Verification of a Neural Network Based Predictive Emission Monitoring Module for an RB211-24C Gas Turbine

2009 ◽  
Author(s):  
K. K. Botros ◽  
C. Selinger ◽  
L. Siarkowski
Keyword(s):  
Neural Network ◽  
Gas Turbine ◽  
Operating Conditions ◽  
Pipeline System ◽  
Emission Data ◽  
Emission Monitoring ◽  
Different Seasons ◽  
Input Variables ◽  
Monitoring Module ◽  
Good Agreement

This paper presents a verification of a Predictive Emission Monitoring (PEM) model developed for a non-DLE RB211-24C gas turbine used at a natural gas compressor station on the TransCanada Pipeline System in Alberta, Canada. The basis and methodology of the PEM model is first described, and its predictions were compared to recent Continuous Emission Monitoring (CEM) data obtained at different engine load conditions varying from 10 to 19 MW (site condition). The PEM model is based on an optimized Neural Network (NN) architecture which takes 6 fundamental engine parameters as input variables. The model predicts NOx (dry) as an output variable. The NN was trained using CEM measurements comprising four sets of actual emission data collected over four different dates in four different seasons during 2000, and at different operating conditions covering the range of the engine operating parameters. The PEM model was then implemented in the station Compressor Equipment Health Monitoring (CEHM) system and NOx predictions were reported online on a minutely basis for several months and NOx emission trends were captured and analyzed. Comparison between predictions and stack measurements shows a fairly good agreement between the PEM and CEM data within ±10 ppm (dry).

Neural Network Based Predictive Emission Monitoring Module for a GE LM2500 Gas Turbine

2010 ◽  
Author(s):  
K. K. Botros ◽  
M. Cheung
Keyword(s):  
Neural Network ◽  
Gas Turbine ◽  
Operating Conditions ◽  
Training Data ◽  
Pipeline System ◽  
Emission Data ◽  
Ambient Temperatures ◽  
Emission Monitoring ◽  
Data Points ◽  
Monitoring Module

A Predictive Emission Monitoring (PEM) model has been developed for a non-DLE GE LM2500 gas turbine used on a natural gas compressor station on the TransCanada Pipeline System in Alberta. The PEM model is based on an optimized Neural Network (NN) architecture which takes four fundamental engine parameters as input variables. The model predicts NOx emission in ppmv-dry-O2 corrected and in kg/hr as NO2. The NN was trained using Continuous Emission Monitoring (CEM) measurements comprising two sets of actual emission data collected over two different dates in 2009, when the ambient ambient temperatures were vastly different (∼1° C and 24 °C), respectively. These training data were supplemented by other emission data generated by GE ‘Cycle-Deck’ tool to generate emission data at different ambient temperatures ranging from −30 to +30 °C. The outcome is a total of 1872 emission data of engine emissions at different operating conditions covering the range of the engine operating parameters (402 data points from CEM and 1470 data points from GE Cycle-Deck). The PEM model comprises a simple single hidden layer perceptron type NN with only two neurons in it. The performance of the NN-based model showed a correlation coefficient greater than 0.99, and error standard deviation of 4.5 ppmv of NOx and 1.4 kg/hr as NO2. Uncertainty analysis was conducted to assess the effects of uncertainties in the engine parameters on the NOx predictions by PEM. It was shown that for uncertainty in the ambient temperature of ±1 °C, the uncertainty in the NOx prediction is ± 0.9 to ±3.5%. Uncertainties of the order of ±1% in the other three input parameters results in uncertainties in NOx predictions by ±2.5 to ±6%. Finally, the PEM model was implemented in the station CEHM (Compressor Equipment Health Monitoring) system and NOx prediction were reported online on a minutely basis. These data are presented here over the first three months since implementation.

