Digital-physical testing system and digital-physical testing procedure

Abstract

Comprehensive digital-physical testing system: a simulation device (1) comprising a numerical calculation model for digital modeling the static crack tip stress of a cracked component and a test device (2) for the physical testing of the dynamic damage and crack propagation behavior of a cracked test specimen (10); wherein the test device (2) comprises a testing machine (20) with a clamping device (21) for fixing and dynamically loading a cracked test specimen (10) by means of the testing machine (20) and a detection device (3) for detecting and tracking at least one cracked area of ​​the test specimen (10), the testing system is trained to automatically execute the following process steps: a) Initial calculation of parameters of the crack tip stress of the cracked component using the simulation device (1) under specification of an initial crack geometry a initial of the overall component, external loads F acting on the overall component and the time course of the acting loads ΔF; b) Applying a number of load cycle increments N to the cracked test specimen (10) by specifying the calculated parameters of the crack tip stress from step a) and a number of load cycles N using the testing machine (20) and measuring the changed crack geometry a N , comprising the crack path geometry as well as the crack tip coordinates a x and a y , on the test specimen (10) by means of the detection device (3) as crack propagation increment due to the imprinted load cycle increment; c) Introducing the modified crack geometry a Nfrom step b) into the numerical calculation model of the overall component and subsequent calculation of parameters of the crack tip stress of the cracked overall component with the modified crack geometry a N via the simulation device (1) under specification of acting external loads F and the time course t of the acting loads ΔF; and d) Optional repetition of process steps b) and c) for any integer number of repetitions, wherein in step b) the calculated parameters of the crack tip stress from the previous step are adopted and wherein in step c) the modified crack geometry a N taken from the previous step and used to determine a progressive fatigue crack.

Description

[0001] The present invention relates to a digital-physical testing system and a digital-physical testing method.

[0002] When inspecting and monitoring structural components and parts subjected to non-constant loads during operation, it is essential to ensure sufficiently accurate detection and tracking of defects, damage, and cracks. Fatigue cracking, or rather the propagation of fatigue cracks, is one of the most relevant types of damage in all technical fields where components and parts are subjected to the aforementioned non-constant loads, and especially dynamic and alternating loads. Particularly in lightweight structures, such as those found in aircraft, the occurrence of fatigue cracks is a critical safety aspect that must be considered during the design phase of components.

[0003] In material mechanics tests for characterizing cracks, particularly for investigating thresholds, crack propagation, fracture toughness, and crack resistance curves, the cracks must be recorded and measured very precisely in terms of their length and orientation. Such tests are conducted at the coupon level with specimen sizes ranging from 1 to 500 mm. It is also known from the prior art to test entire aircraft structures. The aforementioned "full-scale" tests are extremely time-consuming and expensive in practice. Therefore, they should be reduced to the absolute minimum or, where possible, replaced entirely.

[0004] Within the framework of faster and increasingly virtual approval processes, it is a set goal to reduce the number of structural level trials (“Large-Scale” or “Full-Scale” trials) to the necessary minimum.

[0005] Due to the time-consuming and costly nature of large-scale and full-scale tests, the development and approval processes for components and component structures, such as aircraft structures, are very lengthy and require extended development times. Furthermore, replacing these large-scale and full-scale tests, particularly regarding fatigue crack propagation, necessitates comprehensive material models that also incorporate the interaction of local crack tip mechanics and the microstructure. Pure simulation, e.g., using the finite element method according to the state of the art, is often insufficient because the complex structure-property relationships at the microscopic scale around a crack tip can only be partially represented by numerical models, thus not always guaranteeing or achieving the necessary level of safety.

[0006] From STROHMANN, Tobias [et al.]: Automatic detection of fatigue crack paths using digital image correlation and convolutional neural networks. In: Fatigue & Fracture of Engineering Materials & Structures, Vol. 44, 2021, No. 5, pp. 1336-1348, a deep convolutional neural network has been reported for segmenting crack paths and their respective crack tips from displacement fields obtained by digital image correlation during fatigue crack propagation experiments. The neural network was trained to predict the coordinates of the crack paths and crack tips based on the displacement fields around the fatigue crack. The digital image correlation data used for training were supplemented with data from finite element calculations to increase the dataset.

