Multi-objective optimization method and system for multispectral temperature measurement of aero-engine turbine blades

Abstract

This invention discloses a multi-objective optimization method and system for multispectral temperature measurement of aero-engine turbine blades, belonging to the field of aero-engine condition monitoring technology. It addresses the temperature measurement inaccuracies caused by nonlinear emissivity fluctuations, conflicts between temperature and emissivity objectives, and environmental interference in existing technologies. The method includes: acquiring radiation signals from the blade working surface, recording background reflected radiation and thermocouple reference temperature; defining an environmental complexity factor; constructing objective functions for temperature and emissivity errors; constructing a weighted fitness function; and optimizing the measurement process until the iteration number g < a preset threshold G. mid Perform a global particle swarm search, g ≥ G mid This method combines a genetic algorithm for local refinement; injects random perturbations based on the target conflict quantity; dynamically amplifies the penalty term through an environmental complexity factor; and outputs the Pareto front knee solution as the optimal estimate of the real-time blade temperature and emissivity. This method is used for real-time online temperature monitoring of turbine blades in high-temperature environments of aero-engines.

Description

Technical Field

[0001] This invention belongs to the field of aero-engine condition monitoring technology, and in particular relates to a multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades. Background Technology

[0002] Multispectral thermometry is widely used for temperature measurement of turbine blades in aero-engines, especially in high-temperature environments, where it performs temperature inversion by processing the radiation signals of the target object across multiple wavelengths. This method is characterized by high precision and efficiency and is suitable for complex engineering environments. In aero-engines, the surface temperature of turbine blades has a significant impact on engine performance and safety; therefore, high-precision temperature measurement is crucial for stable engine operation and safe aircraft flight. However, in practical applications, existing multispectral thermometry technologies face several challenges, mainly in the following aspects: First, in multispectral thermometry, emissivity variations exhibit highly nonlinear characteristics due to differences in materials, wavelengths, and temperatures. Traditional emissivity models typically assume a constant emissivity or linear interpolation over a fixed wavelength; however, in real-world high-temperature environments, such simplified assumptions often fail to accurately predict emissivity variations, thus affecting the accuracy of temperature inversion. Second, the optimization problem between temperature and emissivity is usually multi-objective, especially in the temperature inversion process for turbine blades, where temperature error and emissivity error need to be optimized simultaneously, and these two objectives are interdependent. Traditional single-objective optimization methods struggle to balance these two objectives, potentially leading to suboptimal compromises in the final result. For example, prioritizing the optimization of temperature inversion results may increase the estimation error of emissivity, and vice versa. To address these challenges, in recent years, Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) have been combined into Hybrid Particle Swarm Optimization-Genetic Algorithm (HPSOGA) to solve optimization problems in temperature inversion. PSO finds a solution with better fitness through global search, while GA enhances local search capabilities through crossover and mutation operations. Although this hybrid optimization algorithm improves search efficiency, it still suffers from problems such as getting trapped in local optima and high computational complexity when facing measurement errors, reflected radiation, and changes in material properties under complex environments, resulting in a lack of stability and robustness in high-precision measurements. Furthermore, few existing optimization methods can effectively consider the relationship between temperature error and emissivity error and simultaneously balance the weights of these two objectives in multi-objective optimization. Especially when dealing with wide temperature ranges or large reflected radiation interference, the adaptability and efficiency of existing methods have room for improvement. Therefore, how to simultaneously optimize temperature and emissivity within a multi-objective optimization framework and effectively address the impact of environmental changes is key to improving the accuracy and stability of multispectral temperature measurement algorithms.

[0003] In summary, under complex conditions such as nonlinear emissivity fluctuations, environmental reflection and radiation interference, and conflicts between temperature and emissivity targets, existing multispectral temperature measurement technologies struggle to achieve the real-time, high-precision, and highly robust temperature inversion required for aero-engine turbine blades. This constitutes a pressing technical challenge in this field. Summary of the Invention

[0004] One object of the embodiments of the present invention is to solve at least the above-mentioned problems and / or defects, and to provide at least the advantages described below.

[0005] Another objective of this invention is to provide a multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades.

[0006] Another objective of this invention is to provide a multi-objective optimization system for multispectral temperature measurement of aero-engine turbine blades.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades includes the following steps: Radiation signals from the working surface of turbine blades are collected, background reflected radiation and thermocouple reference temperature are recorded simultaneously, and an environmental complexity factor is defined based on the background reflected radiation. Based on the preprocessed radiation signal, objective functions for temperature error and emissivity error are constructed respectively. A multi-stage evolutionary strategy is employed to jointly optimize the temperature error objective function and the emissivity error objective function; during this optimization process... Based on the real-time errors of the two objective functions, their weights in the weighted fitness function are dynamically adjusted. Based on the conflict between the two objective functions, random perturbations are injected into the particle update process; The intensity of the penalty term for outlier solutions is dynamically amplified using the aforementioned environmental complexity factor. From the Pareto optimal solution set obtained through optimization, the knee point solution is selected as the optimal estimate of the final real-time temperature and emissivity of the blade.

