A study on the use of machine learning methods to improve reciprocating compressor reliability via torque tailoring

Reciprocating compressors have found popularity in applications where compressed air is required at high pressure levels with moderate flow rates. The mechanical drives used for these compressors are based on the traditional slider-crank linkage which, despite its simplicity, does not lend itself to optimization effort aimed at improving the compressor reliability and performance. The work presented in this paper adopts the notion that the mechanical reliability of the compressor drive is limited by the level and cyclical variability of the loads transmitted through its members and the effectiveness of its cooling system. In the paper, machine learning methods will be employed to craft an objective function suitable to use in a Bayesian optimization effort undertaken to produce a more reliable compressor drive. A numerical example is presented to prove the validity of the presented method and its suitability for use in real life compressor designs.