Method for evaluating residual life of high-temperature structure

Abstract

The invention belongs to the technical field of condition-based maintenance support, and particularly relates to a high-temperature structure residual life evaluation method, which comprises the following steps of: extracting a service part and sampling an in-situ test piece of a dangerous part; determining a test load according to multi-physics field simulation and an equal damage theory, carrying out an in-situ life test under a vacuum condition, and obtaining a surface displacement field, a microstructure morphology, a macroscopic strain rate and life data in the test process; taking the data as input, and training a life prediction model through machine learning; and sampling batch parts of which the residual life needs to be evaluated, and evaluating the residual life by using the trained life prediction model. According to the method for evaluating the residual life of the high-temperature structure, the residual life of the high-temperature structure is accurately predicted by directly sampling the high-temperature structure after service and combining an in-situ test and a machine learning technology in a vacuum environment, and a scientific basis is provided for transformation of equipment from regular maintenance to condition-based maintenance.

Description

Technical Field

[0001] The present invention belongs to the technical field of condition-based maintenance and support for equipment, and particularly relates to a method for evaluating the remaining life of high-temperature structures. Background Art

[0002] In recent years, high-temperature structures such as gas turbine blades, aero-engine turbine blades, and nuclear reactor components have played a crucial role in fields such as aerospace, energy, metallurgy, and chemical engineering. These key components are often in extreme service environments such as high temperature, high pressure, and high stress, and their structural integrity and remaining life are directly related to the safety and reliability of the equipment. Therefore, how to accurately evaluate the remaining life of high-temperature structures has become an important technical issue for ensuring the safe operation of equipment.

[0003] On the other hand, the replacement cost of high-temperature structures is usually relatively high. As domestic equipment reaches its major overhaul age in the future, accurate remaining life assessment can extend the service life of high-value structures and realize the transformation of equipment from "time-based maintenance" to "condition-based maintenance", resulting in significant economic benefits.

[0004] In the field of evaluating the remaining life of high-temperature structures, the existing technologies mainly follow three technical paths, but each has certain limitations.

[0005] The first technical path is to record the data of the structure during service and combine it with a pre-established life model to evaluate the remaining life. For example, Chinese patents with publication numbers CN118036389A and CN118013814A disclose a method for predicting the remaining life of aero-engine blades and a method for predicting the life of high-temperature gas-cooled turbine blades respectively. These literatures judge whether the structure has reached its design life based on the existing life model and load history. However, when evaluating the remaining life using such methods, the current material state of the actual structure is not taken into account, so the service potential of the structure may not be fully exploited, resulting in a conservative evaluation result.

[0006] The second technical path is to simulate the material state after service in the laboratory and conduct corresponding tests to predict the remaining service life of high-temperature structures. Chinese patent with publication number CN117828778A discloses a method for predicting creep life based on the Larson-Miller curve through long-term thermal exposure tests. This solution obtains a life model by artificially preparing deteriorated materials and conducting life tests. However, a significant drawback of this method is that it cannot truly reflect the material and damage state of the actual structure matrix because there may be a large difference between laboratory conditions and actual service environments.

[0007] The third technical approach is to sample from the actual structure and determine the remaining life by analyzing the microstructure morphology of the sample. For example, a Chinese patent with the publication number CN111008495A discloses a method for predicting the creep remaining life of nickel-based single crystal turbine blades, which is to obtain the microstructure morphology by sampling and predict the remaining life based on the established life model. However, there is no one-to-one correspondence between the microstructure morphology of the material and its historical service process. Therefore, this method may not be able to accurately describe the actual damage state of the material, thus affecting the accuracy of the remaining life assessment.

[0008] From the perspective of the macroscopic failure mechanism, the failure of high-temperature structures is mainly caused by component surface damage and the failure of the material matrix inside the structure. Under the condition-based maintenance system, the damage on the component surface can be monitored and eliminated through regular in-situ non-destructive testing. However, the material matrix inside the structure will be subjected to the combined action of multiple factors such as high temperature, mechanical stress, oxidation corrosion, etc. during service. Its failure mechanism is quite complex and is usually accompanied by the evolution of the microstructure and the degradation of material properties. It is impossible to predict the remaining life based on the life results of standard parts.

[0009] Therefore, it is impossible to accurately predict the remaining life of the actual structure through the life test results of standard parts. This situation highlights the deficiencies of the existing technologies in the assessment of the remaining life of high-temperature structures. In addition, there are also some obvious limitations in the existing test methods. When the material is processed into small-sized specimens and subjected to high-temperature tests under atmospheric conditions, the thin-wall effect caused by oxidation will significantly reduce the life of the material, thus unable to truly reflect the actual remaining serviceability of the material matrix. On the other hand, the existing test means have clear requirements for the size of the test specimens, especially for high-temperature tests, which limits the possibility of sampling tests on the characteristic parts of the components. And currently commonly used knurled fixtures mainly rely on friction to fix the test specimens, the installation process is rather difficult and may also affect the accuracy of the test results. In addition, the microstructure and surface damage characteristics of the material usually exist at the micron / sub-micron scale, and conventional observation means cannot observe the changes of these characteristics in real time, which limits our in-depth understanding of the material failure process.

[0010] Therefore, there are many deficiencies in the existing methods for assessing the remaining life of high-temperature structures and they cannot truly and accurately assess the remaining serviceability of the structure. In order to meet the needs of the industrial community for condition-based maintenance, there is an urgent need to provide a new set of remaining life assessment methods and complete processes to overcome the limitations of the existing technologies and improve the accuracy and reliability of the remaining life assessment. Summary of the Invention

[0011] In view of this, the present invention aims to solve the problems of insufficient accuracy and technical limitations in the field of high-temperature structure remaining life assessment. The present invention discloses a method for assessing the remaining life of a high-temperature structure, which obtains a plurality of data through in-situ test piece sampling and in-situ life tests under vacuum conditions, and uses machine learning to train a life prediction model to more realistically and accurately evaluate the remaining serviceability of the high-temperature structure and improve the accuracy and reliability of the assessment.

