Cycle humidification applied to micro Gas Turbines (mGTs) offers a solution to overcome their limited operational flexibility in terms of variable electrical and thermal power production when used in a Combined Heat and Power (CHP) application. Although the positive impact of this cycle humidification on the performance has already been proven numerically and experimentally, very detailed modeling of the system performance remains challenging, especially the determination of the recuperator effectiveness, which has the highest impact on the final cycle performance. Indeed, the recuperator performance depends strongly on the mass flow rate of the air stream and its humidification level, two parameters that are difficult to measure accurately. Accurate modeling of the recuperator performance under both dry and humidified conditions is thus essential for correct assessment of the potential of humidified mGT cycles in Decentralized Energy Systems (DES). In this paper, we present a detailed analysis of the recuperator performance under humidified conditions using averaged experimental data, extended with the application of a Support Vector Regression (SVR) on a time series to improve noise-modeling of the output signal, and thus enhance the accuracy of the monitoring process. In a first step, the missing experimental parameters, air mass flow rate and humidity level, were obtained indirectly, using rotational speed, fuel flow rate, exhaust gas composition and pressure level measurements in combination with the compressor map. Despite the low accuracy, some general trends regarding the recuperator performance could be observed based on these experimental data, indicating that the recuperator, despite having an increased total exchanged heat flux, is actually too small to exploit the full potential of the humidification. In a second step, by means of the SVR model, a first attempt was made to improve the accuracy and reduce the scatter on the recuperator performance determination. The predicted results with the SVR indicated indeed a reduced scatter on the determinations of the air mass flow rate and the amount of introduced water, opening a pathway towards online recuperator performance prediction.

The challenges related to current energy market force gas turbine owners to improve the reliability and availability of gas turbine engines, especially in the ever competitive market of the Oil & Gas sector. Gas turbine trip leads to business interruption and also reduces equipment remaining useful life. Thus, the identification of symptoms of trips is a key factor to predict their occurrence and avoid further damages and costs. Gas turbine transients are tracked by gas turbine operators while they occur, but a database including the complete details of past events for many fleets of engines is not always available. Therefore, a methodology aimed at classifying transients into clusters that identify the type of event (e.g., normal shutdown or trip) is required. Clustering is a data mining technique that addresses the scope of partitioning multi-variate time series into a given number of homogeneous and separated groups. In such a manner, the multi-variate time series belonging to the same cluster are very similar to each other and dissimilar to those of the other clusters. This paper presents a structured methodology composed of a subsequent matching algorithm, a featured-based clustering approach exploiting the unsupervised fuzzy C-means algorithm and a procedure that assigns a label to each cluster for classification purposes. The methodology is applied to a real-word case-study, by investigating transients acquired from a fleet of Siemens gas turbines in operation during three years. The results obtained by using heterogeneous datasets including six measured variables allowed values of Precision, Recall and Accuracy higher than 90 % in almost all cases.

This paper demonstrates that a Large Eddy Simulation (LES) combustion model based on tabulated chemistry and Eulerian stochastic fields can successfully describe the flame dynamics of a premixed turbulent swirl flame. The combustion chemistry is tabulated from one-dimensional burner-stabilized flamelet computations in dependence of progress variable and enthalpy. The progress variable allows to efficiently include a detailed reaction scheme, while the dependence on enthalpy describes the effect of heat losses on the reaction rate. The turbulence-chemistry interaction is modeled by eight Eulerian stochastic fields. A LES of a premixed swirl burner with a broadband velocity excitation is performed to investigate the flame dynamics, i.e. the response of heat release rate to upstream velocity perturbations. In particular, the flame impulse response and flame transfer function are identified from LES time series data. Simulation results for a range of power ratings are in good agreement with experimental data.

A robust and complete uncertainty estimation method is developed to quantify the uncertainty of turbulence quantities measured by hot-wire anemometry at the inlet of a short-duration turbine test rig. The uncertainty is categorized into two macro-uncertainty sources: the measurement related uncertainty (the uncertainty of each instantaneous velocity sample) and the uncertainty stemming from the statistical treatment of the time series. The former is addressed by the implementation of a Monte Carlo method. The latter, which is directly related to the duration of the acquired signal, is estimated using the moving block bootstrap method, a non-parametric resampling algorithm suitable for correlated time series. This methodology allows computing the confidence intervals of the spanwise distributions of mean velocity, turbulence intensity, length scales and other statistical moments at the inlet of the turbine test section.

