Open Access

ARTICLE

Time-Domain Feature Data Generation and Analysis Based on Grouping-Aggregation for Industrial System

Guanfeng Wang^1,2, Xuliang Yao^1,*, Jingfang Wang¹, Yongxin Sun², Jincheng Geng², Erlou Shi², Zhili Zhou^3,*

1 College of Intelligent Systems Science and Engineering, Harbin Engineering University, No. 145, Nantong Street, Nangang District, Harbin, China
2 Department of Intelligent Operation and Maintenance, Harbin Research Institute of Large Electric Machinery Company Limited, No. 51, Sandadongli Road, Xiangfang District, Harbin, China
3 School of Artificial Intelligence, Guangzhou University, 230 Wai Huan Xi Road, Guangzhou Higher Education Mega Center, Guangzhou, China

* Corresponding Authors: Xuliang Yao. Email: ; Zhili Zhou. Email:

(This article belongs to the Special Issue: Industrial Big Data and Artificial Intelligence-Driven Intelligent Perception, Maintenance, and Decision Optimization in Industrial Systems-2nd Edition)

Computers, Materials & Continua 2026, 87(3), 96 https://doi.org/10.32604/cmc.2026.076846

Received 27 November 2025; Accepted 11 March 2026; Issue published 09 April 2026

Abstract

In a real-world industrial system, it is challenging to reduce interference from operating conditions and extract high-quality feature data. To address these issues, this paper proposes a time-domain feature generation and analysis (FGA) scheme, which designs a grouping-aggregation (GA) scheme and an index decomposition (ID) method to extract and analyze high-quality feature data for industrial systems. The FGA represents a designed GA-based hybrid algorithm collaborative architecture, which overcomes the limitations of single algorithms and enables joint feature extraction from multi-condition, multi-scale industrial data. Simultaneously, FGA incorporates a feedback-driven threshold self-calibration mechanism, integrating LSTM-VAE to dynamically adjust detection thresholds based on reconstruction error distributions, significantly enhancing the system’s robustness against data distribution drift based on the FGA scheme. The software model distributes complex computations across scheduled program execution cycles to achieve optimal utilization of system resources. Specifically, the ID decouples complex operating conditions and employs stationarity analysis for valid feature data generation, whereas the GA effectively supplies a substantial set of data samples for data prediction. This approach is suitable for complex nonlinear systems characterized by limited data samples. The experiments conducted on the industrial feature data of a hydropower plant in Hubei, China, demonstrate that the proposed FGA scheme could reduce the data acquisition costs and effectively enhance the industrial system data analysis performance.

---

Keywords

---