Prognostics tools keep machines running smoothly

Sensor-driven measurement in industrial equipment is helping owners to fix potential faults before failures happen.

For Rudie Boshoff, global services director of mining equipment manufacturer Joy Global, automation has transformed the operations of his customers. In the 1990s, a typical mining operation would have had around 500 people working underground. Today, as few as five are needed to supervise underground systems worth hundreds of millions of pounds. And mining operations can be directed from completely different countries. But there is more to do.

“In short, the easy operations have gone and we need to plan, predict, prevent and act faster to improve operational effective­ness in mines,” Boshoff says.

Joy has been spent the last two decades building increasingly sophisticated networked products that rely on a combination of diagnostics and prognostics to optimise the production across entire mines. Prognostics systems use real-time measurements to work out how likely a failure or problem is in the near future and schedule maintenance. They can even guide redesigns.

One of the company’s key systems is designed to excavate some 2000 tonnes of coal an hour from an underground seam hundreds of metres long and take this to the surface on conveyor belts. More than 7000 sensors – including gyroscopes and cameras – link the system of shearers, powered roof supports, face bolters, conveyors, power supplies and pump stations, to provide data on system performance and health. For example, staff may monitor hydraulic pressure across the system, which is crucial to machine operation, alongside other parameters to avoid unscheduled downtime.

Performance alerts can also be sent directly to machine operators, while operating parameters such as speed, cutting height and face alignment of shearers and bolters can be controlled remotely. Joy’s prognostics tools analyse this data to predict possible failures in near real time on motors, drives, clutches and more. At the same time equipment data from any mining operation can be benchmarked against data from the company’s worldwide fleet of assets to highlight process inefficiencies.

Improving design

Boshoff says that in the past this wealth of data has also been used to hone the design of his company’s equipment, from hydraulic integrity to redesigning of relay bars. But as he points out: “this isn’t about having an ‘A-ha moment’ where you analyse a dataset and redesign a component. It’s all about understanding the complete system, optimising it and then focusing on areas to redesign only after everything has been done to maximise operations.”

In the wind-generation industry, prognostics started with the detection of anomalous vibrations within gearboxes.

Adrian Timbus, global head of technology and solutions for wind power and smart grids at ABB, explains: “We have seen many gearbox failures, so manufacturers of these components as well as wind turbine providers needed support to understand this failure.”

Most modern wind turbines are equipped with vibration sensors to provide data for diagnostics and prognostics, but ABB is taking analysis further. For example, the company has adapted a SCADA platform for conventional power generation monitoring and analysis to wind generation. The system monitors wind speed and power output as well as temperatures, currents and voltages across the turbine, and vibrations at the shaft and blades. “Right now, we are monitoring up to 800 signals in any single turbine,” says Timbus.

Data from the SCADA platform can also be fed into ABB’s Asset Health Centre – originally developed for the electricity transmission and distribution industry – to improve performance and avoid outages.

“Data is linked to a condition-based maintenance management system so service personnel can make the correct maintenance steps to maximise production over hours and days,” says Timbus. “But, for example, if a high temperature is detected in a generator, operators can also downgrade the turbine and run it on less power within seconds, to avoid failure.”

Data from operations is beginning to guide design and new installations, too, adds Timbus. “We may start with a specification for a generator bearing at a site,” he says, “but in a second location with different wind conditions, vibrations could be more extreme. So we can use, for example, vibration analysis data and [set] new requirements for the bearing as well as the shaft, rotor and so on. We learn so much about the assets, how everything works in the field, and can improve the design of components so that problems won’t arise in the next installation.”

Shuangwen Sheng, senior engineer from the National Wind Technology Center at the US National Renewable Energy Laboratory, says that to improve the quality of prognostics for wind turbines more reliable and effective monitoring technologies should be developed for main shaft bearings and other low-rotational-speed components, where, as he says, “traditional vibration analysis has challenges”.

