Brüel & Kjær Vibro (B&K Vibro) has been monitoring in the wind industry for more than 40 years and has gained a lot of experience since then. This has fortunately resulted in a lot of success over the years but there also have been a fair share of challenges. Carsten Andersson, Development and Application Advisor at B&K Vibro, is one of our specialists who has been intimately connected to the wind industry for many years. He shares some of his experiences in this interview.
Click a link to navigate to the corresponding section:
Leadership Role within B&K Vibro
Improved Services – Smart Alarms
The Control Room Standard – What Carsten has been up to lately
Artificial Intelligence and Machine Learning
The ADVISOR initiative – An early attempt at using AI and Machine Learning
The Future of Condition Monitoring in the Wind Industry
Hello and welcome! I’m Terry Siggins, Global Marketing and EMEA Sales Director for Brüel & Kjær Vibro, and today I’m delighted to say that I’ve been joined by Carsten Andersson, one of the driving forces behind the success of Brüel & Kjær Vibro. Hello Carsten!
Hello Terry! I’m looking forward to our conversation.
So, Carsten, tell us how your started at Brüel & Kjær Vibro, and over your many years with the company, how your career has taken shape…
I have been with Brüel & Kjær and Brüel & Kjær Vibro since 1979 and have always been part of our Condition Monitoring team. My roles have included software developer, system architect, project manager and product owner.
Today, my title is Development and Application Advisor, and I am gradually winding down into retirement. I am now also spending time on other activities with my family, enjoying my grandchildren, house and garden. I try to keep myself fit by physical exercise, and I’m especially fond of cycling.
During your time with us Carsten, what are some of the major developments that you have been an integral part of?
I’ve been a part of several development projects either, as of which I consider the following as the most important:
- The battery-operated Vibration Analyzer and the related software solution Machine Monitoring Software Type 7616.
- Predecessor of the Compass system, which was a customized development for the Statoil Gullfaks production platforms.
- Compass datalogger type 2526 and the related Compass Off-line Monitoring Software,.
- Advisor Machine Diagnosis Software
- First version of the Windows based Monitoring Workstation for Compass.
- In 2003, I became a member of a small group together with Peter Allpass, Sascha Gutt and Richard Bechelli, who started the Wind turbine Monitoring Business, also known as the Remote Monitoring Group, resulting in development of a series of products we know today as VibroSuite, DDAU2 and DDAU3/VCM-3.
Apart from developing products, one of my other big interests has been development of standards. I have been Chairman of ISO TC108 SC5 Workgroup 16 – Wind turbine Condition Monitoring, where we have developed the standards ISO 16079-1 and ISO 16079-2. I have also taken part in standards development under IEC, where I have been contributing author of IEC61400-25-2 and -6.
Leadership role within B&K Vibro
That’s quite a list of achievements. Could you expand a little more on your leadership role within our wind turbine monitoring solution? What were the challenges that you faced to get this up and running?
BKV was contacted by the wind turbine manufacturer NEG Micon in 2003, who asked whether we could offer a solution for their turbines as they needed a product they could offer to their customers as a response to the requirements from the insurance companies. This was the start of the Remote Monitoring Group in Denmark, in the beginning, only with 4 people.
So, what happened next?
We proposed a solution to NEG Micon based on products we already had in our portfolio at that time. That is, the Compass System as the software solution, and the Vibro IC as the front-end.
The VibroIC located in the wind turbine measured the descriptors and was mounted in a box together with two other hardware products developed by IoTech (now Prevas) consisting of the DDAU1 for acquisition of time waveforms and a WEB server for transfer of data via the mobile network to a data acquisition server located in our server room.
At the same time the remote monitoring group at BKV developed an interface from the data acquisition server to the Compass database and a WEB based wind turbine data management system, a new wind turbine analysis tool – the WTG Analyzer, and reporting tools for the alarm reports.
In the middle of this process, NEG Micon was acquired by Vestas early 2004. We knew Vestas was working on their own condition monitoring solution for wind turbines, so we worked day and night to finish our solution and get it certified before the date when NEG Micon officially merged with Vestas.
We thought our position would be better if NEG Micon had an operational solution which was certified by Germanischer Lloyd the day Vestas took over NEG Micon.
Did you make it in time?
Yes – we managed to reach that goal, and convinced Vestas that they should go for our solution.
Vestas made it a part of their strategy to do condition monitoring in-house and started to offer their own service products. We were in tough competition with other wind turbine monitoring vendors, but Vestas ended basing most of their CMS business on our products, including VibroSuite as their condition monitoring software.
As of today, we have reached nearly 40000 systems in the field. If someone had told us that we would reach that number when we started in 2003, we would never have believed them.
How has the business and customer expectations evolved from the time of the first wind turbine?
Every year since 2003, we have had to make continuous improvements to keep pace with the growing demand for increased turbine monitoring efficiency.
Once our stock of VibroIC’s ran out, we had to develop the new hardware DDAU2 and afterwards the DDAU3, which was developed in parallel with the introduction of the full-blown condition monitoring system that we call “VibroSuite”. Every new development was introduced while we were running our service business. This was a tricky process as we had to effectively change the wheels on the car whilst still driving it!.
Improved Services – Smart Alarms
Have the requirements also changed from a service perspective?
In the beginning, one of our services were to deliver half year status reports talking about the current status of the turbine, including a prognosis for the next half year. In parallel we delivered an alarm report service, where alarm reports were submitted as soon as a fault on a turbine was detected.
The need for the status reports in this form has died out, now the focus is on the alarm report business. Each alarm report has a severity assessment with an associated lead time to failure.