[0008] Preferably, in the multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, radiation signals are collected from the working surface of the aero-engine turbine blades. Simultaneously record background reflected radiation and thermocouple reference temperature ; Define the environment complexity factor , The standard deviation of background radiation; Constructing the objective function of temperature error and emissivity error objective function ; The weights are calculated based on the real-time error through a dynamic weight adjustment mechanism. , Constructing weighted fitness The calculation formula is: ; Through a multi-stage evolutionary strategy, in the iterative algebra g <G mid Perform a global particle swarm search and update particle positions; when g ≥ G mid Combine genetic algorithms to perform local refinement; Through the target conflict handling mechanism, based on the target conflict quantity The formula for calculating the target collision quantity after injecting random perturbations is: ; By leveraging environmental adaptability enhancement mechanisms and environmental complexity factors Dynamically amplified penalty items ; Output Pareto front knee solution As the optimal estimation result of real-time blade temperature and emissivity; The dynamic weight adjustment mechanism satisfies the following: in, and Do not represent the error in temperature and emissivity.

[0009] Preferably, in the multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, for the initial particles: a set number of particles is defined. Maximum number of iterations PSO initialization and termination inertia weights crossover rate in genetic algorithm Variation rate and define the emissivity boundary. ; Randomly generated initial particles and speed ,in ; The multi-stage evolutionary strategy includes: In the iterative algebra g <G mid Perform a global particle swarm search: update speed And obtained a new position ; in, Let be the PSO iterative learning factor, ranging from [1.0 to 1.6]. For random weights in PSO, the value range is (0,1); When g is divisible by the preset crossover period, the first one with the smallest congestion distance... Each particle undergoes a genetic algorithm involving single-point crossover and mutation.

[0010] Preferably, in the multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, the objective conflict handling mechanism, when Above the threshold At this time, the weight of conflicting targets is increased and random perturbations are injected into particle updates. Among these, the threshold... For: Take the collision degree values ​​of Np particles The value at the 80th percentile after sorting by size is used as the threshold. .

[0011] Preferably, in the multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, the environmental adaptability enhancement mechanism includes... This is the magnification factor. This represents the complexity factor of the current environment.

[0012] Preferably, in the multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, when the iteration number g=G max , or max| |< Stop at time, extract the last Pareto front And based on the knee curvature: The solution with the largest curvature is selected as the output.

[0013] Preferably, in the multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, the objective function for temperature error is: For normalized radiation signals, This is a temperature prediction function; Five-point median filtering and dark current correction were performed on the original spectral sequence to obtain the denoised spectral matrix. ; ; The objective function for emissivity error is: in, For the second-order difference of emissivity, The theoretical central value, and These are the weighting coefficients.

[0014] Preferably, in the multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, the following steps are performed before defining the environmental complexity factor: Perform five-point median filtering on the radiated signal V(λ,t); Dark current correction is performed on the filtered signal; Separating background reflected radiation from the correction signal : By time window Calculate the standard deviation .

[0015] Preferably, in the multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, the local refinement of the genetic algorithm includes: When g≥G mid Shrink the particle position boundary to ; Adaptive variability .

[0016] Secondly, a multi-objective optimization system for multispectral temperature measurement of aero-engine turbine blades, used to implement any of the methods described above, includes: The multispectral signal acquisition module is used to acquire the radiation signals of turbine blades. Background reflected radiation and thermocouple reference temperature ; The preprocessing module is configured as follows: For radiation signals Perform five-point median filtering and dark current correction to generate a denoised spectral matrix. ; Calculate the standard deviation of background radiation To define the environmental complexity factor ; Normalized radiation signal ; The dynamic weight adjustment module is configured as follows: Constructing the objective function of temperature error and emissivity error objective function ; Calculate weights based on real-time error. , Constructing weighted fitness ,in, and Do not represent the error in temperature and emissivity; The multi-stage evolution optimization module is configured as follows: In the iterative algebra g <G mid Perform a global particle swarm search and update particle positions: ; in, Let be the PSO iterative learning factor, ranging from [1.0 to 1.6]. For random weights in PSO, the value range is (0,1); When g≥G mid The genetic algorithm is combined with local refinement to find the first few with the smallest crowding distance. Each particle performs single-point crossover and mutation; The target conflict handling module is configured as follows: Calculate the target conflict quantity ; when Above the threshold At the same time, increase the weight of conflicting targets and inject random perturbations into particle updates; The environmental adaptability enhancement module is configured as follows: Utilizing environmental complexity factors Dynamically amplified penalty items ,in This is the magnification factor. This represents the complexity factor of the current environment. The knee point output module is configured as follows: When the iterative algebra g=G max , or max| |< Stop at time, extract the last Pareto front ; According to the curvature formula Select the knee point with the largest curvature. As the final temperature and emissivity estimates.

