Open Access

ARTICLE

Hierarchical Cyber–Physical Symbiosis with Bidirectional State Space Modeling for IIoT Anomaly Diagnosis

Kelan Wang¹, Jianfei Chen^2,*

1 School of Communication and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing, China
2 College of Electronic and Optical Engineering and College of Flexible Electronics (Future Technology), Nanjing University of Posts and Telecommunications, Nanjing, China

* Corresponding Author: Jianfei Chen. Email:

(This article belongs to the Special Issue: Attention Mechanism-based Complex System Pattern Intelligent Recognition and Accurate Prediction)

Computers, Materials & Continua 2026, 88(1), 27 https://doi.org/10.32604/cmc.2026.079644

Received 25 January 2026; Accepted 03 March 2026; Issue published 08 May 2026

Abstract

As 6G-enabled Industrial Internet of Things (IIoT) evolves, green and sustainable industrial monitoring increasingly relies on edge AI to deliver low-latency diagnosis under tight resource constraints. Industrial cyber–physical systems increasingly rely on heterogeneous sensing and communication infrastructures, where network-side attacks can propagate into physical processes and appear as coupled anomalies. Reliable diagnosis therefore requires joint learning from time-synchronized cyber and physical telemetry rather than modeling them as independent signals. This paper develops Cyber–Physical Symbiosis Network (CPSNet), a model designed for edge-AI deployment with a dual-stream architecture for fixed-window multiclass cross-domain anomaly diagnosis in IIoT. CPSNet encodes each modality into hierarchical multi-resolution features and refines them with a Multi-Scale Bidirectional-Recursive (MSBR) block. MSBR couples multi-kernel temporal convolutions with a gated bidirectional state space pathway, capturing transient irregularities while retaining long-range context within the window. Cross-modal dependency is injected at every scale by a symbiosis module that performs bidirectional channel-wise gating and holistic state space fusion to learn unified cross-modal dynamics efficiently. A compact multi-scale pooling head with auxiliary modality supervision preserves discriminative evidence in both streams. On the DataSense benchmark, CPSNet achieves 97.18% Accuracy and 99.04% AUC on Multiclass-8, and 89.07% Accuracy and 94.28% AUC on Multiclass-50, showing consistent improvements over single-modality and multi-modal baselines. Ablation and efficiency analyses further suggest complementary gains from multi-scale refinement and explicit coupling with a favorable accuracy–runtime trade-off. These results suggest that hierarchical cross-modal coupling with state space temporal modeling can improve robust, fine-grained IIoT diagnosis for 6G edge-AI monitoring.

---

Keywords

---