Reports have four severity levels:
- Severity 4 is the first indication of a fault. These reports can have a very long lead time, more than half a year.
- Severity 2 and 3 ???
- Severity 1 is where we advise to stop the turbine.
Our concept has always been to provide alarm reports only when we were certain that a fault was present, and always confirmed by a diagnostic engineer. The goal was to provide highly reliable reports (no false alarms) with as long lead time as possible.
How has this changed today?
As I see it, the picture has changed because operators are driven to run the turbines closer to breakdown, possibly derating the turbine output to reduce load turbine.
One of our most recent developments in the Remote Monitoring group has been to move the “Stop or Go” decision into the monitoring device in the turbine. Our DDAU3 (Diagnostic Unit) has been extended with intelligence that enables the device to send an automatic STOP/GO decision to the turbine controller, thus bypassing the diagnostic engineer and alarm report.
I consider this a very important step for the condition monitoring systems. Condition monitoring is now an integral part of daily operations for wind power plants, sitting at the nerve-centre of daily operations, rather than just sending alarm reports to the maintenance department to use for long-term maintenance planning.
The Control Room Standard
What have you been working on lately?
Recently, I have been a co-author of IEC 61400-25-2, a document which specifies the information model of devices and functions related to wind power plant applications – known as the control room standard. My role has been to add a section describing the data model for communication between the condition monitoring device and the turbine controller.
New failure modes are increasingly showing up as turbines get even bigger. In the new large +5MW turbines, you see faults which are hard to explain by using knowledge from smaller turbines. There are indications that the dynamic behaviour of these large turbines may be different to smaller turbines built using the same principles, so the need for a device capturing sudden event data under a different operating conditions becomes even more important.
Artificial Intelligence and Machine Learning
After many years in the business, we have collected a huge lake of wind turbine monitoring data. Could you share with us how Brüel & Kjær Vibro is using this data to improve analytics, including any work in the fields of Artificial Intelligence or Machine Learning?
The kind of AI I have been working with in the past and what I also think is most relevant for us in the future is what you call narrow AI (NAI) as opposed to General Artificial Intelligence (GAI). In NAI, we use a collection of technologies that rely on algorithms, deterministic and statistical, to simulate intelligence, generally with a focus on a specific task, like making an automatic fault recognition on a machine.
GAI in contrast, is designed to learn and adapt, to make decisions tomorrow that are better than those made today. None of this is easy, which is why most examples of AI you’ll encounter today are the narrow form. Most GAI is still at the research state.
What is known as Machine Learning (ML) is a specific type of NAI with the goal of giving the computer access to some stored data and allowing it to learn from it to do classification, but nowhere near GAI. This may involve searching the data for trends, patterns and anomalies or any information that may not be obvious to a human observer.
As I see it, the fastest way for BKV to utilize AI is in the form of NAI and ML. In this area we can implement fault recognition on machines using the vast amount of knowledge from our own experts and from literature with hundreds of research papers and books.
Machine learning (ML) methods for detecting anomalies I see as a strong tool to detect bad data and detect trend anomalies which could replace our current methods of using alarm limits, which are difficult to maintain. The methods are there, we just need to use them.
Our data scientists have used the wind turbine data collected over many years to gain more knowledge for the benefit of our future systems. We have been successful from analysing data using a refinement of traditional methods by recognizing fault frequencies – especially from bearing data. High focus was given to the planetary stages of the wind turbines.
However, not all our data could be used for that purpose. To examine high resolution data, you need time waveforms of long duration. Many of our older time waveforms do not have sufficient duration to produce a reliable diagnosis, owing to restrictions in network bandwidth and hardware. However, with our new DDAU3 monitoring device this has largely been overcome.
The ADVISOR initiative
I’m sure there is currently a lot of research being done on our big data, which may be confidential. However, I believe that ADVISOR was an early attempt to use AI/ML. Can you tell us a little more about this initiative?
We released the first Automatic Machine Diagnosis Software Advisor in 1995, and the last version was released in 2010, where its main feature was an extension to analyse 5000-line spectra provided by Compass 6000.
Advisor was a NAI product based on an automatic pattern recognition method. Descriptor information was automatically extracted from frequency spectra and matched against known patterns for different failure modes. Apart from the diagnostic results, Advisor provided actionable information as well as justification of the diagnostic results. Justification of the diagnostic result is essential in our market area.
I do not think that AI based Automatic Diagnosis Software will ever be as advanced as the cognitive processes of the human brain. However, a system like Advisor is fully justified by the speed of operation where thousands of datasets can be analysed in a very short time to provide condensed information for an operator, thus saving a lot of routine, repetitive work.
The Future of Condition Monitoring in the Wind Industry
This sounds like a technological achievement years ahead of its time. This brings me to my final question; How do you think condition monitoring and diagnostics will change in the future in terms of requirements, technology and service?
I think it will take two directions. One will be the device- based intelligence providing very fast results for the automatic decision-making process, like I mentioned earlier for the wind turbine business.
I think there will be a growing market for on premise solutions providing actionable condition monitoring information directly to the control room via standard interfaces as the availability of local condition monitoring experts become more and more sparse.
Another will be the cloud-based service where requirements to response time is less critical. In the cloud-based approach, data from much simpler device setups (just a time waveform) can be automatically analysed by cloud-based algorithms.
The advantage of a cloud-based system are that changes and improvements to the automatic diagnosis process is much simpler to deploy as such changes are only deployed to a single place compared to the device based intelligence, where fleets of thousands of devices have to be updated.