scholarly journals Application of Explainable AI (Xai) For Anomaly Detection and Prognostic of Gas Turbines with Uncertainty Quantification.

2021 ◽  
Author(s):  
Ahmad Kamal Mohd Nor ◽  
Srinivasa Rao Pedapati ◽  
Masdi Muhammad
Keyword(s):  
Anomaly Detection ◽  
Uncertainty Quantification ◽  
Gas Turbines ◽  
Health Management ◽  
Detection Task ◽  
Uncertainty Measure ◽  
Real Gas ◽  
Prognostic Performance ◽  
Local Accuracy ◽  
Evaluation Metric

XAI is presently in its early assimilation phase in Prognostic and Health Management (PHM) domain. However, the handful of PHM-XAI articles suffer from various deficiencies, amongst others, lack of uncertainty quantification and explanation evaluation metric. This paper proposes an anomaly detection and prognostic of gas turbines using Bayesian deep learning (DL) model with SHapley Additive exPlanations (SHAP). SHAP was not only applied to explain both tasks, but also to improve the prognostic performance, the latter trait being left undocumented in the previous PHM-XAI works. Uncertainty measure serves to broaden explanation scope and was also exploited as anomaly indicator. Real gas turbine data was tested for the anomaly detection task while NASA CMAPSS turbofan datasets were used for prognostic. The generated explanation was evaluated using two metrics: Local Accuracy and Consistency. All anomalies were successfully detected thanks to the uncertainty indicator. Meanwhile, the turbofan prognostic results show up to 9% improvement in RMSE and 43% enhancement in early prognostic due to SHAP, making it comparable to the best published methods in the problem. XAI and uncertainty quantification offer a comprehensive explanation package, assisting decision making. Additionally, SHAP ability in boosting PHM performance solidifies its worth in AI-based reliability research.

scholarly journals Explainable Artificial Intelligence for Anomaly Detection and Prognostic of Gas Turbines using Uncertainty Quantification with Sensor-Related Data

2021 ◽  
Author(s):  
Ahmad Kamal Mohd Nor ◽  
Srinivasa Rao Pedapati ◽  
Masdi Muhammad ◽  
Víctor Leiva
Keyword(s):  
Artificial Intelligence ◽  
Anomaly Detection ◽  
Uncertainty Quantification ◽  
Gas Turbines ◽  
Health Management ◽  
Mean Square ◽  
Simulation Data ◽  
Related Data ◽  
Local Accuracy ◽  
Explainable Artificial Intelligence

Explainable artificial intelligence (XAI) is in its assimilation phase in the prognostic and health management (PHM). The literature on PHM-XAI is deficient with respect to metrics of uncertainty quantification and explanation evaluation. This paper proposes a new method of anomaly detection and prognostic for gas turbines using Bayesian deep learning and Shapley additive explanations (SHAP). The method explains the anomaly detection and prognostic and improves the performance of the prognostic, aspects that have not been considered in the literature of PHM-XAI. The uncertainty measures considered serve to broaden explanation scope and can also be exploited as anomaly indicators. Real-world gas turbine sensor-related data are tested for the anomaly detection, while NASA commercial modular aero-propulsion system simulation data, related to turbofan sensors, were used for prognostic. The generated explanation is evaluated using two metrics: consistency and local accuracy. All anomalies were successfully detected using the uncertainty indicators. Meanwhile, the turbofan prognostic results showed up to 9% improvement in root mean square error and 43% enhancement in early prognostic due to the SHAP, making it comparable to the best existing methods. The XAI and uncertainty quantification offer a comprehensive explanation for assisting decision-making. Additionally, the SHAP ability to increase PHM performance confirms its value in AI-based reliability research.