Coupled CFD-ANN Procedure for Extending Heat Transfer Correlations Out of Their Range of Validity

2012 ◽  
Author(s):  
Andrea Viano ◽  
Gabriele Ottino ◽  
Luca Ratto ◽  
Giuseppe Spataro
Keyword(s):  
Neural Network ◽  
Heat Transfer ◽  
Experimental Data ◽  
Nusselt Number ◽  
Gas Turbine ◽  
Training Session ◽  
Nonlinear Problems ◽  
Artificial Neural ◽  
Cfd Analyses ◽  
Good Agreement

The heat transfer coefficient and pressure losses are among the main parameters to be evaluated in gas turbine cooling network design. Due to the complexity of these estimates, correlation-based computations are typically used as a result of time-consuming and expensive experimental activities. One of the main problems that the industry has to face is that these correlations, based on non-dimensional experimental data, produce reliable results in a range of validity typically different from that encountered in gas turbine applications. This paper will present preliminary results of an innovative procedure based on CFD analyses and Artificial Neural Networks, able to extend correlation predictions out of their range of validity, without any additional experimental data. Well-known test cases were replicated by building corresponding CAD geometries which were discretized by means of appropriate meshes, resulting from grid-independence studies. CFD analyses, based on the RANS approach, were performed to overlay the computations of the Nusselt number obtained from experimental activities. A preliminary comparison among turbulence models was carried out to find one leading to a good agreement with the experimental data. Then, an optimization method, based on Evolutionary Algorithms, was applied to the CFD analyses in order to find the best set of constant values for the chosen turbulence model, leading to the most accurate prediction of the experimental dataset. The resulting ad hoc CFD model was adopted in order to analyse test case configurations characterized by parameters within and external to the correlation validity field, building a sufficiently wide feeding database. A feed-forward multi-layer neural network was selected among network architectures typically used in engineering applications for prediction analyses. ANNs were chosen because they enable the solution of these complex nonlinear problems by using simple computational operations. The selected Artificial Neural Network was trained by a back-propagation procedure on the CFD results regarding Nusselt number. The validation of the resulting ANN was performed comparing its outputs with experimental data external to the correlation range of validity, which had not been used in the training session. Good agreement has been found. Results are presented and discussed.

Predictive Emission Monitoring Model for LM1600 Gas Turbines Based on Neural Network Architecture Trained on Field Measurements and CFD Data

2001 ◽  
Author(s):  
K. K. Botros ◽  
G. R. Price ◽  
G. Kibrya
Keyword(s):  
Neural Network ◽  
Ambient Temperature ◽  
Gas Turbines ◽  
Network Architecture ◽  
Absolute Error ◽  
Operating Conditions ◽  
Maximum Absolute Error ◽  
Flamelet Model ◽  
Emission Monitoring ◽  
One Year

A Predictive Emission Monitoring (PEM) model has been developed based on an optimized Neural Network (NN) architecture which takes 8 fundamental parameters as input variables. The model predicts both NO and NOx as output variables. The NN is initially trained using a combination of two sets of data: a) measured data at various loads from an LM1600 gas turbine installed at one of the compressor stations on TransCanada Transmission system in Alberta, Canada, b) data generated by a Computational Fluid Dynamics (CFD) at different operating conditions covering the range of the engine operating parameters spanned over one year. The predictions of NOx by CFD employed the ‘flamelet’ model and a set of 8 reactions including the Zeldovich mechanism for thermal NOx along with an empirical correlation for prompt NOx formation. It was found that a Multi Layer Perceptron type Neural Network with two hidden layers was the optimum architecture for predicting NO levels with a maximum absolute error of around 7%, mean absolute error of 2.3% and standard deviation of 1.97%. The model is easy to implement on the station PLC. A set of one year data consisting of 2804 cases was submitted to the above optimized NN architecture with varying ambient temperature from –29.9 °C to 35.7 °C and output power from 570 kW to 16.955 MW. This gave consistent contours of NO levels. As expected, NN architecture shows that NO increases with increasing power or ambient temperature.

