機械学習を用いた配管ヘルスモニタリングシステムの構築

抄録

Piping is an important mechanical member in factories and power or chemical plants. The soundness of pipes must be regularly inspected to prevent accidents such as ruptures due to reduced wall thickness. However, in large-scale plants, the total length of pipes extends over several hundred kilometres, which requires significant costs for the inspection. In recent years, there has been a shortage of engineers and problems with the succession of technology, so labour-saving inspections are required for the sustainable maintenance of the industry. In this study, a method for evaluating the soundness of piping is developed. In this method, vibration waveforms measured by an accelerometer are analysed in frequency and their soundness is evaluated by machine learning. In this paper, the vibrations caused by water flow in pipes with/without failure were measured and the wavelet-transformed data were processed by AI to assess the soundness of the pipes. As a result, it was confirmed that the developed method using the machine learning can detect failures in pipes and the is effective as an health monitoring method for pipes.