Gas turbine air inlet cooling system efficiency test system

Abstract

The invention relates to the technical field of power plant equipment monitoring, and discloses an efficiency testing system for an air inlet cooling system of a gas turbine. The system comprises a data acquisition and preprocessing module which acquires real-time operation parameters of an air inlet cooling system through a sensor network and preprocesses the real-time operation parameters to generate an initial space-time sequence; the feature extraction module decomposes and reconstructs the initial space-time sequence and extracts multi-parameter correlation features; the feature fusion module fuses the multi-parameter associated features into a multi-resolution feature map; the grid division and performance spectrum generation module performs adaptive grid division on the multi-resolution feature map by using a dynamic grid generation method, and outputs a performance spectrum; the finite element simulation module calculates efficiency state distribution of the key monitoring points through finite element simulation based on the efficiency spectrum; an efficiency attenuation analysis module derives an efficiency attenuation period from the efficiency state distribution; the life prediction module predicts the system residual performance life in combination with the material performance degradation model.

Description

Technical Field

[0001] This invention relates to the field of power plant equipment monitoring technology, specifically to a gas turbine intake cooling system performance testing system. Background Technology

[0002] Current performance testing of gas turbine inlet cooling systems primarily employs methods such as periodic shutdown inspections and statistical analysis of data from fixed measurement points. Existing technologies for evaluating cooling system performance often rely on single temperature or pressure parameters, failing to comprehensively consider the coupled effects of multiple parameters such as flow rate and humidity. Data acquisition points are sparse, making it difficult to obtain complete spatiotemporal distribution information within the system. Feature extraction methods are simplistic, and the dynamic correlations between operating parameters are not effectively explored. Performance evaluation models are linearized, failing to accurately reflect the complexity of non-uniform flow fields and heat transfer processes. Fixed mesh generation strategies cannot adapt to changes in system parameter distribution under different operating conditions. Attenuation analysis methods are limited, typically based on simple linear extrapolation, failing to consider the nonlinear effects of material performance degradation. Lifetime prediction accuracy is insufficient, lacking a mechanism to combine real-time operating conditions with material aging physical models. Existing methods need to address key technical challenges such as multi-parameter coupled analysis, spatiotemporal feature extraction, high-precision performance simulation, and lifetime prediction.

[0003] Traditional performance testing systems suffer from significant shortcomings in data analysis and model accuracy. Sensor network deployment density is low, and monitoring data for key areas is lacking. Preprocessing algorithms are simplistic, with limited effectiveness in noise filtering and outlier handling. Fixed feature decomposition methods fail to adaptively extract operational features across different time scales. A multi-resolution feature fusion mechanism is lacking, resulting in ineffective integration of feature information at different levels of detail. Dynamic mesh generation algorithms are computationally complex and struggle to meet real-time requirements. Finite element simulation models are oversimplified, with boundary conditions deviating from actual operating conditions. Performance state distribution calculations do not consider fluid-thermal-structural multi-field coupling effects. Attenuation period derivation relies on empirical formulas, lacking a clear physical mechanism. Material performance degradation model parameters are fixed and cannot adapt to variations in operating environments and material batches. Existing technologies necessitate the development of a high-precision, end-to-end testing solution encompassing data acquisition and lifetime prediction. Summary of the Invention

[0004] The purpose of this invention is to provide a performance testing system for a gas turbine intake cooling system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a performance testing system for a gas turbine inlet cooling system, the system comprising: The data acquisition and preprocessing module acquires real-time operating parameters of the gas turbine intake cooling system through a deployed sensor network, and preprocesses the real-time operating parameters to generate an initial spatiotemporal sequence. The feature extraction module, connected to the data acquisition and preprocessing module, is used to decompose and reconstruct the initial spatiotemporal sequence, thereby extracting multi-parameter correlation features; A feature fusion module, connected to the feature extraction module, is used to fuse the multi-parameter associated features into a multi-resolution feature map; The grid generation and performance spectrum generation module is connected to the feature fusion module. It uses a dynamic grid generation method to adaptively generate a grid for the multi-resolution feature map and outputs a performance spectrum. The finite element simulation module is connected to the mesh generation and performance spectrum generation module, and calculates the performance state distribution of key monitoring points based on the performance spectrum through finite element simulation. The performance decay analysis module is connected to the finite element simulation module and derives the performance decay period from the performance state distribution. The life prediction module is connected to the performance degradation analysis module and predicts the remaining performance life of the gas turbine intake cooling system by combining a preset material performance degradation model.

[0006] Preferably, the step of collecting real-time operating parameters of the gas turbine intake cooling system through a deployed sensor network and preprocessing the real-time operating parameters to generate an initial spatiotemporal sequence includes: Configure distributed sensor nodes to continuously monitor intake air temperature, coolant flow rate, and ambient pressure; perform noise filtering and outlier removal on the monitoring data; align the cleaned data by timestamp to form a standardized time series; decompose the standardized time series into intrinsic mode functions through empirical mode decomposition; reconstruct the intrinsic mode functions to generate an initial spatiotemporal sequence.