A Review on Gas Turbine Anomaly Detection for Implementing Health Management

2016 ◽  
Author(s):  
Ningbo Zhao ◽  
Xueyou Wen ◽  
Shuying Li
Keyword(s):  
Anomaly Detection ◽  
Gas Turbine ◽  
Gas Turbines ◽  
Health Management ◽  
Vibration Signal ◽  
Source Information ◽  
Rapid Improvement ◽  
Recent Developments ◽  
Data Driven Approach ◽  
On Line

With the rapid improvement of equipment manufacturing technology and the ever increasing cost of fuel, engine health management has become one of the most important parts of aeroengine, industrial and marine gas turbine. As an effective technology for improving the engine availability and reducing the maintenance costs, anomaly detection has attracted great attention. In the past decades, different methods including gas path analysis, on-line monitoring or off-line analysis of vibration signal, oil and electrostatic monitoring have been developed. However, considering the complexity of structure and the variability of working environments for engine, many important problems such as the accurate modeling of gas turbine with different environment, the selection of sensors, the optimization of various data-driven approach and the fusion strategy of multi-source information still need to be solved urgently. Besides, although a large number of investigations in this area are reported every year in various journals and conference proceedings, most of them are about aeroengine or industrial gas turbine and limited literature is published about marine gas turbine. Based on this background, this paper attempts to summarize the recent developments in health management of gas turbines. For the increasing requirement of predict-and-prevent maintenance, the typical anomaly detection technologies are analyzed in detail. In addition, according to the application characteristics of marine gas turbine, this paper introduces a brief prospect on the possible challenges of anomaly detection, which may provide beneficial references for the implementing and development of marine gas turbine health management.

Remote Thermal Performance Monitoring and Diagnostics: Turning Data Into Knowledge

2013 ◽  
Author(s):  
Xiaomo Jiang ◽  
Craig Foster
Keyword(s):  
Anomaly Detection ◽  
Gas Turbines ◽  
Performance Monitoring ◽  
Combined Cycle ◽  
Integrated System ◽  
Site Specific ◽  
Performance Improvements ◽  
Maintenance Schedule ◽  
Effectiveness Monitoring ◽  
And Performance

Gas turbine simple or combined cycle plants are built and operated with higher availability, reliability, and performance in order to provide the customer with sufficient operating revenues and reduced fuel costs meanwhile enhancing customer dispatch competitiveness. A tremendous amount of operational data is usually collected from the everyday operation of a power plant. It has become an increasingly important but challenging issue about how to turn this data into knowledge and further solutions via developing advanced state-of-the-art analytics. This paper presents an integrated system and methodology to pursue this purpose by automating multi-level, multi-paradigm, multi-facet performance monitoring and anomaly detection for heavy duty gas turbines. The system provides an intelligent platform to drive site-specific performance improvements, mitigate outage risk, rationalize operational pattern, and enhance maintenance schedule and service offerings via taking appropriate proactive actions. In addition, the paper also presents the components in the system, including data sensing, hardware, and operational anomaly detection, expertise proactive act of company, site specific degradation assessment, and water wash effectiveness monitoring and analytics. As demonstrated in two examples, this remote performance monitoring aims to improve equipment efficiency by converting data into knowledge and solutions in order to drive value for customers including lowering operating fuel cost and increasing customer power sales and life cycle value.

Uncertainty quantification of real gas models in CO2 supersonic flow

2021 ◽  
pp. 101479
Author(s):  
Wenna Raissa dos Santos Cruz ◽  
Fabio Pereira dos Santos ◽  
Ricardo de Andrade Medronho
Keyword(s):  
Supersonic Flow ◽  
Uncertainty Quantification ◽  
Real Gas

Hybrid Modeling of Heavy Duty Gas Turbines for On-Line Performance Monitoring

2014 ◽  
Author(s):  
Thomas Palmé ◽  
Francois Liard ◽  
Dan Cameron
Keyword(s):  
Gas Turbine ◽  
Surrogate Model ◽  
Gas Turbines ◽  
Performance Monitoring ◽  
Performance Model ◽  
Heavy Duty ◽  
Operational Parameters ◽  
Actual Performance ◽  
Real Gas ◽  
The Real

