Vibration analysis software will predict machine failures

A pan-European, UK-led project is developing a software tool that could result in faster and more accurate diagnosis of machinery problems. The EU-funded Vision project has brought together eight academic and industrial institutions - led by Worksop-based Monition, and including ABB and Nestlé - to produce an automated system for analysing vibration data to predict when a machine is likely to fail.

Field trials of the system have suggested that it may be more accurate than human analysts and could cut maintenance costs by 5-20%, depending on the installation.

Although vibration analysis systems already exist, they tend to be less reliable than human analysts. They generally rely on rules drawn up by experts based on their know-how. But their accuracy depends on how well the expert has defined the rules and there are always exceptions that defy the rules and defeat the system.

This means that expert overseers are still needed. This is costly because large users may need to analyse more than 20,000 vibration readings each month, and human experts can handle just 200-300 readings a day.

The new system boosts its accuracy by combining the results from three different modules that analyse the data and combine it with case-based reasoning. The diagnostic engine aggregates the results from all three and produces a consensus.

The system is being tested by several multinational companies on a variety of machines. In a series of tests using previously unseen data, the system returned a correct diagnosis 98.4% of the time. By comparison, human experts are typically correct in just 95% of cases.

The Windows-based tool can be accessed remotely via the Internet. A commercial version is expected to reach the market in about a year from now.