Predictail Othello is a service that Predictail provides for windfarm owners and operators. We search for anomalies in measurements downloaded from turbine SCADA system (Supervisory Control And Data Acquisition) and we describe them in the form of a report. Our clients can use this information to make their maintenance very effective, thus increase total availibility and energy production.

The result of Predictail Othello service is a PDF document – a report of anomalies.

Each anomaly is described by:

• turbine identifier and period when the anomaly was detected
• related component (generator / gearbox / transformer etc. )
• related signal f.e. bearing temperature, oil pressure
• algorithms used in detection f.e. digital-twin and cross-population
• related service logs
• list of similar anomalies from the past
• probable cause and recommendation

We have calculated that statistically, wind farm owners tend to lose around 40 000 Eur of profit per year per 10 turbines because of unplanned stoppages longer than 6 hours. Using Predictail Othello analysis results one can prevent those stoppages and significantly reduce loss – by even more than 60%. So, yes – it does pay off.
There is no wind farm that has perfect availability – is always can be better.

For each component (generator, gearbox, transformer etc. ) of every turbine we use 3 types of algorithms (also called “models”) to detect anomalies:

• “digital-twin” – it is used to find out if given component is behaving in a different way than it used to. One can think of it as a simulator trained to reproduce what is a typical response of a given component. For example, we can detect that generator bearing temperature is higher than it used to be in particular environmental conditions like wind speed and ambient temperature.
• “cross-population” – it is used to find out if a given component is behaving in a different way than the rest of the population. For every measurement sample, we use data from the rest of the turbines to approximate what is an expected response of a given component. For example, we can detect that generator bearing temperature in a given turbine is higher than in a group of most similar turbines in a wind farm population.
• “intra-component” – it is used to find out if given component is behaving in an internally consistent way. For example, using this model we can detect that temperatures of three generator phases are not correlated anymore, thus one of the sensors can be broken.
  For one turbine we use ~20 digital-twins, ~80 cross-population and ~9 intra-component models (numbers depend on the turbine model). We use an automatic scoring procedure to rate the importance of anomalies and select those to be described in the report in detail.

Yes. We have clients that order Predictail Othello despite having an availability guarantee agreement – they treat our service as a 3rd party audit of what is the technical state of their turbines. Also, one can use our list of anomalies to create service tickets for producer’s maintenance.

So far we analysed SCADA data from Vestas, Siemens, Senvion and GE turbines and we are open to analyse any other without additional charge. Algorithms used by Predictail are very generic, they just need to be tuned when we approach a new type of turbine.

To perform our analysis, the following data will be needed:

• SCADA measurement data – all available features, at least 3 years history with at least 10 min sampling frequency
• System statuses data with error codes
• Service history log – to verify detected anomalies we need to know if an anomaly was related to service or not
• Turbine technical specification with producer’s nominal power curve table

You can transfer us data downloaded from SCADA in .csv files via FTP or any other medium that is convinient to you. We can also use OPC or ODBC protocol to download the data from SCADA for you.

We recommend to order our at least analysis twice a year – before every scheduled maintenance, so technicians can use our report during service.

Vibration analysis based systems like CMS are complementary to Predictail Othello – those systems monitor vibrations, while Predictail Othello is used to analyzing all other signals like temperatures, pressures, RPM’s and currents. We recommend using both of solutions in parallel!

We will release an online analysis software tool in 2019 – if you are interested, let us know!

So far we analysed data from 252 turbines, 4+ years history each – it’s over 1000 turbine-years!