[0017] The embodiments of the present invention include at least the following beneficial effects: (1) This invention achieves real-time adaptive balance between temperature error and emissivity error through multi-dimensional dynamic weight adjustment and target conflict measurement, so that the optimization process no longer depends on static empirical weights; compared with the traditional fixed weight HPSOGA, it greatly improves the temperature measurement accuracy and consistency.

[0018] (2) After introducing multi-stage Pareto front evolution and PSO-GA collaborative search, the algorithm can efficiently explore and avoid local optima in the global-local two-level space, and the number of convergence iterations is reduced by about 40%, which meets the real-time requirements of online monitoring of turbine blades.

[0019] (3) By dynamically amplifying the penalty term through the environmental complexity factor, the robustness of the method is significantly improved compared with the traditional method, and it can adapt to complex high temperature and high radiation environments.

[0020] (4) The integrated temperature-emissivity trade-off knee point selection mechanism provides a solution that combines high precision requirements with physical interpretability, offering a reliable and easily mass-producible algorithm for multispectral temperature measurement of aero-engines.

[0021] Other advantages, objectives, and features of the embodiments of the present invention will be apparent in part from the following description, and in part will be understood by those skilled in the art through study and practice of the embodiments of the present invention. Detailed Implementation

[0022] The embodiments of the present invention will be further described in detail below, so that those skilled in the art can implement them based on the description.

[0023] Although existing multispectral thermometry techniques have been widely used in temperature measurement of aero-engine turbine blades, the hybrid particle swarm optimization-genetic algorithm (HPSOGA), which combines particle swarm optimization (PSO) and genetic algorithm (GA), still faces many problems in practical applications, especially in optimizing temperature inversion and emissivity estimation, where it suffers from low efficiency and optimization limitations. Specifically, the application of the Pareto optimization algorithm in multi-objective optimization has the following problems: (1) Local optimal solution problem In multi-objective optimization, the main goal of Pareto optimization is to obtain a set of optimal solutions by balancing multiple objective functions. However, traditional Pareto optimization algorithms are prone to getting trapped in local optima, especially when dealing with complex nonlinear problems, where the global search capability of the optimization process is often limited. This local optima problem means that the final solution may not be suitable for complex application scenarios, especially in turbine blade temperature measurement where reflected radiation has a significant impact.

[0024] (2) High computational complexity While the HPSOGA algorithm combines the advantages of particle swarm optimization and genetic algorithms, improving the search efficiency of the optimization process to some extent, its computational complexity remains high due to the need for each particle to be evaluated and updated on multiple objective functions. This is especially true when dealing with large-scale data, resulting in slow optimization speeds and severely limiting the algorithm's practical application efficiency. In the turbine blade temperature inversion problem, overly complex computational processes can lead to excessively long response times in the measurement system, failing to meet the requirements for high efficiency.

[0025] (3) Insufficient trade-offs between objectives In multi-objective optimization, temperature error and emissivity error are often two interdependent objectives. However, in existing algorithms, balancing the optimization process of these two objectives and dynamically adjusting their weights remains an unresolved problem. Current Pareto optimization algorithms typically rely on fixed weight settings, which may lead to an overemphasis on one objective, ultimately resulting in inaccurate temperature or emissivity estimates. In high-precision temperature measurement applications, this imbalance can affect the final temperature inversion results, failing to provide reliable temperature data.

[0026] (4) Poor adaptability to environmental changes In practical applications, the reflected radiation from turbine blades is affected by factors such as temperature, surface conditions, and radiation from neighboring blades. These environmental changes can introduce errors into the temperature inversion results, and the traditional Pareto optimization algorithm is not robust enough to handle such environmental variations. Although the HPSOGA algorithm can reduce errors to some extent, the complexity caused by environmental changes is not effectively considered in actual measurements, and the optimization process may become unstable, thus affecting the accuracy of temperature measurements.

[0027] Therefore, the technical solution proposed in this invention aims to address the limitations of existing Pareto optimization algorithms in multispectral thermometry, particularly their low optimization efficiency, high computational complexity, insufficient trade-offs between objectives, and poor environmental adaptability when applied in complex environments. This invention introduces a multi-dimensional dynamic weight adjustment mechanism to monitor error changes in each generation of optimization and dynamically adjust the weights of temperature and emissivity errors, enabling real-time adjustments to the optimization focus based on changes in these errors. Simultaneously, a multi-stage evolutionary optimization strategy is employed to overcome the limitation of traditional Pareto optimization algorithms focusing only on a single frontier. First, a coarse global search is performed to obtain a preliminary Pareto frontier; then, a local search refines the frontier, gradually converging to the optimal solution. Furthermore, recognizing the inherent conflict between temperature and emissivity, this invention proposes a Pareto optimization strategy based on target conflict. By establishing a target conflict metric model, the degree of conflict between each pair of targets is calculated, and more optimization resources are allocated to target pairs with greater conflict, effectively mitigating the conflict between objectives. In addition, an environmental adaptability enhancement mechanism is introduced, incorporating the impact of background reflectance radiation changes on the optimization process, further improving the robustness of the Pareto optimization algorithm. Finally, by introducing a coordination factor, global and local searches can be coordinated. The global search expands the search space through particle swarm optimization to ensure that potential optimal solutions are explored, while the local search improves the accuracy of local optimization through crossover and mutation operations in a genetic algorithm.

