Gas Sensor Array Fault Diagnosis Based on Multi-Dimensional Fusion, an Attention Mechanism, and Multi-Task Learning

Area of Science:

• Sensor technology
• Artificial intelligence
• Machine learning

Background:

• Machine olfactory systems are crucial for environmental monitoring and medical diagnosis.
• Accurate gas sensor array fault diagnosis is essential for reliable system operation.
• Existing methods have limitations in diagnostic accuracy and efficiency due to single-dimensional feature extraction.

Purpose of the Study:

• To develop a novel fault diagnosis network, MAM-Net, for gas sensor arrays.
• To address limitations of existing methods by employing multi-dimensional feature fusion, attention mechanisms, and multi-task learning.
• To improve the accuracy and efficiency of gas sensor array fault diagnosis.

Main Methods:

• Developed MAM-Net, a novel fault diagnosis network.
• Applied multi-dimensional feature fusion to extract comprehensive features.
• Utilized a residual network with attention modules and a Bi-LSTM network for spatial and temporal feature capture.
• Integrated features using a concatenation layer and employed multi-task learning for parallel diagnosis.

Main Results:

• MAM-Net demonstrated superior performance compared to existing methods.
• The framework achieved good recognition accuracy and robustness across various datasets and experimental settings.
• Multi-dimensional feature fusion and multi-task learning effectively improved diagnostic capabilities.

Conclusions:

• MAM-Net offers an effective solution for gas sensor array fault diagnosis.
• The proposed framework enhances diagnostic accuracy and efficiency.
• MAM-Net shows significant potential for reliable machine olfactory system operation.