Method and system for generating test specimen specifications for predicting component fatigue life

Abstract

The present invention relates to a method for generating a test specimen (COP) specification (TCS) for predicting the fatigue life of a component (CMP) and a system for performing the method, comprising: (a) determining a load condition (LDC) of the component (CMP), (b) providing a component (CMP) design (CDS), (c) performing a strength analysis of the component (CMP) design under the load condition (LDC) to determine: (i) a critical area (CRA) of the component (CMP), (ii) at least one stress-related parameter (SRP) of the critical area (CRA), and (d) providing at least one material condition (MCD) of the component (CMP) for at least the critical area of ​​the component (CMP). To assist the end user in determining which are the best tests to perform in order to obtain the most relevant data for fatigue prediction of a particular component, the method is characterized by the additional steps of: (e) providing a material model (MTM) (f) providing as input to the material model (MTM): (i) the at least one stress related parameter (SRP) and (ii) the at least one material condition (MCD) (g) the material model (MTM) generating as output: (i) a test specimen specification (TCS) for testing in a testing machine.

Description

Technical Field

[0001] The present invention relates to a method and system for predicting fatigue life of a component, comprising:

[0002] (a) determine the loading conditions of the component,

[0003] (b) provide component design,

[0004] (c) Perform strength analysis on the component design for this load condition to determine:

[0005] (i) Critical areas of components,

[0006] (ii) at least one stress-related parameter of the critical region,

[0007] (d) providing at least one material condition of the component for at least the critical region of the component. Background Art

[0008] This background description may contain concepts related to the present invention that disclose the problem of the system or process, if any, only some parts thereof are disclosed. Therefore, this disclosure should not be interpreted as evidence that the various aspects mentioned herein belong to part of the disclosure or prior art.

[0009] ASTM defines fatigue strength as

[0010] “The stress value at which failure occurs after a specified number of load cycles”.

[0011] ASTM defines the fatigue limit as

[0012] "The stress limit at which failure occurs when the number of cycles becomes very large".

[0013] In other words, this means that the fatigue limit is the stress level below which an infinite number of load cycles can be applied to a material without causing fatigue failure. When a material does not have a distinct limit, the term fatigue strength or endurance strength is used and defined as the maximum value of the fully reversed bending stress that a material can withstand for a specified number of cycles without fatigue failure.

[0014] Characteristics such as surface roughness, porosity, heat treatment and others are known to affect the material's response to fatigue loading.

[0015] A method for determining the dimensions and load of a flat test piece for material testing is known from CN 112 100 765 A. A method for determining the influence of defects on fatigue life is known from CN 109 614 721 A. Both publications assume that the material is a deterministic given property.

[0016] Document WO 2018 / 217903 A1 discloses a method for improving additive manufacturing by using a trained algorithm to classify defects.

[0017] Document EP 3 459 715 A1 shows a method for predicting defects by online monitoring.

[0018] Document WO2020216458A1 [N.Lammens] relates to a method and system for fatigue life prediction of additively manufactured parts taking into account local material properties. A machine learning system is trained with collected data, the collected data including data points related to maximum stress vs. failure cycles at different given processing steps, surface and volume conditions of an element, sample or material structure. In order to collect training data for the machine learning system, many tests need to be performed.

[0019] The machine learning approach proposed in W02020216458A1 relies on separately training the machine learning system covariance function, which determines how two individual test samples are related. A simplified interpretation of this covariance function is to think of it as a "distance" measure that includes all the different influencing parameters (e.g., surface roughness or printing orientation in an additively manufactured part). When two samples are "close" together, their results are similarly combined, while the opposite applies to samples "far away" from each other.

[0020] Typically, this covariance function (there are many different ones) is driven only by the input parameters (surface roughness, construction orientation, ...) and is not substantially affected by the output (the actual fatigue performance of the sample).

[0021] While characterization of fatigue performance is often performed by primary coupon testing, it is often difficult (if not impossible) to reproduce the exact conditions present in the component on these primary coupons. This limitation can be attributed to production limitations (e.g., certain angles cannot be printed in additive manufacturing without additional supports) as well as to different material responses (e.g., due to different thermal behavior of smaller coupons compared to larger components).

[0022] As a result, the end user is often faced with the problem of choosing which specimen to manufacture, since none of them will exactly match the conditions in the part. There are different approaches that are routinely used to deal with such problems:

[0023] 1. The user can use experience to select a set of specimens.

