Gas Turbine Health State Prognostics by Means of Bayesian Hierarchical Models

2019 ◽  
Vol 141 (11) ◽  
Author(s):  
Enzo Losi ◽  
Mauro Venturini ◽  
Lucrezia Manservigi
Keyword(s):  
Gas Turbine ◽  
Field Data ◽  
Health Management ◽  
Theoretical Background ◽  
Health State ◽  
Prediction Errors ◽  
Probabilistic Prediction ◽  
Bayesian Hierarchical ◽  
Prediction Reliability ◽  
Data Representative

Abstract The prediction of time evolution of gas turbine (GT) performance is an emerging requirement of modern prognostics and health management (PHM), aimed at improving system reliability and availability, while reducing life cycle costs. In this work, a data-driven Bayesian hierarchical model (BHM) is employed to perform a probabilistic prediction of GT future behavior, thanks to its capability to deal with fleet data from multiple units. First, the theoretical background of the predictive methodology is outlined to highlight the inference mechanism and data processing for estimating BHM-predicted outputs. Then, the BHM approach is applied to both simulated and field data representative of GT degradation to assess its prediction reliability and grasp some rules of thumb for minimizing BHM prediction error. For the considered field data, the average values of the prediction errors are found to be lower than 1.0% or 1.7% for single- or multi-step prediction, respectively.

Gas Turbine Health State Prognostics by Means of Bayesian Hierarchical Models

2019 ◽  
Author(s):  
Enzo Losi ◽  
Mauro Venturini ◽  
Lucrezia Manservigi
Keyword(s):  
Gas Turbine ◽  
Field Data ◽  
Health Management ◽  
Theoretical Background ◽  
Health State ◽  
Prediction Errors ◽  
Probabilistic Prediction ◽  
Future Health ◽  
Bayesian Hierarchical ◽  
Prediction Reliability

Abstract The prediction of the time evolution of gas turbine performance is an emerging requirement of modern prognostics and health management (PHM), aimed at improving system reliability and availability, while reducing life cycle costs. In this work, a data-driven Bayesian Hierarchical Model (BHM) is employed to perform a probabilistic prediction of gas turbine future health state thanks to its capability to deal with fleet data from multiple units. First, the theoretical background of the predictive methodology is outlined to highlight the inference mechanism and data processing for estimating BHM predicted outputs. Then, BHM is applied to both simulated and field data representative of gas turbine degradation to assess its prediction reliability and grasp some rules of thumb for minimizing BHM prediction error. For the considered field data, the average values of the prediction errors were found to be lower than 1.0 % or 1.7 % for single- or multi-step prediction, respectively.

Autoregressive Bayesian Hierarchical Model to Predict Gas Turbine Degradation

2020 ◽  
Author(s):  
Enzo Losi ◽  
Mauro Venturini ◽  
Lucrezia Manservigi
Keyword(s):  
Gas Turbine ◽  
Hierarchical Model ◽  
Power Output ◽  
Health Management ◽  
Bayesian Hierarchical Model ◽  
Single Step ◽  
Economic Losses ◽  
Prediction Errors ◽  
Bayesian Hierarchical ◽  
Degradation Rates

Abstract Gas turbine industry currently implements prognostic and health management systems as a fundamental task to predict the deteriorated characteristics of a gas turbine at future states and in turn plan maintenance actions. Thus, economic losses caused by system breakdowns and unnecessary repair actions can be reduced. In this work, a data-driven Bayesian Hierarchical Model (BHM) is implemented by means of an innovative autoregressive structure to predict gas turbine progressive deterioration. The novel autoregressive model provides an estimate of the output variable which depends on time and its previous values. In such a model, lagged values of the output are used as predictor variables. The autoregressive BHM, called ARBHM in this paper, is applied to highly heterogeneous field data taken from the literature, characterized by different degradation rates and referred to the power output of a large-size heavy duty gas turbine. The ARBHM tested in this paper includes up to a third-order lag and is compared to a BHM that only uses time as the regression variable. The comparison is carried out by performing both single-step prediction and multi-step prediction of power output. The results demonstrate that, in the considered degradation scenarios, the innovative ARBHM is usually preferable to BHM, since prediction errors decrease up to 2.0 % in the best case.

