Method for determining material properties from a foam sample

Abstract

The invention is in the field of methods for determining material properties from a foam sample. It relates to a computer-implemented method for determining material properties of a foam sample, comprising: (a) providing a representation of the sample, (b) extracting at least one structural feature from the representation, wherein the at least one structural feature comprises a wall, a strut or a node, (c) providing the at least one structural feature to a material model adapted to obtain at least one material property from the structural feature, and (d) outputting the at least one material property received from the material model.

Description

Technical Field

[0001] This invention relates to the field of methods for determining material properties from foam samples, particularly from images of foam samples. Background Technology

[0002] Material properties, such as mechanical, thermal, or chemical properties, are strongly dependent on the material's structure at both the macroscopic and microscopic levels, especially in porous materials such as foams. Determining material properties typically requires extensive testing. Samples must be prepared, appropriate testing equipment must be provided, and testing protocols must be strictly adhered to. For this reason, well-trained personnel are needed. Therefore, there is a desire for a faster method involving fewer personnel and less equipment.

[0003] WO 2015 / 080 912 A1 discloses a method for digitally modeling reservoirs in oil fields from computed tomography images on core samples. Oil field characteristics can be obtained by running simulations on these models. However, many specific assumptions and physical phenomena must be incorporated into the models, making them largely inapplicable to anything outside of oil field analysis.

[0004] WO 2018 / 206 225 A1 discloses a method for modeling objects from images and comparing them with their desired geometry to detect defects. However, material properties are not obtained.

[0005] US 2014 / 044 315 A1 discloses a method for increasing the accuracy of target property values ​​derived from rock samples. However, this method is difficult to transfer to foam samples.

[0006] In his doctoral dissertation, titled "X-Ray Imaging Applied to the Characterization of Polymer Foams' Cellular Structure and Its Evolution," published on March 1, 2014, Samuel Pardo Alonso disclosed a method for generating 3D models from images. However, material properties were not obtained.

[0007] Therefore, the object of this invention is to provide a method for determining material properties effortlessly in terms of apparatus, personnel, and time. This method should be variable for a wide range of different foam materials and size scales. The method is designed to be fast and reliable. Summary of the Invention

[0008] These objectives are achieved through a computer-implemented method for determining the material properties of foam samples, the method comprising:

[0009] (a) Provide a representation of the sample.

[0010] (b) Extract at least one structural feature from the representation, wherein the at least one structural feature includes a wall, a support, or a node.

[0011] (c) Providing the at least one structural feature to a material model suitable for obtaining at least one material property from the structural feature, and

[0012] (d) Output at least one material property received from the material model.

[0013] The present invention also relates to a non-transitory computer-readable data medium storing a computer program including instructions for performing the steps of the method according to any one of the preceding claims.

[0014] The present invention also relates to a production monitoring and / or control system for monitoring and / or controlling the material properties of samples, comprising:

[0015] (a) An input unit configured to receive a representation of the sample.

[0016] (b) A processing unit configured to extract at least one structural feature from the representation, wherein the at least one structural feature includes a wall, a support, or a node.

[0017] (c) A processing unit configured to provide the at least one structural feature to a material model, the material model being adapted to obtain at least one material property from the structural feature, and

[0018] (d) is configured to output the material properties received from the material model.

[0019] Preferred embodiments of the present invention can be found in the specification and claims. Combinations of different embodiments fall within the scope of the present invention. Attached Figure Description

[0020] Figure 1 Possible embodiments of the present invention are described.

[0021] Figures 2a to 2g An example of image processing using the method of the present invention is shown. Detailed Implementation

[0022] The method according to the invention is useful for determining the material properties of foam samples. The sample can be a small piece of material or a complete workpiece. Foam samples include porous materials such as polymer foams, zeolites, and supports for exhaust gas catalysts. The internal structure of the sample refers to the distribution of phase boundaries within the sample, such as the phase boundaries between voids and polymers in a polymer foam. Depending on the resolution of the sample and the image, the features of the internal structure can be at different size scales, for example, in the micrometer range (0.1 to 1000 μm) or at the nanometer scale (1 to 100 nm).

