Validation of Simulated Mechanical Vibration Data for Operational State Recognition System

Accurate real-time models for estimating the current state of an engine-generator set (genset) can be built based on measured mechanical vibration data. There is typically a significant disparity between the amount of data measured during normal and abnormal operation of a genset. Sufficient measured data to build a model for detecting and recognizing abnormal operation is rarely available. The lack of data measured during abnormal operation can be compensated, e.g., by creating more data through simulations. The focus of this study is on producing more realistic simulated vibration data by adding variations in the excitation forces used as input for the simulation. The added variations are based on the integration of previously measured cylinder pressure and rotational speed data from similar gensets as the simulated one. The effect of using varying input data on the simulated vibration responses of a genset is studied by extracting features and training operational state classifier models based on them. The extracted features and the classifier model results are compared with respect to the measured mechanical vibration data from a similar genset as the simulated one. The results show that the simulated responses resemble the measured ones. However, the comparative validation results reveal significant differences between the simulated and measured responses. Thus, further investigation and development is needed regarding production of the simulated mechanical vibration data.