Performance prediction and optimization of nanolubricants for hydrodynamic journal bearings: a data-driven approach for regulating volumetric fraction and aggregate particle size

Abstract

This study aims to set up a data-driven framework for optimization and performance prediction of a nanolubricant. Isoviscous approximation was applied to simplify analytical results. FDM methods were used to generate 1204 data points. A multilayer perceptron (MLP) was trained using this large dataset to achieve high testing accuracy (R = 1). The least mean squared error was found at 987 epochs for 23 hidden layer neurons. Multiple statistical tools were used to analyze the results. The error values in the load prediction (up to 7.34%) were observed to be higher than the friction coefficient (up to 0.041%) and side leakage (up to 0.05%). The JAYA algorithm, a parameter-independent machine learning algorithm was used for zone-specific optimization with the zones being low load region (eccentricity ratio: 0.1 to 0.3), medium load region (eccentricity ratio: 0.3–0.6), high load region (eccentricity ratio: 0.6–0.95) and whole range. The optimized combinations for low, medium, high and whole range of operation gave load capacity of 91 N, 357 N, 1328 N and 124 N, respectively. The optimized results highlight the tendency to minimize the eccentricity ratio and maximize volume fraction and aggregate particle size ratio.

Keywords

• Nanoparticle
• aggregation
• FDM
• hydrodynamic journal bearing
• ANN
• optimization