A Neural Network Based Predictive Emission Monitoring Model for NOx Emission From a Gas Turbine Combustor

2002 ◽  
Author(s):  
G. Kibrya ◽  
K. K. Botros
Keyword(s):  
Neural Network ◽  
Ambient Temperature ◽  
Gas Turbine ◽  
Power Output ◽  
Field Tests ◽  
Absolute Error ◽  
Nox Emission ◽  
Cfd Model ◽  
Gas Turbine Combustor ◽  
Emission Monitoring

A Predictive Emission Monitoring (PEM) model for predicting NOx emission from a gas turbine combustor has been developed by employing an optimized Neural Network (NN) architecture. The Neural Network was trained by using actual field test data and predicted results of a Computational Fluid Dynamics (CFD) model of the combustor. The field tests were performed at a natural gas compressor station driven by a General Electric (GE) LM1600 conventional gas turbine. The model takes eight fundamental parameters (operating and ambient) as input, and predicts NO and NOx as outputs. The data used for training the model covers the entire operating ranges of power and ambient temperature for the site. The CFD model employs a non-equilibrium (flamelet) combustion scheme and a set of 8 reactions including the Zeldovich mechanism for thermal NOx, and an empirical correlation for prompt NOx formation. The results predicted by the CFD model were within 15% of the measured values. Results of the field tests demonstrated that the spool speed ratio of the gas turbine remained constant throughout the tests, the power output of the engine was linearly proportional to the spool speeds, and the NOx emission was proportional to the site power output. A Multi Layer Perceptron type Neural Network with two hidden layers, each with four neurons was found to be the optimum architecture for the model. The NO levels predicted by the PEM model based on the optimized NN had a maximum absolute error of approximately 7%, mean absolute error of 2.3% and standard deviation of 1.97%. One year operating data for the site was submitted to the trained NN model with ambient temperatures varying from −29.9 °C to 35.7 °C and output powers from 5.8 MW to 17 MW. It was found that the model produced consistent contours of NO emissions. As expected, the NO levels were found to increase with increasing power and/or ambient temperature.

A Neural Network Simulator of a Gas Turbine With a Waste Heat Recovery Section

2000 ◽  
Author(s):  
C. Boccaletti ◽  
G. Cerri ◽  
B. Seyedan
Keyword(s):  
Neural Network ◽  
Gas Turbine ◽  
Waste Heat ◽  
Direct Method ◽  
Computing Time ◽  
Operating Conditions ◽  
Network Simulator ◽  
Back Propagation Algorithm ◽  
Waste Heat Boiler ◽  
Computing Technique

The objective of the paper is to assess the feasibility of the neural network (NN) approach in power plant process evaluations. A “feedforward” technique with a back propagation algorithm was applied to a gas turbine equipped with waste heat boiler and water heater. Data from physical or empirical simulators of plant components were used to train such a NN model. Results obtained using a conventional computing technique are compared with those of the direct method based on a NN approach. The NN simulator was able to perform calculations in a really short computing time with a high degree of accuracy, predicting various steady-state operating conditions on the basis of inputs that can be easily obtained with existing plant instrumentation. The optimization of NN parameters like number of hidden neurons, training sample size and learning rate is discussed in the paper.

Application of a 1-D Predicting Tool to the Analysis of Pressure Oscillations in an Industrial Gas Turbine Combustor: GE10 Machine

2007 ◽  
Author(s):  
Antonio Asti ◽  
Luca Mangani ◽  
Antonio Andreini
Keyword(s):  
Gas Turbine ◽  
Gas Turbines ◽  
Operating Conditions ◽  
Equivalent Model ◽  
Design Phase ◽  
Trial And Error ◽  
Combustion Instabilities ◽  
Pressure Oscillations ◽  
Lean Premixed ◽  
Good Agreement