[0028] According to one embodiment of the present invention, a multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades is provided, which is specifically implemented through the following process: First, data acquisition and preprocessing are performed. Radiation signals from the turbine blade working surface are acquired using a spectrometer with a specific wavelength, while background reflected radiation from the environment and reference temperatures measured by thermocouples are also recorded. Median filtering and dark current correction are applied to the raw radiation signals to improve signal quality. Based on the processed background radiation data, an environmental complexity factor is defined by calculating its volatility, thereby quantifying the interference intensity of the current measurement environment.

[0029] Next, a multi-objective optimization model is established. The processed radiation signal is normalized to provide standardized input for subsequent calculations. A temperature prediction model is constructed based on the theory of radiative heat transfer, and two core objective functions are established accordingly: a temperature error objective function to evaluate the degree of agreement between the inverted temperature and the measured spectral data, and an emissivity error objective function to ensure the physical rationality of the emissivity curve by introducing smoothness constraints and theoretical value deviation penalties.

[0030] Then, a collaborative optimization solution process is executed. The parameters of the optimization algorithm are initialized, including the particle swarm size, number of iterations, and reasonable ranges for emissivity. A particle swarm containing initial estimates of temperature and emissivity is randomly generated. The optimization process adopts a phased strategy, initially focusing on global exploration and later shifting to local refined search. Throughout the optimization iterations, three core control mechanisms operate simultaneously: first, dynamically adjusting the weights of temperature and emissivity in the fitness function based on real-time errors; second, adjusting the optimization direction and introducing random perturbations when significant conflicts occur by calculating the target conflict amount; and third, adaptively strengthening the penalty for outlier solutions using an environmental complexity factor to improve the algorithm's stability under disturbed environments.

[0031] Finally, the optimization results are decided and output. When the preset termination conditions are met, the Pareto front is extracted from the final generation of optimized solutions. By calculating the curvature characteristics of each point in the solution set, the knee point solution with the best balance is automatically identified. The temperature value and emissivity curve corresponding to this solution are used as the final output results to complete the high-precision and high-reliability monitoring of the turbine blade's operating status.

[0032] Specifically, this invention also proposes a multispectral temperature measurement algorithm based on an improved Pareto optimization strategy. This method solves the problems of local optima, high computational complexity, insufficient objective trade-offs, and poor environmental adaptability of existing Pareto optimization algorithms in practical applications by combining multi-objective optimization of temperature error and emissivity error and introducing several innovative optimization strategies.

[0033] First, for the multi-objective optimization problem in temperature inversion, the system sets two main objective functions: a temperature error objective function (Obj1) and an emissivity error objective function (Obj2). Specifically, the temperature error objective function calculates the difference between the inverted temperature obtained from multispectral data and the reference temperature, while the emissivity error objective function evaluates the deviation of the emissivity curve from the theoretical value. In traditional Pareto optimization, these two objective functions are usually optimized independently. However, this invention introduces a multi-dimensional dynamic weight adjustment mechanism, allowing the weights of the objective functions to be dynamically adjusted according to the actual error during the optimization process, thereby balancing the optimization progress of temperature and emissivity.

[0034] The objective function for temperature error (Obj1) is defined by the following equation: in, It is the measured voltage signal. This is the reference voltage, and T is the current estimated temperature. This is the thermocouple reference temperature, which is the temperature measured by the thermocouple after it is fixed to the surface of the turbine blade before the experiment, and is used as the reference temperature for subsequent calculations. It's the emissivity. This is Planck's constant. This objective function measures the accuracy of the temperature inversion results, and the accuracy of temperature estimation is improved by minimizing this objective.

[0035] The emissivity error objective function (Obj2) is calculated using the following formula to evaluate the smoothness of emissivity and the deviation from the center value: in, The second-order difference of emissivity, It is the theoretical center value of emissivity (usually set to 0.55), while and These are weighting coefficients used to balance the severity of the penalty for smoothness and center bias. This objective function aims to ensure the smoothness and physical plausibility of the emissivity estimate.