[0024] 2. Users use a very wide range of samples in the hope of capturing the most important behaviors. The disadvantage of this approach is that success depends largely on the experience of the operator. In addition, such testing activities rarely achieve truly optimal results and will result in more testing than is objectively necessary.

[0025] 3. To address this issue, especially when dealing with additive manufacturing, some researchers have proposed printing the actual part and cutting test samples from that part to ensure that the exact material conditions are achieved.

[0026] One problem with this approach is that for many parts the coupons must become impractically small (e.g. due to curvature of the part surface or simply the overall size of the entire part). Furthermore, when the actual part needs to be printed for testing, the benefits of performing material characterization at the coupon level (cheaper and faster) are completely lost.

[0027] The American Society for Testing and Materials [ASTM] has recognized the problem of fatigue characterization of additively manufactured materials in the past. ASTM has expended considerable effort in attempting to define new standards and methods applicable to this approach. Printing ASTM compliant geometries with additive manufacturing involves significant costs that the present invention aims to reduce. Another focus of the present invention is to ensure that the proper material condition is achieved.

[0028] It is an object of the present invention to help the end user determine which are the best tests to perform in order to obtain the most relevant data for fatigue prediction of a specific component. Summary of the invention

[0029] For the purposes of describing the present invention, fatigue behavior or fatigue may be characterized in terms of how long a component can withstand a certain cyclic load without failure.

[0030] The objects of the invention are achieved by the independent claims. The dependent claims describe advantageous developments and modifications of the invention.

[0031] The object of the present invention is to propose a method and system of initial definition type, which comprises the additional steps:

[0032] (e) provide material models,

[0033] (f) providing as input to the material model:

[0034] (i) at least one stress-related parameter and

[0035] (ii) at least one material condition,

[0036] (g) The material model generates as output test specimen specifications for testing in a testing machine.

[0037] The term "material model" according to the invention is different from the strict conventional definition of material model (ie stiffness and Poisson's ratio or similar). The invention extends the meaning of this term to modules according to the definition of claim 1 and the disclosure presented.

[0038] Based on the above-mentioned prior art and associated problems, the present invention is based on a finite element driven process to help the end user determine which are the best tests to perform in order to obtain the most relevant data for fatigue prediction of a specific component. Many advanced materials (such as composite materials or additively manufactured parts) exhibit complex fatigue behavior. The final performance of such materials and their derived components is highly dependent on specific material parameters (which may be, for example, surface conditions or the presence of defects) and the corresponding manufacturing process (curing cycle for composite materials, printing parameters for additive manufacturing). The spatial variability of these different parameters further increases the complexity of component performance evaluation (for example, it may be that the surface roughness in additive manufacturing depends on the orientation of the component surface within the build space of the additive manufacturing machine (which may be a printer)). Although the characterization of fatigue performance is usually performed by basic specimen testing, it may be difficult or even impossible to reproduce the exact conditions present in the component on these basic specimens. This limitation can be attributed both to production limitations (for example, certain angles cannot be printed by additive manufacturing without additional supports) and to different material responses (for example, due to different thermal behavior of smaller specimens to larger components). Therefore, traditionally, end users are often faced with the challenge of choosing which specimen to manufacture, because none of them will exactly match the conditions in the component. The present invention proposes a method in which a machine learning model is combined with finite element analysis and a description of manufacturing constraints to automatically determine which specimens are beneficial or ideal to manufacture and test to achieve improved or even maximum accuracy in fatigue prediction of the component.

[0039] The present invention proposes to use a machine learning model, which can also be understood as a trained covariance function, to calculate how "close" or "far" a potential new test specimen is from a critical condition in a component. According to the present invention, this distance metric is independent of the actual sample performance, and the selection of the best test specimen can be performed purely by computer implementation, respectively, digitally, without the need for testing. After the classification and identification of the best specimens, actual testing can be performed to enrich the material model training data and reduce the uncertainty of the prediction near the critical area of ​​the component.

[0040] Prior art in testing relies on experience (no guarantee of success) or cutting samples from actual parts (high cost, not always feasible). The process according to the present invention allows for the best test campaign to be repeatable and independent of operator choice, and multiple objectives can be considered (e.g., in the above example, focusing on critical areas, but the same approach can be used to focus on the overall accuracy of the material model). A key part of the present invention is the use of a machine learning model (and its covariance function) to identify which samples are most relevant to the conditions of interest.