The development of current industrial gas turbines is strictly constrained by legislative requirements for low polluting emissions. Lean Premixed combustion technology has become through the years the necessary standard to meet such requirements. Premixed technology introduces a new range of problems: combustion instabilities in many operating conditions. Specifically, lean premixed flames pose the threat of pressure oscillations. This phenomenon is the effect of the strong interaction between combustion heat-release and fluid dynamics aspects. The prediction of acoustic oscillations and combustion instabilities is generally difficult because of the complexity of real combustor geometries. As a result, the design phase is usually performed as a trial-and-error task: a specific design is constructed, tested and modified, in a process that continues until acceptable results are found. A specific tool was developed by GE Energy to help predicting the acoustic behaviour of newly designed partially-premixed combustors, avoiding the traditional trial-and-error process: the tool allows the designer to analyze the problem of combustion instabilities since the early design phase, limiting subsequent testing efforts. A mono-dimensional tool based on the 1-D acoustic model was developed by GE Energy and was applied to the single-can combustor of the GE10 machine (a gas turbine in the 10MW class). All the main geometrical features of the GE10 machine, including fuel line geometry, were considered and modeled in a one-dimensional scheme, in order to build an equivalent model for the linear tool analysis. The main frequencies, measured during tests on the GE10 machine, were compared to the numerical results of the tool, showing good agreement between numerical and experimental results and confirming the predictive capability. This good agreement demonstrates that the model can be used for predicting the effects of design changes, with a reduced need of tests.

A Neural Network Simulator of a Gas Turbine With a Waste Heat Recovery Section

2001 ◽  
Vol 123 (2) ◽  
pp. 371-376 ◽  
Author(s):  
C. Boccaletti ◽  
G. Cerri ◽  
B. Seyedan
Keyword(s):  
Neural Network ◽  
Gas Turbine ◽  
Waste Heat ◽  
Direct Method ◽  
Computing Time ◽  
Operating Conditions ◽  
Network Simulator ◽  
Back Propagation Algorithm ◽  
Waste Heat Boiler ◽  
Computing Technique

The objective of the paper is to assess the feasibility of the neural network (NN) approach in power plant process evaluations. A “feed-forward” technique with a back propagation algorithm was applied to a gas turbine equipped with waste heat boiler and water heater. Data from physical or empirical simulators of plant components were used to train such a NN model. Results obtained using a conventional computing technique are compared with those of the direct method based on a NN approach. The NN simulator was able to perform calculations in a really short computing time with a high degree of accuracy, predicting various steady-state operating conditions on the basis of inputs that can be easily obtained with existing plant instrumentation. The optimization of NN parameters like number of hidden neurons, training sample size, and learning rate is discussed in the paper.

scholarly journals Artificial Neural Network based Process History Data Model for Gas Turbine Compressor Systems

2019 ◽  
Vol 8 (4) ◽  
pp. 5069-5077
Keyword(s):  
Neural Network ◽  
Artificial Neural Network ◽  
Electric Power ◽  
Gas Turbine ◽  
Operating Conditions ◽  
System Modelling ◽  
History Data ◽  
Artificial Neural ◽  
Gas Turbine Compressor ◽  
Compressor Efficiency

Gas turbine-based power plants are found to play a vital role in electric power generation and act as spinning reserves for renewable electric power. A robust performance assessment tool is inevitable for a gas turbine system to maintain high operational flexibility, availability, and reliability at different operating conditions. A suitable simulation model of the gas turbine provides detailed information about the system operation under varying ambient and load conditions. This paper illustrates a systematic methodology for process history data-based modelling of a gas turbine compressor system. The ReliefF feature selection method is applied for the proper identification of the parameters influencing the compressor efficiency. Appropriate Artificial Neural Network (ANN) based models are developed for data classification and system modelling of the compressor. The model performance has been validated using actual plant operational data, and the standard deviation of the error in model output was found to be 0.38. A novel approach for suitable integration of data processing methods, machine learning tools and gas turbine domain knowledge has led to the development of a robust compressor model. The model has been utilized for the health assessment of an existing gas turbine compressor, demonstrated through an illustrative case study. The model has been found suitable for parametric analysis of compressor efficiency with operating hours, which is helpful for operational decision-making involving studies on the influence of part-load operation, compressor wash planning, maintenance planning etc

