Predicting precisely when a wind turbine component will fail is the holy grail of condition monitoring (CM). That information would let maintenance crews aggregate the needed heavy work and thereby minimize its costs. Although maintenance predictions are good now, there is room for improvements. To appreciate the state of the art, we spoke with John Coultate, Head of Monitoring and O&M Consultancy, Romax Technology Ltd., a UK and Boulder, Colorado based service consultancy. The company develops hardware and software to complement its turbine diagnostic services.
One instrument the company developed, a data logger for wind turbines, can be set in a nacelle to collect data for a few hours or a few months, and gives analysts an idea of the turbines operating condition. “It’s useful because many turbines don’t have CM systems installed, but operators still want to know the turbine’s health,” says Coultate. A team of two people can collect data on about two turbines per day depending on weather.
The vibration sensors used by the data logger attach by magnetic bases, simple and noninvasive devices that allow putting them on the gearboxes, main bearings, generator bearings, and other places. “We also need the speed and power of the turbine so we use a simple speed sensor, a proximity device. Electric-current sensors fit around the generator cables for a power reading. Electrical current and rotor torque shaft track pretty closely,” he says.
While running, data can be stored on a memory stick for short measurement periods or sent by 3G network to analysts almost anywhere who can report on the turbines.
A good application for the data logger might be an end-of-warranty inspection on a site without CMS, or no access to the CMS data. Coultate says it is a good way to get a snapshot of a turbine’s health. “As long as you get good quality vibration data, you only need a short period of it. We are looking for high-quality data, and that comes from running the turbine at near rated power in a steady state.”
Coultate cites a recent application that called for measuring torque and bending on a turbine’s main shaft. “We installed strain gages on the main shaft to characterize loads during events, such as starts, steady-state running, and stops. This example, in North America and on a 1-MW turbine, was left running on batteries for four months. But we could log into that system from here in the UK to monitor the wind farm,” he says. The analysis is just under way so its lessons are still being learned. However, the loads monitored on other projects surprise him. “We see shock loads during transient events, such as start-ups and stops. Torque occasionally spikes beyond expected limits. Shock loads can have impacts on gears and bearings leading to gearbox damage,” he adds.
Recent projects deal with building a model that predicts the remaining useful life in components. “We have been working at least 10 years on this model of remaining-useful life and how it might help an operator. But computer models cannot mimic everything that happens to a turbine. There are so many ways a drivetrain can fail. A bolt might back out of a connection, oil degrades, and gears might be incorrectly manufactured, or fail due to inclusions in the material. You cannot capture all of these possible failure modes in a computer model, so you have to use empirical data from the site and lots of domain experience from working with this type of machinery,” he says.
Still, he says the company has developed predictive models that simulate elements of a drivetrain and make it work well when combined with data from a site, such as maintenance and inspection reports. The effort focuses on getting longer lead times before the need for maintenance. “Vibration CMS is powerful in the short term, but looking more than three to six months into the future is incredibly powerful.”
The company is working on predictive models with longer term forecasting. With a long enough prediction, and with other turbines with similar problems on the same wind farm, the owner can schedule maintenance for the same crane call out. “Doing work on two or more turbines at the same time is more cost effective. We calculated this method can save about $300,000 on just downtime and crane costs, and by being able to run the turbine during the windy season. So the value of a predictive model is huge,” says Coultate.
Some of today’s hot topics, he says, are main and planet-bearing failures and their root causes. Planet bearings are interesting because they have been difficult to detect. Their vibration signals are complex and changing with the planet-stage rotation. “However, we have developed good signal-processing methods. Recently, for instance, we accurately predicted a three-month lead time before the operator changed out a bearing,” he says.
Oil condition is also getting more attention. Gearboxes create particles and debris that can produce significant damage. “Oil is the lifeblood of the gearbox and one with lots of debris can produce indentations on bearing surfaces that eventually develop into spalls (surface-initiated pitting). Surfaces with an indentation are like a hole in the road hit over and over again by passing cars. After a time, the material cracks away. The surface flaw can sometimes develop and spread quite quickly.” It differs from normal classical fatigue failure which initiates subsurface at peak stresses.
Particles counters, rare in turbines, would provide useful information when it comes with vibration data. Because of the rarity of such counters, laboratory analysis of the oil is useful every three to six month.
Coultate says oil reports are used two ways. In one, the data goes into software for CM that allows tracking and trending the oil-analysis results over time. In the other, it goes into a predictive-life model. Real data in a predictive model captures a lot of operational and physical events.
Wind-farm operators have two options for condition monitoring. They can monitor the site themselves when they employ trained analysts, or they can outsource the task. “Many operators outsource their CM to our company. We monitor over 40% of the UK offshore fleet. We are also doing routine and non-routine monitoring at other sites in North America, Canada, Latin America, and Asia.
One challenge with monitoring many sites is trying to automate processes and alarms. “We have developed sophisticated signal-processing methods for detecting alarm limits, all to find a health index that would say a turbine is well or not. From that we developed software called Insight, a platform for predictive maintenance. Part of the platform is for analyzing vibration data so we can scale the monitoring service and license the software to other operators.” Long term, he expects more operators to transition to self-monitoring, with Romax providing the necessary software tools and training.
Hinting at what might be of interest to U.S. crews, Coultate says the company recently completed the world’s largest offshore end-of-warranty inspection on 88 turbines in the North Sea. Those engineers, trained in offshore survival, successfully completed the project before the deadline.
The inspections took several months because getting to an offshore site is challenging considering bad weather days. Even in good weather, it takes time getting equipment ready, onto and off the boat, and working days can be quite short. The task took multiple teams of people on rotation over a period of several months.
Inspections focused on the drivetrain but crews were responsible for the whole turbine, including blade inspection. “For drivetrain inspections, we used a borescope to inspect gears and bearings. Vibration data told of which things to target during the inspections. When inspecting a gearbox, some faults are easily detected using vibration but hard to find by inspection, and the other way around too. So the two, inspection and vibration, complement each other well.
Good news, he reports: offshore turbines four to five years old are generally running well with high power outputs. What’s more, all new offshore turbines have CM installed as standard equipment.
In the future, Coultate expects many operators will take more responsibility for condition monitoring and will look for a long-term partner to assist in the transitions. He also sees better integration of data. “Today we use vibration CMS alongside SCADA analysis, and longtime predictive life models. The industry will move in that direction,” he suggests. WPE
Filed Under: News