Prediction Reliability of a Statistical Methodology for Gas Turbine Prognostics

2012 ◽  
Vol 134 (10) ◽  
Author(s):  
Mauro Venturini ◽  
Nicola Puggina
Keyword(s):  
Gas Turbine ◽  
Correction Factor ◽  
Gas Turbines ◽  
Threshold Value ◽  
False Alarms ◽  
Prediction Errors ◽  
Past Behavior ◽  
Turbine Performance ◽  
Performance Deterioration ◽  
Prediction Reliability

The performance of gas turbines degrades over time and, as a consequence, a decrease in gas turbine performance parameters also occurs, so that they may fall below a given threshold value. Therefore, corrective maintenance actions are required to bring the system back to an acceptable operating condition. In today’s competitive market, the prognosis of the time evolution of system performance is also recommended, in such a manner as to take appropriate action before any serious malfunctioning has occurred and, as a consequence, to improve system reliability and availability. Successful prognostics should be as accurate as possible, because false alarms cause unnecessary maintenance and nonprofitable stops. For these reasons, a prognostic methodology, developed by the authors, is applied in this paper to assess its prediction reliability for several degradation scenarios typical of gas turbine performance deterioration. The methodology makes use of the Monte Carlo statistical method to provide, on the basis of the recordings of past behavior, a prediction of future availability, i.e., the probability that the considered machine or component can be found in the operational state at a given time in the future. The analyses carried out in this paper aim to assess the influence of the degradation scenario on methodology prediction reliability, as a function of a user-defined threshold and minimum value allowed for the parameter under consideration. A technique is also presented and discussed, in order to improve methodology prediction reliability by means a correction factor applied to the time points used for methodology calibration. The results presented in this paper show that, for all the considered degradation scenarios, the prediction error is lower than 4% (in most cases, it is even lower than 2%), if the availability is estimated for the next trend, while it is not higher than 12%, if the availability is estimated five trends ahead. The application of a proper correction factor allows the prediction errors after five trends to be reduced to approximately 5%.

Gas Turbine Accessory Health Management Evaluation Using a Hybrid Prognostic Approach

2008 ◽  
Author(s):  
Carl S. Byington ◽  
Matthew J. Watson ◽  
Sudarshan P. Bharadwaj
Keyword(s):  
Gas Turbine ◽  
Health Management ◽  
Health Assessment ◽  
Physical Models ◽  
Data Driven ◽  
Health State ◽  
Time To Failure ◽  
Detection Rates ◽  
Hybrid Techniques ◽  
Mission Success

The authors have developed model-based and data-driven techniques aimed at providing a more reliable health assessment of gas turbine engine accessory components, which have contributed to a significant number of events that compromise mission success and equipment availability in military aircraft. As part of this approach, a physical model is used to derive parameters indicative of component-specific faults. Statistical fault classifiers and evolutionary prognostics methods are then used to track these parameters and identify the most likely health state and time to failure for each component. This assessment is fused with the results of independent data-driven routines, which are also used to analyze dynamic signal response and detect faults that would be difficult to incorporate into physical models. The developed approach was demonstrated using an experimental setup representative of aircraft fuel and lubrication systems. Pump leakage, pump gear damage, and valve blockage were seeded on the setup, and the developed routines were trained with high-bandwidth experimental data. The approach produced wide separation between baseline and faulted cases, yielding negligible missed detection rates for moderate faults and reasonable missed detection rates for an incipient valve blockage fault. The demonstration produced a quantifiable estimate of achievable performance using the hybrid techniques.