[0007] Preferably, the step of decomposing and reconstructing the initial spatiotemporal sequence to extract multi-parameter correlation features includes: The singular spectrum analysis algorithm is used to decompose the trend and periodic components of the initial spatiotemporal sequence; Calculate the cross-correlation function between the trend component and the periodic component to identify the coupling relationship between temperature, flow rate and pressure; Principal component analysis is used to reduce the dimensionality of coupled data and extract the main feature vectors. The main feature vectors are then normalized to generate multi-parameter associated features.

[0008] Preferably, fusing the multi-parameter correlated features into a multi-resolution feature map includes: Wavelet transform is performed on multi-parameter correlated features to obtain frequency components at different scales; the wavelet basis function and scale parameters are adjusted according to the data resolution requirements; the frequency components at each scale are fused to generate a multi-resolution feature map; and the integrity of the multi-resolution feature map is verified by inverse wavelet transform.

[0009] Preferably, the step of adaptively dividing the multi-resolution feature map using a dynamic mesh generation method and outputting an efficiency spectrum includes: The time series data of multi-resolution feature maps are analyzed to determine the density and range of the grid division; the initial grid structure is generated using the front-end advancement method; the grid node distribution is dynamically adjusted according to the feature change gradient; the grid quality is optimized through grid smoothing; the performance index of each grid cell is calculated, and the performance spectrum is combined.

[0010] Preferably, the step of calculating the performance status distribution of key monitoring points based on the performance spectrum using finite element simulation includes: A three-dimensional geometric model of the gas turbine intake cooling system is constructed; the discrete three-dimensional geometric model is converted into a finite element mesh; the efficiency spectrum is mapped to the boundary conditions of the finite element mesh; the heat conduction and fluid dynamics equations are solved to simulate the temperature and pressure field distributions; efficiency status data of key monitoring points are extracted to form the efficiency status distribution.

[0011] Preferably, deriving the performance decay period from the performance state distribution includes: performing time-series sampling on the performance state distribution to obtain a performance value sequence; applying an autoregressive integral moving average model to fit the performance value sequence; calculating the variance and autocorrelation function of the fitting residuals; identifying decay frequency components through spectral analysis; and calculating the performance decay period based on the decay frequency components.

[0012] Preferably, the step of predicting the remaining performance life of the gas turbine inlet cooling system using a preset material property degradation model includes: Obtain material parameters for key components of the gas turbine intake cooling system; establish material property degradation curves based on creep and fatigue; input the performance decay period into the material property degradation curves to calculate the cumulative damage index; and predict the remaining performance life using the linear damage accumulation rule.

[0013] Preferably, the step of decomposing the standardized time series into intrinsic mode functions through empirical mode decomposition includes: identifying local extrema of the standardized time series; fitting upper and lower envelopes; calculating the mean of the envelopes; subtracting the mean of the envelopes from the original series to obtain candidate modes; repeating the iteration until the mode conditions are met, and outputting the intrinsic mode functions.

[0014] Preferably, the method of generating the initial mesh structure using the leading edge method includes: defining the boundary points of the computational domain; generating leading edge cells from the boundary points inward; checking the quality and size of the leading edge cells; eliminating invalid cells; optimizing cell connectivity, and completing the initial mesh generation.

[0015] Compared with the prior art, the beneficial effects of the present invention are: The feature extraction module decomposes and reconstructs the initial spatiotemporal sequence to extract multi-parameter correlation features. The decomposition process employs empirical mode decomposition (EMD) or wavelet transform to break down the signal into components of different scales. The reconstruction process selects components that significantly impact system performance for recombination, removing noise and redundant information. Multi-parameter correlation features include correlation coefficients, phase relationships, and mutual information among parameters such as temperature, pressure, flow rate, and humidity. Feature extraction considers time delay effects and nonlinear coupling characteristics between parameters. Correlation features quantify the strength and direction of interactions between different operating parameters. Feature dimensionality is reduced using principal component analysis or an autoencoder to retain key variation information. The extracted feature vectors contain the correlation patterns of parameters in both the time and spatial domains. The feature fusion module fuses the multi-parameter correlation features into a multi-resolution feature map. The fusion process uses a feature pyramid network or U-Net structure to integrate feature information at different resolutions. Low-resolution features capture the global operating state and long-term trends, while high-resolution features retain local details and short-term fluctuations. The multi-resolution feature map achieves scale alignment and information complementarity through upsampling and downsampling operations. Feature-weighted fusion assigns different weights based on feature importance, with key features receiving higher weights. The feature map is organized as raster data, with each raster containing multidimensional feature information for that location. The fusion process considers the spatial correlation between features while preserving their topological structure. The multi-resolution feature map provides a high-dimensional data foundation for subsequent mesh generation and simulation.