[0036] Secondly, in the multi-objective optimization process, there is a certain conflict between temperature and emissivity. This invention dynamically adjusts the weights of temperature error and emissivity error, adjusting the weights of the objective function in each iteration based on the magnitude of the current error. When the temperature error is large, the system automatically increases the weight of the temperature objective; when the emissivity error is large, the system increases the weight of the emissivity objective. This weight adjustment process can be described by the following formula: in, and These represent the errors in temperature and emissivity, respectively. These are adjusted weighting coefficients, enabling the system to dynamically adjust the optimization objective based on the current error during each iteration, ensuring a balance between temperature and emissivity during the optimization process.

[0037] Furthermore, a multi-stage evolutionary optimization strategy using the Pareto front is introduced. First, a coarse global search is performed to obtain a preliminary Pareto front, exploring multiple possible solutions in the global solution space. Then, as the optimization process progresses, the algorithm gradually narrows the search range, entering a local optimization stage. In the local optimization stage, selection, crossover, and mutation operations using a genetic algorithm refine the solution space, gradually approaching the optimal solution. This optimization process is described by the following formula: in, Let represent the i-th solution on the Pareto front, and g represent the current iteration number. In the initial stage, the Pareto front contains a wide range of solutions. As the iterations proceed, the front gradually converges and eventually approaches the optimal solution.

[0038] Next, this invention proposes a Pareto optimization strategy based on target conflict, which calculates the degree of conflict between each pair of targets by establishing a target conflict metric model. When the conflict between targets is high, the algorithm increases the optimization weight of that target, allowing it to be optimized preferentially. The target conflict metric can be calculated using the following formula: Since environmental factors such as background radiation from turbine blades can significantly affect measurement results, this invention introduces an Environmental Complexity Factor (ECF) to dynamically adjust weights and penalty terms during the optimization process. The ECF measures the complexity of environmental changes by calculating the volatility of background radiation; the specific calculation formula is as follows: in, The standard deviation of background radiation data is represented by... This reflects the complexity of environmental changes. When environmental changes are significant, the algorithm automatically increases the penalty for solutions with temperature and emissivity mismatches, thereby improving the algorithm's adaptability to environmental changes.

[0039] Finally, to improve the optimization efficiency of the algorithm, this invention combines the advantages of Particle Swarm Optimization (PSO) and Genetic Algorithm (GA). Through the synergistic effect of global search and local optimization, it improves optimization accuracy and convergence speed. In the global search phase, PSO quickly explores the solution space by updating the velocity and position of particles, avoiding getting trapped in local optima. In the local search phase, GA further refines the solution space through crossover and mutation operations, ensuring the accuracy of the final solution. The synergistic optimization process of global and local search is as follows: in, and These represent the updated particle position and the previous generation particle position, respectively. The change in particle position represents the combined effect of PSO and GA methods. Through the above technical solutions, the optimization algorithm of this invention can provide more accurate and stable temperature inversion results in multispectral thermometry, and has high computational efficiency and good environmental adaptability. The algorithm's innovation lies not only in its in-depth improvement of the traditional Pareto optimization algorithm, but also in addressing the challenges faced by existing optimization methods through the introduction of multi-stage optimization strategies, target conflict handling mechanisms, enhanced environmental adaptability, and global and local search collaborative optimization, further improving the accuracy and stability of temperature measurement.

[0040] In one embodiment of the present invention, a multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades includes the following steps: Radiation signals collected from the working surface of aero-engine turbine blades Simultaneously record background reflected radiation and thermocouple reference temperature ; Define the environment complexity factor , The standard deviation of background radiation; Constructing the objective function of temperature error and emissivity error objective function ; The weights are calculated based on the real-time error through a dynamic weight adjustment mechanism. , Constructing weighted fitness The calculation formula is: ; Through a multi-stage evolutionary strategy, in the iterative algebra g <G mid Perform a global particle swarm search and update particle positions; when g ≥ G mid Combine genetic algorithms to perform local refinement; Through the target conflict handling mechanism, based on the target conflict quantity The formula for calculating the target collision quantity is as follows: (Injecting random perturbations) ; By leveraging environmental adaptability enhancement mechanisms and environmental complexity factors Dynamically amplified penalty items ; Output Pareto front knee solution As the optimal estimation result for real-time temperature and emissivity of turbine blades, a multi-objective optimization core framework for multispectral thermometry of aero-engine turbine blades was established. By collecting radiation signals and background data, an environmental complexity factor was defined to quantify interference intensity. A dual objective function of temperature and emissivity was constructed, and a dynamic weight fitness was introduced to achieve real-time balance between the objectives. A multi-stage evolutionary strategy was adopted to coordinate global particle swarm search and local refinement using a genetic algorithm. Perturbation was injected based on the amount of target conflict to avoid stagnation, and the robustness was enhanced by dynamically adjusting the penalty term using environmental factors. Finally, the Pareto front knee solution was output as the optimal estimate, systematically solving the problem of inaccurate temperature measurement caused by emissivity fluctuations, target conflicts, and environmental interference.

