Distributed Data Filtering and Modeling for Fog and Networked Manufacturing

Abstract

A Fog Manufacturing applies both Fog and Cloud Computing collaboratively and integrates manufacturing equipment and processes into an interconnected network through sensing, actuation, and computation nodes. Therefore, most of the computation services in manufacturing can be provided by using both Fog and Cloud. This is important for future manufacturing, as Fog Manufacturing will provide tremendous promise on reliable and responsive computation services, with the potential for privacy preservation to process data in local computation units. However, due to the relatively limited communication bandwidth to Cloud and computation capabilities of Fog nodes, a large amount of data from the manufacturing network will lead to information redundancy and significant computation time latency for data analysis. In this paper, we propose a distributed data filtering (i.e., subsampling) method to extract small but informative data subsets from the raw data, considering the similarity of the interconnected manufacturing processes. The filtered data from each manufacturing process will later be transmitted to Cloud for manufacturing process modeling. A simulation study and a real Fog Manufacturing testbed for an ingot growth manufacturing have been used to validate the proposed distributed data filtering and modeling method. The results indicate that the proposed distributed data filtering and modeling method will not only reduce the sample size and ensure the modeling performance but also improve the performance of the runtime metrics of the computation, such as time latency of the computation services and the communication bandwidth utilization, in Fog Manufacturing, compared with the centralized Cloud Computing.