Tuesday 12 April 2016

Topic Introduction

Hi everyone again! Long time no see. If I said that a blogger life is easy I take out those words, inspiration is not easy to find these days!

In the second intervention in my blog I intend to talk a bit about my work. oshhhh...boring...

Since my last post my work has been progressing smoothly, in this phase I´m looking only to the load model that loads a wind turbine and its support structures (external variables like the wind, waves...etc) and how to treat it statistically. Recently I have been analyzing offshore data that was collected in Ireland over the last 12 years (a courtesy of Met-Éireann).

Now! Engineering things like offshore wind turbines are built with the expectation of a long, prosperous and peaceful live. If that does not happen, as an engineer prepare to wake up with some investors in front of your door in their classical "stressed" style but instead of holding black cases, holding axes, knives, guns and well... one or another with a flamethrower...

So, how can we guarantee that nothing is going wrong and we have a long life in our turbine?
The first step is guaranteeing that we know what is going to load the turbine very well.
But, in time gaps of 10, 20, 30 years?
Can you imagine if every time we wanted to address a certain site we needed to measure 10, 20 or 30 years of data to accurately characterize the long term behavior of any physical phenomena?
For some purposes we might do it but, on a daily basis, no way!, in the crazy world we live there's no time for that! We really need something that makes the process much faster.

Usually field data are limited. We measure it but like, we cannot be doing it during 30 years the wind prior to any further step....
So, usually the trick is fitting, and applying known statistical distributions to describe uncertain long term phenomena.
In easy words, you do your best with all the data that you have to find a graphical representation like a curve, or surface, that better goes trough all your data. This way, you can know more or less what is happening between those points without needing to detect every possible occurrence in 30 years.

Obviously the less data we have the harder is to have a good representation of the reality, and even harder when we are looking for rare events like extreme events that are in the small tails of the distributions. And if there is some joint dependence, even harder!  For approximating these tails, several techniques and distributions are used as, block maxima, peak over threshold or distributions like the Weibull, Gumbell, or other.

The challenge for the following months is then, diving into this "swamp" of already existing ideas of extreme statistics and come out with a needle from it (or the more mainstream "a needle in the haystack"... which looks easy if we think in a swamp)
Hard, hein? ...but, don´t worry I already have the life-jacket dressed and a rope tied to a tree so that I don't drown myself! I'm kinda of good swimmer too

Already a long post! Bigmouth strikes again..

Next time I'll come with a more light approach to the theme Offshore Wind Turbines. I don't want anyone to have an overdose of boredom while coming here...
.
So I´ll leave the second part that I was thinking on publishing today to the next weeks, the discussion will be one of the most hot topics of the engineering fashion world: Are Wind Turbines in the landscape cool or not? (I'm not a civil engineer, but I already  know that architects are saying nooo)