Keywords: journal bearing, fault diagnosis, artificial neural networks, wear, misalignment, Reynolds equation.

Both detection and monitoring of wear are rather important in tribological research as well as in industrial applications. Some typical examples are: measurement of dynamics of wear processes, engineering surface inspection, coating failure detection, tool wear monitoring and so on. As a result of the dynamic and complex nature of wear process, measurement of wear is usually conducted offline and online detection and monitoring of wear remain challenges for tribological research.

There are many causes that may generate misalignment such as deflection of a rotor under its own weight, geometrical defect of alignment, applied torque on the journal bearing, deflection of the rotor, operating faults and malfunctions such as insufficient lubrication in one or more bearings in a rotor bearing system. Misalignment may be a cause of bearing wear and all machines are always subjected to misalignment conditions.

In this paper, focus is given to bearing misalignment and wear. The basic bearing characteristics, namely the eccentricity, the attitude angle the minimum film thickness are calculated as a function of several wear depths and misalignment angles using a finite element method. An artificial neural network is then submitted to a training process, which eventually provides an ANN-structure capable of identifying the progress of each defect. The calculation of the above parameters is performed with high accuracy and speed so as to be considered as an efficient online monitoring tool for the specific parameters.

The proposed method contributes to the identification of the defects of the rotor bearing systems such as misalignment and wear. The utilization of artificial neural networks provides a very effective tool for on-line monitoring of wear and misalignment due to the fast and accurate calculation. Moreover, when compared to conventional retrieval or heuristic methods, it provides the capability for interpolation among values of existing records or even extrapolation out of the limits that are defined by the elapsed records. Future work will be undertaken to train a dynamic - instead of a static - artificial neural network with values extracted from real systems with actual measurements aimed at investigating the value variation sequences in the proposed theoretic model.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £145 +P&P)