[0016] The mesh generation and performance spectrum generation module utilizes a dynamic mesh generation method to adaptively mesh multi-resolution feature maps and outputs the performance spectrum. Dynamic mesh generation automatically adjusts the mesh density based on the feature gradient distribution, refining the mesh in areas of drastic feature changes. Adaptive mesh generation employs quadtrees or a leading-edge method to achieve a smooth transition in mesh size. Mesh node positions consider feature extrema and abrupt change points to improve meshing accuracy. Performance spectrum calculation converts discrete feature values ​​into a continuous distribution field using interpolation algorithms. Spectral line generation employs Fast Fourier Transform or power spectral density estimation methods. The performance spectrum contains the energy distribution characteristics of the system across different frequency ranges. Spectral peak positions correspond to the system's main operating frequencies, and spectral width reflects operational stability. Performance spectrum updates are dynamically adjusted according to changes in operating status to maintain timeliness. The finite element simulation module calculates the performance state distribution of key monitoring points based on the performance spectrum using finite element simulation. The simulation model considers multi-physics coupling of fluid dynamics, heat transfer, and structural mechanics. Boundary condition settings are automatically optimized based on performance spectrum characteristics to improve simulation accuracy. Key monitoring point selection is based on sensitivity analysis, prioritizing locations with significant impact on system performance. The performance state distribution calculation employs the Galerkin method or least squares method to solve the governing equations. The distribution results include the spatial variations of key parameters such as temperature, pressure, and velocity fields. The simulation process is iterative until the results converge. Through the synergistic effect of feature extraction, multi-resolution fusion, dynamic mesh generation, and finite element simulation, high-precision evaluation of system performance is achieved. Multi-parameter correlation reveals the inherent laws of the system, multi-resolution features preserve complete information, dynamic mesh optimizes computational efficiency, and finite element simulation provides detailed state distribution. This integrated approach significantly improves the accuracy and reliability of performance testing for gas turbine inlet cooling systems. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the working principle of the gas turbine intake cooling system performance testing system described in this invention; Figure 2 This is a schematic diagram of the working principle of the data acquisition and preprocessing module; Figure 3 This is a schematic diagram illustrating the working principle of the feature extraction module. Figure 4 Wavelet transform multi-resolution feature map of multi-parameter correlation features of gas turbine intake cooling system; Figure 5 The temperature field distribution and finite element simulation diagram of key monitoring points in the gas turbine intake cooling system are shown. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides a performance testing system for a gas turbine inlet cooling system. The system includes: a data acquisition and preprocessing module that acquires real-time operating parameters of the gas turbine inlet cooling system, including inlet temperature, coolant flow rate, and ambient pressure, through a deployed sensor network, and preprocesses these real-time operating parameters to generate an initial spatiotemporal sequence; a feature extraction module connected to the data acquisition and preprocessing module that decomposes and reconstructs the initial spatiotemporal sequence to extract multi-parameter correlation features; a feature fusion module connected to the feature extraction module that fuses the multi-parameter correlation features into a multi-resolution feature map; a mesh generation and performance spectrum generation module connected to the feature fusion module that adaptively meshes the multi-resolution feature map using a dynamic mesh generation method and outputs a performance spectrum; a finite element simulation module connected to the mesh generation and performance spectrum generation module that calculates the performance state distribution of key monitoring points based on the performance spectrum using finite element simulation; a performance decay analysis module connected to the finite element simulation module that derives the performance decay period from the performance state distribution; and a lifetime prediction module connected to the performance decay analysis module that predicts the remaining performance lifetime of the gas turbine inlet cooling system by combining a preset material performance degradation model.

[0020] Example 1: See Figure 2In specific implementations, the data acquisition and preprocessing module is configured with distributed sensor nodes to continuously monitor the intake air temperature, coolant flow rate, and ambient pressure of the gas turbine intake cooling system. The distributed sensor nodes collect real-time operating parameters at a fixed sampling frequency, and the monitoring data is transmitted to the central processing unit via wired or wireless communication. In specific implementations, noise filtering and outlier removal are performed on the monitoring data. Noise filtering uses a low-pass digital filter to handle high-frequency interference components, and outlier removal uses statistical methods to identify and remove data points exceeding a preset threshold range. In some embodiments, the cleaned data is aligned by timestamps to form a standardized time series. Timestamp alignment synchronizes the data streams from each sensor using a unified clock source, and the standardized time series maps data values ​​to a unified numerical range using a min-max scaling method. It can be understood that the standardized time series eliminates dimensional differences, facilitating subsequent processing. The specific implementation of the min-maximum scaling method includes identifying the minimum and maximum values ​​in the data sequence. The minimum value is the smallest numerical value among the data points, and the maximum value is the largest numerical value among the data points. A linear transformation maps each data point to a preset unified numerical interval. The lower and upper limits of the unified numerical interval are set according to system requirements. The transformation process calculates the proportion of the difference between each data point and the minimum value. This proportion is divided by the difference between the maximum and minimum values, and the result is scaled to the target interval. This mapping process converts parameters with different physical dimensions, such as intake air temperature, coolant flow rate, and ambient pressure, into dimensionless values. The unified numerical interval ensures that all parameters are on the same scale, facilitating subsequent decomposition and reconstruction operations by the feature extraction module. After data value conversion, the relative distribution characteristics of the original data are maintained, eliminating analytical biases caused by dimensional differences.

