Case study: Validation of vibration data augmented finite element analysis
Kuitunen, Jarno (2024)
2024
Konetekniikan DI-ohjelma - Master's Programme in Mechanical Engineering
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. Only for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-03-15
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202402072192
https://urn.fi/URN:NBN:fi:tuni-202402072192
Tiivistelmä
The aim of this thesis was to study the accuracy of simulated stress results obtained with hybrid modelling. The research was conducted as a case study of a Wärtsilä 18V50SG engine-generator set. Stress data was collected from the running engine-generator set for use as comparison data to the hybrid model stress results. Also, comparison data was collected by simulating the stresses of the engine-generator set during operation by carrying out an excitation stress response simulation using time-amplitude excitations. This was done using the Abaqus finite element analysis software.
In addition to collecting comparison data, operational deflection shape measurements were carried out for the engine part of the engine-generator set. The vibration measurement data acquired from the measurements were used in hybrid modelling to compute measurement data expansions. To obtain as accurate results as possible, the work aimed to find an optimal mode shape selection for the hybrid model.
Stress results obtained using the optimal set of mode shapes were then compared to the measured stresses as well as stresses simulated using time-amplitude excitations. The conducted comparison revealed that hybrid modelling can yield results which are very comparable to the actual behaviour of the structure. However, the stress results obtained with hybrid modelling are very error sensitive to irregularities in the simulation model as well as mode shape selection. Nonetheless, the acquired stress results gave a good depiction of the stresses in studied locations of the structure.
The most notable error source which made it difficult to get accurate results was the inaccuracy of the simulation model. At lowest engine running orders, compared to measured stresses there were considerably higher responses observed in both hybrid modelling results and time-amplitude excitation simulation results. These unexpectedly high stress amplitudes were caused by highly localized deformations which were produced by too high contributions of a few erroneous modes. This was presumably due to differences in the base frame support chute weld joints between the simulation model and the real structure. The problem of the erroneous mode shapes created was more pronounced in the hybrid modelling results, because there was no vibration measurement data in close proximity to the studied strain gage locations. Also, the problem was exacerbated in the hybrid model as the number of mode shapes was considerably smaller than what was used in the time-amplitude excitation simulation.
The inclusion of residual modes to the measurement data expansion proved challenging. No such residual mode selections which would improve results at all the studied engine running orders were found.
Further research would be useful to make the hybrid modelling method more reliable. Because the number of measured degrees of freedom restricts the number of modes that can be used for a measurement data expansion, the accuracy of obtained results could possibly be improved by adding more vibration measurement data points.
In addition to collecting comparison data, operational deflection shape measurements were carried out for the engine part of the engine-generator set. The vibration measurement data acquired from the measurements were used in hybrid modelling to compute measurement data expansions. To obtain as accurate results as possible, the work aimed to find an optimal mode shape selection for the hybrid model.
Stress results obtained using the optimal set of mode shapes were then compared to the measured stresses as well as stresses simulated using time-amplitude excitations. The conducted comparison revealed that hybrid modelling can yield results which are very comparable to the actual behaviour of the structure. However, the stress results obtained with hybrid modelling are very error sensitive to irregularities in the simulation model as well as mode shape selection. Nonetheless, the acquired stress results gave a good depiction of the stresses in studied locations of the structure.
The most notable error source which made it difficult to get accurate results was the inaccuracy of the simulation model. At lowest engine running orders, compared to measured stresses there were considerably higher responses observed in both hybrid modelling results and time-amplitude excitation simulation results. These unexpectedly high stress amplitudes were caused by highly localized deformations which were produced by too high contributions of a few erroneous modes. This was presumably due to differences in the base frame support chute weld joints between the simulation model and the real structure. The problem of the erroneous mode shapes created was more pronounced in the hybrid modelling results, because there was no vibration measurement data in close proximity to the studied strain gage locations. Also, the problem was exacerbated in the hybrid model as the number of mode shapes was considerably smaller than what was used in the time-amplitude excitation simulation.
The inclusion of residual modes to the measurement data expansion proved challenging. No such residual mode selections which would improve results at all the studied engine running orders were found.
Further research would be useful to make the hybrid modelling method more reliable. Because the number of measured degrees of freedom restricts the number of modes that can be used for a measurement data expansion, the accuracy of obtained results could possibly be improved by adding more vibration measurement data points.