Development of a Real-Time Anomaly Detection System for Dry Vacuum Pumps Using Low-Cost IoT Devices and Machine Learning.

Autor: Kim, Byung-Sub, Ro, Seung-Kook
Zdroj: International Journal of Precision Engineering & Manufacturing; Oct2024, Vol. 25 Issue 10, p2133-2143, 11p
Abstrakt: Data-driven health diagnosis methods based on machine learning have been receiving a lot of attention with the hope that they can open a new era for mechanical researchers and engineers. Machine learning algorithms including deep learning can show insights or state trends through complicated electro-mechanical sensor signals, but it is difficult for machine learning algorithms to go beyond the given data boundaries and to generalize their inferring rules from one case to the other. Researchers have even artificially damaged common mechanical parts in their experiments to collect fault signals for machine learning training. Creating and gathering the sensor signals covering a variety of machine status may be time-consuming or impossible in reality. Anomaly detection for dry vacuum pumps started from the recognition that the artificially induced fault signals may not reflect the natural malfunctioning state of the pumps. Anomaly detection algorithms use only normal data for training, so abnormal state data is not necessarily required. In this research several anomaly detection algorithms were tested and compared with time-domain input features and frequency-domain input features when acceleration signals were injected in for training. A long short-term memory based autoencoder (LSTM-autoencoder) scheme with discrete wavelet transformed (DWT) input signals was chosen for the finalist due to its superior capability capturing the nature of normal state characteristics. The output of LSTM-autoencoder's loss could be used as an indicator of the machine's health deterioration. Furthermore, we successfully demonstrated that the developed algorithm worked well on low-cost Internet of Things devices such as Raspberry Pi 4 and Arduino Mega 2560 boards for real-time monitoring. [ABSTRACT FROM AUTHOR]
Databáze: Complementary Index
Externí odkaz: