.png){kind=icon}
.png){kind=icon}
Return
Inception-SECA-BiLSTM: a multi-scale feature extraction modeling framework with attention enhancement for industrial process fault diagnosis
.png){kind=icon}
PRE
AI
C
L
F
R
DOI：10.1016/j.psep.2026.108964.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Industrial process data are often characterized by strong nonlinearity, temporal dependence, severe noise interference, and the difficulty of extracting weak fault features, which pose significant challenges to accurate fault diagnosis. To address these issues, this paper proposes an Inception Self-Adaptive Efficient Channel Attention Bidirectional Long Short-Term Memory (Inception-SECA-BiLSTM) model for industrial process fault diagnosis. The proposed model integrates multi-scale feature extraction, channel attention enhancement, and temporal dependency modeling into a unified framework. Specifically, an Inception-SECA module is designed to capture multi-scale fault features through parallel convolutions, while the embedded Self-Adaptive Efficient Channel Attention (SECA) mechanism adaptively recalibrates channel-wise feature responses, enhancing critical fault-related information and suppressing noise. Furthermore, a Bidirectional Long Short-Term Memory (BiLSTM) network is employed to model bidirectional temporal dependencies, combined with a single-head scaled dot-product self-attention mechanism to further emphasize informative features. Finally, a feature aggregation strategy is introduced to fuse global representations for improved fault discrimination. Experimental results on the Tennessee Eastman (TE) process and a real-world industrial coke furnace dataset demonstrate that the proposed method achieves superior classification accuracy, as well as favorable robustness to noise disturbances and stable training behavior. These findings indicate that the proposed model provides an effective and reliable solution for industrial process fault diagnosis.
Keywords:
multi-scale feature extraction
channel attention
bidirectional LSTM
fault diagnosis
industrial process
Journal
IF:
7.8
Papers:
9.4K
Citations:
3.8W
Organization
Cited Papers