[0023] Material properties can be mechanical properties, such as Young's modulus, elasticity, tear resistance, wear resistance, and coefficient of friction; thermal properties, such as heat capacity and thermal conductivity; electrical properties, such as electrical conductivity or resistivity and dielectric constant; or optical properties, such as transparency, refractive index, and diffusivity.

[0024] In the context of this invention, a representation of a sample is a data structure containing the internal structure of the sample. The representation associates each location in the sample with information about the material or voids at that location. The representation can be two-dimensional or three-dimensional, preferably three-dimensional. It is possible that the representation already exists in previous work, is received from a remote computer or cloud, or is generated on the same computer as the steps of the method according to the invention. Preferably, the representation is generated from an image showing the internal structure of the sample. The representation can have various formats, including bitmaps like pixel or voxel data, point clouds, and triangular surface models such as those in Standard Triangulation Language (STL). The format of the representation must be suitable for including information about the material and / or voids.

[0025] Images of the sample can be provided in various ways. For example, they can be obtained directly from the measuring device or from a database containing previous measurements. Various measurements are available, such as photography of cut or broken samples, optical microscopy, or electron microscopy, or non-destructive methods such as computed tomography, magnetic resonance imaging, ultrasound, or confocal microscopy. Images are typically taken from parallel planes within the sample, but these planes can also be at an angle to each other. The distance between the planes taking the images generally does not exceed the largest feature of the sample's internal structure. The number of images can vary. More images generally lead to more accurate results. However, the number of images increases computation time, especially if the image resolution is high. Therefore, it is generally best to use the minimum number of images that produces sufficient accuracy for a given sample. The number can range from 5 to 1000, for example, 10 to 50 or 100 to 400. The image resolution should be high enough that features can be clearly identified, but not so high that computation time becomes excessively long. Typical image resolutions range from 10x10 to 1024x1024 pixels, where the image does not need to be quadratic, so 768x1024 or 512x288 pixels can also be used. Preferably, the image is in grayscale or has been converted to grayscale.

[0026] Preferably, the image is preprocessed before generating a representation from the image to facilitate phase boundary detection. Preprocessing may include adjusting brightness, contrast, noise removal, applying a threshold, or a combination thereof. Even more preferably, the preprocessing parameters that produce optimal results in the method according to the invention are saved and automatically suggested to the user or directly applied to further images to be preprocessed.

[0027] Representations can be generated from images in various ways. Most of these involve determining the edges or surface detection or segmentation of corresponding phase boundaries. Edge detection converts 3D voxel data into 3D surface data, for example, by assigning threshold gray values ​​to edge voxels, interpolating between voxel gray values, searching for the derivative of the maximum gray value, finding intermediate gray values ​​between light air voxels and dark matter voxel levels, or using locally adaptive gray thresholding. Noise and artifact reduction, as well as interpolation, rely on numerous publications known to those skilled in the art.

[0028] Preferably, the representation is generated from the image by segmenting the grayscale image using a thresholding algorithm, thereby converting the grayscale into an image where each color represents a phase, i.e., a certain material or void. For example, in foam, the material can be white, and the voids can be black. In a composite material containing three materials, the first material can be white, the second material can be gray, and the third material can be black. In some cases, the segmented image may be sufficient for representation. However, it is often useful to apply further methods to reliably extract structural features from the representation. Preferably, the segmented image relies on a distance function that assigns the distance of each pixel or voxel to the nearest pixel or voxel with a different color. Preferably, after applying the distance function, a watershed algorithm is applied to identify objects such as holes, embedded particles, walls, pillars, or nodes. It is also possible to determine the center of walls, pillars, and nodes when the watershed algorithm overflows.

[0029] The method according to the invention includes (b) extracting at least one structural feature from a representation. A structural feature is all the features of a sample directly related to its internal structure. Structural features may relate to walls, i.e., materials between two particles of different materials or between two holes, such as their thickness, curvature, planar expansion, or moment. Structural features may also relate to supports, i.e., materials between three particles of different materials or between three holes, such as their cross-section, length, curvature, or moment. Structural features may also relate to nodes, i.e., materials between four particles of different materials or between four holes, such as their volume or moment. Structural features may also relate to holes or particles, such as their volume, sphericity, or moment. Structural features may also relate to units, such as their local density or moment. Structural features may also relate to diagrams (such as foam diagrams), such as the connections between holes, walls, units, supports, and nodes.