[0021] In practical implementation, the standardized time series is decomposed into intrinsic mode functions (EMFs) through empirical mode decomposition (EMD). The EMD process includes identifying local extrema of the standardized time series, defined as values ​​in the series that are greater than or less than their adjacent data points. Upper and lower envelopes are fitted using cubic spline interpolation to connect all local maxima to form the upper envelope, and all local minima to form the lower envelope. The mean of the envelopes is calculated by averaging the upper and lower envelopes. Candidate modes are obtained by subtracting the mean of the envelopes from the original series. This iterative process is repeated until the candidate modes meet the conditions of the EMFs. The conditions for the EMFs include that the difference between the number of extrema and the number of zero-crossing points is no more than one, and that the mean of the envelope is close to zero throughout the series. In some embodiments, the iterative process of EMD uses the following formula to calculate the mean of the envelopes: in: This represents the mean of the envelope. Indicates the upper envelope. The lower envelope is represented, and the intrinsic mode functions are output as the decomposition result. Optionally, the intrinsic mode functions are reconstructed to generate an initial spatiotemporal sequence. The reconstruction process is achieved by linearly superimposing all intrinsic mode functions. The initial spatiotemporal sequence stores time and parameter dimension data in matrix form.

[0022] In the intake air temperature monitoring of the intake cooling system of a 20MW industrial gas turbine, the upper envelope of the standardized time series within a certain period is obtained. The values ​​at three consecutive time points were 35℃, 36℃, and 34℃, respectively. The lower envelope... The corresponding time points were 28℃, 29℃, and 27℃. (The last sentence appears to be incomplete and possibly refers to a different time point.) and Substituting into the iterative process of empirical mode decomposition, the mean of the envelope at the first time point is The second time point is The third time point is Based on these mean values, candidate modes are obtained by subtracting them from the original temperature sequence. After three iterations, the candidate modes satisfy the conditions that the difference between the number of extreme points and zero crossing points does not exceed 1 and the mean of the envelope is close to zero. The intrinsic mode function is successfully output. This function can be used for the reconstruction of the subsequent initial spatiotemporal sequence, providing accurate temperature data components for multi-parameter correlation feature extraction.

[0023] It is understandable that Empirical Mode Decomposition (EMD) is suitable for nonlinear and non-stationary data characteristics. In specific implementation, the data acquisition and preprocessing module transmits the initial spatiotemporal sequence to the feature extraction module for further analysis. The deployment locations of distributed sensor nodes cover the key monitoring areas of the gas turbine intake cooling system. Noise filtering parameters are adjusted according to the actual operating environment, outlier removal thresholds are set based on historical data statistics, and the generation of standardized time series ensures data consistency and comparability. The number of iterations of EMD is controlled by a preset convergence criterion. The output order of the intrinsic mode functions reflects the components of the data from high frequency to low frequency. The reconstruction process retains the main features of the original sequence, and the initial spatiotemporal sequence serves as the basis for multi-parameter correlation feature extraction.

[0024] Example 2: See Figure 3In practical implementation, the feature extraction module receives the initial spatiotemporal sequence from the data acquisition and preprocessing module. It uses a singular spectrum analysis (SSA) algorithm to decompose the trend and periodic components of the initial spatiotemporal sequence. The SSA algorithm constructs a trajectory matrix by embedding dimensions and performs singular value decomposition on the trajectory matrix to obtain eigenvalues ​​and eigenvectors, thereby separating the trend and periodic components. The cross-correlation function between the trend and periodic components is then calculated. The cross-correlation function quantifies the linear correlation between the two components at different time lags, identifying the coupling relationship between temperature, flow rate, and pressure. The coupling relationship is characterized by the peak position and amplitude of the cross-correlation function. Principal component analysis (PCA) is used to reduce the dimensionality of the coupled data. PCA calculates the eigenvalues ​​and eigenvectors of the covariance matrix, selects the main eigenvectors whose cumulative contribution rate exceeds a preset threshold, and normalizes the main eigenvectors. The normalization process uses a min-max scaling method to map the eigenvalues ​​to the range of zero to one, generating multi-parameter associated features. These multi-parameter associated features are stored in vector form for subsequent feature fusion.

[0025] In some embodiments, the feature fusion module performs wavelet transform on the multi-parameter associated features. The wavelet transform uses a mother wavelet function to convolve the multi-parameter associated features to obtain frequency components at different scales. The wavelet basis function and scale parameter are adjusted according to the data resolution requirements. The wavelet basis function can be either the Daubechies wavelet or the Haar wavelet, and the scale parameter controls the level of detail of the frequency components. Frequency components at each scale are fused to generate a multi-resolution feature map. The fusion process integrates high-frequency and low-frequency components through a weighted summation method. The multi-resolution feature map represents time-frequency domain information in matrix form. The integrity of the multi-resolution feature map is verified by inverse wavelet transform. The inverse wavelet transform reconstructs the multi-resolution feature map into the original domain data and compares the reconstruction error with the input multi-parameter associated features. It can be understood that wavelet transform provides time-frequency localization analysis capabilities. In the generation of multi-resolution feature maps, the continuous form of the wavelet transform can be expressed as: in: Represents wavelet coefficients, Indicates multi-parameter correlation features, This represents the wavelet basis functions after scaling and translation. Indicates the scale parameter. This represents the translation parameter.

