KANS: Knowledge Discovery Graph Attention Network for Soft Sensing in Multivariate Industrial Processes

TL;DR

Soft sensing in complex industrial processes is challenged by non-linear dynamics and non-Euclidean sensor dependencies. The authors propose KANS, which combines latent sensor embeddings, unsupervised contrastive graph structure learning to infer sensor topology, and a graph attention-based predictor to estimate hard-to-measure outputs, trained end-to-end with $L_{\text{MSE}}$. Key contributions include learning sensor graphs without predefined topology, parallel graph attention representation learning, and knowledge discovery analyses for interpretability. On the Cranfield MFP dataset, KANS consistently outperforms baselines and state-of-the-art methods across metrics, and identifies sensor relationships that align with process structure. The approach enables accurate, interpretable soft sensing without heavy reliance on domain knowledge, with potential extension to hypergraphs.

Abstract

Soft sensing of hard-to-measure variables is often crucial in industrial processes. Current practices rely heavily on conventional modeling techniques that show success in improving accuracy. However, they overlook the non-linear nature, dynamics characteristics, and non-Euclidean dependencies between complex process variables. To tackle these challenges, we present a framework known as a Knowledge discovery graph Attention Network for effective Soft sensing (KANS). Unlike the existing deep learning soft sensor models, KANS can discover the intrinsic correlations and irregular relationships between the multivariate industrial processes without a predefined topology. First, an unsupervised graph structure learning method is introduced, incorporating the cosine similarity between different sensor embedding to capture the correlations between sensors. Next, we present a graph attention-based representation learning that can compute the multivariate data parallelly to enhance the model in learning complex sensor nodes and edges. To fully explore KANS, knowledge discovery analysis has also been conducted to demonstrate the interpretability of the model. Experimental results demonstrate that KANS significantly outperforms all the baselines and state-of-the-art methods in soft sensing performance. Furthermore, the analysis shows that KANS can find sensors closely related to different process variables without domain knowledge, significantly improving soft sensing accuracy.

Figures (4)

• Figure 1: Overview of our proposed framework (KANS). A represents data preprocessing module that converts raw data to sliding window format. B illustrates a latent sensor embedding module that extracts the characteristics of each sensor. C demonstrate an unsupervised contrastive graph structure learning module that learns the relationship between sensors. D shows a graph attention-based representation learning module that predicts the soft sensor output.
• Figure 2: Diagram of Cranfield Multiphase Flow (MFP) facility
• Figure 3: Plots of prediction results compared to the ground truth, for soft sensor output variables 5, 8, 15, 16, 19, 20.
• Figure 4: Heatmaps of data correlation, embedding correlation, and attention matrix for all six different soft sensor output variables. They are in the range from -1 to 1.