[0041] The concept of identifying the impact of hypothetical new training points on machine learning predictions is a well-known area in AI / machine learning.

[0042] Test specimens usually have a predetermined geometry (so they fit the testing machine and / or according to some standard) and are usually printed in near net shape. These specimens are tested in a common testing machine that applies a unidirectional load along the length of the specimen.

[0043] The stress-related parameter may be any parameter related to the stress condition determined in the strength analysis. The stress-related parameter may be an "equivalent stress" value, which may include effects such as residual stress (which may substantially change the "mean stress" value). In this way, the residual stress may not be treated as a separate parameter, but may be indirectly included in the stress parameter. According to a preferred embodiment, the stress-related parameter may be an equivalent stress determined from the mean stress or residual stress and variable stress. Such calculations are known from conventional durability solvers.

[0044] According to a preferred embodiment, the temperature history of the component may be a material condition. In this case, the material model may include a relationship that predicts how fatigue life varies with different temperature histories.

[0045] The component design may preferably be provided as a CAD file in a computer engineering design platform or database.

[0046] The strength analysis of the component may be a finite element analysis performed on a computer with engineering design software.

[0047] These steps can be sequenced automatically or have at least one manual input possibility in between to decide to proceed with one or several options.

[0048] According to another preferred embodiment of the method of the present invention, the material model is a machine learning fatigue model, wherein the machine learning model is provided by a method comprising the following steps:

[0049] (i) providing a set of test specimen specifications to the material model, wherein the test specimen specifications each include a difference in at least one material condition of the specimen specifications,

[0050] (ii) Collect data points of maximum stress versus cycles to failure for the specimen specification,

[0051] (iii) training a material model with the collected data to select a test specimen specification from a set of test specimen specifications upon receiving as input:

[0052] (i) at least one stress-related parameter and

[0053] (ii) at least one material condition.

[0054] Such a model is known from document WO2020216458A1 including a training option.

[0055] A further preferred embodiment provides that the component is additively manufactured at least in critical areas.

[0056] A further preferred embodiment provides that the component is produced from a composite material at least in critical regions.

[0057] Especially for these two material types, the method according to the invention is most suitable, since conventional methods of this type are not sufficiently accurate or involve many tests, wherein the invention reduces the number of tests due to the application of the material model.

[0058] A further preferred embodiment provides that, in the case of additive manufacturing of the component at least in the critical region, at least one material condition is at least one of the following:

[0059] - surface condition or surface roughness,

[0060] -defect,

[0061] - inclusions,

[0062] - microstructure,

[0063] - Orientation of printed layers of additively manufactured parts,

[0064] - Temperature history

[0065] - residual stresses,

[0066] -Printing parameters for additive manufacturing.

[0067] In general, a "material condition" may also be a "stress-related parameter". In particular, "residual stress" affects the properties of the material and can therefore be considered a material condition. Residual stress is also a stress-related parameter. In principle, the user can include residual stress (e.g., mean stress compensation equation) or choose to ignore it (e.g., Von Mises stress of the load amplitude). Depending on the material used, different stress-related parameters can be defined. It is beneficial to know exactly the residual stress that is included as a stress-related parameter. According to another preferred embodiment, residual stress can be applied as a parameter under the material condition as a classification value. (Example: Consider different heat treatments, all of which introduce or release a certain amount of residual stress. Since it is impractical to measure the residual stress of different treatments, the residual stress can be advantageously considered as a material condition by defining different categories for different heat treatments. In this case where the actual value is not known, the residual stress is preferably not included as a stress-related parameter input.

[0068] According to a preferred embodiment, it is preferred and sufficient to include parameters as inputs of material conditions or stress-related parameters. The most accurate option is to consider that it can be in two input groups.

[0069] Another preferred embodiment provides that, in the case of the component at least in the critical region of the composite material, at least one material condition is at least one of the following:

[0070] - surface condition or surface roughness,

[0071] -defect,

[0072] - Texture,

[0073] - inclusions,

[0074] - microstructure,

[0075] - Temperature history

[0076] -Residual stress

[0077] -Curing cycle.

[0078] Another preferred embodiment provides that the method according to the invention may comprise an additional step before step (d) of providing at least one material condition, the additional step comprising:

[0079] - Provide manufacturing process plans for components,

[0080] - Monitoring and / or simulating a manufacturing process plan for a component according to the manufacturing process plan.