The aerospace industry has already taken advantage of prognostics technology for its own airborne turbines. Rolls-Royce now tracks the health of more than 5000 civil aircraft around the world. Andy Mills, researcher in the department of automatic control and systems engineering at the University of Sheffield on work supported by Rolls Royce, points out: “Aerospace products are already inherently safe without using advanced diagnostics and prognostics, so the role of monitoring is to reduce disruption to aircraft operation and optimise maintenance activities.”

To this end Mills and colleagues are exploring algorithms and fault modelling, as well as using Bayesian forecasting to determine the useful life of civil aerospace gas turbine engines.

Typically, pressure, temperature and speed have been measured in an aircraft engine, as this data is deemed essential to safe system control. At the same time, vibration, thermal or even chemical data can be monitored via advanced sensing to provide information on product health and degradation.

“The fusion of these data streams can be used to build individual models of health of each and every engine as it flies around the world, [and combined] to improve maintenance decisions,” Mills adds. “Scalable and distributed computing concepts, such as cloud computing, give us the opportunity to integrate different types of data, including maintenance, repair and measured data, in real time.”

Even before machines go into operation they are being monitored. Prognostics technology is being deployed in factories to make better use of the data from SCADA systems they already have.

In a recent project at a large Cork-based manufacturer of orthopedic products, employing around 900 workers, Dominic O’Sullivan, head of the intelligent efficiency research group at University College Cork, designed with colleagues a system to pool data from its industrial information systems. “This has been all about extracting data from the many different systems at the plant and bringing it to a single repository where we can run analysis and build models,” he says.

Legacy systems

Some systems give up their information more readily than others when it comes to prognostics. “Getting data out of legacy systems is a huge challenge. Today’s manufacturing facilities and technologies will still exist in five-to-ten years’ time, so we have to know how to deal with these systems,” O’Sullivan says. “In an ideal world, it would be great to build a greenfield site using only smart technologies,” he says. “But the reality is we really have to deal with legacy systems, and often it’s about using older protocols to extract data from these systems as well as smarter analytics.”

The researchers incorporated a ‘site manager’ within their industrial data pipeline at the Cork facility. This resides on a cloud server and stores metadata about the facility’s state-of-the-art and legacy data sources, ranging from programmable logic controllers to instrument log files. The metadata is used by software agents running in systems around the factory to capture and pre-process data that then gets fed to the prognostics systems. The researchers now intend to develop additional prognostics for the facility.

“For example we could bring in production data and look at that alongside energy data to determine in real time if production should be ramped up or down, according to environmental conditions and energy prices at the time,” says O’Sullivan. “We’ve been building the foundations for prognostics, but process scheduling is our aspiration here and joining more data from different systems is our next challenge,” he adds.

Although a number of lead customers across a variety of industries have embraced the idea of prognostics, it’s still early days. Boshoff says: “The reality is, I think, that the uptake within the mining industry has been slower than I would have anticipated.

Timbus sees an opportunity for prognostics to improve wind farm efficiency, but the use is currently at the level of individual turbines. “Turbine manufacturers such as Vestas, Siemens and General Electric have developed basic tools and are adding functionality but when we look at the broader level of wind farms, these solutions are not yet fully in place,” he says.

Ken Pipe, managing director of UK prognostics development company Humaware, claims: “An underlying truth across all industries is that prognostics has yet to break through.”

Pipe believes that although projects have been set up in recent years, more extensive government-funded demonstrator programmes are needed to persuade manufacturers and operators to invest more heavily in systems that will ultimately streamline maintenance. “It’s a huge risk to take on board a new technology, so this needs to be subsidised via such a programme to permit the end user to take this risk.”

Timbus adds: “We have so much more data flowing between assets now, and it’s not expensive to have data in turbines, or in any other industry for that matter. For example, sensors are getting cheaper and more and more people are realising the value of what this data can do.”

The challenge is pulling the data from those increasingly cheap sensors into actionable information that companies can use to drive operations. 

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them