With the wide applications of Unmanned Aerial Vehicle (UAV) in the civilian and military fields, its operational safety has drawn much attention. A series of fault detection methods are studied to avoid disasters. Due to the capabilities of strong feature extraction and massive flight data processing, the deep learning-based methods have received extensive attention. However, restricted by UAV airborne size, weight, and power consumption, a significant challenge is posed to deploy these complicated detection methods in the airborne application which requires to run in real-time. In this paper, a Fault Detection Model Acceleration Engine (FDMAE) for UAV real-time fault detection is realized under the airborne constraint. Firstly, a high-performance detection model is designed based on stacked long short-term memory network, and fault detection is achieved by a statistical threshold in this method. Secondly, a model pruning method based on principal component analysis is proposed to improve computing efficiency. Finally, the pruned fault detection method is optimized and integrated as a flexible acceleration engine through high-level synthesis, and deployed on an airborne embedded computing platform based on field-programmable gate array. Real UAV flight data are used to verify the proposed FDMAE. By comparing accuracy, the area under the receiver operating characteristic curve, speed, and power consumption, the effectiveness of the FDMAE is proved.

Accurate fault detection for unmanned aerial vehicle (UAV) actuators is essential for ensuring flight safety and mission completion. Without the requirement of modeling complex physical mechanism, data-driven actuator fault detection approaches have attracted much attention. Among them, the long short-term memory (LSTM) approach has shown superior performance due to its capability of modeling complex spatial-temporal features. However, the modeling uncertainty of LSTM is actually changeable under different flight conditions, which has not been well considered in existing researches. In this paper, a novel uncertainty-aware LSTM (UA-LSTM) based dynamic flight condition fault detection approach for UAV actuator is proposed. A prediction based fault detection model is first set up based on LSTM, utilizing the information from both the flight action modes and the actuator effect. Its inputs are specifically selected by both physical mechanism and data correlation analysis. Furthermore, time series features indicating the prediction uncertainty of the model are constructed on the selected inputs to characterize dynamic flight conditions more accurately. Then, an adaptive threshold estimation space is set up based on an enhanced distribution based condition clustering approach. Fault detection thresholds for various flight conditions are obtained, and are smoothed to reduce the influence of disturbances. Finally, the fault detection model with step-wise adaptive detection threshold is acquired. Experimental results on both simulation and real flight data illustrate that the proposed approach is superior for actuator fault detection under dynamic flight conditions.

Fault detection for sensors of unmanned aerial vehicles is essential for ensuring flight security, in which the flight control system conducts real-time control for the vehicles relying on the sensing information from sensors, and erroneous sensor data will lead to false flight control commands, causing undesirable consequences. However, because of the scarcity of faulty instances, it still remains a challenging issue for flight sensor fault detection. The one-class support vector machine approach is a favorable classifier without negative samples, however, it is sensitive to outliers that deviate from the center and lacks a mechanism for coping with them. The compactness of its decision boundary is influenced, leading to the degradation of detection rate. To deal with this issue, an optimized one-class support vector machine approach regulated by local density is proposed in this paper, which regulates the tolerance extents of its decision boundary to the outliers according to their extent of abnormality indicated by their local densities. The application scope of the local density theory is narrowed to keep the internal instances unchanged and a rule for assigning the outliers continuous density coefficients is raised. Simulation results on a real flight control system model have proved its effectiveness and superiority.

The aluminum electrolytic capacitor (AEC) is one of the most important parts of a power electronic converter. Throughout its lifespan, the equivalent series resistance (ESR) of the AEC will increase, which will affect the performance of the power electronic converter. To guarantee the reliability of a converter, it is necessary to monitor the health state of the AEC, for which one common technique is to estimate the ESR of the AEC. Equipment with a high sampling rate is usually required because converter usually works at a high frequency. This requirement will increase the cost of ESR monitoring. To address this issue, this article proposes a novel method of estimating the ESR for an AEC. The proposed method uses an innovative compressed sensing approach to acquire the ripple voltage and current. Instead of complete reconstruction of the ripple voltage and current, only the high-frequency features in the wavelet subspace are reconstructed and used to calculate the ESR. The proposed method reduces the cost of data sampling, transmission, and storage. This method is not limited by the converter circuit structure and can be applied in various environments. Simulations and experiments were carried out to verify the effectiveness of the proposed method.

Effective anomaly detection of sensing data is essential for identifying potential system failures. Because they require no prior knowledge or accumulated labels, and provide uncertainty presentation, the probability prediction methods (e.g., Gaussian process regression (GPR) and relevance vector machine (RVM)) are especially adaptable to perform anomaly detection for sensing series. Generally, one key parameter of prediction models is coverage probability (CP), which controls the judging threshold of the testing sample and is generally set to a default value (e.g., 90% or 95%). There are few criteria to determine the optimal CP for anomaly detection. Therefore, this paper designs a graphic indicator of the receiver operating characteristic curve of prediction interval (ROC-PI) based on the definition of the ROC curve which can depict the trade-off between the PI width and PI coverage probability across a series of cut-off points. Furthermore, the Youden index is modified to assess the performance of different CPs, by the minimization of which the optimal CP is derived by the simulated annealing (SA) algorithm. Experiments conducted on two simulation datasets demonstrate the validity of the proposed method. Especially, an actual case study on sensing series from an on-orbit satellite illustrates its significant performance in practical application.

DC/DC converters play an important role in electrical systems. The anomalous state of a DC/DC converter has a major impact on the operation of the back-end components and the entire electrical system. To effectively recognize an anomalous state, particularly for DC/DC converters with unknown circuit structures, an online anomaly detection method that involves statistical feature estimation using Gaussian process regression (GPR) and a genetic algorithm (GA) is proposed. In the proposed method, the normal output range is built upon the DC/DC normal output signal using GPR, and seven statistical features are considered as the estimation indexes. In the detection process, the output signal of the DC/DC converter is acquired, and the corresponding statistical features are calculated by using GA, which can effectively reduce the operation time and the calculation consumption. The working state of the DC/DC converter is distinguished according to these features. Simulation and hardware experimental results validate the practicability and effectiveness of the proposed method.