DATABASE FOR CLASSIFICATION - MACHINE LEARNING MAIN DIESEL ENGINE USING TORSIONAL VIBRATION

Abstract

This paper proposed a method to build the database for diagnosing the common technical states by the torsional vibrations of the main diesel engine (MDE) using machine-learning toolkit of LabVIEW. The set of technical states includes R classes D_k corresponding with: D₀- All of cylinders are working normally; D_1..z - One of  cylinders is misfiring. The database was constructed to easily apply the analysis machine learning (AML) toolkit for classification and diagnosing. The database was created based on the Design of Experiment (DoE) containing R fundamental experiments. Each basic experiment was totally executed N times including: (i) m repeat times for the noises of diagnosing speed regimes of MDE n_avr(rpm) with dn=±5%; (ii)N_s repeat times for the noises of firing/misfiring states with Cf(i)=±5%. Specifically, Cf(i)=[0.95 - 1.05] when i^th- cylinder is working normally, and Cf(i)=[0-0.05] when this cylinder is misfiring. In the verified case study for MDE 6S46MCC7 installed on MV.HR34000DWT: at n_avr=73(rpm), the selected working speed range is [69-77] (rpm). These speed values satisfy the conditions: far from resonant speed regimes and dn=±5%. The made DoE had N=7.2⁶.9=4032 experiments.  The Torsional Vibration signal (TVs) was calculated using Software for Automatic Torsional Vibration Calculation (SATVC), which was made in VietNam Maritime University. The result database was used for illustrating Classification conditional states by Machine Learning (CML) method for this verified object.