Journal of Engineering, Project, and Production Management, 2026, 16(5), 2025-292
Improved Multi-Scale CNN with Adaptive Weight Normalization for Closed-Loop Fault Diagnosis and Intelligent Recovery in Centralized Control Systems
1 Senior
engineer, State Grid Anhui Ultra High Voltage Company, Hefei, 231131,
China, E-mail: xiayilol@163.com (corresponding author).
2 Senior engineer, State Grid Anhui
Ultra High Voltage Company, Hefei, 231131, China, E-mail: xuy221x@ah.sgcc.com.cn
3 Senior engineer, State Grid Anhui
Ultra High Voltage Company, Hefei, 231131, China, E-mail: guolongg0317@ah.sgcc.com.cn
4 State Grid Anhui Electric Power
Co., Ltd., Hefei 230022, China, E-mail: xiec0222@ah.sgcc.com.cn
5 Engineer, State Grid Anhui Ultra
High Voltage Company, Hefei, 231131, China, E-mail: ahdlmahuan@163.com.cn
6 Senior engineer, State Grid Anhui
Ultra High Voltage Company, Hefei, 231131, China, E-mail: wengl0611@ah.sgcc.com.cn
Project Management
Received November 27, 2025; revised January 6, 2026; accepted January 14, 2026
Available online June 17, 2026
Abstract: Traditional centralized control systems often struggle to extract discriminative features from complex multi-source signals, leading to delays in fault diagnosis and suboptimal recovery performance. To address this, we propose an improved Convolutional Neural Network (CNN) architecture that integrates multi-scale convolutional kernels to capture heterogeneous spatiotemporal patterns. An adaptive weight normalization mechanism recalibrates feature channel responses, enhancing the detection of subtle anomalies. Extracted features are aggregated via global average pooling and classified by a softmax layer to identify fault types. The diagnosis results are then forwarded to an intelligent control module, where a parameter reconfiguration algorithm adjusts control strategies and regenerates execution commands, thereby establishing a closed-loop, self-adaptive recovery framework. Unlike existing studies that focus solely on either fault diagnosis or fault-tolerant control, this work explicitly integrates multi-scale feature learning and adaptive parameter reconfiguration within a unified closed-loop architecture, forming a system-level coupling between perception and recovery rather than introducing an isolated algorithmic improvement. Experiments on a centralized control simulation platform show that the proposed method achieves a fault classification accuracy of 95.2%, with an average diagnostic response time of 0.42s. Under extreme anomaly conditions, the system maintains a recovery stability index of 0.89. This study shifts centralized control from passive monitoring to proactive intelligent perception and self-healing control, offering a viable technical pathway for high-reliability industrial operation.
Keywords: Convolutional neural network, centralized control system, fault diagnosis, adaptive weight normalization, intelligent recovery.
Copyright © Journal of Engineering, Project, and Production Management (EPPM-Journal).
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
Requests for reprints and permissions at eppm.journal@gmail.com.
Citation: Xia, Y., Xu, Y., Guo, L., Xie, C., Ma, H., and Weng, L. (2026). Improved Multi-Scale CNN with Adaptive Weight Normalization for Closed-Loop Fault Diagnosis and Intelligent Recovery in Centralized Control Systems. Journal of Engineering, Project, and Production Management, 16(5), 2025-292.
