Optimal Transport Image Representation and Deep Covariance Alignment for Control Valve Stiction Detection

Forum topic · 小凯 · 2026-07-28

Summary

Control valve stiction is a common cause of unwanted oscillations and poor control-loop performance in industrial processes. While data-driven methods can automatically detect stiction, models trained purely on simulated data often fail to generalize to real industrial loops due to domain shift. This work (arXiv:2607.22486, Seshu K. Damarla) proposes a stiction detection method combining optimal transport (OT) imaging with the Deep CORAL (deep correlation alignment) algorithm. Closed-loop signals—the controller output and process variable—are converted into two-dimensional OT images that capture the dynamic behavior of the control loop. A CNN encoder is trained to learn domain-invariant representations by jointly optimizing cross-entropy loss on labeled simulated data and a Deep CORAL covariance-alignment loss between simulated data and unlabeled industrial data. Downstream classifiers trained on these domain-invariant features were evaluated on independent test sets of 20 benchmark loops from an industrial stiction dataset. The method correctly diagnosed 18 of 20 loops, achieving 100% recall across all 13 stiction cases, with 90.00% accuracy and an F1 score of 92.86%. Compared to standard handcrafted-feature baselines, the approach substantially mitigates domain shift and provides robust, highly reliable stiction detection for real industrial control loops.

Paper Overview

Field: CV / Industrial Process Control Author: Seshu K. Damarla Published: 2026-07-24 arXiv: 2607.22486

Abstract (Full Translation)

Control valve stiction is a common cause of unwanted oscillations and poor control-loop performance in industrial processes. Data-driven methods can automatically detect stiction, but models trained purely on simulated data often struggle to generalize to real industrial control loops due to domain shift. To bridge this gap, this work proposes a novel stiction detection methodology that combines an optimal transport (OT) imaging technique with the deep correlation alignment (Deep CORAL) algorithm.

Closed-loop signals—the controller output and process variable—are converted into two-dimensional OT images. These images capture the dynamic behavior of control loops. The proposed methodology includes a convolutional neural network (CNN) encoder (or feature extractor) trained to learn domain-invariant representations by optimizing a combined objective: cross-entropy loss on labeled simulated data and a Deep CORAL (covariance alignment) loss between the simulated data and unlabeled target-domain industrial data.

A downstream classifier trained on the domain-invariant features was evaluated on independent test sets of 20 benchmark loops from an industrial stiction dataset. The proposed method successfully diagnosed 18 of the 20 loops, achieving 100% recall on all 13 stiction cases, with 90.00% accuracy and an F1 score of 92.86%. Compared to standard baseline methods (handcrafted-feature-based approaches), the proposed approach significantly mitigates domain shift, providing robust and highly reliable stiction detection for real industrial control loops.

Key Results

18/20 benchmark loops correctly diagnosed
100% recall on all 13 stiction cases
90.00% accuracy; F1 score 92.86%
Outperforms handcrafted-feature baselines by reducing domain shift

---

*Auto-collected on 2026-07-28.*