A Computational Framework for Cloud-based Machine Prognosis

Prognosis of machine degradation and failure propagation is essential to preventative maintenance scheduling and sustainable manufacturing. Emerging technologies such as Internet of Things (IoT) and cloud computing offer new opportunities for scaling up computing performance and capacity for machine monitoring and prognosis. This paper addresses challenges in machine prognosis due to high-speed data streaming from real-time sensing by leveraging parallel computing on the cloud. A framework for cloud-based prognosis is then presented to model the relationships between hidden machine states and sensor measurements under varying operating conditions and maintenance actions. To account for uncertainties associated with model representation and/or measurement quality, each relationship is modeled as a probability distribution and estimated through either model-based (e.g. particle filtering) or data-driven algorithms (e.g. support vector machine), according to the available physical/mathematical description of the relationship. A complete prognostic model of the machine is then constructed by merging the individual probability distributions. The computational process is implemented on the MapReduce-based cloud computing platform. Prognosis of the entire machine is accomplished by aggregating prognosis results of the individual components, through a separate parallel computing process. The proposed framework is experimentally demonstrated using tool data collected from CNC machines.