[0041] In one embodiment of the present invention, preferably, the dynamic weight adjustment mechanism satisfies: in, and This section defines the errors in temperature and emissivity. It implements a dynamic and adaptive allocation of weights for these errors. The weight coefficients are automatically adjusted based on the real-time ratio of temperature and emissivity errors in the current iteration, enabling the optimization process to dynamically respond to changes in system state. When temperature error dominates, the weight of the temperature objective is increased; when emissivity error is significant, emissivity optimization is emphasized, ensuring a reasonable balance between the two objectives throughout the iteration process and avoiding optimization bias caused by fixed weights.

[0042] In one embodiment of the present invention, preferably, for the initial particles: a set number of particles is provided. Maximum number of iterations PSO initialization and termination inertia weights crossover rate in genetic algorithm Variation rate and define the emissivity boundary. ; Randomly generated initial particles and speed ,in ; The multi-stage evolutionary strategy includes: In the iterative algebra g <G mid Perform a global particle swarm search: update speed And obtained a new position ; When g is divisible by the preset crossover period, the first one with the smallest congestion distance... Individual particles undergo single-point crossover and mutation using a genetic algorithm. A specific execution scheme for multi-stage collaborative optimization is provided. Limiting particle swarm parameters and emission rate boundaries ensures the rationality of the initial solution. Global wide-area exploration is achieved through the particle swarm velocity update formula. Crossover and mutation operations are introduced into the genetic algorithm at preset periods to enhance local search capabilities. Non-dominated sorting and crowding distance are combined to select elite particles, forming a high-quality Pareto front. The search range is narrowed in stages, and the velocity upper limit is reduced, achieving a smooth transition from global exploration to local refinement and improving convergence efficiency.

[0043] In one embodiment of the present invention, preferably, in the target conflict handling mechanism, when Above the threshold At the same time, increase the weight of conflict targets and inject random perturbations into particle updates.

[0044] In one embodiment of the present invention, preferably, in the environmental adaptability enhancement mechanism, This is the magnification factor. This is the current environmental complexity factor. It enhances the algorithm's adaptability to environmental disturbances. Based on the fluctuation of background radiation, the environmental complexity factor dynamically amplifies the penalty term for particles that violate constraints or experience sudden increases in error. The penalty mechanism is automatically strengthened when reflected radiation changes drastically, suppressing the impact of outliers on the optimization process and significantly improving the algorithm's stability and fault tolerance in complex high-temperature environments. Dynamic intervention is implemented for target conflict problems. The degree of conflict is monitored in real time by quantifying the ratio of the temperature to the emissivity objective function. When the conflict exceeds a threshold, the corresponding target weight priority is actively increased, and random perturbations are injected into the particle update process. This mechanism effectively breaks the optimization stagnation caused by mutual constraints between targets, maintains the population's evolutionary vitality, and ensures the continuous optimization process of the Pareto front.

[0045] In one embodiment of the present invention, preferably, when the iterative algebra g=G max , or max| |< Stop at time, extract the last Pareto front And based on the knee curvature: The solution with the largest curvature is selected as the output. Optimization termination conditions and optimal solution selection criteria are established. The timing of algorithm termination is controlled by the maximum number of iterations or a fitness change threshold to ensure reasonable allocation of computational resources. Based on the curvature formula, the trade-off characteristics of each point in the Pareto solution set are quantified, and the knee point solution with the largest curvature is automatically identified as the final output. This mechanism ensures that the selected solution achieves an optimal balance between temperature accuracy and emissivity rationality, avoiding subjective selection bias.

[0046] In one embodiment of the present invention, preferably, the objective function for the temperature error is: For normalized radiation signals, This is a temperature prediction function; Five-point median filtering and dark current correction were performed on the original spectral sequence to obtain the denoised spectral matrix. ; ; The objective function for emissivity error is: in, For the second-order difference of emissivity, The theoretical central value, and The weighting coefficients are defined. A clear method for quantifying temperature and emissivity errors is established. The temperature error objective function establishes the correlation between spectral data and temperature estimation by normalizing the sum of squared residuals from the radiation signal and the temperature prediction model. The emissivity error objective function integrates second-order difference smoothing constraints and theoretical center value deviation penalties to ensure that the emissivity curve conforms to physical characteristics. Both provide a computable mathematical basis for multi-objective optimization.

[0047] In one embodiment of the present invention, preferably, the following is performed before defining the environment complexity factor: Perform five-point median filtering on the radiated signal V(λ,t); Dark current correction is performed on the filtered signal; Separating background reflected radiation from the correction signal : By time window Calculate the standard deviation Standardize the radiation signal preprocessing procedure. Median filtering suppresses impulse noise in the spectral data, and dark current correction eliminates detector background errors, improving the quality of the original signal. Background reflection components are separated from the corrected signal, and their standard deviation is calculated to provide accurate input for the environmental complexity factor, laying the data foundation for subsequent optimization modules.