Fault Detection and Identification by Gas Path Analysis Approach in Heavy Duty Gas Turbine

2006 ◽  
Author(s):  
Stefania Della Gatta ◽  
Paolo Adami
Keyword(s):  
Path Analysis ◽  
Gas Turbine ◽  
Diagnostic System ◽  
Theoretical Background ◽  
Critical Discussion ◽  
Diagnostic Approach ◽  
Health State ◽  
Heavy Duty ◽  
Fault Detection And Identification ◽  
Heavy Duty Gas Turbine

The possibility to get information about gas turbine “health” state is largely based on availability and reliability of operational data and on-line acquisition systems. However, further instruments are needed in order to deduce useful information in maintenance scheduling from actually measured data. In Gas Path Analysis approaches, a model to simulate the engine behavior is required. Furthermore, in order to individuate, locate and evaluate faulty conditions, a diagnostic approach needs to be developed and introduced. This paper presents a critical discussion of the problem to highlight the main requirements of a diagnostic approach. Furthermore, some procedures to verify the ability of a diagnostic tool can be obtained directly from the theoretical background of GPA. To demonstrate how these procedures work, an application case has been examined. An engine model has been specifically developed for the monitored heavy-duty gas turbine. It allows to calculate thermodynamics data and to identify performance parameters through a mathematical modeling process. The suitability of this model to be introduced in a diagnostic system has been investigated. An exhaustive description of the procedures and discussion of results are reported.

Prediction Reliability of a Statistical Methodology for Gas Turbine Prognostics

2012 ◽  
Author(s):  
Mauro Venturini ◽  
Nicola Puggina
Keyword(s):  
Gas Turbine ◽  
Correction Factor ◽  
Gas Turbines ◽  
Threshold Value ◽  
False Alarms ◽  
Prediction Errors ◽  
Past Behavior ◽  
Turbine Performance ◽  
Performance Deterioration ◽  
Prediction Reliability

The performance of gas turbines degrades over time and, as a consequence, a decrease in gas turbine performance parameters also occurs, so that they may fall below a given threshold value. Therefore, corrective maintenance actions are required to bring the system back to an acceptable operating condition. In today’s competitive market, the prognosis of the time evolution of system performance is also recommended, in such a manner as to take appropriate action before any serious malfunctioning has occurred and, as a consequence, to improve system reliability and availability. Successful prognostics should be as accurate as possible, because false alarms cause unnecessary maintenance and non-profitable stops. For these reasons, a prognostic methodology, developed by the authors, is applied in this paper to assess its prediction reliability for several degradation scenarios typical of gas turbine performance deterioration. The methodology makes use of the Monte Carlo statistical method to provide, on the basis of the recordings of past behavior, a prediction of future availability, i.e. the probability that the considered machine or component can be found in the operational state at a given time in the future. The analyses carried out in this paper aim to assess the influence of the degradation scenario on methodology prediction reliability, as a function of a user-defined threshold and minimum value allowed for the parameter under consideration. A technique is also presented and discussed, in order to improve methodology prediction reliability by means a correction factor applied to the time points used for methodology calibration. The results presented in this paper show that, for all the considered degradation scenarios, the prediction error is lower than 4% (in most cases, it is even lower than 2%), if the availability is estimated for the next trend, while it is not higher than 12%, if the availability is estimated five trends ahead. The application of a proper correction factor allows the prediction errors after five trends to be reduced to approximately 5%.

scholarly journals Online Health Management for Complex Nonlinear Systems Based on Hidden Semi-Markov Model Using Sequential Monte Carlo Methods

2012 ◽  
Vol 2012 ◽  
pp. 1-22
Author(s):  
Qinming Liu ◽  
Ming Dong
Keyword(s):  
Monte Carlo ◽  
Nonlinear System ◽  
Nonlinear Systems ◽  
Markov Model ◽  
Sequential Monte Carlo ◽  
Health Management ◽  
Health State ◽  
Health States ◽  
Complex Nonlinear System ◽  
Complex Nonlinear Systems

Health management for a complex nonlinear system is becoming more important for condition-based maintenance and minimizing the related risks and costs over its entire life. However, a complex nonlinear system often operates under dynamically operational and environmental conditions, and it subjects to high levels of uncertainty and unpredictability so that effective methods for online health management are still few now. This paper combines hidden semi-Markov model (HSMM) with sequential Monte Carlo (SMC) methods. HSMM is used to obtain the transition probabilities among health states and health state durations of a complex nonlinear system, while the SMC method is adopted to decrease the computational and space complexity, and describe the probability relationships between multiple health states and monitored observations of a complex nonlinear system. This paper proposes a novel method of multisteps ahead health recognition based on joint probability distribution for health management of a complex nonlinear system. Moreover, a new online health prognostic method is developed. A real case study is used to demonstrate the implementation and potential applications of the proposed methods for online health management of complex nonlinear systems.