Due to their complex physics, accurate modeling of modern heavy duty gas turbines can be both challenging and time consuming. For online performance monitoring, the purpose of modeling is to predict operational parameters to assess the current performance and identify any possible deviation between the model’s expected performance parameters and the actual performance. In this paper, a method is presented to tune a physical model to a specific gas turbine by applying a data-driven approach to correct for the differences between the real gas turbine operation and the performance model prediction of the same. The first step in this process is to generate a surrogate model of the 1st principle performance model through the use of a neural network. A second “correction model” is then developed from selected operational data to correct the differences between the surrogate model and the real gas turbine. This corrects for the inaccuracies between the performance model and the real operation. The methodology is described and the results from its application to a heavy duty gas turbine are presented in this paper.

scholarly journals Methods for the Calculation of Thermoacoustic Stability Margins and Monte Carlo-Free Uncertainty Quantification

2017 ◽  
Author(s):  
Georg A. Mensah ◽  
Luca Magri ◽  
Jonas P. Moeck
Keyword(s):  
Monte Carlo ◽  
Factor Analysis ◽  
Risk Factor ◽  
Uncertainty Quantification ◽  
Frequency Domain ◽  
Gas Turbines ◽  
Network Models ◽  
System Stability ◽  
Risk Factor Analysis ◽  
Model Parameters

Thermoacoustic instabilities are a major threat for modern gas turbines. Frequency-domain based stability methods, such as network models and Helmholtz solvers, are common design tools because they are fast compared to compressible CFD computations. Frequency-domain approaches result in an eigenvalue problem, which is nonlinear with respect to the eigenvalue. Nonlinear functions of the frequency are, for example, the n–τ model, impedance boundary conditions, etc. Thus, the influence of the relevant parameters on mode stability is only given implicitly. Small changes in some model parameters, which are obtained by experiments with some uncertainty, may have a great impact on stability. The assessment of how parameter uncertainties propagate to system stability is therefore crucial for safe gas turbine operation. This question is addressed by uncertainty quantification. A common strategy for uncertainty quantification in thermoacoustics is risk factor analysis. It quantifies the uncertainty of a set of parameters in terms of the probability of a mode to become unstable. One general challenge regarding uncertainty quantification is the sheer number of uncertain parameter combinations to be quantified. For instance, uncertain parameters in an annular combustor might be the equivalence ratio, convection times, geometrical parameters, boundary impedances, flame response model parameters etc. Assessing also the influence of all possible combinations of these parameters on the risk factor is a numerically very costly task. A new and fast way to obtain algebraic parameter models in order to tackle the implicit nature of the eigenfrequency problem is using adjoint perturbation theory. Though adjoint perturbation methods were recently applied to accelerate the risk factor analysis, its potential to improve the theory has not yet been fully exploited. This paper aims to further utilize adjoint methods for the quantification of uncertainties. This analytical method avoids the usual random Monte Carlo simulations, making it particularly attractive for industrial purposes. Using network models and the open-source Helmholtz solver PyHoltz it is also discussed how to apply the method with standard modeling techniques. The theory is exemplified based on a simple ducted flame and a combustor of EM2C laboratory for which experimental validation is available.

Detection and Classification of Sensor Anomalies in Gas Turbine Field Data

2018 ◽  
Author(s):  
Giuseppe Fabio Ceschini ◽  
Lucrezia Manservigi ◽  
Giovanni Bechini ◽  
Mauro Venturini
Keyword(s):  
Anomaly Detection ◽  
Gas Turbine ◽  
Gas Turbines ◽  
Field Data ◽  
Time Correlation ◽  
Detection Algorithm ◽  
Anomaly Classification ◽  
Monitoring And Diagnostics ◽  
Faulty Sensors