[0012] To achieve the above object, the technical solution of the present invention is realized as follows:

[0013] A method for assessing the remaining life of a high-temperature structure, comprising:

[0014] S1: According to the in-situ test loading scheme of the micro-sized sampling test piece, extract the service components and conduct in-situ test piece sampling at the dangerous parts;

[0015] S2: Determine the test load according to multi-physical field simulation and the equal damage theory, conduct in-situ life tests under vacuum conditions, and obtain the surface displacement field, microstructure morphology, macroscopic strain rate and life data during the test through non-contact mechanics methods;

[0016] S3: Take multiple surface displacement fields, microstructure morphologies, macroscopic strain rates and life data obtained from the in-situ test as inputs, and train a life prediction model through machine learning;

[0017] S4: Sample the batch of components whose remaining life needs to be evaluated, and conduct remaining life assessment through the life prediction model.

[0018] Further, in step S1, based on historical failure records and non-destructive testing records, combined with full three-dimensional multi-physical field simulation, judge the dangerous parts, determine the sampling position and direction of the micro in-situ test piece, and determine the sampling scheme according to the local structure geometric characteristics.

[0019] Further, in step S1, design and check the fixture according to the in-situ test loading scheme to form the geometric structure of the micro in-situ test piece, then sample from the batch of components whose remaining life needs to be determined, remove the deformed layer of the material, and check the microstructure of the sampled piece.

[0020] Further, the loading scheme includes two asymmetric snap clamps and a heating table. The asymmetric snap clamps are arranged on both sides of the test piece in the length direction. The asymmetric snap clamps are used to provide the axial tensile load of the test piece, and the heating table is used to heat the assessment part and limit the displacement in the normal direction of the test surface.

[0021] Further, the heating table includes:

[0022] Heating ceramics, which are used to generate heat and heat the assessment section of the in-situ test piece;

[0023] Two thermocouples are set, one of which is connected to the heating ceramic to form a negative feedback channel to ensure accurate application and maintenance of temperature, and the other is on the upper surface of the test piece to determine whether the test part has reached thermal equilibrium;

[0024] Graphite sheet, used to transfer heat to the upper surface of the test piece;

[0025] An elastic base is arranged in a clamping slot of the asymmetric clamping fixture and is used to limit the normal phase displacement of the surface of the test piece;

[0026] The heat insulation baffle is used to protect the electron gun of the scanning electron microscope during high temperature experiments. It is cooled by a serpentine water cooling channel and its horizontal movement is controlled by a servo motor.

[0027] Furthermore, in step S2, a non-contact mechanical testing method is used to obtain the constitutive relationship of the local material after service, and the full three-dimensional multi-physical field simulation results are dynamically corrected based on the constitutive relationship of the material after service, and the in-situ test load spectrum is determined based on the equal damage principle. The test load spectrum is any one of the creep damage design load spectrum, the creep-low cycle interaction load spectrum, and the variable load creep load spectrum.

[0028] Furthermore, the in-situ test load spectrum is determined based on the equal damage principle, including the following steps:

[0029] Extract creep duration and low-cycle load cycles based on actual service load spectrum;

[0030] Determine the creep load and low cycle load cycle of the characteristic part by combining the characteristic part load under each working condition determined by the multi-physics field simulation in step S1;

[0031] When it is detected that the damage of the specimen under creep load is greater than the order of magnitude of low-cycle fatigue damage, only the creep damage design load spectrum is considered; or, when it is detected that the low-cycle fatigue damage of the specimen is greater than the order of magnitude of damage under creep load, only the low-cycle fatigue damage design load spectrum is considered; otherwise, the creep-low-cycle interaction load spectrum is designed according to the damage ratio of the two.

[0032] Furthermore, in step S2, the non-contact mechanical method includes a digital image correlation method or a moiré method, and during the in-situ life test of the sample, the strain-time curve and surface displacement distribution of the sample are obtained by the non-contact mechanical testing method, and the microstructure information is extracted by using a digital image processing scheme.

[0033] Furthermore, in step S2, a scanning electron microscope with high magnification and large depth of field is combined with the test to observe the microstructure and surface damage evolution of the sample in situ, and the scanning electron microscope uses a controlled electron beam to scan point by point for imaging.

[0034] Further, in step S3, a hybrid model that combines a long short-term memory network and a convolutional neural network is used for data processing and feature extraction to identify displacement field features, microstructure features, creep rate, and their implicit relationship with the remaining life. The long short-term memory network can capture dependencies in time series, and the convolutional neural network can automatically extract local spatial features.

[0035] Compared with the prior art, the method for evaluating the remaining life of a high-temperature structure described in the present invention has the following advantages:

[0036] 1. The method for evaluating the remaining life of a high-temperature structure described in the present invention accurately locates the dangerous parts of the high-temperature structure through a combination of multi-physical field simulation analysis, directly samples at the dangerous parts of the in-service components, uses non-contact mechanical detection in a vacuum environment to realize real-time acquisition of key data such as the displacement field, microstructure morphology, and macroscopic strain rate on the surface of the specimen, and accurately determines the test load by combining multi-physical field simulation and the equal damage theory. On this basis, a machine learning model is used to deeply analyze and mine the internal laws of material degradation and life evolution from a large amount of in-situ test data, so as to accurately predict the remaining life of the high-temperature structure, providing a scientific basis for the transformation of high-temperature equipment from time-based maintenance to condition-based maintenance.

[0037] 2. The method for evaluating the remaining life of a high-temperature structure described in the present invention is simple to operate, low in cost, and high in data acquisition and processing efficiency, and can be widely applied to the remaining life prediction of high-temperature service structures such as gas turbine blades, aero-engine turbine blades, and nuclear reactor components. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a logical block diagram of the method for evaluating the remaining life of a high-temperature structure according to an embodiment of the present invention;

[0039] Figure 2 is a schematic diagram of the asymmetric bayonet structure according to an embodiment of the present invention;

[0040] Figure 3 is an overall schematic diagram of the test piece loading scheme according to an embodiment of the present invention;

[0041] Figure 4 is an example of continuously obtaining the surface damage and speckle images of the test piece in-situ under a scanning electron microscope;

[0042] The markings in the figure are indicated as:

[0043] 100 - Asymmetric snap clamp; 200 - Test piece; 300 - Heating ceramic; 400 - Graphite sheet; 500 - Thermocouple; 600 - Heat insulation baffle. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] Next, the technical solutions in the embodiments of the present application will be clearly described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present application.

[0045] In the description of the present application, it should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit the exemplary embodiments of the present application. For the convenience of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationship. Technologies, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but in appropriate cases, the said technologies, methods, and devices should be regarded as part of the authorization specification. In all the examples shown and discussed here, any specific value should be interpreted as merely exemplary, rather than as a limitation. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that: similar reference numerals and letters denote similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0046] It should be noted that the terms "first", "second", etc. in the description and claims of the present application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and do not limit the number of objects. For example, the first object can be one or more. In addition, "and / or" in the description and claims means at least one of the connected objects, and the character " / " generally means an "or" relationship between the associated objects before and after.

