cost savings from replacing a full-blown finite element simulation with an identified model,
• the three step recipe described here of getting from stress as a function of time to damage as a function of time, and
• a "power as a function of time" curve which can be attributed to a real wind turbine

, you can generate a plot like Figure 9a below:

Figure 9a: Power versus time displayed together with damage versus time

A modification of thermal properties may extend the life of the horseshoe

The calculation procedure leading to Figure 9a above may be modified so that lessons may be learned. For instance, this figure shows that the fillet stress follows the heat load with a time delay so that fast load cycles lead to lower stress amplitudes than slow load cycles.

One way to decrease at least some of the load cycles is therefore to increase the time constant of the thermal problem. Just as an experiment, an additional heat capacity was therefore included next to the heat sink bottom left in this figure. Figure 9a is then modified to look like this:

Figure 11a: Power versus time displayed together with damage versus time - now with reduced damage

Compared to Figure 9a, the damage after 30 years of the same operation load history has decreased from 1.04 to 0.92 just from a modification of the thermal properties of the horseshoe.