Anomaly detection and classification is a key challenge for gas turbine monitoring and diagnostics. To this purpose, a comprehensive approach for Detection, Classification and Integrated Diagnostics of Gas Turbine Sensors (named DCIDS) was developed by the authors in previous papers. The methodology consists of an Anomaly Detection Algorithm (ADA) and an Anomaly Classification Algorithm (ACA). The ADA identifies anomalies according to three different levels of filtering. Anomalies are subsequently analyzed by the ACA to perform their classification, according to time correlation, magnitude and number of sensors in which an anomaly is contemporarily identified. The performance of the DCIDS approach is assessed in this paper based on a significant amount of field data taken on several Siemens gas turbines in operation. The field data refer to six different physical quantities, i.e. vibration, pressure, temperature, VGV position, lube oil tank level and rotational speed. The analyses carried out in this paper allow the detection and classification of the anomalies and provide some rules of thumb for field operation, with the final aim of identifying time occurrence and magnitude of faulty sensors and measurements.

scholarly journals Methods for the Calculation of Thermoacoustic Stability Boundaries and Monte Carlo-Free Uncertainty Quantification

2018 ◽  
Vol 140 (6) ◽  
Author(s):  
Georg A. Mensah ◽  
Luca Magri ◽  
Jonas P. Moeck
Keyword(s):  
Monte Carlo ◽  
Uncertainty Quantification ◽  
Gas Turbines ◽  
Network Models ◽  
System Stability ◽  
Model Parameters ◽  
Geometrical Parameters ◽  
Parameter Uncertainties ◽  
Flame Response ◽  
Mode Stability

Thermoacoustic instabilities are a major threat for modern gas turbines. Frequency-domain-based stability methods, such as network models and Helmholtz solvers, are common design tools because they are fast compared to compressible flow computations. They result in an eigenvalue problem, which is nonlinear with respect to the eigenvalue. Thus, the influence of the relevant parameters on mode stability is only given implicitly. Small changes in some model parameters, may have a great impact on stability. The assessment of how parameter uncertainties propagate to system stability is therefore crucial for safe gas turbine operation. This question is addressed by uncertainty quantification. A common strategy for uncertainty quantification in thermoacoustics is risk factor analysis. One general challenge regarding uncertainty quantification is the sheer number of uncertain parameter combinations to be quantified. For instance, uncertain parameters in an annular combustor might be the equivalence ratio, convection times, geometrical parameters, boundary impedances, flame response model parameters, etc. A new and fast way to obtain algebraic parameter models in order to tackle the implicit nature of the problem is using adjoint perturbation theory. This paper aims to further utilize adjoint methods for the quantification of uncertainties. This analytical method avoids the usual random Monte Carlo (MC) simulations, making it particularly attractive for industrial purposes. Using network models and the open-source Helmholtz solver PyHoltz, it is also discussed how to apply the method with standard modeling techniques. The theory is exemplified based on a simple ducted flame and a combustor of EM2C laboratory for which experimental data are available.

Using Diagnostics and Prognostics to Minimize the Cost of Ownership of Gas Turbines

2006 ◽  
Author(s):  
Hans R. DePold ◽  
Jason Siegel
Keyword(s):  
Value Chain ◽  
Gas Turbines ◽  
Failure Modes ◽  
Parametric Design ◽  
Health Management ◽  
System Component ◽  
Value Chain Analysis ◽  
Gas Generator ◽  
Material Condition ◽  
Cost Of Ownership

In general, health management technologies observe features associated with anomalous system behavior and relate these features to useful information about the system’s condition. In the case of prognostics, this information is then related to the expected condition at some future time. The ability to estimate the time to conditional or to mechanical failure is of great benefit in health management systems. Inherently probabilistic in nature, prognostics can be applied to system/component failure modes governed by material condition and by functional loss. Like diagnostic algorithms, prognostic algorithms tend to be generic in design but specific in application. Today, elements of turbine gas generator condition based maintenance, module and part life analysis, and soft removal times play essential roles in sustaining safe operations and effective equipment maintenance. When intelligently combined with value chain analysis they provide the decision support system needed to undertake the maintenance actions which minimize total cost of ownership. The methodologies and mathematical constructs for performing optimization require the system designer to clearly define a useful cost or objective function, which when minimized mathematically produces the parametric design combination that we call optimized. In the specific cases where parametric constraints exist, our optimized system typically will be found along those boundary conditions.

Sign in / Sign up

Export Citation Format

Share Document