[0047] It should be noted that in the description of the present application, the orientation or positional relationship indicated by the orientation terms such as "front, back, up, down, left, right", "horizontal, vertical, perpendicular, horizontal" and "top, bottom" is usually based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present application and simplifying the description. Without contrary description, these orientation terms do not indicate and imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, so it cannot be understood as a limitation on the protection scope of the present application; the orientation terms "inside, outside" refer to the inside and outside relative to the contour of each component itself.

[0048] It should be noted that, in the present application, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be noted that the scope of the method and device in the embodiment of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0049] The applicant has found in current technical practice that taking out small-sized specimens from the service structure and conducting life tests is considered to be a method that can effectively reflect the true remaining service capacity and remaining life of the material matrix in a specific part. The advantage of this method is that it directly evaluates the actual material state during the service process and has high accuracy. On the other hand, with the continuous advancement of observation methods, it is now possible to record the damage and microstructural state of the material in real time during the test, which provides strong support for in-depth analysis of the failure mechanism of the material. In addition, the rapid development of machine learning makes it possible to mine implicit life laws from complex test phenomena, which further improves the accuracy and reliability of remaining life assessment.

[0050] Based on this, the present application discloses a method for assessing the remaining life of a high-temperature structure, comprising:

[0051] S1: According to the in-situ test loading plan of micro-size sampling test pieces, extract the service parts and conduct in-situ test piece sampling of dangerous parts;

[0052] S2: Determine the test load based on multi-physics field simulation and equal damage theory, carry out in-situ life test under vacuum conditions, and obtain the surface displacement field, microstructure morphology, macroscopic strain rate and life data during the test by non-contact mechanics methods;

[0053] S3: Multiple surface displacement fields, microstructure morphology, macroscopic strain rate and life data obtained from in-situ tests are used as input to train the life prediction model through machine learning;

[0054] S4: Sampling the batch of components whose remaining life needs to be evaluated, and evaluating the remaining life using the life prediction model.

[0055] The high-temperature structure remaining life assessment method disclosed in this application, as Figure 1 shown, selects representative dangerous parts for sampling in service components according to the in-situ test loading scheme of micro-sized sampling test pieces. During the sampling process, the key areas are accurately positioned according to the local geometric characteristics and actual service conditions of the components. By directly sampling from the high-temperature structure after service, it can truly reflect the deterioration effect of the service environment on the material matrix, avoiding the problem of mismatch between the laboratory simulation method and the actual service state. Then, the sampled test pieces are subjected to in-situ tests in a vacuum environment, effectively avoiding the oxidation effect of small-sized specimens at high temperatures and making the test results closer to the true performance of the material. During the test process, non-contact mechanical testing means are used to obtain dynamic data such as surface displacement distribution, microstructure morphology, and macroscopic strain rate. A large amount of collected data is preprocessed and used as input to train a life prediction model using machine learning methods to achieve intelligent analysis and prediction driven by data. This method effectively overcomes the local deviations caused by specimen size limitations, test environment differences, and processing damage in traditional assessment methods on the basis that the test data reflects the true service state, realizing seamless connection from the accurate collection of service state specimens, the efficient acquisition of dynamic data to the training and prediction of intelligent models, thus providing an accurate and reliable basis for the remaining life assessment of subsequent batches of components. The overall process of this method makes full use of advanced simulation technology, in-situ test means, and data intelligent analysis to comprehensively quantify the performance degradation and damage evolution law of high-temperature structure materials, providing real-time and effective data support and scientific guidance for equipment maintenance decision-making, and is especially suitable for the rapid assessment of a large number of in-service equipment.

[0056] The high-temperature structure remaining life assessment method described in the present invention uses the in-situ test method of micro-sized specimens at the dangerous parts of service components, and uses non-contact mechanical detection in a vacuum environment to realize the real-time collection of key data such as the surface displacement field, microstructure morphology, and macroscopic strain rate of the specimens. Combining multi-physics field simulation and equal damage theory to accurately determine the test load. On this basis, a machine learning model is used to deeply analyze a large amount of in-situ test data to explore the internal laws of material degradation and life evolution, so as to achieve accurate prediction of the remaining life of high-temperature structures, effectively solving the problems existing in traditional life assessment technologies such as damage introduced by specimen processing, mismatch between laboratory simulation and actual service environment, and data collection limitations. It can not only truly reflect the actual degradation state of the material matrix under high temperature, high pressure, and complex load conditions, but also timely capture the material damage evolution process in dynamic monitoring, providing a scientific basis for the transformation of equipment from time-based maintenance to condition-based maintenance. The process is simple, the data processing is efficient, the model prediction is accurate, and the applicable range is wide, and it can assess the remaining life of a large number of high-temperature components.

[0057] As a preferred example of the present application, in step S1, based on historical failure records and non-destructive testing records, combined with full three-dimensional multi-physics field simulation, the dangerous parts are judged, the sampling positions and directions of the micro in-situ test pieces are determined, and the sampling scheme is determined according to the local structural geometric characteristics. In a specific example of the present application, based on historical failure records and non-destructive testing records, combined with full three-dimensional multi-physics field simulation, a comprehensive analysis of the service structure is carried out. The simulation results are used to determine the areas in the structure with risk factors such as stress concentration, temperature anomaly, and load superposition, and then the representative dangerous parts are judged. According to the geometric characteristics of the local structure, such as thickness, curvature, boundary, and connection method, etc., the optimal sampling position and sampling direction of the micro in-situ test piece are determined. At the same time, the sampling scheme is further optimized in terms of angle and orientation by combining the first principal stress direction and the actual load action direction. That is, in the example of the present application, the sampling direction and the normal direction of the test piece plane are determined according to the first principal stress direction of the local dangerous part of the structure. The local geometric characteristics are finely measured and compared through finite element simulation to determine the size, shape, and sampling angle of the specimen, so as to ensure that the collected specimen can truly reflect the distribution of internal damage and deterioration of the material. During the sampling process, an optimized processing technology is adopted to reduce the additional damage caused by mechanical cutting or polishing, and the multi-physics field simulation results and on-site detection data are used to verify each other to achieve the accurate formulation of the sampling scheme for the dangerous parts.

