Fusion of heterogeneous industrial data using polygon generation & deep learning

Abstract

Analysis of industrial data imposes several challenges. These data are acquired from heterogeneous sources such as sensors, cameras, IoT, etc, and are stored in different structures and formats with different sampling frequencies. They are also stored in isolated silos in different locations which hinders their exploitation. Therefore, there is a clear need to integrate these disconnected data silos at different processing levels and make them clean, easily accessible, and fully exploitable. This paper proposes a data fusion method that merges heterogeneous sources of data at raw, information, and decision levels using polygon generation and deep learning (DL) techniques. An innovative polygon generation technique is proposed to preprocess each data source and convert it into powerful representations that capture all possible relationships in the data, thus extracting the maximum knowledge and achieving better prediction accuracy of the corresponding DL method. The proposed method is targeting challenging data modeling problems found in industrial processes. It is validated successfully using a case study in the realm of process system engineering. The results obtained demonstrate that the proposed fusion method is more accurate, with a minimum of 20% improvement, compared to other methods previously used in the literature.

Keywords

• Data fusion
• Decision fusion
• Deep learning
• Polygon generation
• Energy efficiency
• Process system engineering