[0007] Melching, David [et al.]: Explainable machine learning for precise fatigue crack tip detection. In: Scientific Reports, Vol. 12, 2022, Art.No. 9513, p.1-4 describes a novel architecture of a neural network that combines segmentation and regression of crack tip coordinates.

[0008] Based on the aforementioned prior art, the present invention aims to simplify the methods or testing systems for characterizing fatigue and / or cracking processes in components or component structures and to significantly reduce the required time expenditure, while simultaneously ensuring the necessary accuracy and reliability of the methods and testing systems.

[0009] According to the invention, the problem according to a first aspect of the invention is solved by a digital-physical testing system, in particular for material mechanics testing.

[0010] The digital-physical testing system according to the invention comprises a simulation device including a numerical calculation model for digitally modeling the static crack tip stress of a cracked component and a test device for physically testing the dynamic damage and crack propagation behavior of a cracked specimen. The test device includes a testing machine with a clamping device for fixing and dynamically loading the cracked specimen by means of the testing machine and a detection system for detecting and tracking at least one cracked area of ​​the specimen.

[0011] The testing system is designed to automate the following process steps: a) Initial calculation of parameters of the crack tip stress of the cracked component using the simulation device, specifying an initial crack geometry ainitial of the overall component, external loads F acting on the overall component and the time course of the acting loads ΔF; b) Applying an initial number of load cycles to the cracked test specimen by specifying the calculated parameters of the crack tip stress from step a) and a number of load cycles N using the testing machine and measuring the changed crack geometry a N , comprising the crack path geometry as well as the crack tip coordinates a x and a y , on the test specimen by means of the detection device as crack progression increment due to the imprinted load cycle increment; c) Introducing the modified crack geometry a N from step b) into the numerical calculation model of the overall component and subsequent calculation of parameters of the crack tip stress of the cracked overall component with the modified crack geometry a Nvia the simulation device under specified external loads F and the time course of the acting loads ΔF; and d) Optional repetition of process steps b) and c) for any integer number of repetitions, wherein in step b) the calculated parameters of the crack tip stress from the previous step are adopted and wherein in step c) the modified crack geometry a N taken from the previous step and used to determine a progressive fatigue crack.

[0012] In the testing system according to the invention, the connection between the experiment performed on the test specimen using the testing machine and the numerical simulation of the entire component is achieved through the mutual exchange of measured and calculated values. This creates a link between the digital / numerical calculation model and the physical test in the testing machine. Through this parameter exchange, the experiment in the testing machine and the digital calculation model are thus integrated, allowing them to mutually control and regulate each other. The testing system according to the invention, through this integration, forms a digital-physical control loop.

[0013] The cracked test specimen, which is clamped in the testing machine, replicates the local damage and cracking behavior of the entire cracked component. In particular, according to the invention, the cracked test specimen can be a standardized test specimen, preferably having a length between 1 mm and 1000 mm. In any case, the dimensions of the test specimen according to the present invention are smaller than the dimensions of the entire component, and the geometry of the test specimen according to the present invention is less complex than the geometry of the entire component.

[0014] The simulation device can further be designed to calculate parameters of the crack tip stress for a stationary crack depending on external loads F acting on the overall component and the time course of the loads ΔF and the crack geometry a.

[0015] The parameters to be calculated for the crack tip stress can be selected from at least one element from the group of: - the voltage intensity factors K I ,K II ,K III ; - the T-voltage; and / or - Higher-order terms of the Williams series - the stresses σ x ,σ y ,τ xy ·

[0016] The testing machine can be designed to impose a mixed-mode load on the test specimen or to introduce defined stresses parallel and perpendicular to an introduced crack into the test specimen via a biaxial load application.

