Development of Digital Twin for Reciprocating Compressor Using Machine Learning Methodic

Reciprocating compressors are used to compress gas in several common processes in industry. Good prediction and control of the compressor interstage temperatures and pressures are of great concern when operating near their limits. These limits can be encountered when the unit production rate is being maximized and cooling limitations are reached. In addition to operating limits, there are several mechanical constraints in the compressor. Prediction of these constraints is critical for safe operating of the process units. Our research is dedicated to the development and investigation of digital twin asset for reciprocating (or also called positive displacement) type of compressor machines with use of advanced pattern recognition (APR) technique which is one of machine learning methodic. Developed digital twin consists of set of models which monitor and optimize energy consumption, performance, and health of physical reciprocating compressor asset.