Neural Networks as a Tool to Characterise Oil State After Porous Bearings Prolonged Tests

In this article we use artificial neural networks (ANN) in durability analysis of porous bearings. First, we present briefly the results of durability tests of porous sleeves impregnated with synthetic ester oil under different duration of the tests (100, 500, and 1000 h) and bearing temperature (60, 80, and 130 °C) at a rotational speed of 1000 rpm. After each durability test oil samples were removed from the bearings and some chosen parameters were checked (Fourier Transform Infrared spectra and total acid number). In the second stage, the collected data were used in the design of ANN, i.e. work parameters as the inputs and oil properties as the outputs. The tests of various ANN types were performed to achieve the smallest training error and the best performance. The best parameters were achieved for multilayer perceptrons neural networks, and also quite good prediction of oil parameters after the test was observed. The achieved results, i.e. the ANN designed, algorithms, and oil parameters used, were compared with those observed in the previous tests for mineral oil.

Polymer bearings are widely used due to dry-lubrication mechanism, low weight, corrosion resistance and free maintenance. They are applied in different tribological pairs, i.e. household appliances, mechatronics systems, medical devices, food machines and many more. However their use is limited by high coefficient of thermal expansion and softening at elevated temperature, especially when working outside recommended pv factors. The modification of bearing design to achieve better characteristics at more demanding conditions, requires full understanding of mechanical and thermal phenomena of bearing work. The first step was to observe a thermal behavior of polymer bearing under real test conditions (50, 100, 150 rpm and 350 and 700N) until constant values were achieved, i.e. temperature and moment of friction. Subsequently collected data were used in a design of temperature distribution model. Thermal simulations of the polymer bearing were done using commercial software package ANSYS Fluent, which is based on finite volume method. All calculations were performed for 3D geometrical model that included polymer bearing, its housing, shaft and some volume of the surrounding air. The heat generation caused by friction forces was implemented by volumetric heat source. All three main heat transfer mechanism (conduction, convection and radiation) were included in heat transfer calculations and the air flow around the bearing and adjacent parts was directly solved. The unknown parameters of the numerical model were adjusted by comparison of the results from computer calculations with the measured temperature rise. In the presented work the calculations were limited to steady state conditions only, but the model may be also used in transient analysis. DOI: http://dx.doi.org/10.5755/j01.ms.21.3.7342