Nowadays, gas turbines are equipped with an increasing number of sensors, of which the acquired data are used for monitoring and diagnostic purposes. Therefore, anomaly detection in sensor time series is a crucial aspect for raw data cleaning, in order to identify accurate and reliable data. To this purpose, a novel methodology based on Bayesian Hierarchical Models (BHMs) is proposed in this paper. The final aim is the exploitation of information held by a pool of observations from redundant sensors as knowledge base to generate statistically consistent measurements according to input data. In this manner, it is possible to simulate a “virtual” healthy sensor, also known as digital twin, to be used for sensor fault identification. The capability of the novel methodology based on BHM is assessed by using field data with two types of implanted faults, i.e. spikes and bias faults. The analyses consider different numbers of faulty sensors within the pool and different fault magnitudes, so that different levels of fault severity can be investigated. The results demonstrate that the new approach is successful in most fault scenarios for both spike and bias faults and provide guidelines to tune the detection criterion based on the morphology of the available data.

Operation and maintenance costs are a major driver for levelized cost of energy of wind power plants and can be reduced through optimized operation and maintenance practices accomplishable by various prognostics and health management (PHM) technologies. In recent years, the wind industry has become more open to adopting PHM solutions, especially those focusing on diagnostics. However, prognostics activities are, in general, still at the research and development stage. On the other hand, the industry has a request to estimate a component’s remaining useful life (RUL) when it has faulted, and this is a key output of prognostics. Systematically presenting PHM technologies to the wind industry by highlighting the RUL prediction need potentially helps speed up its acceptance and provides more benefits from PHM to the industry. In this paper, we introduce a PHM for wind framework. It highlights specifics unique to wind turbines and features integration of data and physics domain information and models. The output of the framework focuses on RUL prediction. To demonstrate its application, a data domain method for wind turbine gearbox fault diagnostics is presented. It uses supervisory control and data acquisition system time series data, normalizes gearbox temperature measurements with reference to environmental temperature and turbine power, and leverages big data analytics and machine-learning techniques to make the model scalable and the diagnostics process automatic. Another physics-domain modeling method for RUL prediction of wind turbine gearbox high-speed-stage bearings failed by axial cracks is also discussed. Bearing axial cracking has been shown to be the prevalent wind turbine gearbox failure mode experienced in the field and is different from rolling contact fatigue, which is targeted during the bearing design stage. The method uses probability of failure as a component reliability assessment and RUL prediction metric, which can be expanded to other drivetrain components or failure modes. The presented PHM for wind framework is generic and applicable to both land-based and offshore wind turbines.

This study is based on time-series data from the combined cycle utility gas turbines consisting of three-gas turbine units and one steam turbine unit. We construct a multi-stage vector autoregressive model for the nominal operation of powerplant assuming sparsity in the association among variables and use this as a basis for anomaly detection and prediction. This prediction is compared with the time-series data of the plant-operation containing anomalies. Granger causality networks, which are based on the associations between the time series streams, are learned as an important implication from the vector autoregressive modelling. Anomaly is detected by comparing the observed measurements against their predicted value.

This study assesses and compares two alternative approaches to determine the acoustic scattering matrix of a pre-mixed turbulent swirl combustor: 1) The acoustic scattering matrix coefficients are obtained directly from a compressible Large Eddy Simulation (LES). Specifically, the incoming and outgoing characteristic waves f and g extracted from the LES are used to determine the respective transmission and reflection coefficients via System Identification techniques. 2) The flame transfer function (FTF) is identified from LES time series data of upstream velocity and heat release rate. The transfer matrix of the reactive combustor is then derived by combining the FTF with the Rankine-Hugoniot relations across a compact heat source and a transfer matrix of the cold combustor, which is deduced from a linear network model. Linear algebraic transformation of the transfer matrix consequently yields the combustor scattering matrix. A cross-comparison study that includes comprehensive experimental data shows that both approaches successfully predict the scattering matrix of the reactive turbulent swirl combustor.

This paper studies the behavior of Industrial Gas Turbines (IGTs) based on time-series measurements with low sampling rates. The aim is to find the most suitable set of statistical/time-domain features derived from the measurements, which can represent the characteristic behavior of the IGTs, or alternatively, which can discriminate between different engines or different states of an engine. For this end, a scheme of optimal feature selection process is proposed in the paper. For cross-fleet analysis, signals from a group of inter-duct thermocouples on IGT engines are studied. A feature matrix is formulated at each sliding time step, by calculating the statistical features of the sensor group, after the time-domain features of the individual sensor measurements are calculated. Feature matrix values from different engines are then clustered, and a modified Davies–Bouldin index is introduced to measure the quality of the clusters. Finally, grid search is run to find the optimal set of the features, which form the clusters with the most similarity, or otherwise, the most discrepancy across the IGT engines. The window size effect is also investigated. To demonstrate that the optimal feature selection process is also useful for fault diagnosis of IGTs, the proposed scheme is then applied on a group of different measurements on an IGT, i.e. from burner tip thermocouples, in a fault diagnostic scenario, which is subsequently validated using a k-nearest neighbor classification algorithm. The case studies have demonstrated that, ultimately, the developed techniques can be broadly applied to other groups of measurements for both cross-fleet analysis and fault diagnosis of IGTs.

This paper addresses the challenge of forecasting the future values of gas turbine measureable quantities. The final aim is the simulation of “virtual sensors” capable of producing statistically coherent measurements aimed at replacing anomalous observations discarded from the time series. Among the different available approaches, the Bayesian Forecasting Method (BFM) adopted in this paper uses the information held by a pool of observations as knowledge base to forecast the values at a future state. The BFM algorithm is applied in this paper to Siemens field data to assess its prediction capability, by considering two different approaches, i.e. single-step prediction (SSP) and multi-step prediction (MSP). While SSP predicts the next observation by using true data as base of knowledge, MSP uses previously predicted data as base of knowledge to perform the prediction of future time steps. The results show that BFM single-step average prediction error can be very low, when filtered field data are analyzed. On the contrary, the average prediction error achieved in case of BFM multi-step prediction is remarkably higher. To overcome this issue, the BFM single-step prediction scheme is also applied to clusters of time-wise averaged data. In this manner, an acceptable average prediction error can be achieved by considering clusters composed of 60 observations.

The thermoacoustic behavior of a combustion system can be determined numerically via acoustic tools such as Helmholtz solvers or network models coupled with a model for the flame dynamic response. Within such a framework, the flame response to flow perturbations can be described by a Finite Impulse Response (FIR) model, which can be derived from LES time series via system identification. However, the estimated FIR model will inevitably contain uncertainties due to e.g., the statistical nature of the identification process, low signal-to-noise ratio or finite length of time series. Thus, a necessary step towards reliable thermoacoustic stability analysis is to quantify the impact of uncertainties in FIR model on the growth rate of thermoacoustic modes. There are two practical considerations involved in this topic. First, how to efficiently propagate uncertainties from the FIR model to the modal growth rate of the system, considering it is a high dimensional uncertainty quantification (UQ) problem? Second, since longer CFD simulation time generally leads to less uncertain FIR model identification, how to determine the length of the CFD simulation required to obtain satisfactory confidence? To address the two issues, a dimensional reduction UQ methodology called “Active Subspace approach” is employed in the present study. For the first question, Active Subspace approach is applied to exploit a low-dimensional approximation of the original system, which allows accelerated UQ analysis. Good agreement with Monte Carlo analysis demonstrates the accuracy of the method. For the second question, a procedure based on Active Subspace approach is proposed, which can serve as an indicator for terminating CFD simulation. The effectiveness of the procedure is verified in the paper.

Maintenance costs are a substantial contributor to airline operating costs. In this context, understanding, analyzing, and predicting engine performance deterioration is crucial. While diagnostic methods for analyzing the current module and overall engine condition are established in state-of-the-art engine condition monitoring (ECM) systems, deterioration modeling and prognosis are still fields of research. The identification of the contribution of deterioration mechanisms, such as fouling, erosion, and abrasion, to the in-service deterioration poses a key challenge in this area. This paper focuses on a top-down approach for the high pressure compressor (HPC) module. The selected approach is to quantify the contribution of individual deterioration mechanisms to the overall HPC efficiency deterioration based on in-flight measurements. This is accomplished by first using the in-flight measurements to analyze the HPC efficiency loss. Then, the resulting time series of the analyzed HPC efficiency loss are preprocessed. Finally, models of the deterioration mechanisms are fitted to the preprocessed time series. The deterioration models are chosen based on literature references to the respective deterioration mechanisms. As multiple influencing factors affect the deterioration mechanisms, a fleet analysis is conducted to select the model inputs. The fitting process involves a parametric nonlinear regression problem. The outcome is an estimation of the evolution of the deterioration mechanisms over time. This methodology is used to evaluate all available in-service engines of the same type and fleet and to define a fleet model. In the final step, benefits and limitations of the fleet model are investigated.