[0017] According to the invention, the terms biaxiality and biaxial load application are used, but multiaxial load conditions can also be applied to a test specimen according to the invention.

[0018] In the testing machine, a standardized test specimen can preferably be subjected to the crack tip load.

[0019] The test specimen can, in particular, depict the local cracked section of the overall component.

[0020] The test device can be configured to perform the following further sub-process steps in process step b): b1) Generating at least one first macroscopic reference image using a first stationary imaging device and a plurality of microscopic reference images using a second positionable imaging device of the unloaded cracked test specimen; b2) Applying a load or load regime to the cracked test specimen; b3) Generating at least one macroscopic image of the cracked test specimen under applied load according to procedure step b2) and / or after unloading via the first stationary imaging device b4) Calculating the current localization of at least one area of ​​a crack tip in the test specimen, preferably using a trained neural network for crack detection; b5) preferably aligning the second positionable image acquisition device based on the at least one region of a crack tip in the test specimen according to process step b4) and generating at least one microscopic image of the at least one region of a crack tip in the test specimen using the second image acquisition device under applied load according to process step b2) and / or after unloading; and b6) Optional repetition of process steps b2) to b5) by any integer number of repetitions, wherein in process step b2) a load or load regime different from the previously applied load is applied.

[0021] According to the invention, in process step b4), at least one region of a crack tip in the test specimen can be determined using any automated method. The preferred use of a trained neural network is therefore only one example of several possibilities for the automated calculation of the localization.

[0022] The trained neural network for crack detection can include a preferred neural network (CNN) with U-network architecture with multiple encoding and decoding blocks as a segmentation branch; and a fully connected neural network (FCNN) as a regression branch.

[0023] It may be provided that the decoding blocks are connected to the encoding blocks via a base block, with the encoding and decoding blocks of the same level being directly connected to each other by a "skip" connection. where the regression branch is connected to the base block of the segmentation branch, and the computer-implemented training procedure for maintaining the trained neural network for damage detection comprises the following steps: 1) Initialization of the neural network with random weights; 2) Calculating the mean squared error between the predicted crack tip position and the actual crack tip position ŷ = (ŷ1,ŷ2) E [-1,1] 2 about the formula: MSE(y,y^)=(y1−y^1)2+(y2−y^2)2 as an error in the regression branch of the neural network; 3) Calculating the crack tip segmentation error (Dice error) for the segmentation task using the formula: Dice(z,z^)=1−2∑ijzijz^ij+ε∑ij(zij+z^ij)+ε where z = (z ij ) with z ij ∈ [0,1] denotes the segmentation output (after sigmoid activation) and ẑ = (ẑ ij ) represents the fundamental truth. With ε > 0 as a small constant introduced to cover the case z = ẑ ≡ 0, preferably ε = 10 -6 is chosen as an error in the segmentation branch of the neural network; 4) Calculating a weighted total loss function ω (z,y,ẑ,ŷ) via the formula: Lossω(z,y,z^,y^)=Dice(z,z^)+ωMSE(y,y^) where ω ≥ 0 is a weighting factor that adjusts the training influence of the FCNN / segmentation branch of the neural network; 5) Optimizing the model parameters of the neural network using the "backpropagation" algorithm and the weighted total loss function. ω (z,y,ẑ,ŷ) to maintain the trained neural network for damage detection.

[0024] According to the invention, it may be provided that the sub-process before process step b2) further comprises the step: b1-1) Calibrating the positioning of the second image acquisition device by aligning the current spatial position of the second image acquisition device with respect to the first stationary image acquisition device.

[0025] Furthermore, it may be provided that in process step b1) macroscopic reference images of the entire unloaded cracked test specimen are generated using a first stationary image acquisition device and a plurality of microscopic reference images of the entire unloaded test specimen are generated using the second positionable image acquisition device.

[0026] It may also be provided that the procedure in step b5) additionally includes the following step: b51) optical focusing of the second image acquisition device on the area of ​​the damage or the crack tip.