[0026] When performing wavelet transform on multi-parameter correlated features (including coupled data of intake air temperature and coolant flow) under a certain operating condition, the Daubechies wavelet is selected as the mother wavelet function. Set scale parameters (Corresponding to intermediate frequency components), translation parameters (Corresponding to the start offset of the time axis). Multi-parameter correlation features The values ​​at five consecutive time points were 0.8, 0.9, 0.7, 0.8, and 0.9 (normalized). and Perform convolution operations to obtain wavelet coefficients. The results at the corresponding time points were 0.12, 0.15, 0.11, 0.13, and 0.14, respectively. These coefficients constitute the frequency components at this scale, and... (high frequency), After fusing the (low-frequency) components, a multi-resolution feature map is generated. Then, the feature map is reconstructed into the original domain data through inverse wavelet transform. The reconstructed data has minimal deviation from the input multi-parameter correlation features, verifying the integrity of the multi-resolution feature map, which can be used for adaptive grid density determination in the subsequent grid generation module.

[0027] Optionally, in the principal component analysis (PCA) dimensionality reduction process, the cumulative contribution rate threshold is set to 95% to ensure that most of the data variance is preserved. The scale parameter of the wavelet transform is dynamically adjusted according to the actual application requirements to balance frequency resolution and time resolution. During the inverse wavelet transform verification process, the multi-resolution feature map is considered complete when the reconstruction error is below the preset tolerance. It can be understood that multi-resolution feature maps can simultaneously capture the global trend and local details of a signal. In some embodiments, the calculation of the cross-correlation function is accelerated using Fast Fourier Transform to improve computational efficiency. The eigenvector selection for PCA is based on the eigenvalue ranking, and normalization processing prevents certain features from dominating the fusion process. Optionally, the selection of wavelet basis functions depends on the data characteristics; for example, Daubechies wavelets are suitable for non-stationary signals. The generation of multi-resolution feature maps supports the processing of subsequent grid partitioning modules.

[0028] See Figure 4This graph is one of the core outputs of the feature fusion module in the gas turbine intake cooling system performance testing system. Based on wavelet transform technology, it performs time-frequency domain decomposition and reconstruction on the extracted multi-parameter correlated features. The vertical axis wavelet scale corresponds to the detail levels of different frequency components and can be adaptively adjusted to balance frequency and time resolution. The horizontal axis time covers the complete cycle of system operation, ensuring the capture of features across all time periods; the color bars on the right quantify the amplitude of wavelet coefficients, intuitively reflecting the feature energy distribution. This graph, through the fusion of multi-scale frequency components, forms a multi-resolution feature map: low-scale components accurately capture local details and short-term fluctuations in system operation, while high-scale components effectively reflect global trends and long-term patterns. This multi-dimensional feature representation provides crucial input for the subsequent mesh generation and performance spectrum generation modules. The mesh generation module can adaptively adjust the mesh density based on the gradient distribution of the feature map, densifying the mesh in areas of drastic feature changes to improve computational accuracy. The efficiency spectrum generation module quantifies the system's efficiency contribution in different frequency domains based on the energy distribution of multi-resolution features, ultimately providing high-precision data support for finite element simulation and efficiency decay analysis. It is one of the core technology carriers for realizing multi-parameter coupling effect analysis, spatiotemporal feature extraction and high-precision efficiency evaluation of gas turbine intake cooling system.

[0029] Example 3: In specific implementation, the mesh generation and performance spectrum generation module receives multi-resolution feature maps from the feature fusion module, parses the time-series data of the multi-resolution feature maps, the time-series data contains time-frequency domain information organized in matrix form, determines the density and range of mesh generation, the mesh generation density is dynamically set according to the local rate of change of data points, and the mesh generation range covers the complete time span of the multi-resolution feature maps. The advancing front method is used to generate the initial mesh structure. The advancing front method defines the boundary points of the computational domain. The boundary points are composed of key time nodes of the multi-resolution feature maps. Advancing front cells are generated from the boundary points inward. The advancing front cells are the basic geometric units in the mesh generation process. The quality and size of the advancing front cells are checked. The quality of the advancing front cells is evaluated by the cell side length ratio and interior angle. Invalid cells are eliminated. Invalid cells refer to cells that do not meet the size or geometric quality standards. Cell connectivity is optimized. Cell connectivity optimization is achieved by adjusting the node sharing relationship of adjacent cells. The initial mesh generation is completed. It can be understood that the advancing front method is suitable for mesh generation in complex geometric regions.

[0030] In some embodiments, the distribution of grid nodes is dynamically adjusted based on the feature change gradient, which is obtained by calculating the difference between adjacent data points in the multi-resolution feature map. The grid node density is increased in regions with large feature change gradients and decreased in regions with small feature change gradients. The grid quality is optimized through grid smoothing, which uses the Laplace smoothing algorithm to reposition the nodes. The performance index of each grid cell is calculated using the following formula: in: This represents the performance index of the i-th grid cell. Represents the time point within the grid cell The amplitude of the wavelet coefficients at that point, This represents the mean value of the wavelet coefficients within the grid cell. The total number of time points contained within a grid cell is represented by the combined performance spectrum, which is a two-dimensional distribution map of the performance indicators of each grid cell arranged according to their spatial location.

[0031] After dividing a multi-resolution feature map into a grid, a specific grid cell is selected. This unit contains wavelet coefficient amplitudes at 5 time points. The values ​​are 1.2, 1.3, 1.1, 1.2, and 1.2, respectively. First, calculate the mean value of the wavelet coefficient amplitude within this unit. Then each and The sum of the absolute values ​​of the differences is obtained by summing them. Substitute the values ​​into the performance index formula to calculate the performance index of this grid cell. The performance indicators of this grid cell and the other 20 grid cells are arranged spatially to form a complete performance spectrum. Smaller mesh cells correspond to regions with gentle feature fluctuations, providing stable boundary condition inputs for finite element simulation modules.

