LOPdM: A Low-Power On-Device Predictive Maintenance System Based on Self-Powered Sensing and TinyML
Zijie Chen, Yiming Gao, Junrui Liang
Abstract
Predictive maintenance (PdM) has emerged as a prominent strategy that can recognize the current state and predict the future trend of machines. It helps prevent disastrous breakdowns. Such systems were mostly realized based on artificial intelligence (AI) models that run on resource-rich and power-hungry servers. To meet the ultralow-power, low-cost, and on-device-inferring demands, in this paper, we introduce a self-contained low-power on-device predictive maintenance (LOPdM) system based on the cutting-edge self-powered sensor (SPS) and tiny machine learning (TinyML) techniques. A rich dataset is collected with an SPS in a simulated vibration environment. The collected data is analyzed using six established AI models. Under an ultra-short data length, small data number, and low sampling rate condition, the random forest (RF) and the deep neural network (DNN) stand out with up to 99% precision. The trained model is then deployed on an embedded system for in-situ inferring and condition-based PdM. Power measurement is carefully conducted to compare the power consumption using an inertial measurement unit (IMU) or an SPS, respectively. It shows that the SPS-based system can save up to 66.8% of energy. An all-in-one prototype is assembled and utilized in field tests. It makes a high accuracy in malfunctions identification. As an interdisciplinary study, the development of LOPdM provides valuable guidance for future ubiquitous AI applications.