[0081] This embodiment enables the simulation option to first manufacture the test coupon and test whether the requirements are met before manufacturing the part. In general, this feature enables accurate consideration of the options in the manufacturing process parameters and test coupon selection.

[0082] In order to be able to fully take into account the load spectrum during the strength analysis without over-analysis, the method may comprise, before step (a) of determining the load condition of the component, an additional step,

[0083] Additional steps include:

[0084] - Provide the working load spectrum of the components,

[0085] -Determine the load conditions based on the component's operating load spectrum.

[0086] The present invention also comprises a system adapted to perform the method according to any of the preceding claims, the system comprising at least one processing unit adapted to perform computer-implemented steps (a), (c), (g). According to another preferred embodiment, at least one processing unit of the system may also be adapted to perform computer-implemented steps (b) and / or (d) and / or (e) and / or (f). BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Embodiments of the present invention are herein described, by way of example only, with reference to the accompanying drawings, in which:

[0088] Figure 1 shows a simplified schematic diagram of a system for predicting the fatigue life of a component by applying the method according to the invention;

[0089] Figure 2 A schematic flow chart of the method according to the invention is shown.

[0090] The illustrations in the drawings are in schematic form. It should be noted that in different drawings similar or identical elements may be provided with the same reference numerals. DETAILED DESCRIPTION

[0091] While the present invention has been described in detail with reference to preferred embodiments, it should be understood that the invention is not limited to the disclosed examples and that numerous additional modifications and variations may be made thereto by those skilled in the art without departing from the scope of the present invention.

[0092] It should be noted that the use of "a" or "an" throughout this application does not exclude a plurality, and "comprising" does not exclude other steps or elements. Furthermore, elements described in connection with different embodiments may be combined. It should also be noted that reference signs in the claims should not be interpreted as limiting the scope of the claims.

[0093] Figure 1 A simplified schematic diagram of a system SYS for predicting the fatigue life of a component CMP by applying the method according to the invention is shown.

[0094] Starting from placing the design CDS of the component CMP under the load condition LDC, a strength analysis FEM of the component CMP is performed. During the strength analysis SVM, the critical area CRA and at least one stress-related parameter SRP are determined. Together with at least one material condition MCD, these results are provided to the material model MTM. The at least one material condition MCD can be at least one of the following:

[0095] - surface condition or surface roughness,

[0096] -defect,

[0097] - inclusions,

[0098] - microstructure,

[0099] - Temperature history

[0100] - residual stresses,

[0101] - orientation of the AM printed layers or AM printing parameters (in the case of AM critical areas CRA),

[0102] - Cure cycle or texture when the critical area CRA is a composite material.

[0103] The analysis of these parameters provided by the material model MTM, preferably by calculation of covariances, can generate a ranked list of specimens, where the specimens with the highest correlation are the specimens to be tested. The material model MTM generates as output a test specimen COP specification TCS for testing in the test machine TST. The specimen COP specification TCS can include complete geometrical specifications, preferably complete material specifications, and preferably a complete manufacturing process plan.

[0104] Figure 2 A schematic flow chart of the method according to the invention is shown.

[0105] As a preparatory step (a0), the method according to the invention may comprise an additional step before step (a) by determining the load condition LDC of the component CMP, which may be accomplished by:

[0106] - Provide the working load spectrum OLS of the component CMP,

[0107] - Determination of the load condition LDC from the operating load spectrum of the component CMP. This preparation makes the subsequent analysis more efficient. The load condition can also be called equivalent operating load.

[0108] Then follow these steps:

[0109] (a) Determine the load condition LDC of the component CMP,

[0110] (b) Provide component CMP design CDS,

[0111] (c) Perform a strength analysis FEM of the component CMP design under load condition LDC to determine:

[0112] (i) Critical Area (CRA) of component CMP,

[0113] (ii) at least one stress-related parameter SRP of the critical area CRA.

[0114] Optionally, the method may include an additional step (d0) (shown by a dashed box) before the step (d) of providing at least one material condition MCD, the additional step (d0) comprising:

[0115] - Provide manufacturing process plan AMP for component CMP,

[0116] - Monitoring MON and / or simulation SIM are performed according to the manufacturing process plan AMP of the component CMP. This additional step (d0) enables the recorded parameters of the manufacturing process [or a simulation thereof] to be considered as the material condition MCD in the next step (d).

