skip to Main Content

How Predictive Analytics is impacting O&M

The 5th ONYX InSight Wind Turbine Technical Symposium was held entirely online in 2020 due to the COVID pandemic. This shift to digital allowed us to connect with more people in our industry and share knowledge easily. Divided over three days, with times to suit our global audience, our symposium was attended by over 500 industry professionals.

Our first panel consisted of Annette Andersen, Engineering Director at bp Wind, Joris Peeters Head of Digitalization at ZF Wind Power, and Robert Gomez Director of Asset Performance at Ørsted, and was chaired by our VP of Engineering in North America, Dr Zhiwei Zhang.

Dr Zhang led the session by asking specific questions to our panelists to get an insight into the day-to-day realities of using predictive analytics in operations and maintenance. Answers have been condensed and edited for this article.

Watch the full panel discussion recording here >>

What predictive analytics do your companies use, and what benefits have you seen so far?

Annette Andersen (bp Wind)

At bp Wind, we use predictive analytics to monitor our vibration analyses and database analytics. This has saved us millions in terms of maintenance for our fleet.

The three key benefits we have seen are:

  1. Less severe repairs
  2. Room to optimise the deployment of maintenance crane and crew
  3. Minimal downtime. In some cases, the turbine runs right up to the time the cranes arrive on site.

Robert Gomez (Ørsted)

We use predictive analytics in vibration and scatter data to predict failures upfront, and this is part of our LTSA  (Long Term Service Agreement) with our OEM (Original Equipment Manufacturer).

It’s essential to use data to predict failures upfront, which reduces downtime for us as well as component costs. This, in turn, has an impact on the LCOE (Levelised Cost of Electricity) of the operating fleet.

My team goes a step further by focusing on the performance of assets while in operation. We can improve performance by looking at outlier detection, where there is some misalignment or underperformance. We can use this to improve response times and calculate metrics such as mean time-to-failure and failure duration. Additionally, this helps us drive conversations with our providers, as we can identify our pain points because we have this level of visibility.

Joris Peeters (ZF Wind Power)

As a component provider, our analytics are focused on creating value for the customer. We use it in condition monitoring for identifying anomalies and developing reliability analytics to predict failure and improve prognostics.

We also use statistics, such as scatter loads, that helps us calculate consumed lifetime, which is correlated with failure modes that are fatigue driven.

We build reliability models based on statistics of actual failures. It’s important to know what’s inside your powertrain, the quality of data as well, to build confidence into reliability analytics.

Another part is keeping track of service history, from upgrades to repairs. As a supplier, we try and make this data readily available and in a useable format for the industry.

Can you give an example of a problem that predictive analytics is solving in your organisation?

Annette Andersen (bp Wind)

Beyond the drivetrain, we are looking to applied predictive analytics to the sub-components in the wind turbines. After bringing a data scientist in-house, we looked into a failure happening with a cooling fan. This is usually a straightforward repair and is inexpensive, but the turbine would fault and shutdown and would not run until it could be repaired, which could take a few months.

Using the data from a sub-set of turbines and looking at these sub-components, we were able to predict 27 failures ahead of time. This helped us combine trips and repairs with other work, therefore reduce the safety risk that comes with confined space entry and reducing downtime.

Robert Gomez (Ørsted)

We use analytics to predict what our assets should be producing in a given amount of time, then identifying assets that are not meeting expectations and diving into why they are not meeting those expectations. Sometimes it’s parameter settings or issues with the asset itself.

Historically, certain assets have been curtailed, but when you look into the data history, we can see why they are not meeting expectations.

Unlock potential by looking at performance analytics.

Joris Peeters (ZF Wind Power)

From the drivetrain perspective, we’ve already been able to add value in the domain of spares management. The importance of asset managers balancing their spare-part stock and supply chain availability was highlighted in the opening talk by Dr Evgenia Golysheva from ONYX.

It’s common industry knowledge that having short term availability of necessary spare parts is important but is not always a given. We use reliability analytics to predict which failures will occur and which solutions we will need to have in stock. This helps our customers reduce unnecessary inventory, but we assure them that they have the right spares in stock for their fleet. With predictive analytics, availability of stock is guaranteed, downtime is decreased, and unnecessary live inspections are reduced.

Where do you see predictive analytics providing the most value for wind turbine O&M and why?

Annette Andersen (bp Wind)

We see the most value in the drivetrain, but a successful analytics program should extend to wherever there are business needs.

That’s been the focus of our program: having those data skillsets in-house that’s able to work with our wind farm managers and our engineers. This has enabled us to build analytics to suit needs like tracking transformer health, earlier failure detection, the health of the main bearings and having insight over the spectrum of health across the fleet.

Robert Gomez (Ørsted)

As Annette said, the most value in is predicting drivetrain failures, preventing unnecessary downtime and changing component failures to sub-component failures.

From an offshore perspective, my colleagues have been focused on trip efficiencies. They are making sure that they solve problems the first time. They track metrics on repeat failures and return visits. There are efficiencies to be gained in these scenarios, especially when access is an issue.

Joris Peeters (ZF Wind Power)

We look at lost energy production during downtime and reducing it as the main benefit for our customers. We also look at reducing the operational expenditure in the drivetrain. We try to avoid failures, and if that’s not possible, we avoid a gearbox exchange, and if that’s not possible, we have the gearbox exchange planned in advance which is important for offshore fleets.

What limitation have you observed when it comes to predictive analytics?

Annette Andersen (bp Wind)

Is the data available? Sometimes even if there is a sensor installed, the data stream might not be available to us.

The second limitation is around data fidelity. Do we have the right frequency of data?

Some data from our components cannot be captured, or our data historians don’t have the capacity to store it. There has to be some innovation when it comes to ways of capturing new data because there is value in it.

Robert Gomez (Ørsted)

We have a relatively young fleet onshore. We’re lucky to have data access, data fidelity and that’s because we have functional Condition Monitoring Systems (CMS) installed.

Ageing fleets can have data availability issues or a CMS standpoint.

Vibration analytics has developed tremendously in the last few years, but we are still trying to refine the amount of time we have to make the repair. We can have 18-24 months of lead time on a bearing failure, but we want to be able to use a component efficiently, identify its remaining useful life and refine that six-month range of failure to something smaller.

Joris Peeters (ZF Wind Power)

The industry still needs to be improved when it comes to data quality. This can be done by standardising data across brands of wind turbines and implementing taxonomies for assets and component failures.

The other limitation is the availability of data. Data sharing is not always constant with the component supplier. Predictive analytics is sometimes hampered by not having access to data.

A challenge faced by those using predictive analytics today is understanding the root causes of failures. It’s not known, even with a new machine, what can occur and what can do wrong from the beginning. Predictive analytics needs a training period, and that requires domain knowledge.

For more insights, watch the full-length panel discussion. Our panelists delve into their experience with CMS, how data access and be improved, as well as share insights on how the wind industry can improve with the help of data.

Back To Top