[0058] The high-temperature structure remaining life assessment method disclosed in the present application realizes the comprehensive capture of the internal damage and degradation process of the material through the accurate identification of the dangerous parts of the service components and the optimized sampling of the micro-specimens, providing reliable data support for the condition-based maintenance of the equipment. It not only provides a solid data foundation for obtaining real surface displacement fields, microstructural morphologies, and macroscopic strain rate data through non-contact mechanical methods in subsequent in-situ tests, but also lays a foundation for the acquisition of high-quality input data for machine learning models, significantly improving the accuracy and real-time performance of the remaining life prediction.

[0059] As a preferred example of the present application, in step S1, the fixture is designed and checked according to the in-situ test loading scheme to form the geometric structure of the micro in-situ test piece, and then samples are taken by sampling from the batch components whose remaining life is to be determined, the deformed layer of the material is removed, and the microstructure of the sampled parts is inspected. In the example of the present application, in step S1, the special fixture is designed and checked according to the in-situ test loading scheme to form the accurate geometric structure of the micro in-situ test piece, and representative specimens are randomly selected from the batch components whose remaining life is to be determined for processing to remove the deformed layer caused by mechanical cutting and polishing. At the same time, the microstructure of the sampled parts is inspected in detail to ensure that the specimen truly reflects the service state of the material.

[0060] This setting significantly improves the accuracy and real-time performance of the remaining life assessment of high-temperature structures by optimizing the fixture loading scheme and the fine sampling and processing technology. It uses full three-dimensional multi-physics field simulation combined with historical failure records and non-destructive testing data to accurately determine the dangerous parts existing in the service components, thereby guiding the precise sampling of the micro-in-situ test specimens, and ensuring the authenticity of the microstructure of the specimens by removing the deformation layer introduced during the processing. It not only solves the problem of data distortion caused by small specimen size, processing damage and atmospheric oxidation in traditional sampling tests, but also effectively reduces the dispersion of the evaluation results, providing scientific and reliable data support for the transformation of equipment from scheduled maintenance to condition-based maintenance.

[0061] As a preferred example of this application, in step S1, when inspecting the microstructure of the sampling piece, for the nickel-based single-crystal superalloy used in advanced gas turbine blades, the inspection includes quantitative and qualitative analysis of the volume fraction, size and morphology of the strengthening phase (γ'), as well as the presence of microvoids and TCP phases. As a specific example of this application, the inspection of the microstructure of the sampling piece depends on the material characteristics. For the mainstream nickel-based single-crystal superalloy of advanced gas turbine blades, the microstructure mainly includes the volume fraction, size and morphology of the strengthening phase (γ'), microvoids, TCP phases, etc. High-resolution scanning electron microscopy and digital image processing are used to obtain the microstructure data of the sampling piece to ensure that the collected microstructure information can truly reflect the structural degradation and performance deterioration generated during the service of the material, significantly improving the ability to accurately reflect the actual service state of the material, and dynamically correcting the material constitutive relationship in combination with the full three-dimensional multi-physics field simulation results, effectively solving the problem of data distortion caused by processing damage and environmental interference during the traditional specimen sampling and testing process.

[0062] As a specific example of this application, the loading scheme includes two asymmetric snap clamps 100 and a heating table. The asymmetric snap clamps 100 are arranged on both sides of the test piece 200 in the length direction. The asymmetric snap clamps 100 provide the tensile load in the axial direction of the test piece 200, and the heating table is used to heat the assessment part and limit the displacement in the normal direction of the test surface. In the example of this application, by using the asymmetric snap clamps 100, it is convenient to install and clamp the test piece 200. Among them, the design scheme of the asymmetric snap clamps 100 is as follows:

[0063] The mounting groove in one clamp can closely fit all the clamping sections at one end of the in-situ test piece 200, and the mounting groove in the other clamp is only used to prevent the displacement of the in-situ test piece 200 in the tensile direction at the other end.

[0064] As Figure 2As shown in the figure, for the asymmetric snap clamp fixture 100 described in this application, the displacement perpendicular to the surface of the test piece 200 is jointly restricted by the bottom of the mounting slot and the heating table. When necessary, the flat slot of the mounting slot can be changed to a dovetail slot. This design reduces the number of fixtures and the installation steps. Moreover, since the centering of the test piece 200 is controlled by the slot surface during the installation process, the process of manual centering is avoided, and the installation accuracy of the test piece can be better controlled, thereby reducing the dispersion caused by the test process.

[0065] In the example of this application, the heating table includes:

[0066] A heating ceramic 300, which is used to generate heat and heat the assessment section of the in-situ test piece 200;

[0067] Two thermocouples 500 are provided. One is connected to the heating ceramic 300 and is used to form a negative feedback channel to ensure the accurate application and maintenance of temperature. The other is on the upper surface of the test piece 200 and is used to judge whether the assessment part reaches thermal equilibrium;

[0068] A graphite sheet 400, which is used to transfer heat to the upper surface of the test piece 200;

[0069] An elastic base is arranged in the mounting slot of the asymmetric snap clamp fixture 100 and is used to restrict the normal displacement of the surface of the test piece 200;

[0070] A heat insulation baffle 600 is used to protect the scanning electron microscope electron gun during high-temperature tests. It is cooled through a serpentine water-cooling channel and its horizontal movement is controlled by a servo motor.