[0027] According to another aspect of the present invention, the invention relates to a digital-physical testing method which comprises the following process steps: a) Initial calculation of parameters of the crack tip stress of the cracked component as a whole via a simulation device comprising a numerical calculation model for the digital modeling of the static crack tip stress of a cracked component as a whole, given an initial crack geometry a initial of the overall component, external loads F acting on the overall component and the time offset t of the acting loads F; b) Applying a number of load cycles N to a cracked test specimen using a testing machine by specifying the calculated parameters of the crack tip stress from step a) and a number of load cycles N and measuring the changed crack geometry a N by means of a detection device for detecting and tracking at least one cracked area on the test specimen as a crack progression increment based on the imprinted load cycle increment; c) Introducing the modified crack geometry a N from step b) into the numerical calculation model of the overall component and subsequent calculation of parameters of the crack tip stress of the cracked overall component with the modified crack geometry a N via the simulation device under specified external loads F and the time offset t of the acting loads F; and d) Optional repetition of process steps b) and c) for any integer number of repetitions, wherein in step b) the calculated parameters of the crack tip stress from the previous step are adopted and wherein in step c) the modified crack geometry a N taken from the previous step and used to determine progressive dynamic crack behavior.

[0028] In the testing method according to the invention, as in the testing system according to the invention, a linkage between the experiment in the testing machine and the digital calculation model takes place through parameter exchange, whereby these mutually control and regulate each other. Through this linkage, the testing method according to the invention forms a digital-physical control loop.

[0029] In the following, an exemplary embodiment of the digital physical testing system according to the invention is explained schematically with reference to the attached figure.

[0030] It shows: Fig. 1 A schematic view of a digital physical testing system according to the present invention.

[0031] The digital physical testing system according to Fig. 1 initially comprises a simulation device 1, which in turn comprises a numerical calculation model for the digital modeling of the static crack tip stress of a cracked component. As in the Fig. As schematically represented in Figure 1, the simulation device can be any computer hardware, for example, a single computer or distributed and networked computers, including those on which a numerical computational model is stored and on which corresponding numerical simulations can be performed, for example, using the finite element method. Furthermore, the digital physical testing system comprises a test device 2 for the physical testing of the dynamic damage and crack propagation behavior of a cracked test specimen 10. The test device 2, in turn, comprises a testing machine 20 with a clamping device 21 for fixing and dynamically loading a cracked test specimen 10 by means of the testing machine 20. In the schematic embodiment shown, the testing machine 20 is simplified to an axial testing machine, whereas according to the invention, biaxial or...Multiaxial load conditions can be applied to the test specimen 10 by means of the testing machine 20. According to the invention, the testing machine 20 can therefore be a biaxial testing machine or a uniaxial testing machine with a corresponding specimen clamping device that introduces multiaxial stress conditions into the specimen. The test device 2 further comprises a detection device 3 for the automatic detection and tracking of at least one cracked area of ​​the test specimen 10, which in the illustrated embodiment is represented in the form of two optical image acquisition devices 31.

[0032] The test specimen 10 subjected to the loads in the testing machine 20 can in particular be a standardized test specimen.

[0033] The detection device 3 can, in particular, comprise a first stationary image acquisition device 31, by means of which macroscopic reference images 30 of the cracked test specimen 10 can be generated. Furthermore, optionally, the detection device 3 can comprise a second positionable image acquisition device 5, by means of which microscopic images 50 of at least one crack tip or damage area of ​​the cracked test specimen 10 can be generated. The spatial orientation of the second positionable image acquisition device 5 is, as shown in the Fig. 1 shown, which can be changed translationally and rotationally in three-dimensional space via a positioning and adjustment device such as a robot arm 7.

[0034] As this is shown in the Fig.As shown by the dashed arrows in Figure 1, both the macroscopic images 30 and the microscopic images 50 are forwarded to the simulation device 1 for evaluation. The simulation device 1 can also control the testing machine 20, the corresponding recording and recognition devices 3 and 5, as well as the test device 2 and / or the robot arm 7, and is connected to them via the signal flow, as indicated by the double arrows. The signals between the aforementioned devices and equipment can, of course, be transmitted via wired and / or wireless connections.

