Establish real-time corrosion map through dual-driven data and knowledge neural network

Abstract

The real-time corrosion map is an effective method for studying atmospheric corrosion. Traditional coupons exposure data are difficult to establish the real-time corrosion maps due to the long experimental period. The Atmospheric Corrosion Monitor (ACM) technology, which provides continuous and real-time data, can be used to support the establishment of real-time corrosion maps. However, the high cost restricts the extensive application of ACM sensor. Additionally, power outages, equipment failures and lifespan limitations are limited factors for ACM sensors to provide sufficient data for unlimited period. It is an unsolved problem to generate a real-time corrosion map based on the ACM data from few locations and limited period. This research work focusses on establishing a new method, the dual-driven data and knowledge neural network (DKNN) model, to establish a real-time corrosion map. Firstly, A DKNN framework that includes multi-objective optimization loss function is constructed by the corrosion data which is amplified by common corrosion knowledge. Next, the DKNN model is evaluated by comparing the consistency between the corrosion electric quantity map and corrosion depth maps from the coupons exposure data. Finally, this paper provides a month-based and a quarter-based corrosion electric quantity maps for different areas in China. This approach could provide a solution to establish real-time corrosion map based on limited ACM data.