scholarly journals Gas Turbine Health State Determination: Methodology Approach and Field Application

2012 ◽  
Vol 2012 ◽  
pp. 1-14 ◽  
Author(s):  
Michele Pinelli ◽  
Pier Ruggero Spina ◽  
Mauro Venturini
Keyword(s):  
Gas Turbine ◽  
Gas Turbines ◽  
Field Measurements ◽  
Field Application ◽  
Diagnostic Process ◽  
Health State ◽  
Gas Compression ◽  
State Determination ◽  
Measurement Validation ◽  
Machine Health

A reduction of gas turbine maintenance costs, together with the increase in machine availability and the reduction of management costs, is usually expected when gas turbine preventive maintenance is performed in parallel to on-condition maintenance. However, on-condition maintenance requires up-to-date knowledge of the machine health state. The gas turbine health state can be determined by means of Gas Path Analysis (GPA) techniques, which allow the calculation of machine health state indices, starting from measurements taken on the machine. Since the GPA technique makes use of field measurements, the reliability of the diagnostic process also depends on measurement reliability. In this paper, a comprehensive approach for both the measurement validation and health state determination of gas turbines is discussed, and its application to a 5 MW gas turbine working in a natural gas compression plant is presented.

Comparison of Linear and Non-Linear Gas Turbine Performance Diagnostics

2003 ◽  
Author(s):  
Ph. Kamboukos ◽  
K. Mathioudakis
Keyword(s):  
Gas Turbine ◽  
Condition Monitoring ◽  
Theoretical Background ◽  
Diagnostic Methods ◽  
Turbine Performance ◽  
Different Types ◽  
Non Linear ◽  
Computational Procedures ◽  
Monitoring And Diagnostics ◽  
Performance Diagnostics

The features of linear performance diagnostic methods are discussed, in comparison to methods based on full non-linear calculation of performance deviations, for the purpose of condition monitoring and diagnostics. First, the theoretical background of linear methods is overviewed to establish a relationship to the principles used by non-linear methods. Then computational procedures are discussed and compared. The effectiveness of determining component performance deviations by the two types of approaches is examined, on different types of diagnostic situations. A way of establishing criteria to define whether non-linear methods have to be employed is presented. An overall assessment of merits or weaknesses of the two types of methods is attempted, based on the results presented in the paper.

Health Monitoring and Degradation Prognostics in Gas Turbine Engines Using Dynamic Neural Networks

2015 ◽  
Author(s):  
A. Vatani ◽  
K. Khorasani ◽  
N. Meskin
Keyword(s):  
Neural Network ◽  
Gas Turbine ◽  
Health Monitoring ◽  
Network Architecture ◽  
Information Criterion ◽  
Turbine Engines ◽  
Prediction Errors ◽  
Gas Turbine Engines ◽  
Dynamic Neural Networks ◽  
Turbine Engine

In this paper two artificially intelligent methodologies are proposed and developed for degradation prognosis and health monitoring of gas turbine engines. Our objective is to predict the degradation trends by studying their effects on the engine measurable parameters, such as the temperature, at critical points of the gas turbine engine. The first prognostic scheme is based on a recurrent neural network (RNN) architecture. This architecture enables ONE to learn the engine degradations from the available measurable data. The second prognostic scheme is based on a nonlinear auto-regressive with exogenous input (NARX) neural network architecture. It is shown that this network can be trained with fewer data points and the prediction errors are lower as compared to the RNN architecture. To manage prognostic and prediction uncertainties upper and lower threshold bounds are defined and obtained. Various scenarios and case studies are presented to illustrate and demonstrate the effectiveness of our proposed neural network-based prognostic approaches. To evaluate and compare the prediction results between our two proposed neural network schemes, a metric known as the normalized Akaike information criterion (NAIC) is utilized. A smaller NAIC shows a better, a more accurate and a more effective prediction outcome. The NAIC values are obtained for each case and the networks are compared relatively with one another.

Sign in / Sign up

Export Citation Format

Share Document