Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Prediction of Volumetric and Adiabatic Performance of Screw Compressors Using Gaussian Process Regression

Jonsson, Ivar LU (2025) In Master's Theses in Mathematical Sciences FMSM01 20251
Mathematical Statistics
Abstract
Compressed air systems account for roughly 10 % of industrial electricity use, suggesting that even
marginal gains in compressor efficiency yield substantial cost and emission savings. Although screw
compressors are valued for their compact and reliable design, the process of optimizing the performance of a machine for a specific application is typically time-consuming due to manual testing under
numerous operating conditions. This thesis is dedicated to the statistical modeling of volumetric and
adiabatic efficiency measures under different measurement circumstances. Due to the compressor’s
non-linear response with respect to input parameters as well as the need to build a model using
only a few data points, this thesis focuses on... (More)
Compressed air systems account for roughly 10 % of industrial electricity use, suggesting that even
marginal gains in compressor efficiency yield substantial cost and emission savings. Although screw
compressors are valued for their compact and reliable design, the process of optimizing the performance of a machine for a specific application is typically time-consuming due to manual testing under
numerous operating conditions. This thesis is dedicated to the statistical modeling of volumetric and
adiabatic efficiency measures under different measurement circumstances. Due to the compressor’s
non-linear response with respect to input parameters as well as the need to build a model using
only a few data points, this thesis focuses on Gaussian processes. The data used for the thesis are
real test data from machines at Svenska Rotor Maskiner as well as simulated data. The real data
consist of 290 test runs, covering pressure ratios, speeds, inlet temperatures, and oil flows. Two
simulated datasets are generated by a multi-chamber model to examine the behavior of the GPR
method compared to real data. Firstly, the kernel choice is investigated, including the Radial Basis
Function, Matérn and the Rational Quadratic kernel. The RBF + White-noise kernel is selected for
its combination of promising performance measures and simplicity. On real test data, GPR achieves
significantly better prediction metrics for all evaluated kernels in comparison to linear models, suggesting it is able to capture the non-linear behavior of the machine. Furthermore, experimentation
on sample-efficiency simulations indicates that the GP model shows strong predictive ability for as
few as 120 data samples. Simulated data experiments support the flexibility of GPR. The robustness
of the model in terms of input noise is investigated, suggesting the resilience against uncertainties in
the measurements. The simulated data suggest irregularities in variable importance. The findings
support that GPR can substantially reduce experimental testing while still offering solid uncertainty
estimates. This allows for reduction of both the development costs and energy use, helping drive the
development towards significant emission savings. (Less)
Please use this url to cite or link to this publication:
author
Jonsson, Ivar LU
supervisor
organization
course
FMSM01 20251
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Screw compressors, Gaussian Process Regression (GPR), Volumetric efficiency, Adiabatic efficiency, Radial Basis Function (RBF) kernel
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMS-3542-2025
ISSN
1404-6342
other publication id
2025:E98
language
English
id
9211349
date added to LUP
2025-09-01 14:44:25
date last changed
2025-10-02 17:21:09
@misc{9211349,
  abstract     = {{Compressed air systems account for roughly 10 % of industrial electricity use, suggesting that even
marginal gains in compressor efficiency yield substantial cost and emission savings. Although screw
compressors are valued for their compact and reliable design, the process of optimizing the performance of a machine for a specific application is typically time-consuming due to manual testing under
numerous operating conditions. This thesis is dedicated to the statistical modeling of volumetric and
adiabatic efficiency measures under different measurement circumstances. Due to the compressor’s
non-linear response with respect to input parameters as well as the need to build a model using
only a few data points, this thesis focuses on Gaussian processes. The data used for the thesis are
real test data from machines at Svenska Rotor Maskiner as well as simulated data. The real data
consist of 290 test runs, covering pressure ratios, speeds, inlet temperatures, and oil flows. Two
simulated datasets are generated by a multi-chamber model to examine the behavior of the GPR
method compared to real data. Firstly, the kernel choice is investigated, including the Radial Basis
Function, Matérn and the Rational Quadratic kernel. The RBF + White-noise kernel is selected for
its combination of promising performance measures and simplicity. On real test data, GPR achieves
significantly better prediction metrics for all evaluated kernels in comparison to linear models, suggesting it is able to capture the non-linear behavior of the machine. Furthermore, experimentation
on sample-efficiency simulations indicates that the GP model shows strong predictive ability for as
few as 120 data samples. Simulated data experiments support the flexibility of GPR. The robustness
of the model in terms of input noise is investigated, suggesting the resilience against uncertainties in
the measurements. The simulated data suggest irregularities in variable importance. The findings
support that GPR can substantially reduce experimental testing while still offering solid uncertainty
estimates. This allows for reduction of both the development costs and energy use, helping drive the
development towards significant emission savings.}},
  author       = {{Jonsson, Ivar}},
  issn         = {{1404-6342}},
  language     = {{eng}},
  note         = {{Student Paper}},
  series       = {{Master's Theses in Mathematical Sciences}},
  title        = {{Prediction of Volumetric and Adiabatic Performance of Screw Compressors Using Gaussian Process Regression}},
  year         = {{2025}},
}