Claims

[1] Digital-physical testing system comprising: a simulation device (1) comprising a numerical calculation model for digital modeling the static crack tip stress of a cracked component and a test device (2) for the physical testing of the dynamic damage and crack propagation behavior of a cracked test specimen (10); wherein the test device (2) comprises a testing machine (20) with a clamping device (21) for fixing and dynamically loading a cracked test specimen (10) by means of the testing machine (20) and a detection device (3) for detecting and tracking at least one cracked area of ​​the test specimen (10), the testing system is trained to automatically execute the following process steps: a) Initial calculation of parameters of the crack tip stress of the cracked component using the simulation device (1) under specification of an initial crack geometry a initial of the overall component, external loads F acting on the overall component and the time course of the acting loads ΔF; b) Applying a number of load cycle increments N to the cracked test specimen (10) by specifying the calculated parameters of the crack tip stress from step a) and a number of load cycles N using the testing machine (20) and measuring the changed crack geometry a N , comprising the crack path geometry as well as the crack tip coordinates a x and a y , on the test specimen (10) by means of the detection device (3) as crack propagation increment due to the imprinted load cycle increment; c) Introducing the modified crack geometry a Nfrom step b) into the numerical calculation model of the overall component and subsequent calculation of parameters of the crack tip stress of the cracked overall component with the modified crack geometry a N via the simulation device (1) under specification of acting external loads F and the time course t of the acting loads ΔF; and d) Optional repetition of process steps b) and c) for any integer number of repetitions, wherein in step b) the calculated parameters of the crack tip stress from the previous step are adopted and wherein in step c) the modified crack geometry a N taken from the previous step and used to determine a progressive fatigue crack. [2] Test system according to claim 1, wherein the simulation device (1) is designed to calculate parameters of the crack tip stress for a stationary crack depending on external loads F acting on the overall component, the crack geometry a and the time course of the acting external loads ΔF. [3] Test system according to claim 1 or 2, wherein the parameters to be calculated for the crack tip stress are selected from at least one element from the group of: - the voltage intensity factors K I ,K II ,K III , - the T-voltage, and / or - Higher-order terms of the Williams series, - the stresses σ x ,σ y ,τ xy . [4] Testing system according to one of the preceding claims, wherein the testing machine (20) is configured to apply a mixed-mode load to the test specimen (10) or to apply defined stresses parallel and perpendicular to a crack introduced into the test specimen (10) via a biaxial load application. [5] Testing system according to one of the preceding claims, wherein a standardized test specimen is subjected to the crack tip load in the testing machine (20). [6] Test system according to one of the preceding claims, wherein the test specimen (10) represents the cracked section of the overall component. [7] Test system according to one of the preceding claims, wherein the test device (2) is configured to perform the following further sub-process steps in process step b): b1) Generating at least one first macroscopic reference image (30) using a first stationary imaging device (31) and a plurality of microscopic reference images (50) using a second positionable imaging device (5) of the unloaded cracked test specimen (10); b2) Applying a load or load regime to the cracked test specimen (10); b3) Generating at least one macroscopic image (30) of the cracked test specimen (10) under applied load according to procedure step b2) and / or after unloading via the first stationary image acquisition device (31); b4) Calculating the current localization of at least one area of ​​a crack tip in the test specimen, preferably using a trained neural network for crack detection; b5) preferably aligning the second positionable image acquisition device (5) based on the at least one region of a crack tip in the test specimen (10) according to process step b4) and generating at least one microscopic image (50) of the at least one region of a crack tip in the test specimen (10) using the second image acquisition device (5) under applied load according to process step b2) and / or after unloading; and b6) Optional repetition of process steps b2) to b5) by any integer number of