[0030] By evaluating these features, more complex structural characteristics can be obtained. An example is an isolated evaluation of absolute values ​​and their spatial distribution, such as determining the gradient of strut lengths in a sample. Another example is an evaluation considering structural features typically relevant in physical models, such as strut lengths related to adjacent hole sizes. Yet another example is the evaluation of foam diagrams, such as the number of walls per hole or the number of struts per wall.

[0031] According to the invention, at least one structural feature includes a wall, a support, or a node. If only one structural feature is extracted, it must be any one of a wall, a support, or a node. Typically, more than one structural feature is extracted. In this case, at least one of the structural features is a wall, a support, or a node, while the others can be remaining features among the walls, supports, or nodes, or one or more of other structural features as described above. Preferably, the structural features include at least two of walls, supports, or nodes; in particular, the structural features include all of them, i.e., walls, supports, and nodes.

[0032] Structural features can be extracted in various ways. Suitable algorithms include watershed, distance evaluation, component analysis, local voxel evaluation, or preferably combinations thereof, such as combinations of at least two of these. These methods work well in image processing libraries such as SciKit-image, OpenCV, SimpleCV, NumPy, SciPy, PIL / Pillow, Mahotas, ITK, GraphicsMagick, or Cairo. Extracting structural features from representations has several advantages over other methods, such as measurement methods: it is more flexible because it can extract very different features that would otherwise only be accessible through different measurement methods. In many cases, it is more accurate because it tends to be less susceptible to artifacts or sources of interference.

[0033] The method according to the invention includes (c) providing at least one structural feature to a material model suitable for obtaining at least one material property from the structural feature. It is possible that only one material property is determined, or preferably, more than one material property is determined, such as at least two or at least three.

[0034] A material model is generally a model that takes structural features as input and outputs associated material properties. Material models include physical models and data-driven models. Physical models use natural laws, such as thermodynamics or classical mechanics, to translate structural features into material properties. For example, L. Gibson and M. Ashby summarized a considerable number of physical models for materials with honeycomb structures in *Cellular Solids*, Cambridge University Press, ISBN 978-0-521-49911-8. Physical models can be validated using experimental data.

[0035] A data-driven model is a trained mathematical model that is parameterized based on training data to take structural features as input and output associated material properties without involving any knowledge of physical laws. The data-driven model is preferably a data-driven machine learning model. The material model can be a linear or multinomial regression model, a random forest model, a Bayesian network, or a neural network. Preferably, the data-driven model is simplified by considering only those structural features that significantly influence the material properties of interest after training with historical data. In this way, the amount of historical data required is reduced while still obtaining a robust model with high accuracy.

[0036] In the context of this invention, historical data refers to a dataset comprising at least one structural feature and at least one material property. Such data is typically obtained through measurements of historical samples, with the corresponding material properties usually obtained directly or indirectly by appropriate methods, such as mechanical tests like indentation. Associated structural features can be obtained by analyzing images of samples using methods similar to those described above, or they can be obtained through analytical methods specific to each structural feature.

[0037] The material model can run on the same system as the other steps in the process, or it can run on a remote system, such as on a server or in the cloud. In this case, structural features are sent to the remote system that executes the material model, and results are received from the remote system. This is typically achieved via a communication interface.

[0038] Generally, the type of material, i.e., the chemical composition of each phase in a sample, influences the relationship between structural characteristics and material characteristics. When samples differ from one another in their material type, it is preferable to consider information about the material type used for each phase in the sample. In this way, one can usually achieve more accurate material characteristics. The type of material can be given as a general material category, such as ceramics, resins, viscoelastic polymers such as rubber, or metals. The type of material can also be given more specifically by referring to its chemical composition, such as polystyrene, zeolite, melamine-formaldehyde resin, oak, borosilicate glass. Therefore, it is preferable to provide the material type used for each phase in the sample.

[0039] Typically, the type of material is considered by selecting an appropriate model. It is possible to set up a separate model for each type of material. This makes sense if only a few different types of materials are of interest, such as in a factory producing a small number of different products. Alternatively, the model can use the material type for each phase as additional input parameters. Clearly, for data-driven models, the historical data used to train the model needs to be appropriately labeled with the material type for each phase in the samples. For physical models, properties of bulk materials are typically chosen because these properties are readily available from databases. The error caused by the fact that material properties in small structures deviate from those of bulk materials is acceptable for most applications.

