Real-time anomaly detection framework using a support vector regression for the safety monitoring of commercial aircraft

Lee, Hyunseong; Li, Guoyi; Rai, Ashwin; Chattopadhyay, Aditi

doi:10.1016/j.aei.2020.101071

Cited by 68 publications

(15 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Because of the difficulty of obtaining labels for aviation data, unsupervised-learning approaches for anomaly detection have been another thrust of research. The field of unsupervised approaches is diverse and includes proximity-based methods [9,10], clusteringbased methods [11,12], kernel-based methods [13][14][15], and deeplearning-based methods [16,17]. Bay and Schwabacher [9] define anomaly as a point in a feature space with remote nearest neighbors.…”

Section: Related Workmentioning

confidence: 99%

“…Li et al [18] developed the Cluster-AD-Flight method that transforms FOQA data into high-dimensional vectors, making different flights comparable by sampling each flight parameter at fixed temporal or distance-based intervals starting from an anchoring event (e.g., time from takeoff or distance from touchdown) with subsequent clustering using a density-based spatial clustering algorithm. Kernelbased methods based on support vector regression [15] and one-class support vector machines (OC-SVMs) [13,14] have also been developed to identify anomalies in FOQA data. An OC-SVM constructs an optimal hyperplane that segregates normal data in a high-dimensional reproducing kernel Hilbert space by maximizing the margin between the origin and the hyperplane.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Multiclass Anomaly Detection in Flight Data Using Semi-Supervised Explainable Deep Learning Model

Memarzadeh

Matthews

Templin

2022

Journal of Aerospace Information Systems

View full text Add to dashboard Cite

The identification of precursors to safety incidents in aviation data is a crucial task, yet extremely challenging. The main approach in practice leverages domain expertise to define expected tolerances in system behavior and flags exceedances from such safety margins. However, this approach is incapable of identifying unknown risks and vulnerabilities. Various machine-learning approaches have been investigated and deployed to identify anomalies, with the great challenge of procuring enough labeled data to achieve reliable and accurate performance. This paper presents an explainable deep semi-supervised model for anomaly detection in aviation, building upon recent advancements described in the machine-learning literature. The proposed model combines feature engineering and classification in feature space, while leveraging all available data (labeled and unlabeled). Our approach is validated with case studies of anomaly detection during the takeoff and landing phases of commercial aircraft. Our model outperforms the state-of-the-art supervised anomaly-detection model, reaching significantly higher accuracy and fewer false alarms, even if only small proportion of data in the training set is labeled.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Multiclass Anomaly Detection in Flight Data Using Semi-Supervised Explainable Deep Learning Model

Memarzadeh

Matthews

Templin

2022

Journal of Aerospace Information Systems

View full text Add to dashboard Cite

show abstract

“…An automobile engine startup failure is used as a case study, but fails to show any validation of the results obtained. In [24], the authors propose a framework to detect anomalies in aircraft systems during flight. However, it does not provide a systematic methodology for the resolution of the identified failures.…”

Section: Related Workmentioning

confidence: 99%

Self-Adaptative Troubleshooting for to Guide Resolution of Malfunctions in Aircraft Manufacturing

et al. 2021

View full text Add to dashboard Cite

The increasing complexity of systems and the heterogeneous origin of the possible malfunctions bring about the necessity of redefining the troubleshooting processes. Troubleshooting comprises the set of steps for the systematic analysis of the symptoms after the detection of a malfunction. The complexity of certain systems, such as aircraft, means the origin of that malfunction can be any of several reasons, where diagnosis techniques support engineers in determining the reason for the unexpected behaviour. However, derived from the high number of components involved in an aircraft, the list of possible fault origins can be extremely long, and the analysis of every element on the list, until the element responsible is found, can be very time-consuming and error-prone. As an alternative, certain input/output signals can be read to prevent the substitution of a correctly functioning component, by validating its behaviour in an indirect way. In order to optimise the actions to perform, we have identified the relevant parts of the model to propose a troubleshooting process to ascertain the signals to read and the components to substitute, while striving to minimise the action cost in accordance with a combination of structural analysis, the probability of malfunction associated to the components, and the cost associated to each extra signal read and component substituted. The proposal has been validated in a system taken from a real scenario obtained in collaboration with the Airbus Defence and Space company. A statistical analysis of the degree of improvement of the troubleshooting process has also been included. INDEX TERMS Decision-making process, Model-based Diagnosis, multi-objective function, Troubleshooting.

show abstract

“…The impacts of aviation include CO 2 effects and non-CO 2 effects ranging from NO x and water vapour to contrails and aerosols. While CO 2 and water vapour are direct greenhouse gases, NO x emissions influence the atmospheric composition via chemical processes by depleting methane (cooling effects) and forming ozone (warming effects) [20].…”

Section: Introductionmentioning

confidence: 99%

“…The significance of the aforementioned individual effect on the overall climate change has been studied thoroughly by climate scientists [13] and quantified in the form of Radiative Forcing (RF) change [20,21]. A transport carbon footprint methodology has been proposed to identify unit carbon footprints [22].…”

Section: Introductionmentioning

confidence: 99%

Multi-Objective Flight Altitude Decision Considering Contrails, Fuel Consumption and Flight Time

Xue

Hsu

2020

Sustainability

View full text Add to dashboard Cite

The rapid growth of air travel and aviation emissions in recent years has contributed to an increase in climate impact. Contrails have been considered one of the main factors of the aviation-induced climate impact. This paper deals with the formation of persistent contrails and its relationship with fuel consumption and flight time when flight altitude and true airspeed vary. Detailed contrail formation conditions pertaining to altitude, relative humidity and temperature are formulated according to the Schmidt–Appleman criterion. Building on the contrail formation model, the proposed model would minimise total travel time, fuel consumption and contrail length associated with a given flight. Empirical data (including pressure, temperature, relative humidity, etc.) collected from seven flight information regions in Chinese observation stations were used to analyse the spatial and temporal distributions of the persistent contrail formation area. The trade-off between flight time, fuel consumption and contrail length are illustrated with a real-world case. The results provided a valuable benchmark for flight route planning with environmental, flight time, sustainable flight trajectory planning and fuel consumption considerations, and showed significant contrail length reduction through an optimal selection of altitude and true airspeed.

show abstract

Real-time anomaly detection framework using a support vector regression for the safety monitoring of commercial aircraft

Cited by 68 publications

References 34 publications

Multiclass Anomaly Detection in Flight Data Using Semi-Supervised Explainable Deep Learning Model

Multiclass Anomaly Detection in Flight Data Using Semi-Supervised Explainable Deep Learning Model

Self-Adaptative Troubleshooting for to Guide Resolution of Malfunctions in Aircraft Manufacturing

Multi-Objective Flight Altitude Decision Considering Contrails, Fuel Consumption and Flight Time

Contact Info

Product

Resources

About