AIを用いた機器の予知保全に関する研究

抄録

Predictive maintenance is attracting attention as a low-cost and efficient maintenance method compared to preventive maintenance and after-the-fact maintenance. However, one of the challenges in predictive maintenance is the need to accurately detect and identify equipment anomalies. Therefore, this research aims to perform predictive maintenance with high accuracy using AI. In this paper, from the viewpoint of vibration engineering, the condition of the target equipment is monitored through vibration monitoring to predict the possibility of detecting abnormalities. FRS analysis of the time history acceleration data of the equipment makes it easier to extract the characteristics of the vibration. However, there are many issues to be considered, such as the possibility of erasing important vibration features if the attenuation setting is too large when using FRS, and the relationship between the number of dimensions and accuracy when performing machine learning. Taking them into consideration, whether the machine learning of changes in waveforms due to damping and differences in discrimination results depending on the number of dimensions can accurately detect abnormalities was verified. As a result, the use of FRS with a small amount of damping in the feature extraction confirmed the reduction of extra noise and waveform deviations.