How to diagnose the wear of rolling element bearings based on indirect condition monitoring methods

In maintenance it is of greatest importance to know what should be done and when. With condition monitoring it is possible to reduce the number of unplanned stoppages, which cost a lot of money compared to planned maintenance actions. Condition monitoring of rotating machinery, i.e. detection of wear of the components of the machinery, is usually based on indirect methods or monitoring because it is very difficult to monitor or measure wear as such in practise. The reason for this is simply that no such practical methods exist that could be used for measuring the wear of such machinery components as bearings or gears or impellers etc. because these are hidden behind supporting structure or covers. Today the diagnosis of the needed maintenance actions does not usually give a prediction of how much time there is left for maintenance prior the component in question will break. The paper tries to tackle this question in case of rotating machinery and especially in case of rolling bearings. The ultimate goal is to be able to give prognosis of how much time there is before the component will suffer catastrophic failure. The paper starts with a discussion of the wear of rolling element bearings. How does it start, how does it proceed and how does it increase towards the end of the life of the components? The link between the indirect monitoring methods such as oil analysis techniques, vibration measurements and measurement of acoustic emission is covered into some extent. The developed approach starts from the idea of modelling the wear of the component. In case of rotating machinery components the wear often takes place progressively. The reason for this is that when a fault is initiated it increases with increasing speed because the loads that are the cause of wear increase as a function of the size of the fault. In the approach a limited number of condition monitoring parameters are used for diagnosis of the fault. These parameters are then used as input in higher order polynomial regression functions with a limited number of terms. The purpose of using higher order polynomial regression functions is to be able to mimic the development of the fault and also to be able to save the history, i.e. the trend of the development of these parameters, in a very compact form. The regression functions can give prognosis of the development of the fault. The severity of the situation is analyzed using simplified fuzzy logic. A number of measured and analyzed examples are given. All the examples concern rolling bearings, which are probably the most widely monitored component of rotating machinery in the industry. In the tested cases the bearing fault can be diagnosed when about three or four percent of the lifetime of the bearing still remains.