[0117] The subsequent step (d) comprises providing at least one material condition MCD of the component CMP at least for a critical area of ​​the component CMP.

[0118] According to the invention, a material model MTM is provided, which receives as input [step (f)]:

[0119] (i) at least one stress-related parameter SRP and

[0120] (ii) at least one material condition MCD.

[0121] In step (g), the material model MTM generates as output a test specimen COP specification TCS for testing in the test machine TST. As explained below, when received as input, the material model MTM may have been trained to select a test specimen specification TCS from a set of test specimen specifications SCS:

[0122] (i) at least one stress-related parameter SRP and

[0123] (ii) at least one material condition MCD.

[0124] like Figure 1 Additional subsequent steps shown are generating a test coupon COP and conducting coupon testing.

[0125] The material model MTM may be a machine learning fatigue model MLS. The machine learning model may be provided by a method comprising the following steps [(e)]:

[0126] (i) providing a set of test specimen specifications SCS to the material model MTM, wherein the test specimen specifications SCS respectively comprise a difference in at least one material condition MCD of the specimen specifications SCS,

[0127] (ii) Collect the data points DPT of the maximum stress of the test specimen SCS versus the number of cycles to failure,

[0128] (iii) Train the material model MTM using the collected data.

[0129] Each of the steps (a)-(g) - individually or in their entirety - may be computer-implemented to be performed by at least one processing unit CPU of the system SYS. In particular, the system SYS may comprise at least one processing unit CPU adapted to perform the computer-implemented steps (a), (c), (g).

Claims

1. A method for generating a test specimen (COP) specification (TCS) for predicting fatigue life of a component (CMP), comprising: (a) determining the load condition (LDC) of the component (CMP), (b) provide component (CMP) design (CDS), (c) performing a strength analysis (FEM) of the component (CMP) design under the load condition (LDC) to determine: (i) the critical area (CRA) of the component (CMP), (ii) at least one stress related parameter (SRP) of said critical area (CRA), (d) providing at least one material condition (MCD) of said component (CMP) for at least said critical region of said component (CMP), It is characterized by the following additional steps: (e) Provide material model (MTM), (f) providing as input to the material model (MTM): (i) said at least one stress related parameter (SRP) and (ii) at least one material condition (MCD), (g) said material model (MTM) generates as output said test specimen (COP) specification (TCS) for testing in a testing machine (TST), Wherein, the material model (MTM) is a machine learning fatigue model (MLS), Wherein, the machine learning model is provided by a method comprising the following steps: (i) providing a set of test specimen specifications (SCS) to the material model (MTM), wherein the test specimen specifications (SCS) respectively comprise a difference of the at least one material condition (MCD) of the specimen specifications (SCS), (ii) collecting data points (DPT) of maximum stress versus cycles to failure for said test specimen specification (SCS), (iii) training the material model (MTM) with the collected data to select a test specimen specification (TCS) from the set of test specimen specifications (SCS) upon receiving as input: (i) said at least one stress related parameter (SRP) and (ii) at least one material condition (MCD).

2. The method according to claim 1, wherein: The component (CMP) is additively manufactured at least in the critical area (CRA).

3. The method according to claim 1, wherein: The component (CMP) is made of a composite material at least in the critical area (CRA).

4. The method according to claim 2, wherein: The at least one material condition (MCD) is at least one of: - surface condition or surface roughness, -defect, - inclusions, - microstructure, - Orientation of printed layers of additively manufactured parts, - Temperature history, - residual stresses, -Printing parameters for additive manufacturing.

5. The method according to claim 3, wherein: The at least one material condition (MCD) is at least one of: - surface condition or surface roughness, -defect, - Textures, - inclusions, - microstructure, - Temperature history, - residual stresses, -Curing cycle.

6. The method according to any one of the preceding claims, comprising an additional step before step (d) of providing at least one material condition (MCD), the additional step comprising: - providing a manufacturing process plan (AMP) for said component (CMP), - performing a monitoring (MON) and / or a simulation (SIM) according to said manufacturing process plan (AMP) of said component (CMP) according to said manufacturing process plan (AMP).

7. The method according to any one of the preceding claims, comprising, before step (a) of determining the load condition (LDC) of the component (CMP), an additional step comprising: - providing an operating load spectrum (OLS) of said component (CMP), - determining a load state (LDC) based on an operating load spectrum of the component (CMP).

Images