[0048] In one embodiment of the present invention, preferably, the local refinement of the genetic algorithm includes: When g≥G mid Shrink the particle position boundary to ; Adaptive variability The implementation strategy for the local optimization stage is refined. In the later stages of the algorithm, the particle position boundaries are narrowed, focusing on high-potential solution regions for deep searching. An adaptive mutation rate mechanism is adopted, dynamically adjusting the mutation intensity according to the evolutionary stage to avoid excessive perturbation of converged solutions in the later stages. This measure significantly improves the targeting and efficiency of the local refinement process, ensuring the accuracy of the final solution.

[0049] This invention provides a multi-objective optimization system for multispectral temperature measurement of aero-engine turbine blades, comprising: The multispectral signal acquisition module is used to acquire the radiation signals of turbine blades. Background reflected radiation and thermocouple reference temperature ; The preprocessing module is configured as follows: For radiation signals Perform five-point median filtering and dark current correction to generate a denoised spectral matrix. ; Calculate the standard deviation of background radiation To define the environmental complexity factor ; Normalized radiation signal ; The dynamic weight adjustment module is configured as follows: Constructing the objective function of temperature error and emissivity error objective function ; Calculate weights based on real-time error. , Constructing weighted fitness ,in, and Do not represent the error in temperature and emissivity; The multi-stage evolution optimization module is configured as follows: In the iterative algebra g <G mid (G) mid = 0.5 × G max G max Perform a global particle swarm search (to maximize the number of iterations) and update the particle positions: ; in, Let be the PSO iterative learning factor, ranging from [1.0 to 1.6]. This is the random weight in PSO, used to increase the randomness of the search and avoid early convergence. Its value range is (0,1). When g≥G mid The genetic algorithm is combined with local refinement to find the first few with the smallest crowding distance. Each particle performs single-point crossover and mutation; The target conflict handling module is configured as follows: Calculate the target conflict quantity ; when Above the threshold At the same time, increase the weight of conflicting targets and inject random perturbations into particle updates; The environmental adaptability enhancement module is configured as follows: Utilizing environmental complexity factors Dynamically amplified penalty items ,in This is the magnification factor. This represents the complexity factor of the current environment. The knee point output module is configured as follows: When the iterative algebra g=Gmax , or max| |< Stop at time, extract the last Pareto front ; According to the curvature formula Select the knee point with the largest curvature. As the final temperature and emissivity estimates, a complete multispectral thermometry multi-objective optimization system architecture is constructed. A multispectral signal acquisition module synchronously acquires radiation signals, background reflections, and reference temperatures, ensuring data source reliability. A preprocessing module performs signal filtering, correction, and normalization, and calculates an environmental complexity factor to quantify environmental interference. A dynamic weight adjustment module dynamically allocates dual-objective weights based on real-time errors, enabling the fitness function to be adaptive. A multi-stage evolutionary optimization module integrates global particle swarm optimization and local refinement using genetic algorithms, ensuring efficient exploration of the solution space. A target conflict handling module monitors the degree of target conflict and injects perturbations to prevent optimization stagnation. An environmental adaptability enhancement module dynamically strengthens the penalty mechanism for abnormal solutions based on environmental complexity. A knee point solution output module automatically selects the optimal compromise solution through curvature analysis, ultimately forming a systematic solution integrating data acquisition, dynamic optimization, conflict resolution, and environmental adaptation.

[0050] In one embodiment of the present invention, a multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades includes the following steps: (1) Radiation signals were collected on the working surface of the turbine blade using a spectrometer in the 1100 nm-1600 nm band. Simultaneously record background reflected radiation and thermocouple reference temperature The blade model and adjacent blade models are then imported into the temperature measurement system to obtain geometric information.

[0051] (2) Perform five-point median filtering and dark current correction on the original spectral sequence to obtain the denoised spectral matrix. By time window Calculate the standard deviation Define an environmental complexity factor to measure the magnitude of the current change in reflected radiation: (3) Normalize the radiation at each time point to a relative value: And establish a prediction function for subsequent calculation of temperature error: (4) Set the number of particles Maximum number of iterations PSO initialization and termination inertia weights crossover rate in genetic algorithms Variation rate and define the emissivity boundary. .

[0052] (5) Randomly generated initial particles and speed ,in (6) Calculate the objective functions for temperature error and emissivity error in each generation g: (7) Calculate dynamic weights based on the current two types of errors. And construct a weighted fitness (8) Introduce target conflict quantity when Above the threshold At that time, the weight of the corresponding target is increased and random perturbations are injected into the particle update to alleviate the stagnation caused by the conflict. Specifically, in the g-th generation, the conflict degree values ​​of all particles are collected. There are Np numbers in total. The number at the 80th percentile after sorting these numbers by size is used as the threshold. .

[0053] (9) Perform a global PSO search: update speed And obtained a new position ; in, Let be the PSO iterative learning factor, ranging from [1.0 to 1.6]. This is the random weight in PSO, used to increase the randomness of the search and avoid early convergence. Its value range is (0,1).