It is necessary to quantitatively resolve the flow field generated in compressors at various operating conditions including the surge. In this study, fast-responding pressure-sensitive-paint (PSP) and temperature-sensitive-paint (TSP) was applied to measurement of steady and unsteady pressure and temperature fields in a turbocharger compressor. The surface pressure of a compressor impeller and the diffuser downstream of the impeller was measured under the rotating speed of 133,000 rpm with various mass flow rate of the inlet. Blue LED strobes were used as the excitation light source for PSP/TSP. The luminescence from PSP/TSP were captured by a high-speed camera. These instruments were controlled using the signal synchronized with a fixed rotation angle of the impeller, which was detected using a gap sensor installed near the impeller. The pressure was calculated from the time-series of the PSP signal. The output of PSP/TSP was compared with the signal from semiconductor pressure transducers and the thermal image taken by an infrared camera. For the analysis of the diffuser, the power spectra of the pressure field were obtained using the fast Fourier transform of the time-series data of the pressure fluctuation. The pressure field on the impeller was successfully obtained with the aid of adequate temperature compensation by TSP. It was confirmed that the time-averaged pressure on the impeller blades increased as the inlet mass flow rate was decreased. Large-amplitude points were distributed on the extension line of the impeller blades in the diffuser when the surge occurs.

Statistical parametric methodologies are widely employed in the analysis of time series of gas turbine sensor readings. These methodologies identify outliers as a consequence of excessive deviation from a statistically-based model, derived from available observations. Among parametric techniques, the k-σ methodology demonstrates its effectiveness in the analysis of stationary time series. Furthermore, the simplicity and the clarity of this approach justify its direct application to industry. On the other hand, the k-σ methodology usually proves to be unable to adapt to dynamic time series, since it identifies observations in a transient as outliers. As this limitation is caused by the nature of the methodology itself, two improved approaches are considered in this paper in addition to the standard k-σ methodology. The two proposed methodologies maintain the same rejection rule of the standard k-σ methodology, but differ in the portions of the time series from which statistical parameters (mean and standard deviation) are inferred. The first approach performs statistical inference by considering all observations prior to the current one, which are assumed reliable, plus a forward window containing a specified number of future observations. The second approach proposed in this paper is based on a moving window scheme. Simulated data are used to tune the parameters of the proposed improved methodologies and to prove their effectiveness in adapting to dynamic time series. The moving window approach is found to be the best on simulated data in terms of True Positive Rate ( TPR ), False Negative Rate ( FNR ) and False Positive Rate ( FPR ). Therefore, the performance of the moving window approach is further assessed towards both different simulated scenarios and field data taken on a gas turbine.

The reliability of gas turbine health state monitoring and forecasting depends on the quality of sensor measurements directly taken from the unit. Outlier detection techniques have acquired a major importance, as they are capable of removing anomalous measurements and improve data quality. To this purpose, statistical parametric methodologies are widely employed thanks to the limited knowledge of the specific unit required to perform the analysis. The backward and forward moving window (BFMW) k-σ methodology proved its effectiveness in a previous study performed by the authors, to also manage dynamic time series, i.e. during a transient. However, the estimators used by the k-σ methodology are usually characterized by low statistical robustness and resistance. This paper aims at evaluating the benefits of implementing robust statistical estimators for the BFMW framework. Three different approaches are considered in this paper. The first methodology, k-MAD, replaces mean and standard deviation of the k-σ methodology with median and mean absolute deviation (MAD), respectively. The second methodology, σ-MAD, is a novel hybrid scheme combining the k-σ and the k-MAD methodologies for the backward and the forward windows, respectively. Finally, the bi-weight methodology implements bi-weight mean and bi-weight standard deviation as location and dispersion estimators. First, the parameters of these methodologies are tuned and the respective performance is compared by means of simulated data. Different scenarios are considered to evaluate statistical efficiency, robustness and resistance. Subsequently, the performance of these methodologies is further investigated by injecting outliers in field data sets taken on selected Siemens gas turbines. Results prove that all the investigated methodologies are suitable for outlier identification. Advantages and drawbacks of each methodology allow the identification of different scenarios in which their application can be most effective.

Anomaly detection in sensor time series is a crucial aspect for raw data cleaning in gas turbine industry. In addition to efficiency, a successful methodology for industrial applications should be also characterized by ease of implementation and operation. To this purpose, a comprehensive and straightforward approach for Detection, Classification and Integrated Diagnostics of Gas Turbine Sensors (named DCIDS) is proposed in this paper. The tool consists of two main algorithms, i.e. the Anomaly Detection Algorithm (ADA) and the Anomaly Classification Algorithm (ACA). The ADA identifies anomalies according to three different levels of filtering based on gross physics threshold application, inter-sensor statistical analysis (sensor voting) and single-sensor statistical analysis. Anomalies in the time series are identified by the ADA, together with their characteristics, which are analyzed by the ACA to perform their classification. Fault classes discriminate among anomalies according to their time correlation, magnitude and number of sensors in which an anomaly is contemporarily identified. Results of anomaly identification and classification can subsequently be used for sensor diagnostic purposes. The performance of the tool is assessed in this paper by analyzing two temperature time series with redundant sensors taken on a Siemens gas turbine in operation. The results show that the DICDS is able to identify and classify different types of anomalies. In particular, in the first dataset, two severely incoherent sensors are identified and their anomalies are correctly classified. In the second dataset, the DCIDS tool proves to be capable of identifying and classifying clustered spikes of different magnitudes.

