Engine crankshaft torsional vibration analysis for anomalies detection

Thumbnail Image

Files

master_Palestini_Cesare_2018.pdf (4.64 MB)

URL

Journal Title

Journal ISSN

Volume Title

Insinööritieteiden korkeakoulu | Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2018-06-11

Department

Major/Subject

Mechanical Engineering

Mcode

ENG25

Degree programme

Master's Programme in Mechanical Engineering (MEC)

Language

en

Pages

61+4

Series

Abstract

Reliability is a key factor in medium speed engines. Since they are employed in power plants and vessels, there is the constant need to increase the accuracy of their anomaly detection loops. Here I investigated a novel idea aimed at decreasing the number of sensors deployed. The goal was to understand whether using only one speed sensor, located at the flywheel, is enough to detect and possibly classify different abnormal operating conditions. A torsional vibration model of a V20 spark-ignited gas engine was developed and validated with field data. The model was then used to simulate normal, heavy knock, misfire and overpressure conditions. Subsequently, the speed signals of the crankshaft, measured at the flywheel, were analysed in the time- and frequency-domains, in order to identify possible patterns that could be linked to the four scenarios. The spectra of the signals were then processed with two machine-learning approaches: pattern recognition and a neural network. Both approaches showed accuracies up to 97% in the classification of the scenarios. Amongst the pattern recognition algorithms, the Ensemble with Subspace kNN proved to be the most accurate one, with a remarkably low rate of false negatives (<1%). The neural network had a slightly lower accuracy in the binary identification between normal and abnormal situations, but it performed better in classifying the different anomalies. The results are promising. They indicate that, with some further improvement, the method tested and developed in this thesis could be a reliable alternative to the safety systems currently in use in engines.

Description

Supervisor

Kuosmanen, Petri

Thesis advisor

Gallici, Irene
Könnö, Juho

Keywords

multi-body dynamics, anomalies detection, data analytics, pattern recognition, artificial neural network, torsional vibration

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-201806293854

Collections

[dipl] Insinööritieteiden korkeakoulu / ENG