Fault Detection Method Using Auto-Associative Shared Nearest Neighbor Kernel Regression for Industrial Processes

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(5), P. 2251 - 2251

Published: Feb. 20, 2025

As industrial systems grow larger and more interconnected, timely fault detection is essential to minimize downtime, enhance reliability, reduce costs. However, conventional methods focus on reactive maintenance, limiting their ability detect faults before escalation. Additionally, propagation in large-scale can degrade performance. To address these challenges, we propose an auto-associative shared nearest neighbor kernel regression method for complex processes. Inspired by attention mechanisms, the proposed approach assigns higher weights relevant training data. Shared used assess similarity between data, rescaling distances accordingly. These adjusted are then utilized detection. The performance of evaluated applying it benchmark data from Tennessee Eastman Process a real-world, unplanned shutdown case concerning circulating fluidized bed boiler. experimental results show that anomalies up 2 h earlier than methods.

Language: Английский

Multiscale Interaction Purification-Based Global Context Network for Industrial Process Fault Diagnosis

Mathematics, Journal Year: 2025, Volume and Issue: 13(9), P. 1371 - 1371

Published: April 23, 2025

The application of deep convolutional neural networks (CNNs) has gained popularity in the field industrial process fault diagnosis. However, conventional CNNs primarily extract local features through convolution operations and have limited receptive fields. This leads to insufficient feature expression, as neglect temporal correlations data, ultimately resulting lower diagnostic performance. To address this issue, a multiscale interaction purification-based global context network (MIPGC-Net) is proposed. First, we propose refinement (MFIR) module. module aims enriched with combined information while refining representations by employing efficient channel attention mechanism. Next, develop wide dependency extraction sub-network (WTD) integrating MFIR network. can capture correlation from input, enhancing comprehensive perception information. Finally, MIPGC-Net constructed stacking multiple WTD sub-networks perform diagnosis processes, effectively capturing both proposed method validated on Tennessee Eastman Continuous Stirred-Tank Reactor confirming its effectiveness.

Language: Английский