This paper documents the set-up and validation of nonlinear autoregressive exogenous (NARX) models of a heavy-duty single-shaft gas turbine. The considered gas turbine is a General Electric PG 9351FA located in Italy. The data used for model training are time series data sets of several different maneuvers taken experimentally during the start-up procedure and refer to cold, warm and hot start-up. The trained NARX models are used to predict other experimental data sets and comparisons are made among the outputs of the models and the corresponding measured data. Therefore, this paper addresses the challenge of setting up robust and reliable NARX models, by means of a sound selection of training data sets and a sensitivity analysis on the number of neurons. Moreover, a new performance function for the training process is defined to weigh more the most rapid transients. The final aim of this paper is the set-up of a powerful, easy-to-build and very accurate simulation tool which can be used for both control logic tuning and gas turbine diagnostics, characterized by good generalization capability.

Centrifugal compressor is a piece of key equipment for factories. Among the components of centrifugal compressor, impeller is a pivotal part as it is used to transform kinetic energy to pressure energy. But it usually leads to blade crack or failure as irregular aerodynamic load effect on the blade. Therefore, early crack feature extraction and pattern recognition is important to prevent it from failure. Although time series analysis for monitored signal can be used on feature extraction, incipient weak feature extraction method should be investigated. In this research, pressure pulsation sensors arranged in close vicinity to crack area are used to monitor the blade crack and feature extraction. As there are different kinds of flow interference, the pressure pulsation signal for centrifugal compressor is full of nonlinear characteristics. Therefore, how to obtain the weak information from monitored signal is investigated. Although FFT and envelope analysis have been widely used for rotating equipment, they are not suitable for the determination of incipient crack of a blade as the signal modulation and noise interference. In this research, stochastic resonance is used for the pressure pulsation signal. The results show that it is an effective tool to blade incipient crack classification on centrifugal compressor.

The present work aims at investigating a new methodology developed at Ansaldo Energia, for the transient finite element modelling of the whole engine with an axisymmetric approach. The strong coupling and non linearity in the heat transfer process during transient thermal analyses are handled by a partly coupled scheme. The 2D axisymmetric finite element model includes a dedicated thermal fluid network where fluid-metal temperatures are computed. In the overall procedure the selected finite element solver is a customized version of CalculiX ® , while mass flow rates and pressure distributions in each thermal fluid network element are provided by external fluid network solvers in terms of customized time series. This paper represents a first insight about a fully integrated WEM (Whole Engine Modelling) procedure currently under development. Geometrical changes during operation, lead to different fluid properties affecting heat transfer coefficients too. These modified conditions in their turn impact the material temperature and displacements. The future implementation steps will be oriented on the adoption of a customized version of the native CalculiX ® fluid network solver with the aim of developing a fully integrated procedure able to take into account the interaction between the secondary air system and the modifications in the clearances and gaps due to the thermal and mechanical loads. In this paper, a detailed description of the procedure will be reported with comprehensive discussions about some fundamental modelling aspects. Preliminary results, related to the first application of the new methodology to the transient thermal modelling of a simplified test case representative of real engine geometries, will be presented.

For the evaluation and improvement of fan noise reduction notions and the verification of broadband sound power measurement in flow ducts, special interest was given to the analysis of the broadband noise fields in duct. Two axial sensor arrays were wall-flush mounted upstream of a single-stage axial fan in the circular duct section of the fan inlet, staggered by 180° in the circumferential direction. During the tests, the rotating drum was rotated by step of 6 degrees to give a total 840 measurement points. There are two static reference microphones mounted upstream of rotating measurement section. Time-series of 12 seconds duration were recorded at each of the 30 azimuthal positions of the rotating duct in order to allow for a statistically accurate cross-correlation data analysis. This method is able to discriminate the sound waves propagating in upstream and downstream directions. Special attention was given to the blade passing frequencies of the axial fan. Rotor-stator interaction mode dominates the incident sound field, while modes with low azimuthal order play an important role in the reflected sound field. The reflected broadband sound power is almost 10dB lower than the transmitted sound field in a broad-frequency range. On the whole, this method behaves robustly in decomposition of broadband noise in flow duct and delivers physically meaningful broadband mode amplitudes.