Hybrid Continuous Density Hmm-Based Ensemble Neural Networks for Sensor Fault Detection and Classification in...

Area of Science:

• Computer Science
• Electrical Engineering
• Network Security

Background:

• Sensor devices in wireless sensor networks (WSNs) are prone to faults in hazardous environments.
• Existing fault detection methods often neglect the dynamic state changes during fault occurrences.
• Capturing these state dynamics is crucial for identifying rapidly evolving faults.

Purpose of the Study:

• To propose a novel fault detection and classification method for WSNs.
• To investigate the dynamics of sensor data states during fault events.
• To evaluate the efficacy of hybrid CDHMM and neural network models for sensor fault analysis.

Main Methods:

• Utilized Continuous Density Hidden Markov Models (CDHMM) to model state transition dynamics.
• Employed various Neural Networks (NNs), including Learning Vector Quantization (LVQ), Probabilistic Neural Network (PNN), Adaptive Probabilistic Neural Network (APNN), and Radial Basis Function (RBF) for fault classification.
• Developed an ensemble NN framework with a majority voting scheme for enhanced decision-making.

Main Results:

• The hybrid CDHMM-LVQ model demonstrated superior detection accuracy compared to other NN classifiers.
• The ensemble CDHMM classifier effectively captured state change dynamics, proving vital for detecting instant faults.
• Performance was evaluated using detection accuracy, false positive rate, F1-score, and Matthews correlation coefficient.

Conclusions:

• Hybrid CDHMM and NN models offer a robust solution for sensor fault detection and classification in WSNs.
• The proposed ensemble framework excels at identifying rapidly evolving faults by analyzing state transition dynamics.
• This approach enhances the reliability and resilience of sensor networks operating in challenging conditions.