[0054] (10) When g is divisible by the preset crossover period, the nearest congestion distance is selected. Each particle undergoes single-point crossover and mutation to generate offspring, which are then merged with the PSO results. The offspring are then retained based on non-dominated sorting and crowding distance. Elite particles form the current generation of the Pareto frontier. This refers to the non-dominated sorting level during the optimization process.

[0055] (11) If g <G mid The algorithm maintains a large speed limit and wide-area weights. Perform a global exploration; when g ≥ G mid By gradually reducing the speed limit and narrowing the position boundary, the search focus is shifted to a local refinement stage, achieving multi-stage evolution.

[0056] (12) Using environmental complexity factors according to Amplify the penalty term for particles that violate the boundary or experience a sudden increase in error. This ensures that convergence stability is maintained even when background reflected radiation fluctuates drastically.

[0057] (13) Iteration of g=G max , or max| |< Stop at time, extract the last Pareto front And based on the knee curvature: Determine the most compromise-oriented solution .

[0058] (14) Output This serves as the result of estimating the real-time temperature and emissivity of the blade.

[0059] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the embodiments of the present invention. Other modifications can be readily implemented by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the embodiments of the present invention are not limited to the specific details.

Claims

1. A multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades, characterized in that, Includes the following steps: Collecting radiation signals from the working surface of aero-engine turbine blades Simultaneously record background reflected radiation and thermocouple reference temperature Based on the background reflected radiation, an environmental complexity factor is defined. , The standard deviation of background radiation; Based on the preprocessed radiation signal, temperature error objective functions are constructed respectively. and emissivity error objective function ; A multi-stage evolutionary strategy is employed to jointly optimize the temperature error objective function and the emissivity error objective function, with an iteration number g. <G mid Perform a global particle swarm search and update particle positions; when g ≥ G mid Combine genetic algorithms to perform local refinement; During the optimization process, Based on the real-time errors of the two objective functions, their weights in the weighted fitness function are dynamically adjusted. , Constructing weighted fitness The calculation formula is: ; Through a target conflict handling mechanism, based on the target conflict quantity between two objective functions... To inject random perturbations into the particle update process, the formula for calculating the target collision quantity is: ; Through an environmental adaptability enhancement mechanism, the intensity of the penalty term for outlier solutions is dynamically amplified using the aforementioned environmental complexity factor. ; From the set of optimal Pareto solutions obtained through optimization, select the Pareto front knee solution. As the optimal estimate of the final real-time temperature and emissivity of the blade; The dynamic weight adjustment mechanism satisfies the following: ； in, and These represent the errors in temperature and emissivity, respectively.

2. The multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades as described in claim 1, characterized in that, For the initial particles: set the number of particles. Maximum iteration algebra PSO initialization and termination inertia weights crossover rate in genetic algorithm Variation rate and define the emissivity boundary. ; Randomly generated initial particles and speed ,in ; The multi-stage evolutionary strategy includes: In the iterative algebra g <G mid Perform a global particle swarm search: update speed ， And obtained a new position ; When g is divisible by the preset crossover period, the first one with the smallest congestion distance... Each particle undergoes a genetic algorithm involving single-point crossover and mutation.

3. The multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades as described in claim 1, characterized in that, In the aforementioned target conflict handling mechanism, when Above the threshold At the same time, increase the weight of conflict targets and inject random perturbations into particle updates.

4. The multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades as described in claim 1, characterized in that, In the aforementioned environmental adaptation enhancement mechanism, ， This is the magnification factor. This represents the complexity factor of the current environment.

5. The multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades as described in claim 2, characterized in that, When the iterative algebra g=G max , or max| |< Stop at time, extract the last Pareto front And based on the knee curvature: ， The solution with the largest curvature is selected as the output.

6. The multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades as described in claim 2, characterized in that, The objective function for the temperature error is: ， ， ， For normalized radiation signals, For temperature prediction function, For the current estimated temperature, c2 is Planck's constant; Five-point median filtering and dark current correction were performed on the original spectral sequence to obtain the denoised spectral matrix. ; ; The objective function for emissivity error is: ， in, For the second-order difference of emissivity, The theoretical central value, and These are the weighting coefficients.

7. The multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades as described in claim 1, characterized in that, Execute before defining the environment complexity factor: Perform five-point median filtering on the radiated signal V(λ,t); Dark current correction is performed on the filtered signal; Separating background reflected radiation from the correction signal : By time window Calculate the standard deviation .

8. The multi-objective optimization method for multispectral temperature measurement of aero-engine turbine blades as described in claim 5, characterized in that, Local refinement in genetic algorithms includes: When g≥G mid Shrink the particle position boundary to ; Adaptive variability .

9. A multi-objective optimization system for multispectral temperature measurement of aero-engine turbine blades, characterized in that, It is used to implement the method described in any one of claims 1 to 8.

Images