MRS-Pyramid-TimeXer: A quality-related distributed monitoring framework for multi-rate industrial processes

Dong, Jie, Zhang, Minjie, Zhang, Qichun and Peng, Kaixiang (2026) MRS-Pyramid-TimeXer: A quality-related distributed monitoring framework for multi-rate industrial processes. Measurement, 267. p. 120658. ISSN 02632241

Full text not available from this repository. (Request a copy)

Abstract

Modern industrial processes are increasingly large-scale and multi-unit, which renders quality-related fault detection under multi-rate sampling a challenging task. To address this, MRS-Pyramid-TimeXer is proposed,which is a quality-related distributed fault detection framework that integrates pyramid-based multi-scale representations with an enhanced TimeXer architecture. First, it employs a knowledge collaborative strategy based on mutual information–mechanism to partition multi-rate variables into interpretable subsystems, incorporating process parameters as auxiliary variables to enhance multi-rate coupling characterization. Second, Pyramid-TimeXer models are constructed within local subsystems, utilizing patch-wise self-attention and variate-wise cross-attention to capture multi-scale dynamic correlations between quality and process variables. Finally, Bayesian posterior probability fuses statistics from each subsystem to derive global process metrics. Experiments on a Hot Strip Milling Process (HSMP) show that the proposed framework achieves an average fault detection rate of 98% with an average false alarm rate below 2% across three typical fault scenarios, consistently outperforming state-of-the-art baselines and validating its effectiveness and engineering applicability in large-scale manufacturing.

Item Type:	Article
Depositing User:	Research and Knowledge Exchange Office Admin 1
Date Deposited:	12 Feb 2026 11:05
Last Modified:	24 Feb 2026 11:14
URI:	https://bnu.repository.guildhe.ac.uk/id/eprint/20829

Actions (login required)

Edit Item