[0032] Optionally, the mesh density is positively correlated with the gradient of feature changes. The mesh smoothing process is iterated multiple times until the node position change is less than a set threshold. It can be understood that dynamic mesh node distribution can more accurately capture key features in multi-resolution feature maps. In some embodiments, the quality check of the leading edge element includes calculating whether the minimum interior angle of the element is greater than a set angle threshold. Invalid elements are eliminated by merging excessively small elements or dividing excessively large elements. Element connectivity optimization ensures that there are no isolated nodes or elements in the mesh. Optionally, the standard deviation of wavelet coefficient amplitude is used to characterize the degree of data fluctuation within the mesh element in the efficiency index calculation. The efficiency spectrum serves as the input condition for finite element simulation, providing a quantitative basis for subsequent analysis.

[0033] Example 4: In specific implementation, the finite element simulation module constructs a three-dimensional geometric model of the gas turbine intake cooling system. This model is based on the actual structural dimensions of the gas turbine intake cooling system and includes the intake duct, cooling units, and outlet section components. The discrete three-dimensional geometric model is a finite element mesh. The discretization process divides the three-dimensional geometric model into multiple small elements, such as tetrahedral or hexahedral elements, forming a finite element mesh structure. The performance spectrum is mapped to the boundary conditions of the finite element mesh. The performance spectrum comes from the output of the mesh generation and performance spectrum generation module. The boundary conditions include temperature, pressure, and flow parameters, which are assigned to the corresponding nodes or surfaces of the finite element mesh. The heat conduction and fluid dynamics equations are solved. The heat conduction equation describes the evolution of temperature distribution over time, while the fluid dynamics equation describes fluid flow and pressure changes, simulating the temperature and pressure field distributions. The temperature field represents the temperature value at each point in the system, and the pressure field represents the pressure value at each point in the system. The performance status data of key monitoring points are extracted. The key monitoring points are located at specific locations in the gas turbine intake cooling system, such as the air inlet, coolant inlet and outlet, forming a performance status distribution. The performance status distribution stores the spatial changes of parameters such as temperature and pressure in the form of a data field.

[0034] In some embodiments, the design drawings of the gas turbine intake cooling system are imported into computer-aided design software to construct the three-dimensional geometric model. The three-dimensional geometric model includes the precise dimensions and connection relationships of all key components. The discrete three-dimensional geometric model is a finite element mesh generated using an automatic mesh generation algorithm. The mesh size is adjusted according to the geometric complexity. When solving the heat conduction and fluid dynamics equations, the following formula is used to describe the heat conduction process: in: Represents the temperature field. Indicates time, Indicates the thermal diffusivity. The Laplace operator is used to calculate the temperature change over time. The fluid dynamics equations include the mass conservation and momentum conservation equations. The temperature and pressure field distributions are simulated by solving a set of partial differential equations using numerical iteration methods. The locations of key monitoring points are set based on engineering experience. The performance state distribution output is a two-dimensional or three-dimensional data array.

[0035] In the finite element simulation of the intake cooling system of an industrial gas turbine, water is used as the coolant, and its thermal diffusivity is... Pick (Based on the material properties of water). The temperature field of a key area in the intake duct is simulated at the initial time ( Temperature in this area (Intake temperature 15°C higher than ISO standard), after Then, by solving the heat conduction equation, the temperature of the region was calculated. The temperature dropped to 28°C. Simulation continued until... ,temperature The temperature stabilized at 27℃. Comparing this result with the actual monitored temperature of the corresponding node in the finite element mesh, the deviation was within an acceptable range. Furthermore, the simulated temperature field distribution was consistent with the temperature data of key monitoring points (such as MP-01, coordinates 0.5m, 0.2m, 1.0m), verifying the accuracy of the heat conduction equation application. This temperature field distribution can be used for time-series sampling in the subsequent performance degradation analysis module.

[0036] Optionally, the element type of the finite element mesh is selected as tetrahedral to adapt to complex geometries. Boundary condition mapping distributes the performance spectrum data to the mesh nodes through interpolation. The equations are solved using finite element analysis software to set material properties and initial conditions. It can be understood that the coupled solution of heat conduction and fluid dynamics equations provides a comprehensive performance view of the system. In some embodiments, the performance status data of key monitoring points are extracted and stored in tabular form. See Table 1, which shows the coordinates and parameter types of key monitoring points.

[0037] Table 1: Key Monitoring Point Parameter Table Optionally, the thermal diffusivity is set according to the coolant material properties. The Laplace operator is approximated on a discrete mesh using the finite element method. Numerical iteration methods include Gauss-Seidel iteration or the conjugate gradient method. The performance state distribution can be used for subsequent performance degradation analysis. The specific implementation of the finite element method includes discretizing the three-dimensional geometric model into a finite element mesh, which consists of multiple small elements such as tetrahedrons or hexahedrons. The Galerkin method is applied to each element to transform the governing equations into a weak form, and the variables such as temperature and pressure fields are approximated using element shape functions. The Laplace operator discretization is achieved by calculating the gradient of the element shape function and the integral of the test function, forming the stiffness matrix and load vector. Boundary conditions are mapped from the performance spectrum to mesh nodes or element surfaces. After integrating the global system equations, the linear equations are solved using Gauss-Seidel iteration or the conjugate gradient method to obtain numerical solutions for simulating the temperature and pressure field distributions.