[0071] In step S1 of this application, a loading scheme consisting of two asymmetric snap clamps 100 and a heating table is adopted. The asymmetric snap clamps 100 are respectively arranged on both sides of the test piece 200. The loading slot of one clamp can closely fit all the clamping sections at one end of the test piece 200 to ensure that the specimen is firmly fixed when an axial tensile load is applied. The loading slot of the other clamp is only used to prevent the specimen from displacing along the tensile direction, and the displacement of the specimen perpendicular to its surface is restricted through the cooperation of the bottom of the loading slot and the heating table. When necessary, the flat slot is changed to a dovetail slot to further improve the loading alignment accuracy. At the same time, the heating table is composed of a heating ceramic 300, two thermocouples 500, a graphite sheet 400, an elastic base, and a heat insulation baffle 600. The heating ceramic 300 generates uniform heat through voltage control to heat the assessment part of the test piece 200. One thermocouple 500 is connected to the heating ceramic 300 to form a negative feedback channel to ensure accurate temperature application. The other thermocouple 500 is arranged on the surface of the test piece 200 to monitor the thermal equilibrium state in real time, and the closed-loop control of temperature is achieved through the two thermocouples; the graphite sheet 400 is used for heat transfer and can be appropriately trimmed according to the area to be heated. For example, when the total length of the test piece is less than the size of the heating table, the size of the graphite sheet can be reduced to reduce the heating power at the clamping end and improve the design margin of the clamping end; the elastic base is used to restrict the normal displacement of the surface of the test piece 200, and its equilibrium position is located on the symmetry plane of the loading device to ensure the overall balance of the loading device. The heat insulation baffle 600 protects the electron gun of the scanning electron microscope through a serpentine water-cooling channel and is controlled by a servo motor to move horizontally, so as to collect key data such as the surface displacement field, microstructure morphology, and macroscopic strain rate of the test piece 200 during the test in real time through a non-contact mechanical testing method under vacuum conditions, and input a large amount of collected data into a machine learning model for life prediction after preprocessing. In a specific example of this application, the width-to-thickness ratio of the assessment part of the micro-sized sampling test piece 200 in step S1 is >2. To avoid thermal deformation caused by machining during sampling, the thickness of the micro-sized sampling test piece 200 should not be less than 0.6 mm; the strength of the test piece 200 and the fixture as a whole should be checked by finite element simulation. For high-temperature tests, since the temperature drops rapidly outside the assessment part, the strength requirements for the contact surface of the test piece and the fixture can be appropriately relaxed.

[0072] By adopting an optimized loading scheme consisting of two asymmetric snap clamps 100 and a heating table, this application effectively improves the specimen loading accuracy and loading stability. The overall process realizes seamless connection from specimen preparation, precise loading, dynamic data collection to data-driven intelligent life prediction, providing scientific and reliable data support for equipment condition-based maintenance.

[0073] As a preferred example of this application, in step S2, a non-contact mechanical testing method is used to obtain the constitutive relationship of local materials after service, and the full three-dimensional multi-physics field simulation results are dynamically corrected based on the constitutive relationship of the materials after service. The in-situ test load spectrum is determined based on the equal damage principle, and the test load spectrum is any one of the creep damage design load spectrum, the creep-low cycle interaction load spectrum, and the variable load creep load spectrum. As a specific example of this application, determining the in-situ test load spectrum based on the equal damage principle includes the following steps:

[0074] According to the actual service load spectrum, extract the creep duration and low-cycle load cycles;

[0075] Combined with the loads of the characteristic parts determined by the multi-physics field simulation in step S1 under various working conditions, determine the creep load and low-cycle load cycles of the characteristic parts;

[0076] When it is detected that the damage of the specimen under the creep load is greater than the order of magnitude of the low-cycle fatigue damage, only consider the creep damage design load spectrum. Or, when it is detected that the low-cycle fatigue damage of the specimen is greater than the order of magnitude of the damage under the creep load, only consider the low-cycle fatigue damage design load spectrum. Otherwise, design it as a creep-low cycle interaction load spectrum according to the damage ratio of the two.

[0077] In step S2 of this application, the load spectrum is determined by multi-physics field simulation and the equal damage theory. Among them, the multi-physics field simulation determines the magnitude of the load, and the equal damage theory determines the form of the load. According to the load characteristics of the service components, it can be designed as creep, creep-low cycle interaction, variable load creep, etc. When necessary, an accelerated test spectrum can be designed.

[0078] In step S2, the constitutive relationship of the local material after service is obtained in real time by using a non-contact mechanical testing method, and the obtained constitutive data is used to correct the full three-dimensional multi-physics field simulation results to accurately reflect the actual performance changes of the material under high temperature, high pressure and complex loads. At the same time, according to the creep duration and low-cycle load cycle data extracted from the actual service load spectrum, combined with the load information of each characteristic part determined by multi-physics field simulation in step S1, the equal damage principle is used to calculate the damage ratio of the material under creep and low-cycle fatigue conditions, and based on this, an in-situ test load spectrum that can reflect both the individual creep damage and the low-cycle fatigue interaction effect is designed, wherein when it is detected that the creep damage is significantly greater than the low-cycle fatigue damage, only the creep damage is considered, and vice versa. Otherwise, the creep and low-cycle fatigue interaction load spectrum is designed according to the damage ratio. This application organically combines non-contact mechanical testing methods, full three-dimensional multi-physics field simulation, equal damage principle and deep learning technology to achieve real-time acquisition and precise correction of local material constitutive relations after service, ensuring that the design of the in-situ test load spectrum can fully reflect the actual damage accumulation of the material under high temperature, high pressure and complex loads, thereby significantly improving the accuracy and real-time performance of the remaining life prediction, and providing reliable technical support for the transformation of equipment from scheduled maintenance to condition-based maintenance.

[0079] As a preferred example of the present application, in step S2, if the actual load spectrum of the service component structure is complex, multiple sets of in-situ test loads are set. In the example of the present application, by comprehensively collecting and analyzing the load data of the service structure under different operating conditions, identifying the various load cycles, temperature fluctuations and load interaction effects experienced by the structure in actual working conditions, and combining the full three-dimensional multi-physics field simulation, historical failure records and on-site non-destructive testing data to finely divide the duration, amplitude and action frequency of various loads, the complex load spectrum of the structure during service is determined. The complex load spectrum includes not only long-term continuous creep loads, but also short-term low-cycle fatigue cycles and the mixed load effects formed by the interaction between the two. For example, the complex load The load spectrum includes a combination of load cycles, temperature fluctuations and load interaction effects under various operating conditions. In step S2, multiple groups of in-situ test loads are set, and each group of load spectra is independently designed for pure creep loads, low-cycle fatigue cycles, and interactions between creep and low-cycle fatigue. In addition, for structures with intermittent overloads, severe load fluctuations, and diverse operating conditions, the load duration, amplitude, and frequency under each operating condition are carefully divided to achieve comprehensive coverage of various load states under complex service environments, thereby avoiding the remaining life prediction deviation caused by the use of a single load spectrum for extrapolation prediction.

[0080] As a preferred example of the present application, in step S2, the non-contact mechanics method includes digital image correlation method or moiré method. During the in-situ life test of the sampled part, the strain-time curve, surface displacement distribution of the specimen are obtained respectively through the non-contact mechanics test method, and the microstructure information is extracted by using a digital image processing scheme. In a specific example of the present application, the non-contact mechanics test method adopted in step S2 is limited to the digital image correlation method or the moiré method. This method uses a high-resolution digital camera to continuously collect the image sequence generated when the specimen is loaded during the in-situ life test, and through advanced image processing algorithms, feature tracking and quantitative analysis are carried out on the specimen surface, so as to obtain the strain-time curve and surface displacement distribution of the specimen in real time, and the microstructure information is extracted through the digital image processing scheme, realizing the dynamic monitoring of the surface damage and microstructure evolution process of the specimen. This test method avoids the interference and installation errors introduced by traditional contact sensors, and can accurately reflect the true degradation state of the material under high temperature, high pressure and complex loads, thus significantly improving the accuracy and repeatability of data acquisition.

