Intelligent Predictive Maintenance For Sustainable Industrial IOT

Abstract

Predictive maintenance has emerged as a promising technique for condition monitoring, newlinefault diagnosis, and minimizing risks and catastrophes in Industry 4.0. In this newlineparadigm, the industrial Internet of Things (IoT) framework is used to collect both the newlineprocess parameters of systems (e.g. coal-fired boiler) and the vibration-based health newlinedata of associated machines (e.g., three-phase induction motor/TIM). These data are newlinethen evaluated for intelligent predictive maintenance using machine learning (ML)/deep newlinelearning (DL) approaches. Although significant progress is observed in machine condition newlinemonitoring, predictive maintenance based on process data is still limited. Furthermore, newlinethe success of data-driven predictive maintenance depends on sustainability newlineof the remotely deployed IoT hardware and sensor s performance, which is still a challenging newlinetask. It is of great importance to monitor the health of the IoT sensor node s newline(SN) components particularly, the rechargeable batteries and low-cost sensors as they newlineoften suffer from faults. Moreover, both the machine and the associated sensor may get newlinefaulty simultaneously; hence, isolating the machine fault in the presence of the sensor newlinefault is also important. Therefore, the aim of the present dissertation is to develop a newlineholistic predictive maintenance framework employing both industrial process data and newlineaccompanying machine health data collected by a self-sustaining IoT platform. newlineAt first, the dissertation proposes an intelligent fault diagnosis technique with process newlinedata by taking clinkering fault in a coal-fired boiler as a case study. For this purpose, newlinea suitable monitoring indicator is extracted using a stacked denoising autoencoder newlinenetwork. The results exhibit online clinkering detection and prognosis with 99.8% accuracy newlineand 0.0103 RMSE, respectively. Subsequently, to collect health data of associated newlinemachines, a self-sustainable IoT sensor node with onboard battery health selfmonitoring newlineand sensor fault detection and calibration capabilities is presented. The newlinepr