Data-Centric Helicopter Failure Anticipation: The MGB Oil Pressure Virtual Sensor Case

Daouayry, Nassia; Mechouche, Ammar; Maisonneuve, Pierre-Loic; Scuturici, Vasile-Marian; Petit, Jean-Marc

doi:10.1109/bigdata47090.2019.9006532

“…The predicted value is compared to the one given by the physical sensor: an alert is raised if the difference between the two values is too high. An example of such a virtual sensor has been proposed by AH for the oil pressure of the helicopter Main Gear Box (MGB) [2]. We reuse the data from this study to perform the experiments of this paper.…”

Section: Ah Classification Datasetsmentioning

confidence: 99%

“…We applied our contextualization technique in the context of predictive maintenance for helicopters, a growing topic in the industry. Virtual sensors such as the ones used for the experiments of this paper [2] are interesting solutions in this context. Similarly, [3] proposes a virtual sensor to anticipate failures on photo-voltaic systems.…”

Section: Related Workmentioning

confidence: 99%

Contextualisation of Datasets for Better Classification Models: Application to Airbus Helicopters Flight Data

Guilly

¹

,

Daouayry

²

,

Maisonneuve

³

et al. 2020

Communications in Computer and Information Science

Self Cite

0

View full text Add to dashboard Cite

For helicopters, anticipating failures is crucial. To this end, the analysis of flight data allows to develop predictive maintenance approaches, for which Airbus Helicopters (AH) has proposed several solutions, some based on machine learning using predictive models. One recurrent problem in this setting is the contextualization of the data, that is to identify the data better fitting the phenomenon being modeled. Indeed, helicopters are complex systems going through different flight phases. Experts therefore have to identify the adequate ones, in which the selected flight parameters are stable and consistent with the studied problem. In this paper, we propose a generic solution to contextualize classification data, and present an experimental study on AH flight data: the results are encouraging and allow to keep domain experts involved the process.

show abstract

“…Data series 1 anomaly detection is a crucial problem with application in a wide range of domains [6,46]. Examples of such applications can be found in manufacturing, astronomy, engineering, and other domains [44,46], including detection of abnormal heartbeats in cardiology [27], wear and tear in bearings of rotating machines [5], machine degradation in manufacturing [41], hardware and software failures in data center monitoring [47], mechanical faults in vehicle operation monitoring [17] and identification of transient noise in gravitational wave detectors [7]. This implies a real need by relevant applications for developing methods that can accurately and efficiently achieve this goal.…”

Section: Introductionmentioning

confidence: 99%

Series2Graph: Graph-based Subsequence Anomaly Detection for Time Series

Boniol,

Palpanas

2022

Preprint

View full text Add to dashboard Cite

Subsequence anomaly detection in long sequences is an important problem with applications in a wide range of domains. However, the approaches that have been proposed so far in the literature have severe limitations: they either require prior domain knowledge that is used to design the anomaly discovery algorithms, or become cumbersome and expensive to use in situations with recurrent anomalies of the same type. In this work, we address these problems, and propose an unsupervised method suitable for domain agnostic subsequence anomaly detection. Our method, Series2Graph, is based on a graph representation of a novel low-dimensionality embedding of subsequences. Series2Graph needs neither labeled instances (like supervised techniques), nor anomaly-free data (like zero-positive learning techniques), and identifies anomalies of varying lengths. The experimental results, on the largest set of synthetic and real datasets used to date, demonstrate that the proposed approach correctly identifies single and recurrent anomalies without any prior knowledge of their characteristics, outperforming by a large margin several competing approaches in accuracy, while being up to orders of magnitude faster. This paper has appeared in VLDB 2020.

show abstract

Series2Graph

Boniol¹,

Palpanas

²

2020

View full text Add to dashboard Cite

Subsequence anomaly detection in long sequences is an important problem with applications in a wide range of domains. However, the approaches that have been proposed so far in the literature have severe limitations: they either require prior domain knowledge that is used to design the anomaly discovery algorithms, or become cumbersome and expensive to use in situations with recurrent anomalies of the same type. In this work, we address these problems, and propose an unsupervised method suitable for domain agnostic subsequence anomaly detection. Our method, Series2Graph, is based on a graph representation of a novel low-dimensionality embedding of subsequences. Series2Graph needs neither labeled instances (like supervised techniques), nor anomaly-free data (like zero-positive learning techniques), and identifies anomalies of varying lengths. The experimental results, on the largest set of synthetic and real datasets used to date, demonstrate that the proposed approach correctly identifies single and recurrent anomalies without any prior knowledge of their characteristics, outperforming by a large margin several competing approaches in accuracy, while being up to orders of magnitude faster.

show abstract

Data-Centric Helicopter Failure Anticipation: The MGB Oil Pressure Virtual Sensor Case

Cited by 3 publications

References 16 publications

Contextualisation of Datasets for Better Classification Models: Application to Airbus Helicopters Flight Data

Contextualisation of Datasets for Better Classification Models: Application to Airbus Helicopters Flight Data

Series2Graph: Graph-based Subsequence Anomaly Detection for Time Series

Series2Graph

Contact Info

Product

Resources

About