Investigating hermetic reciprocating compressor performance by using various machine learning methods


Bacak A., Çolak A. B., DALKILIÇ A. S.

Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, cilt.238, sa.11, ss.5369-5384, 2024 (SCI-Expanded) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 238 Sayı: 11
  • Basım Tarihi: 2024
  • Doi Numarası: 10.1177/09544062231213276
  • Dergi Adı: Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Applied Science & Technology Source, Communication Abstracts, Compendex, INSPEC, Metadex, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.5369-5384
  • Anahtar Kelimeler: gradient boosted, Hermetic reciprocating compressor, machine learning, PNN, polynomial regression, random forest, simple regression
  • Yıldız Teknik Üniversitesi Adresli: Evet

Özet

Due to their durability and efficiency, hermetic reciprocating compressors (HRCs) are used in refrigeration and air conditioning. Compressor performance and reliability concerns reduce system efficiency and raise maintenance costs. Machine learning (ML) is being used to improve hermetic reciprocating compressor performance, reliability, and energy economy. ML is used in hermetic reciprocating compressors for issue identification, performance improvement, predictive maintenance, and energy management. This research compared HRC performance factors such as mass flow rate, cooling capacity, compression power, coefficient of performance, exhaust line losses, and volumetric efficiency. Simple regression, probabilistic neural network, gradient boosted, polynomial regression, and random forest (RF) were used to examine and evaluate these parameters as outputs. Over three cycles, the Fluid-Structure Interaction (FSI) approach assessed compressor performance parameters. For compressor speeds of 1300, 2100, and 3000 rpm, mass flow rate, compression power, cooling efficiency coefficient, and exhaust line energy losses varied by 10%, 4%, 5%, and 6%. To gather ML algorithm inputs, the research used experimental, fluid-structure interaction, and ML methodologies. Experimental and FSI approaches produced 108 data points. These data points were randomly assigned, with 70% for learning and 30% for prediction. The mean convergence criterion for mass flow rate, cooling capacity, compression power, cooling efficiency coefficient, exhaust line energy losses, and volumetric efficiency parameters was 0.9966, 0.9969, 0.9572, 0.0561, 0.9925, and 0.4640 for all ML methods. Simple regression, probabilistic neural networks, gradient boosted, polynomial regression, and RF convergence criteria were 0.8978, 0.9999, 0.6016, 0.4439, and 0.7761.