Anomaly Detection in Gas Turbine Time Series by Means of Bayesian Hierarchical Models

Losi, Enzo; Venturini, Mauro; Manservigi, Lucrezia; Ceschini, Giuseppe Fabio; Bechini, Giovanni

doi:10.1115/1.4044781

Cited by 15 publications

(13 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The fault detection is based on a distance measure utilizing recursively calculated means and standard deviations. A similar approach utilizing Bayesian hierarchical models and the same fault-detection scheme is demonstrated in [18] for evaluating gas turbines. A utilization of data-driven models based on LSTMs in combination with a statistical evaluation of the residuals is demonstrated in [19] for a three-shaft marine gas turbine.…”

Section: Fault Detectionmentioning

confidence: 99%

Steady-State Fault Detection with Full-Flight Data

et al. 2022

View full text Add to dashboard Cite

Aircraft engine condition monitoring is a key technology for increasing safety and reducing maintenance expenses. Current engine condition monitoring approaches use a minimum of one steady-state snapshot per flight. Whilst being appropriate for trending gradual engine deterioration, snapshots result in a detrimental latency in fault detection. The increased availability of non-mandatory data acquisition hardware in modern airplanes provides so-called full-flight data sampled continuously during flight. These datasets enable the detection of engine faults within one flight by deriving a statistically relevant set of steady-state data points, thus, allowing the application of machine-learning approaches. It is shown that low-pass filtering before steady-state detection significantly increases the success rate in detecting steady-state data points. The application of Principal Component Analysis halves the number of relevant dimensions and provides a coordinate system of principal components retaining most of the variance. Consequently, clusters of data points with and without engine fault can be separated visually and numerically using a One-Class Support Vector Machine. High detection rates are demonstrated for various component faults and even for a minimum instrumentation suite using synthesized datasets derived from full-flight data of commercially operated flights. In addition to the tests conducted with synthesized data, the algorithm is verified based on operational in-flight measurements providing a proof-of-concept. Consequently, the availability of continuously sampled in-flight measurements combined with machine-learning methods allows fault detection within a single flight.

show abstract

Section: Fault Detectionmentioning

confidence: 99%

Steady-State Fault Detection with Full-Flight Data

et al. 2022

View full text Add to dashboard Cite

show abstract

“…The first one is the size of the CPTs, which grows exponentially with a large number of nodes and states, making the problem intractable if we want to identify the correct fault and magnitude among several possible conditions. This limitation has been addressed by proposing hierarchical models [13,14] and will be tackled in Part 2 of this paper [27]. The second shortcoming concerns the choice of prior probability distribution, especially when sufficient data are not available from the system of interest.…”

Section: Bayesian Networkmentioning

confidence: 99%

“…To overcome the limitations related to model complexity, when multiple faults have to be analyzed simultaneously, hierarchical models have been proposed. Hierarchical Bayesian models were proven to successfully detect various sensor faults [13] and to isolate single and multiple faults also in the presence of sensor biases [14]. Despite their promising features, BNs have been employed mostly for fault isolation.…”

Section: Introductionmentioning

confidence: 99%

Assessment of Dynamic Bayesian Models for Gas Turbine Diagnostics, Part 1: Prior Probability Analysis

2021

View full text Add to dashboard Cite

The reliability and cost-effectiveness of energy conversion in gas turbine systems are strongly dependent on an accurate diagnosis of possible process and sensor anomalies. Because data collected from a gas turbine system for diagnosis are inherently uncertain due to measurement noise and errors, probabilistic methods offer a promising tool for this problem. In particular, dynamic Bayesian networks present numerous advantages. In this work, two Bayesian networks were developed for compressor fouling and turbine erosion diagnostics. Different prior probability distributions were compared to determine the benefits of a dynamic, first-order hierarchical Markov model over a static prior probability and one dependent only on time. The influence of data uncertainty and scatter was analyzed by testing the diagnostics models on simulated fleet data. It was shown that the condition-based hierarchical model resulted in the best accuracy, and the benefit was more significant for data with higher overlap between states (i.e., for compressor fouling). The improvement with the proposed dynamic Bayesian network was 8 percentage points (in classification accuracy) for compressor fouling and 5 points for turbine erosion compared with the static network.

show abstract

“…Hybrid techniques that apply a collective problem-solving approach have shown promising performance when the methods are integrated on the basis of their complementary strengths [15,16]. In general, the current literature on gas turbine diagnostics, for instance [17][18][19], shows that advanced diagnostic method development is still a subject of considerable research effort.…”

Section: Introductionmentioning

confidence: 99%

Aircraft Engine Performance Monitoring and Diagnostics Based on Deep Convolutional Neural Networks

2021

View full text Add to dashboard Cite

The rapid advancement of machine-learning techniques has played a significant role in the evolution of engine health management technology. In the last decade, deep-learning methods have received a great deal of attention in many application domains, including object recognition and computer vision. Recently, there has been a rapid rise in the use of convolutional neural networks for rotating machinery diagnostics inspired by their powerful feature learning and classification capability. However, the application in the field of gas turbine diagnostics is still limited. This paper presents a gas turbine fault detection and isolation method using modular convolutional neural networks preceded by a physics-driven performance-trend-monitoring system. The trend-monitoring system was employed to capture performance changes due to degradation, establish a new baseline when it is needed, and generatefault signatures. The fault detection and isolation system was trained to step-by-step detect and classify gas path faults to the component level using fault signatures obtained from the physics part. The performance of the method proposed was evaluated based on different fault scenarios for a three-shaft turbofan engine, under significant measurement noise to ensure model robustness. Two comparative assessments were also carried out: with a single convolutional-neural-network-architecture-based fault classification method and with a deep long short-term memory-assisted fault detection and isolation method. The results obtained revealed the performance of the proposed method to detect and isolate multiple gas path faults with over 96% accuracy. Moreover, sharing diagnostic tasks with modular architectures is seen as relevant to significantly enhance diagnostic accuracy.

show abstract

Anomaly Detection in Gas Turbine Time Series by Means of Bayesian Hierarchical Models

Cited by 15 publications

References 19 publications

Steady-State Fault Detection with Full-Flight Data

Steady-State Fault Detection with Full-Flight Data

Assessment of Dynamic Bayesian Models for Gas Turbine Diagnostics, Part 1: Prior Probability Analysis

Aircraft Engine Performance Monitoring and Diagnostics Based on Deep Convolutional Neural Networks

Contact Info

Product

Resources

About