Early warning

In industry we would rather know when something is about to break so it can be replaced before becoming inconvenient, causing personal damage and incurring expensive plant downtime costs.

With an ever-increasing amount of cash being spent on condition monitoring and predictive maintenance, the techniques themselves are becoming smarter at what they do with all types of sensor data.

It has been known for some time that vibrations in components such as bearings and shafts were a precursor to failure. even before the advent of data collection from a wide range ofsensors, engineers recognised something was more likely to go wrong if theirmachinery started to make unusual clanking or whirring sounds.

Once, finding the solution to these problems relied upon pressing your ear to the casing. But with today's sophisticated sensory monitoring equipment it is possible to examine in intricate detail vibration and sound patterns to give advance indication of something amiss in the works. One of the most advanced of these is

Novus detection software.

One of the key features of Novus is the ability to sort through the huge amounts of data generated by sophisticated sensor arrays to identify the most useful data and use it as information so that maintenance decisions can be taken.

It is important to distinguish between 'data' and 'information.' Data is only useful if it can be applied practically, such as identifying when a plant component is likely to fail.

As Peter Knight, head of engineering software and controls at Alstom said: 'Too many engineers collect data fromall over their equipment and thendon't use it properly. They take data,but never make use of it as useable information.'

This viewpoint was echoed by Manchester-based

business development engineer, Andy Smith. 'A condition monitoring system which reports every piece of diagnostics data available is guilty of committing information overload, thus reducing its effectiveness to almost zero,' he said.

'A reliable and useful system must therefore be engineered with the required end result in mind. Condition monitoring will never be an 'off-the-shelf' product, but always an engineered solution.'

Special attention is given to training Novus to recognise so-called 'normal' behaviour as well as identifying behaviours outside of 'normal' parameters that may be indicators of impending failure. Once a valid plant model is defined, the software can be used to investigate real plant data to look for variations in behaviour. If a potential problem is identified an email can be automatically sent to an engineer so that he or she can take appropriate action.

This system has proved itself in the identification of possible combustion problems on gas turbines. Conventional monitoring of exhaust gas temperature (EGT) isused from the outlet of a turbine to estimate empirically the firingtemperature, which is then used to control the flow of fuel and overheating. Overheating results in a reduction of fuel when some hard 'safe' temperature limit is reached.

In a case where six combustors are employed, individual variancesin combustor behaviour cannot readily be detected until it is too late. Novus is designed to detect subtle changesin the EGT measurements viamultivariate analysis methods, and the use of pattern recognition techniques means that subtle differences inindividual combustor performance can be determined.

The end result is that the development of faults can be identified from abnormal behaviour, even though tighter threshold limits on turbine performance are in place. And as the user has control of parameter thresholds through the training phase, the ratio of 'missed anomalies' to 'false alarms' can be optimised.

Learning is also a key feature of Turkey-based

Machine Condition Monitoring (MCM) and diagnostics system. The system uses a learning period to assess a power spectral density (PSD) frequency spectrum. Any abnormal peaks in the spectrum allied with information relating to the frequency at which they occur, indicates a possible fault developing.

The novel feature of the system is the use of a motor as a transducer without the need for external sensor arrays. Data is produced from the motor's own electrical supply.

This technology has been trialled by a Korean semiconductor manufacturer, where it is claimed to successfully identify potential faults in transmission elements — belts, pulleys and gearboxes — and can make practical recommendations such as checking to see if the belt is loose, worn out or vibrating or checking for eccentric pulleys.

One of the main components prone to failure in engineering plant are the bearings used to transmit forces. In industries such as steelmaking, these are under great stresses due to the movement of heavy crucibles, billets and othersemi-finished products.

The Praxis partnership is a collaboration between

(CNES) and

who have combined their engineering and condition monitoring expertise. This includes vibration monitoring, acoustic emission monitoring, oil analysis, thermographic imaging and training and consultant services.

This collaboration has resulted in a system that encompasses CNES' Aquilla AE Pro acoustic emission fixed monitoring system, that is particularly effective at slow rotational speeds from 80rpm down to 0.25rpm and FAG ProCheck, an online monitoring system that detects plant or machinery faults early, preventing costly breakdowns.

The system, which was developed in partnership with

, helps companies monitor vibration levels on critical rotating plant or machinery, including electric motors, drives, bearing arrangements, gearboxes, pumps, generators, ventilators, fans and excavators. This multiple expertise solution is one way of offering extended predictive maintenance strategies.

It's certain that the more sophisticated machinery becomes, condition monitoring will have to keep up — or even stay one move ahead.

Endress + Hauser's Smith summed it up: 'As advanced onboard diagnostics develop and assets become ever-more intelligent, maintenance staff will become dependent on the high-quality information they produce to determine maintenance schedules.

'This will ultimately reduce the necessity for corrective or planned maintenance, and move into the area of predictive maintenance.'