Spatio-Temporal Feature Attention Fusion for Fault Detection in Complex Industrial Processes

In the practice of maintaining industrial production safety, fault detection in industrial processes plays a crucial role. The data generated during industrial production is influenced by numerous operating conditions, resulting in process data often exhibiting spatiotemporal complexity. Existing fault detection methods face significant challenges in effectively handling these characteristics simultaneously. To address this, this paper proposes a Spatio-temporal Feature Attention Fusion (ST-FAF) framework designed to extract both temporal and spatial features from time-series data. Specifically, the paper utilizes a one-dimensional convolutional neural network to extract spatial features from the raw time-series data, while employing the Gramian Angular Summation Field (GASF) method to convert time-series data into two-dimensional images for better capture of temporal features. Furthermore, a channel weighting mechanism is introduced to enhance the expressive power of both temporal and spatial features. Ultimately, through a cross-fusion strategy, the organic integration of temporal and spatial features is achieved. The effectiveness of the proposed method is validated through numerical examples and Tennessee-Eastman process. Experimental results confirm that the proposed method outperforms existing technologies in fault detection accuracy, thereby demonstrating its significant application value in the field of industrial safety production.