Optimal active unsupervised fault detection in cascaded h-bridge inverters based on machine learning

Area of Science:

• Electrical Engineering
• Power Electronics
• Machine Learning

Background:

• Multi-Level Inverters (MLIs) are crucial for high-voltage, high-power applications, but their reliability is challenged by increasing fault occurrences with more switches.
• Effective fault detection is essential for MLI reliability in industrial settings, yet traditional physics-based and model-based methods face challenges due to parameter uncertainties.
• The complexity of MLI systems necessitates advanced fault detection techniques to ensure operational integrity.

Purpose of the Study:

• To propose and evaluate a highly efficient, unsupervised machine learning model for fault detection in MLIs.
• To demonstrate the effectiveness of the Isolation Forest (IF) algorithm in identifying faults within a 17-level Cascaded H-Bridge (CHB) inverter.
• To compare the performance of the IF model against other methods using key performance indicators.

Main Methods:

• Implementation of the Isolation Forest (IF) algorithm, an unsupervised machine learning technique for anomaly detection.
• Simulation of a 17-level Cascaded H-Bridge (CHB) inverter under various fault conditions.
• Comparative analysis of the IF model's performance using F1-Score, Precision, Recall, and Accuracy metrics.

Main Results:

• The Isolation Forest (IF) model demonstrated high efficiency and accuracy in detecting faults in the simulated 17-level CHB inverter.
• The IF model achieved superior performance compared to other methods based on F1-Score, Precision, Recall, and Accuracy.
• The proposed unsupervised approach requires minimal computational complexity for MLI systems.

Conclusions:

• The Isolation Forest (IF) model provides an accurate and efficient unsupervised fault detection solution for MLIs.
• This method paves the way for fully automated and self-healing industrial power systems.
• The IF model's effectiveness in fault detection enhances the overall reliability and performance of MLI systems.