[0081] As a preferred example of the present application, in step S2, a scanning electron microscope with high magnification and large depth of field is combined with the test for in-situ observation of the microstructure and surface damage evolution of the specimen. The scanning electron microscope uses a controlled electron beam to scan point by point for imaging. In the example of the present application, in order to achieve accurate in-situ observation of the microstructure and surface damage evolution of high-temperature structural materials during service, in step S2, a scanning electron microscope with high magnification and large depth of field characteristics is combined with the in-situ test system, which can realize in-situ observation of the microstructure and surface damage evolution during the experiment, and the cost is relatively low. This scanning electron microscope uses its excellent imaging resolution and wide depth of field to continuously collect the surface images of the specimen by controlling the electron beam to scan point by point during the test. Although the shooting speed is slow, under the service load conditions of high-temperature components, due to the low stress level borne by the specimen and the plastic deformation of the material during imaging can be ignored, the adverse effects brought by the slow imaging speed are avoided.

[0082] By combining a scanning electron microscope with high magnification and large depth of field characteristics with the in-situ test system in step S2 of the present application, not only the advantages of the scanning electron microscope in capturing microstructure details are fully utilized, but also the problem of slow imaging speed of the electron beam scanning point by point is effectively avoided by using the characteristic of small deformation of the material under low-stress service loads, thus realizing real-time, continuous and non-contact acquisition of the surface displacement field, microstructure evolution and surface damage of the specimen. The high-quality image data collected can truly reflect the degradation state of the material under high temperature, high pressure and complex load conditions after standardized processing and automated digital image analysis, and provide high-precision, stable and reliable input data for the accurate prediction of the remaining life based on the deep learning model.

[0083] As a preferred example of the present application, in step S3, a hybrid model that combines a long short-term memory network (LSTM) and a convolutional neural network (CNN) is used for data processing and feature extraction to identify displacement field features, microstructure features, creep rate, and their implicit relationship with the remaining life. Among them, LSTM can capture the dependencies in the time series, and CNN can automatically extract local spatial features. In the example of the present application, in step S3, by using a hybrid model that combines a long short-term memory network (LSTM) and a convolutional neural network (CNN), continuous real-time and non-contact acquisition of the surface displacement field, macroscopic strain rate, and microstructure morphology data of the specimen collected by digital image correlation method or moiré method during the in-situ test is realized. After obtaining the displacement data by comparing the process images with the initial no-load state image as a reference and performing normalization processing on it, at the same time, an open-source digital image processing program is used to automatically extract parameters such as the volume fraction, size, and morphology of the γ / γ' two phases in the nickel-based single crystal superalloy. Then, in the hybrid model constructed by tensorflow.keras, the CNN uses multiple convolutional layers and pooling layers to automatically extract local spatial features, and the LSTM captures the dependencies in the time series data through multiple LSTM layers. Subsequently, the high-dimensional features output by each layer are fused through a fully connected layer. Finally, the Adam optimizer and the mean squared error loss function are used to train and cross-validate the model to form a comprehensive description of the degradation behavior of the material under high temperature, high pressure, and complex load conditions, and then the remaining life of the material is accurately predicted using the trained deep learning model to obtain an accurate prediction of the remaining life of the subsequent batch of structures.

[0084] The method of training the life prediction model through machine learning in the present application not only overcomes the prediction deviation caused by discontinuous specimen data acquisition, insufficient feature extraction, and limitations of single models in traditional methods, but also the model obtained through training can accurately reflect the material degradation law under complex service conditions, providing a scientific, reliable, and real-time decision-making basis for equipment maintenance and condition-based maintenance.

[0085] In the example of the present application, the inputs of the life model in step S3 include the surface displacement distribution, microstructure morphology, macroscopic strain rate, and remaining life at a certain state during the test, where the remaining life = total test life - current test acquisition time.

[0086] As a preferred example of the present application, in step S4, sampling is carried out based on statistical principles to ensure that the selected samples can represent the service state of the overall batch of components, and the remaining life assessment results in step S4 can be used to adjust the maintenance plan of subsequent service components to achieve condition-based maintenance.

[0087] The high-temperature structure remaining life assessment method disclosed in this application takes small-sized specimens from the in-service structure and conducts in-situ life tests. By combining full three-dimensional multi-physical field simulation with historical failure records and non-destructive testing data, it accurately locates and optimally samples dangerous parts in the in-service components, such as those with stress concentration, temperature anomalies, and load superposition. It uses advanced processing techniques to remove the deformation layers introduced during mechanical cutting, grinding, and electrolytic polishing to ensure that the sampled parts truly reflect the service state of the material. At the same time, in a vacuum environment, key dynamic data such as the surface displacement field, microstructure morphology, and macroscopic strain rate of the specimen are collected in real-time through non-contact mechanical testing methods. Combining digital image correlation method, moiré method, and scanning electron microscope with high magnification and large depth of field, it realizes in-situ observation of the microstructure and surface damage evolution of the specimen. An automated digital image processing program is used to accurately extract microscopic features such as the volume fraction, size, and morphology of the γ / γ' two phases in nickel-based single crystal superalloys. Then, a hybrid deep learning model that combines long short-term memory network and convolutional neural network is used to perform standardized preprocessing, local spatial feature extraction, and time series dependence analysis on the collected continuous data, and the Adam optimizer and mean square error loss function are used for model training and cross-validation, thereby realizing dynamic correction of the degradation behavior of the material under high temperature, high pressure, and complex loads and accurate prediction of the remaining life. When facing complex service load spectra, covering various load cycles, temperature fluctuations, and load interaction effects, by setting multiple groups of in-situ test loads to simulate pure creep, low-cycle fatigue, and their interactions respectively, it comprehensively covers various service conditions, avoiding prediction deviations caused by extrapolation of a single load spectrum, thus providing scientific, real-time, and reliable data support for the transformation of equipment from time-based maintenance to condition-based maintenance, significantly extending the service life of key components of high-temperature structures, reducing maintenance costs, and improving the operation safety and overall efficiency of equipment. The high-temperature structure remaining life assessment method disclosed in this application, by taking out micro-sized sampling test pieces from the post-service structure and conducting life tests, fully considers the performance degradation of the material matrix in the service environment; by designing the load spectrum according to the equal damage principle, it can reflect the remaining life law of the structure under complex load conditions; by using machine learning to process a large amount of data obtained from in-situ tests and obtaining a life model, it can support the remaining life assessment of a large batch of high-temperature structures, with simple operation, low cost, and high data collection and processing efficiency, and can be widely applied to the remaining life prediction of high-temperature service structures such as gas turbine blades, aero-engine turbine blades, and nuclear reactor components.

