Abstract:
Rotating machines play a strategic role in a manufacturing process; it is the case of a cement mill. These machines are made of fragile organs (bearings and gears, etc.) subjected to mechanical stress and harsh industrial environments. There are multiple causes of deterioration of a machine element: normal wear and tear, overload (or under load), poor lubrication, mounting problems, etc… Depending on the extent of degradation, the surfaces in contact present spalling a more or less important.
In general, the wear can be seen as associating failure mechanisms (shear joints, fatigue ...) to interactive phenomena such as thermal effects, volume phenomena (plastic deformation, phase change, diffusion) and naturally surface effects (reaction, adsorption, segregation ...).
In this work the one hand, a study of the characterization of the bearing wear (QJ1244N2MA and SKF 22248 CC / N1W33C3) of the gear unit DMGH 25.4 of a horizontal cement mill was made.
On the other hand, the wear can be seen as a loss of functionality of a system, which affects the image of the components of the vibratory system. Indeed, this change of image vibration will be used as an indicator of defects that will make a diagnosis of an industrial system.
However, many currently available techniques require considerable expertise for successful implementation: it requires new techniques that allow relatively unskilled operators to make reliable decisions without knowing the mechanism of the system and analyze the data. The artificial neural networks (ANN) are suitable for this kind of problem.
It is within this context that our investigations. On the one hand a first study is dedicated to an overview of more detailed knowledge about the wear and tools to understand this phenomenon,especially the concept of third body as the medium at the interface between two bodies in contact, in which wear can be considered as a complex competition between the phenomena of detachment of particles of the contact surfaces and final ejection of these particles out of contact.
On the other hand, a diagnostic approach to implementing multilayer artificial neural networks from measurements performed on a horizontal gear DMGH 25.4.
To study the performance of neural networks vis-a-vis the problems of system diagnostic gear and bearings, cases of failures have been minimized, two types of defects are considered: wear the ring and outside QJ1244 breaking of a tooth of the intermediate gear.