[0040] The method according to the invention includes (d) outputting at least one material property received from a material model. Output may refer to writing the material property to a non-transitory data storage medium, displaying it on a user interface, or sending it to another program on a local or remote system; preferably, at least one material property is output to the user interface.

[0041] The preferred methods for determining the material properties of a sample include:

[0042] (a1) Provides an image of a foam sample.

[0043] (a2) Provide the material type for each phase in the sample.

[0044] (a2) Convert the image into a representation.

[0045] (b) Extract at least one structural feature from the representation, wherein the at least one structural feature includes a wall, a support, or a node.

[0046] (c) Providing at least one structural feature to a material model, which is either a physical model or a data-driven model trained on historical data including structural features and material types, and

[0047] (d) Output at least one material property received from the material model.

[0048] Figure 1 An example of how the invention can be implemented is described. Samples can be produced in factory 10. This relies on a microscope device 11 that generates images of the samples. These images are converted into a representation by a processing unit 12. This representation is provided to a processing unit 13, which extracts at least one structural feature from the representation. The at least one structural feature is provided to a processing unit 14, which then provides it to a model. This model has been trained using historical data obtained from a data storage device 15. The model obtains material properties that are provided to an output device 16. The output device 16 can output the material properties to factory 10, for example, to adjust production parameters.

[0049] The present invention also relates to a non-transitory computer-readable data medium storing a computer program including instructions for performing steps of the method according to the invention. The computer-readable data medium includes, for example, a hard disk drive on a server, a USB storage device, a CD, DVD, or Blu-ray disc. The computer program may contain all the functions and data required to perform the method according to the invention, or it may provide an interface to allow portions of the method to be processed on a remote system, such as a cloud system.

[0050] This invention also relates to a production monitoring and / or control system for monitoring and / or controlling the material properties of foam samples. Unless explicitly described differently below, the description relating to the method, including preferred embodiments, also applies to the system. The system may be a computing device, such as a computer, tablet, or smartphone. Typically, the computing device has a network connection to communicate with other computing devices, such as servers or cloud networks. Production may refer to large-scale production in a factory or the production of several samples within the context of a research program. Monitoring is typically conducted in the context of quality management to ensure that products remain consistently within a set range of given material properties, or to classify products based on different specifications, such as high-quality products and average-quality products. Control may refer to the process of selecting optimal samples to facilitate and expedite the research and development process.

[0051] According to the invention, the system includes (a) an input unit configured to receive an image showing the internal structure of a sample. Preferably, the input unit includes a user interface that allows a user to select an image to be processed, for example, from a local or remote storage medium or directly from a measuring device analyzing the sample. Preferably, the input unit is configured to receive the material type for each phase in the sample. The input unit can be implemented as a web service or a standalone software package. The input unit can form a presentation or application layer. Preferably, the input unit includes a user interface.

[0052] According to the present invention, the system includes (b) a processing unit configured to extract at least one structural feature from a representation. The processing unit may be a local processing unit, including a central processing unit (CPU) and / or a graphics processing unit (GPU) and / or an application-specific integrated circuit (ASIC) and / or a tensor processing unit (TPU) and / or a field-programmable gate array (FPGA). The processing unit may also be an interface to a remote computer system, such as a cloud service.

[0053] According to the invention, the system includes (c) a processing unit configured to provide at least one structural feature to a material model, the material model being adapted to obtain at least one material property from the structural feature. The processing unit may be the same as or different from that in (b), for example, the processing unit in (b) may be on a local machine, while the processing unit in (c) is an interface to a cloud service.

[0054] According to the invention, the system includes (d) an output unit configured to output material properties received from a material model. The output unit can be implemented as a web service or a standalone software package. The output unit can form a presentation or application layer. Preferably, the output unit is a user interface configured to display the material properties of samples. The user can then take necessary actions, such as adjusting production parameters if the samples do not meet specifications, or selecting samples with the highest quality in a research project. Alternatively, the output unit may include or have an interface for automatically adjusting production parameters or classifying samples based on their material properties.