[0089] This application discloses a high-temperature structure remaining life assessment method, specifically for assessing the remaining life of a certain type of aero-gas turbine turbine blade, including the following steps:

[0090] S1-1: For the main flight conditions, conduct a full three-dimensional multi-physics field simulation of the gas turbine blade to obtain the temperature, stress / strain fields of the structure under service conditions, and preliminarily select the positions with higher temperature, stress / strain as the dangerous parts;

[0091] S1-2: According to the inspection results of the retired blades in the field, increase or decrease and adjust the dangerous parts as appropriate. Taking the position of 1 / 3 of the blade height as an example, since the direction of the first principal stress is approximately parallel to the blade height direction, the sampling direction is determined to be parallel to the blade height direction;

[0092] S1-3: Combine the local geometric model to determine the size of the assessment section of the micro-sized sampling part as 0.7 mm × 1.5 mm;

[0093] S1-4: According to the size of the assessment section, design the clamping end of the test piece and the corresponding loading system as Figure 3 shown.

[0094] S1-5: Use slow wire electrical discharge machining to cut out a 0.8-mm-thick test piece from the specified position of the structure after service, and finish machining the geometry of the side of the test piece by finishing cut; the surface roughness obtained by slow wire electrical discharge machining is <1 μm. After manual polishing with 3000# SiC water sandpaper and mechanical polishing with 3-μm, 1-μm, and 0.3-μm alumina polishing fluid, the test piece is electro-polished with an electrolyte of 7% perchloric acid and 93% ethanol to remove the surface deformation layer caused by wire cutting; then, measure the size of the assessment section of the in-situ test piece, where the thickness is measured with a precision thickness gauge and the width is measured with a scale in a scanning electron microscope.

[0095] S2-1: Use the digital image correlation method with relatively simple sample pretreatment to measure the surface deformation of the test piece. First, determine the speckle particle size according to the required field of view size, and the field of view size should be comprehensively evaluated by considering typical damage size, surface microstructure evolution characteristic size, etc.; for the nickel-based single-crystal superalloy in this example, zirconia particles with an average particle size of 2 μm are selected to prepare the speckle. First, evenly distribute the particles in the alcohol solution through an ultrasonic cleaner, drop the suspension onto the surface, and quickly spin-dry it evenly through a spin coater, so that the zirconia particles are randomly and evenly distributed on the surface to form a speckle pattern.

[0096] S2-2: Calibrate the image. During the imaging process of the scanning electron microscope, there may be unfavorable factors such as distortion that are detrimental to subsequent analysis. Before the formal start of the experiment, it is necessary to correct the image distortion. First, select appropriate parameters such as the shooting time, current magnitude, working distance, and current magnitude. This can be judged by shooting multiple groups of images and comparing the noise level and data repeatability, and keep them consistent in subsequent shootings; use the method of translation experiment to verify the measurement accuracy. Move the test piece up and down and left and right multiple times and obtain images. The length of each translation is about 1 / 4 of the field of view; use the digital image correlation program to evaluate the distortion level and strain measurement accuracy, and judge whether it is necessary to correct the results. Preferably, the strain measurement accuracy should be better than 1e-3 strain.

[0097] S2-3: Conduct an in-situ tensile test to obtain the Young's modulus and Poisson's ratio of the material after service in the sampling direction in the linear elastic section. Install the micro-sized sampling test piece into the experimental equipment, continuously control the tensile force and heat it to the specified temperature, and stretch it to the local maximum load determined in S1-1; obtain the surface pictures before and after stretching, and analyze the surface displacement field through an open-source program; through the selected marked points, determine the average elongation in the tensile direction and the transverse direction based on the displacement changes before and after stretching, and then calculate the average strain. The Young's modulus and Poisson's ratio can be calculated according to the corresponding definitions based on the obtained average strain.

[0098] S2-4: Compare the tensile properties of the material in the material handbook and after service. In this example, for the nickel-based single-crystal superalloy, due to the performance degradation caused by the rafting of the strengthening phase, the influence on the material's constitutive is usually negligible. If the test results deviate significantly from the results in the material handbook, it is necessary to re-calibrate the simulation results and analyze the reasons for the change in the material's constitutive parameters.

[0099] S2-5: Conduct an in-situ life test according to the load spectrum determined by the full three-dimensional multi-physics field simulation results and the equal damage principle. Regularly obtain the surface images of the test piece during the experiment. As Figure 4 shown, when obtaining images with the scanning electron microscope, stop the crossbeam of the testing machine to avoid image distortion caused by vibration.

[0100] S3-1: Process the acquired in-situ data to obtain the machine learning input. In this example, the open-source code Ncorr is used to analyze the surface deformation displacement field. To avoid cumulative errors, the image obtained in the initial unloaded state is used as the reference image, and the process image is used as the target image. After obtaining the displacement data, it is standardized, with the displacement at the image center being 0. Based on the obtained surface displacement distribution - time history, the strain - time history is obtained through the method in S4 to obtain the macroscopic strain rate. In this example, since the evolution research on the microstructure evolution under high-temperature service conditions is relatively sufficient and the evolution process and key parameters are also relatively clear, to reduce the training cost, the γ / γ' two-phase morphology image is processed by the open-source digital image processing program ImageJ to extract the γ' volume fraction, length, and width dimensions.

[0101] S3-2: Build a hybrid model using tensorflow.keras; take the surface displacement distribution, macroscopic strain rate, microstructure morphology, test load conditions, and the real-time remaining life obtained after the test obtained in S3-1 as inputs; use CNN to extract image features, including two convolutional layers and a pooling layer, and finally extract high-dimensional features through a fully connected layer; use LSTM to process time series data, including two LSTM layers, and finally extract high-dimensional features through a fully connected layer; fuse the image features with the macroscopic strain rate, microstructure morphology, and load condition features, and further process through a fully connected layer, and finally output the predicted remaining life.