repetitions, wherein in process step b2) a load or load regime different from the previously applied load is applied. [8] Testing system according to claim 7, wherein the trained neural network for crack detection comprises a convolutional neural network (CNN) with U-network architecture with multiple encoding and decoding blocks as a segmentation branch; and a fully connected neural network (FCNN) as a regression branch; where the decoding blocks are connected to the encoding blocks via a base block, where the encoding and decoding blocks of the same level are directly connected to each other via a jump link, where the regression branch is connected to the base block of the segmentation branch, and the computer-implemented training procedure for maintaining the trained neural network for damage detection comprises the following steps: 1) Initialization of the neural network with random weights; 2) Calculating the mean squared error between the predicted crack tip position and the actual crack tip position ŷ = (ŷ1,ŷ2) ∈ [-1,1] 2 about the formula: MSE(y,y^)=(y1−y^1)2+(y2−y^2)2 as an error in the regression branch of the neural network; 3) Calculating the crack tip segmentation error (Dice error) for the segmentation task using the formula: Dice(z,z^)=1−2∑ijzijz^ij+ε∑ij(zij+z^ij)+ε where z = (z ij ) with z ij ∈ [0,1] denotes the segmentation output after sigmoid activation and ẑ = (ẑ ij ) represents the fundamental truth. With ε > 0 as a small constant introduced to cover the case z = ẑ ≡ 0, preferably ε = 10 -6 is chosen as an error in the segmentation branch of the neural network; 4) Calculating a weighted total loss function ω (z,y,ẑ,ŷ) via the formula: Lossω(z,y,z^,y^)=Dice(z,z^)+ω MSE(y,y^) where ω ≥ 0 is a weighting factor that adjusts the training influence of the FCNN / segmentation branch of the neural network; 5) Optimizing the model parameters of the neural network using a backpropagation algorithm and the weighted total loss function. ω (z,y,ẑ,ŷ) to maintain the trained neural network for damage detection. [9] Testing system according to claim 7 or 8, wherein the sub-method prior to process step b2) further comprises the step: b1-1) Calibrating the positioning of the second image acquisition device (5) by aligning the current spatial position of the second image acquisition device (5) with respect to the first stationary image acquisition device (31). [10] Testing system according to one of claims 7 to 9, wherein in process step b1) macroscopic reference images (30) of the entire unloaded cracked test specimen (10) are generated by means of a first stationary image acquisition device (31) and a plurality of microscopic reference images (50) of the entire unloaded test specimen (10) are generated by means of the second positionable image acquisition device (5). [11] Testing system according to one of claims 7 to 10, wherein the method in step b5) additionally comprises the step: b51) optically focusing the second image acquisition device (5) on the area of ​​the damage or the crack tip. [12] Digital-physical testing procedure comprising the following process steps: a) Initial calculation of parameters of the crack tip stress of the cracked component using a simulation device (1) comprising a numerical calculation model for the digital modeling of the static crack tip stress of a cracked component under specification of an initial crack geometry a initial of the overall component, external loads F acting on the overall component and the time course of the acting loads ΔF; b) Applying a number of load cycle increments N to a cracked test specimen (10) using a testing machine (20) by specifying the calculated parameters of the crack tip stress from step a) and a number of load cycles N and measuring the changed crack geometry a N , comprising the crack path geometry as well as the crack tip coordinates a x and a y, by means of a detection device (3) for detecting and tracking at least one cracked area on the test specimen (10) as a crack propagation increment due to the imprinted load cycle increment; c) Introducing the modified crack geometry a N from step b) into the numerical calculation model of the overall component and subsequent calculation of parameters of the crack tip stress of the cracked overall component with the modified crack geometry a N via the simulation device (1) under specification of acting external loads F and the time offset t of the acting loads F; and d) Optional repetition of process steps b) and c) for any integer number of repetitions, wherein in step b) the calculated parameters of the crack tip stress from the previous step are adopted and wherein in step c) the modified crack geometry a Ntaken from the previous step and used to determine progressive dynamic crack behavior.

Images