Hybrid Machine Learning-Based Fault-Tolerant Sensor Data Fusion and Anomaly Detection for Fire Risk Mitigation in...

Area of Science:

• Industrial Internet of Things (IIoT)
• Sensor Networks
• Artificial Intelligence

Background:

• Fire detection systems in oil and gas IIoT rely on sensor data, which is vulnerable to inaccuracies from faulty sensors.
• Sensor issues like noise, outliers, and drift can cause delayed or missed fire predictions, increasing safety and operational risks.

Purpose of the Study:

• To present an advanced approach for handling faulty sensors in edge servers within IIoT environments.
• To enhance the reliability and accuracy of fire prediction systems through multi-sensor fusion, machine learning, and uncertainty handling.

Main Methods:

• Real-time anomaly detection and statistical assessment for sensor data preprocessing.
• Machine learning (ML)-driven probabilistic model adjustment with dynamic thresholding and belief mass assignment.
• Uncertainty quantification using Hellinger and Deng entropy with Dempster-Shafer Theory for sensor fusion.

Main Results:

• Improved accuracy in fire prediction even with unreliable sensor data through dynamic adjustment of probabilistic models.
• Enhanced decision-making by effectively managing sensor discrepancies and reducing the impact of faulty readings.
• Increased reliability of fire detection systems in challenging IIoT environments.

Conclusions:

• The proposed approach provides a robust solution for real-time fire prediction in oil and gas IIoT settings, even with faulty sensor inputs.
• Mitigation of fire risks is achieved by enhancing the accuracy and reliability of fire detection systems through intelligent sensor data management.