[0102] S3-3: Compile the model using the Adam optimizer and the mean squared error (MSE) loss function; in this example, take 4 groups of test results as the training set, 1 group as the validation set, and use cross-validation to enhance the training effect.

[0103] S4-1: Use the trained model for remaining life prediction. For other batches of structures, through in-situ sampling, obtain the microstructure information and conduct single-piece short-term tests. After obtaining the required prediction inputs, the remaining life of this batch of structures can be quickly evaluated; the newly added test results can also be used to improve the model.

[0104] Through the method of taking out small-sized sampling test pieces from the structure after service and conducting life tests, the present invention fully considers the performance degradation of the material matrix due to the service environment; by designing the load spectrum according to the equal damage principle, it can reflect the remaining life law of the structure under complex load conditions, and the solution has high feasibility.

[0105] The present invention processes a large amount of surface damage information and microstructure evolution information obtained from in-situ tests under a scanning electron microscope through machine learning and obtains a life assessment model, which can be used to quickly and accurately evaluate the remaining life of the other batches of structures.

[0106] The embodiments of the present application have been described above in conjunction with the accompanying drawings. Without conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present application, those of ordinary skill in the art can also make many forms without departing from the purpose of the present application and the scope protected by the claims, and all of them belong to the protection scope of the present application.

Claims

1. A method for assessing the remaining life of a high temperature structure, characterized in that ,include: S1: According to the in-situ test loading plan of micro-size sampling test pieces, extract the service parts and conduct in-situ test piece sampling of dangerous parts; S2: Determine the test load based on multi-physics field simulation and equal damage theory, carry out in-situ life test under vacuum conditions, and obtain the surface displacement field, microstructure morphology, macroscopic strain rate and life data during the test by non-contact mechanics methods; S3: Multiple surface displacement fields, microstructure morphology, macroscopic strain rate and life data obtained from in-situ tests are used as input to train the life prediction model through machine learning; S4: Sampling the batch of components whose remaining life needs to be evaluated, and evaluating the remaining life using the life prediction model.

2. The method for assessing the remaining life of a high temperature structure according to claim 1, characterized in that: In step S1, based on historical failure records and nondestructive testing records, combined with full three-dimensional multi-physics field simulation, the dangerous parts are judged, the sampling position and direction of the micro in-situ test piece are determined, and the sampling plan is determined according to the local structural geometric characteristics.

3. The method for assessing the remaining life of a high temperature structure according to claim 2, characterized in that: In step S1, a fixture is designed and calibrated according to the in-situ test loading scheme to form the geometry of the miniature in-situ test piece, and then a sample is taken from the batch of parts whose remaining life is to be determined, the deformed layer of the material is removed, and the microstructure of the sampled piece is inspected.

4. The method for assessing the remaining life of a high temperature structure according to claim 2, characterized in that: The loading scheme comprises two asymmetric snap-fit ​​fixtures (100) and a heating platform, wherein the asymmetric snap-fit ​​fixtures (100) are arranged on both sides of the length direction of the test piece (200), the asymmetric snap-fit ​​fixtures (100) are used to provide an axial tensile load to the test piece (200), and the heating platform is used to heat the test part and limit the displacement of the normal phase of the test surface.

5. The method for assessing the remaining life of a high temperature structure according to claim 4, characterized in that: The heating stage comprises: A heating ceramic (300) is used to generate heat to heat the test section of the in-situ test piece (200); Two thermocouples (500) are provided, one of which is connected to the heating ceramic (300) to form a negative feedback channel to ensure accurate application and maintenance of temperature, and the other is on the upper surface of the test piece (200) to determine whether the test part has reached thermal equilibrium; A graphite sheet (400) for transferring heat to the upper surface of the test piece (200); An elastic base, arranged in a clamping slot of the asymmetric clamp (100), and used to limit the normal phase displacement of the surface of the test piece (200); The heat insulation baffle (600) is used to protect the electron gun of the scanning electron microscope during high temperature tests. The heat insulation baffle (600) is cooled by a serpentine water cooling channel and is controlled to move horizontally by a servo motor.

6. The method for assessing the remaining life of a high temperature structure according to claim 1, characterized in that: In step S2, a non-contact mechanical testing method is used to obtain the constitutive relationship of the local material after service, and the full three-dimensional multi-physical field simulation results are dynamically corrected based on the constitutive relationship of the material after service. The in-situ test load spectrum is determined based on the equal damage principle. The test load spectrum is any one of the creep damage design load spectrum, the creep-low cycle interaction load spectrum, and the variable load creep load spectrum.

7. The method for assessing the remaining life of a high-temperature structure according to claim 6, characterized in that: Determining the load spectrum of the in-situ test based on the equal damage principle includes the following steps: Extract creep duration and low-cycle load cycles based on actual service load spectrum; Determine the creep load and low cycle load cycle of the characteristic part by combining the characteristic part load under each working condition determined by the multi-physics field simulation in step S1; When it is detected that the damage of the specimen under creep load is greater than the order of magnitude of low-cycle fatigue damage, only the creep damage design load spectrum is considered; or, when it is detected that the low-cycle fatigue damage of the specimen is greater than the order of magnitude of damage under creep load, only the low-cycle fatigue damage design load spectrum is considered; otherwise, the creep-low-cycle interaction load spectrum is designed according to the damage ratio of the two.

8. The method for assessing the remaining life of a high temperature structure according to claim 1, characterized in that: In step S2, the non-contact mechanical method includes a digital image correlation method or a moiré method, and during the in-situ life test of the sample, the strain-time curve and surface displacement distribution of the sample are obtained by the non-contact mechanical testing method, and the microstructure information is extracted by using a digital image processing scheme.

9. The method for assessing the remaining life of a high-temperature structure according to claim 8, characterized in that: In step S2, a scanning electron microscope with high magnification and large depth of field is combined with the test to observe the microstructure and surface damage evolution of the sample in situ, and the scanning electron microscope uses a controlled electron beam to scan point by point for imaging.

10. The method for assessing the remaining life of a high temperature structure according to claim 1, characterized in that: In step S3, a hybrid model combining a long short-term memory network and a convolutional neural network is used for data processing and feature extraction to identify displacement field characteristics, microstructure characteristics, creep rate and their implicit relationship with remaining life. The long short-term memory network can capture the dependencies in the time series, and the convolutional neural network can automatically extract local spatial features.

Images