Predictive maintenance requires as much data as possible to predict when a particular part or device will need to be serviced or replaced. The data base for durable products in particular is however insufficient for making statements with an acceptably low probability of error.

As an article in the online magazine Fraunhofer InnoVations explains, researchers at the Fraunhofer Institute for Factory Operation and Automation IFF have thus begun to supplement the usual mathematical methods such as fuzzy models with the experience of employees who, for example, operate a machine. To do so, they question the workers about the conditions at the production site prior to the failure of the machine and include the results in the equation as additional factors.

However, the information provided by employees consists of qualitative classifications such as "good", "average" or "bad" rather than exact measurements. Examples include changes in the process flow or a specific noise that was heard in the run-up to the failure. The researchers say they were nevertheless able to significantly improve the predictive quality by including these factors. The method is apparently applicable to a wide variety of industrial components.