[0055] Example

[0056] Figures 2a to 2g Examples of how steps (a) and (b) can be implemented are shown. Figure 2a This shows the raw data when it is obtained, for example, from an X-ray computed tomography device. After applying filters to prepare for binarization, the following is obtained: Figure 2b The image in the image. Figure 2c The result of applying a threshold to binarize the image is shown. For Figure 2d Distance filters were applied to both phases, with opposite negative signs in the aperture phase and positive signs in the material phase. Local minima were then identified, and a watershed algorithm with lines between elements and no mask was applied. Figure 2e The results are shown. Figure 2f It shows the use of from Figure 2c The binarized data of the image is obtained from the mask marking cells to obtain the marked holes. Figure 2e The boundary line depicted in the watershed results represents the skeleton of the foam, and in Figure 2g As shown in the diagram. From there, voxels in the skeleton can be labeled by the number of their adjacent units: a voxel with two adjacent units represents a wall, a voxel with three adjacent units represents a pillar, and a voxel with four or more units represents a node. Connected voxels are labeled with the same type as individual walls, pillars, or nodes.

[0057] From there, a small portion of the material located at the unit wall φ = 0.249 was extracted, along with a description of the foam density (ρ). foam ) and the density of bulk materials (ρ) bulkThe relative density is the ratio between ρ* and 0.531. The relative Young's modulus E* = 0.344 is obtained using the following formula.

[0058]

Claims

1. A computer-implemented method for determining the material properties of a foam sample, comprising: (a) Provide a representation of the sample and the material type of each phase in the sample, wherein the representation is a data structure containing the internal structure of the sample. (b) Extract at least one structural feature from the representation, wherein the at least one structural feature includes a wall, a support, or a node. (c) Providing the at least one structural feature to a material model suitable for obtaining at least one material property from the structural feature, wherein the material model is a data-driven model trained based on historical data of the material type including the structural feature and each phase, wherein the material property is a mechanical property, thermal property, electrical property, or optical property, and wherein the mechanical property includes at least one of Young's modulus, elasticity, tear resistance, abrasion resistance, and coefficient of friction. (d) Output at least one material property received from the material model.

2. The computer-implemented method according to claim 1, wherein, The representation is generated from an image showing the internal structure of the sample.

3. The computer-implemented method according to claim 2, wherein, The representation is generated from the image by applying a thresholding algorithm to segment the grayscale image.

4. The computer-implemented method according to claim 3, wherein, The segmented image depends on the distance function and the watershed algorithm.

5. The computer-implemented method according to any one of claims 2 to 4, wherein, Before generating a representation from the image, the image is preprocessed by automatically applying saved preprocessing parameters.

6. The computer-implemented method according to any one of claims 1 to 4, wherein, The representation is a three-dimensional representation.

7. The computer-implemented method according to any one of claims 1 to 4, wherein, The material properties are displayed on the user interface.

8. The computer-implemented method according to any one of claims 1 to 4, wherein, At least two structural features are extracted from the representation and provided to the material model.

9. The computer-implemented method according to any one of claims 1 to 4, wherein, The sample is a polymer foam.

10. A non-transitory computer-readable data medium storing a computer program, the computer program comprising instructions for performing the steps of the method according to any one of claims 1 to 9.

11. A production monitoring and / or control system for monitoring and / or controlling the material properties of foam samples, comprising: (a) An input unit configured to receive a representation of the sample and the material type of each phase in the sample, wherein the representation is a data structure containing the internal structure of the sample. (b) A processing unit configured to extract at least one structural feature from the representation, wherein the at least one structural feature includes a wall, a support, or a node. (c) A processing unit configured to provide the at least one structural feature to a material model adapted to obtain at least one material property from the structural feature, wherein the material model is a data-driven model trained based on historical data of the material type including the structural feature and each phase, wherein the material property is a mechanical property, thermal property, electrical property, or optical property, and wherein the mechanical property includes at least one of Young's modulus, elasticity, tear resistance, abrasion resistance, and coefficient of friction. (d) is configured to output the material properties received from the material model.

12. The production monitoring and / or control system according to claim 11, wherein, The processing unit (b) and / or (c) is an interface to a remote computer system.

13. The production monitoring and / or control system according to claim 11 or 12, wherein, The input unit and / or the output unit is a user interface.

Images