Fundamental Combustion Calculations With DARS and CHEMKIN™ Software Packages: A Comparison

2013 ◽  
Author(s):  
Venkata Nori ◽  
Fang Xu
Keyword(s):  
Gas Turbine ◽  
Reaction Mechanisms ◽  
Ignition Delay ◽  
Flame Temperature ◽  
Operating Conditions ◽  
Combustion Chemistry ◽  
Software Packages ◽  
Extinction Strain Rate ◽  
Speed Accuracy ◽  
Good Agreement

An investigation was conducted within Honeywell to assess the speed, accuracy, and robustness of DARS for performing basic combustion calculations. In this study, the results for the fundamental combustion characteristics from DARS are compared with those obtained by CHEMKIN, the most well known and widely used software for combustion chemistry calculations. These characteristics include the adiabatic flame temperature, flame speed, ignition delay and extinction strain rate. The operating conditions (pressure and temperature) are chosen that are relevant to aircraft gas turbine combustors. For the ignition delay studies, Jet-A is used whereas Methane is the fuel for the other simulations. Well validated reaction mechanisms for Jet-A and Methane are used for the study. Temperature and species values (profiles) are presented and compared between DARS and CHEMKIN, wherever applicable. In general, very good agreement has been found between the results of DARS and CHEMKIN. The features provided by both tools with an emphasis on gas turbine combustion application were assessed. In conclusion, it was found that DARS has equivalent accuracy and capability as CHEMKIN for this type of application.

Adsorption Isotherm Predictions for Multiple Molecules in MOFs Using the Same Deep Learning Model

2019 ◽  
Author(s):  
Ryther Anderson ◽  
Achay Biong ◽  
Diego Gómez-Gualdrón
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Molecular Simulation ◽  
Large Scale ◽  
Learning Model ◽  
Operating Conditions ◽  
Small Subset ◽  
Screening Methods ◽  
Large Set ◽  
Metal Organic

<div>Tailoring the structure and chemistry of metal-organic frameworks (MOFs) enables the manipulation of their adsorption properties to suit specific energy and environmental applications. As there are millions of possible MOFs (with tens of thousands already synthesized), molecular simulation, such as grand canonical Monte Carlo (GCMC), has frequently been used to rapidly evaluate the adsorption performance of a large set of MOFs. This allows subsequent experiments to focus only on a small subset of the most promising MOFs. In many instances, however, even molecular simulation becomes prohibitively time consuming, underscoring the need for alternative screening methods, such as machine learning, to precede molecular simulation efforts. In this study, as a proof of concept, we trained a neural network as the first example of a machine learning model capable of predicting full adsorption isotherms of different molecules not included in the training of the model. To achieve this, we trained our neural network only on alchemical species, represented only by their geometry and force field parameters, and used this neural network to predict the loadings of real adsorbates. We focused on predicting room temperature adsorption of small (one- and two-atom) molecules relevant to chemical separations. Namely, argon, krypton, xenon, methane, ethane, and nitrogen. However, we also observed surprisingly promising predictions for more complex molecules, whose properties are outside the range spanned by the alchemical adsorbates. Prediction accuracies suitable for large-scale screening were achieved using simple MOF (e.g. geometric properties and chemical moieties), and adsorbate (e.g. forcefield parameters and geometry) descriptors. Our results illustrate a new philosophy of training that opens the path towards development of machine learning models that can predict the adsorption loading of any new adsorbate at any new operating conditions in any new MOF.</div>

Sign in / Sign up

Export Citation Format

Share Document