[0038] See Figure 5This figure is the core output of the finite element simulation module of the gas turbine intake cooling system performance testing system. Based on the performance spectrum generated by the mesh generation and performance spectrum generation module, it constructs a three-dimensional geometric model of the system and discretizes it into a finite element mesh. The performance spectrum is then mapped to mesh boundary conditions, and the temperature field distribution is generated after solving the heat conduction equation. The X and Y coordinates in the figure represent spatial dimensions, and the color bars quantify temperature, intuitively presenting the spatial gradient distribution of temperature within the system. The placement of key monitoring points is based on sensitivity analysis, focusing on regions critical to system performance. This figure provides the basic data for the performance degradation analysis module, which uses time-series sampling of monitoring points. By extracting the dynamic temperature changes at each point and combining model fitting and spectral analysis, the frequency components of performance degradation can be identified to derive the degradation period. Simultaneously, it provides thermal load input for the life prediction module, supporting the cumulative damage calculation of the material performance degradation model. It achieves numerical simulation of multi-physics coupling of heat conduction and fluid dynamics, solving the problems of large deviations between boundary conditions and actual operating conditions and the lack of spatial details in the performance state distribution in traditional testing. It is a key technology carrier for high-precision analysis of the entire system process from performance spectrum to state distribution to life prediction.

[0039] Example 5: In specific implementation, the performance decay analysis module receives the performance state distribution from the finite element simulation module and performs time series sampling on the performance state distribution. Time series sampling extracts performance values ​​from the performance state distribution at fixed time intervals to obtain a performance value sequence. An autoregressive integral moving average (ARM) model is applied to fit the performance value sequence. The ARM model establishes a statistical model of the time series through autoregressive terms, difference terms, and moving average terms, and calculates the variance and autocorrelation function of the fitting residuals. The fitting residuals are the difference between the observed values ​​and the model's predicted values; the variance represents the dispersion of the fitting residuals; and the autocorrelation function describes the correlation of the fitting residual sequence under different time delays. Attenuation frequency components are identified through spectral analysis, which transforms the fitting residual sequence from the time domain to the frequency domain. The attenuation frequency components correspond to the peak frequencies in the spectrum. The performance decay period is calculated based on the attenuation frequency components, and the performance decay period is the reciprocal of the attenuation frequency components.

[0040] In some embodiments, the life prediction module acquires material parameters of key components of the gas turbine inlet cooling system, including elastic modulus, fatigue limit, and creep rate. It then establishes material performance degradation curves based on creep and fatigue, describing the performance degradation trajectory of the material under cyclic loading and high temperature. The performance degradation cycle is input into the material performance degradation curve. A cumulative damage index is calculated, characterizing the total damage suffered by the material under a given load history. The remaining performance life is predicted using a linear damage accumulation rule, which assumes that the damage caused by each load cycle can be linearly added. The following formula is used to calculate the cumulative damage index: in: Indicates the cumulative damage index. This represents the actual number of cycles at the i-th stress level. This represents the number of cycles that led to failure at the i-th stress level. This indicates the number of different stress levels.

[0041] When predicting the lifespan of the cooling unit (a critical component made of stainless steel) in the intake cooling system of a gas turbine, two stress levels were identified. (Low load) and (High load). In At that time, the actual number of loops. The number of cycles that lead to failure under this stress. Next. In At that time, the actual number of loops. The corresponding failure cycle count The cumulative damage index was calculated once. .because If the failure threshold is less than 1, and combined with the material performance degradation curve (established based on stainless steel creep and fatigue data), the remaining performance life of the cooling unit can be predicted by the linear damage accumulation law. The results can guide the power plant to formulate component maintenance plans and avoid system performance degradation due to material failure.

[0042] It is understandable that the autoregressive integral moving average model can effectively handle non-stationary time series data, the spectral analysis uses the fast Fourier transform algorithm to achieve time-frequency conversion, the material performance degradation curve is obtained by fitting accelerated life test data, and the linear damage accumulation rule provides a theoretical basis for remaining life prediction. Optionally, the time series sampling interval is set according to the system operating characteristics, the parameters of the autoregressive integral moving average model are determined by the Akaike information criterion, the autocorrelation function of the fitting residuals is used to test the model's adaptability, and the frequency resolution of the spectral analysis is determined by the sampling length. In some embodiments, material parameters are obtained from material handbooks or experimental tests, the material performance degradation curve considers the temperature and stress coupling effect, the cumulative damage index calculation considers the interaction between creep damage and fatigue damage, and the failure threshold is set to 1 in the linear damage accumulation rule. Optionally, the performance decay period is used as the input load frequency of the material performance degradation curve, and the remaining performance life prediction result is expressed in time units as the system's remaining service life.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A performance testing system for a gas turbine inlet cooling system, characterized in that, The system includes: The data acquisition and preprocessing module acquires real-time operating parameters of the gas turbine intake cooling system through a deployed sensor network, and preprocesses the real-time operating parameters to generate an initial spatiotemporal sequence. The feature extraction module, connected to the data acquisition and preprocessing module, is used to decompose and reconstruct the initial spatiotemporal sequence, thereby extracting multi-parameter correlation features; A feature fusion module, connected to the feature extraction module, is used to fuse the multi-parameter associated features into a multi-resolution feature map; The grid generation and performance spectrum generation module is connected to the feature fusion module. It uses a dynamic grid generation method to adaptively generate a grid for the multi-resolution feature map and outputs a performance spectrum. The finite element simulation module is connected to the mesh generation and performance spectrum generation module, and calculates the performance state distribution of key monitoring points based on the performance spectrum through finite element simulation. The performance decay analysis module is connected to the finite element simulation module and derives the performance decay period from the performance state distribution. The life prediction module is connected to the performance degradation analysis module and predicts the remaining performance life of the gas turbine intake cooling system by combining a preset material performance degradation model.

