UAV Sensor Fault Detection Using a Classifier without Negative Samples: A Local Density Regulated Optimization...

Area of Science:

• Aerospace Engineering
• Machine Learning
• Control Systems

Background:

• Sensor fault detection is critical for unmanned aerial vehicle (UAV) flight security.
• Erroneous sensor data can lead to critical flight control command failures.
• Scarcity of faulty instances presents a significant challenge for traditional fault detection methods.

Purpose of the Study:

• To address the limitations of standard one-class support vector machines (OC-SVM) in handling outliers for UAV sensor fault detection.
• To enhance the detection rate and robustness of fault detection systems in the presence of scarce faulty data.
• To propose an optimized OC-SVM approach that effectively manages outliers based on their local density characteristics.

Main Methods:

• Developed an optimized one-class support vector machine (OC-SVM) approach.
• Incorporated local density estimation to regulate the decision boundary's tolerance to outliers.
• Defined a rule for assigning continuous density coefficients to outliers, preserving internal data integrity.

Main Results:

• The proposed optimized OC-SVM demonstrated superior performance in simulation.
• Effectively regulated the decision boundary's sensitivity to outliers based on local density.
• Achieved improved fault detection rates compared to standard OC-SVM methods.

Conclusions:

• The optimized OC-SVM approach effectively enhances UAV sensor fault detection.
• Local density regulation provides a robust mechanism for handling outliers.
• The method offers a promising solution for improving flight security in UAVs.