2. The gas turbine inlet cooling system performance testing system according to claim 1, characterized in that, The process of acquiring real-time operating parameters of the gas turbine intake cooling system through a deployed sensor network and preprocessing these parameters to generate an initial spatiotemporal sequence includes: Configure distributed sensor nodes to continuously monitor intake air temperature, coolant flow rate, and ambient pressure; perform noise filtering and outlier removal on the monitoring data; align the cleaned data by timestamp to form a standardized time series; decompose the standardized time series into intrinsic mode functions through empirical mode decomposition; reconstruct the intrinsic mode functions to generate an initial spatiotemporal sequence.

3. The gas turbine inlet cooling system performance testing system according to claim 1, characterized in that, The process of decomposing and reconstructing the initial spatiotemporal sequence to extract multi-parameter correlation features includes: The singular spectrum analysis algorithm is used to decompose the trend and periodic components of the initial spatiotemporal sequence; Calculate the cross-correlation function between the trend component and the periodic component to identify the coupling relationship between temperature, flow rate and pressure; Principal component analysis is used to reduce the dimensionality of coupled data and extract the main feature vectors. The main feature vectors are then normalized to generate multi-parameter associated features.

4. The gas turbine inlet cooling system performance testing system according to claim 1, characterized in that, The process of fusing the multi-parameter correlated features into a multi-resolution feature map includes: Wavelet transform is performed on multi-parameter correlated features to obtain frequency components at different scales; the wavelet basis function and scale parameters are adjusted according to the data resolution requirements; the frequency components at each scale are fused to generate a multi-resolution feature map; and the integrity of the multi-resolution feature map is verified by inverse wavelet transform.

5. The gas turbine inlet cooling system performance testing system according to claim 1, characterized in that, The step of adaptively dividing the multi-resolution feature map using a dynamic mesh generation method and outputting an efficiency spectrum includes: The time series data of multi-resolution feature maps are analyzed to determine the density and range of the grid division; the initial grid structure is generated using the front-end advancement method; the grid node distribution is dynamically adjusted according to the feature change gradient; the grid quality is optimized through grid smoothing; the performance index of each grid cell is calculated, and the performance spectrum is combined.

6. The gas turbine inlet cooling system performance testing system according to claim 1, characterized in that, The calculation of the performance status distribution of key monitoring points based on the performance spectrum through finite element simulation includes: A three-dimensional geometric model of the gas turbine intake cooling system is constructed; the discrete three-dimensional geometric model is converted into a finite element mesh; the efficiency spectrum is mapped to the boundary conditions of the finite element mesh; the heat conduction and fluid dynamics equations are solved to simulate the temperature and pressure field distributions; efficiency status data of key monitoring points are extracted to form the efficiency status distribution.

7. The gas turbine inlet cooling system performance testing system according to claim 1, characterized in that, The step of deriving the performance decay period from the performance state distribution includes: performing time series sampling on the performance state distribution to obtain a performance value sequence; applying an autoregressive integral moving average model to fit the performance value sequence; calculating the variance and autocorrelation function of the fitting residuals; identifying the decay frequency components through spectral analysis; and calculating the performance decay period based on the decay frequency components.

8. The gas turbine inlet cooling system performance testing system according to claim 1, characterized in that, The prediction of the remaining performance life of the gas turbine inlet cooling system using a pre-defined material property degradation model includes: Obtain material parameters for key components of the gas turbine intake cooling system; establish material property degradation curves based on creep and fatigue; input the performance decay period into the material property degradation curves to calculate the cumulative damage index; and predict the remaining performance life using the linear damage accumulation rule.

9. The gas turbine inlet cooling system performance testing system according to claim 2, characterized in that, The step of decomposing a standardized time series into intrinsic mode functions through empirical mode decomposition includes: identifying local extrema of the standardized time series; fitting upper and lower envelopes; calculating the mean of the envelopes; subtracting the mean of the envelopes from the original series to obtain candidate modes; repeating the iteration until the mode conditions are met, and outputting the intrinsic mode functions.

10. The gas turbine inlet cooling system performance testing system according to claim 5, characterized in that, The method of generating the initial mesh structure using the leading edge method includes: defining the boundary points of the computational domain; generating leading edge cells from the boundary points inward; checking the quality and size of the leading edge cells; eliminating invalid cells; optimizing cell connectivity, and completing the initial mesh generation.

Images