Device for inspecting composite material, method for inspecting composite material, and program for inspecting composite material

Abstract

To provide a device for inspecting a composite material capable of using the device for evaluating the composite material by inferring mechanical properties of the composite material without measuring mechanical properties.SOLUTION: An inspection device comprises a storage section for storing a mechanical property inference model generated by machine learning on the basis of mechanical property information and nondestructive inspection information of a first composite material, which is reinforced with reinforcing fibers and for which the mechanical property information and the nondestructive inspection information are known, the mechanical property inference model inferring mechanical property information of a second composite material which is reinforced with reinforcing fibers and for which the mechanical property information is unknown, with nondestructive inspection information of the second composite material as input. The inspection device acquires the nondestructive inspection information of the second composite material, inputs the nondestructive inspection information to the mechanical property inference model, acquires the mechanical property information of the second composite material from the mechanical property inference model, and outputs on the basis of the mechanical property information.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an inspection device, an inspection method, and an inspection program for a fiber-reinforced composite material. [Background technology]

[0002] Carbon fiber-reinforced composite materials are able to reinforce the brittleness of the matrix resin with high-strength fibers, and are therefore widely used as lightweight, high-performance materials. Conventionally, in the production process of fiber-reinforced composite materials, ultrasonic inspection is performed to inspect defective products during the manufacture of the composite materials. For example, in Patent Document 1, the following inspection is performed in the process of impregnating carbon fibers with thermoplastic resin. First, an ultrasonic transmitter and receiver having directivity are placed facing the object to be inspected (a composite material in which thermoplastic resin is impregnated into carbon fibers) at a certain distance. Then, ultrasonic waves are emitted from one of the ultrasonic transmitters, the ultrasonic waves are received by the opposing receiver of the object to be inspected, and the propagation time of the ultrasonic waves is measured by a signal processing circuit, and internal defects of the object to be inspected are detected in a non-contact manner. The data of the inspection using ultrasonic waves here is converted into an image, and the pass / fail judgment of the object to be inspected can be made based on the image. Patent documents 2 and 3 disclose a device that automatically performs highly accurate searches to efficiently separate blood vessels, feathers, etc. during the production process of processed foods. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-158459 A [Patent Document 2] International Publication No. 2019 / 151393 [Patent Document 3] International Publication No. 2019 / 151394 Summary of the Invention [Problem to be solved by the invention]

[0004] In recent years, due to rising raw material prices and labor costs, it has become increasingly difficult to reduce production costs while maintaining high quality, and there is a demand for low-cost, highly accurate inspection of fiber-reinforced composite materials. The composite material sorting work by ultrasonic inspection described in Patent Document 1 relies on visual inspection of the obtained images. For this reason, it is difficult to grasp the state of the composite material in detail. In particular, it is difficult to set objective evaluation criteria when judging pass / fail by visually inspecting the images, and it is difficult to determine what criteria should be used to calculate the pass / fail criteria for the material. Furthermore, the food inspection systems described in Patent Documents 2 and 3 are devices for humans to locate bones, and are different from technologies for inspecting composite materials. These food inspection systems merely use neural networks to replace what humans can judge by looking at images or objects to be measured.

[0005] The present invention aims to provide an inspection device, an inspection method, and an inspection program for a composite material that can estimate the mechanical properties of a composite material without measuring the mechanical properties, and can be useful for evaluating the composite material. [Means for solving the problem]

[0006] The above object can be achieved by the following aspects. The inspection device according to one embodiment of the present invention is a mechanical property estimation model generated by machine learning based on mechanical property information and non-destructive inspection information of a fiber-reinforced first composite material, the mechanical property information and non-destructive inspection information of which are known, a processor that can access a model storage unit that stores a mechanical property estimation model that estimates mechanical property information of a fiber-reinforced second composite material by inputting non-destructive testing information of the second composite material, the mechanical property information of which is unknown; The processor acquires non-destructive testing information of the second composite material, inputs the non-destructive testing information to the mechanical property estimation model, acquires mechanical property information of the second composite material from the mechanical property estimation model, and performs output based on the mechanical property information, where the non-destructive testing information is vibration property information or acoustic property information, and the sampling frequency for acquiring the non-destructive testing information of the second composite material is greater than 0 Hz.

[0007] An inspection method according to one aspect of the present invention includes the steps of: generating a mechanical property estimation model for estimating mechanical property information of a fiber-reinforced first composite material, whose mechanical property information and non-destructive testing information are known, by machine learning the mechanical property information and non-destructive testing information of the second composite material using non-destructive testing information of a fiber-reinforced second composite material, whose mechanical property information is unknown, as an input, acquiring non-destructive testing information of the second composite material, inputting the acquired non-destructive testing information into the mechanical property estimation model, acquiring mechanical property information of the second composite material from the mechanical property estimation model, and performing output based on the mechanical property information, where the non-destructive testing information is vibration property information or acoustic property information, and the sampling frequency for acquiring the non-destructive testing information of the second composite material is greater than 0 Hz.

[0008] An inspection program according to one embodiment of the present invention causes a processor to execute the steps of: generating a mechanical property estimation model for estimating mechanical property information of a fiber-reinforced first composite material, whose mechanical property information and non-destructive testing information are known, by machine learning the mechanical property information and non-destructive testing information of the second composite material using non-destructive testing information of a fiber-reinforced second composite material, whose mechanical property information is unknown, as an input; acquiring non-destructive testing information of the second composite material; inputting the acquired non-destructive testing information into the mechanical property estimation model, acquiring mechanical property information of the second composite material from the mechanical property estimation model, and performing output based on the mechanical property information; wherein the non-destructive testing information is vibration property information or acoustic property information, and the sampling frequency for acquiring the non-destructive testing information of the second composite material is greater than 0 Hz. Effect of the Invention

[0009] According to the present invention, it is possible to estimate the mechanical properties of a composite material only from non-destructive testing information without measuring the mechanical properties, and to use the estimated properties in the evaluation of the composite material. According to the present invention, it is possible to instantly estimate mechanical property information with high accuracy, which cannot be estimated from non-destructive testing information no matter how hard a person tries. This reduces waste in the production of composite materials, and allows high-quality composite materials to be provided at low cost. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of an inspection system. [Diagram 2] 13 is a flowchart of a learning process. [Diagram 3] A diagram showing an example of a neural network that outputs three response values. [Figure 4] FIG. 1 is a diagram showing the calculation process between units of a neural network. [Diagram 5] 13 is a flowchart of an estimation process. [Figure 6] The classification surface and distribution of response values ​​when RBF is used as the activation function. [Figure 7] The discrimination surface and distribution of response values ​​when a sigmoid function is used as the activation function. [Figure 8] (a)(b)(c)(d) Vibration inspection images taken using the analysis tool BIOMAS from Elmec. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] An inspection system including an inspection device according to one embodiment of the present invention will be described below, but the present invention is not limited thereto.

[0012] [Outline of the inspection system] The inspection system of this embodiment takes a fiber-reinforced composite material (second composite material) whose mechanical properties are unknown as the object to be inspected, and estimates the mechanical properties of this second composite material without actually measuring it. The mechanical property information of a composite material is information indicating the mechanical properties of the composite material, for example, information on the fracture or elasticity of the composite material, such as strength. Examples of the mechanical property information include information on fracture strength, such as tensile strength or bending strength, and information on the elastic modulus associated with each of fracture strength, compressive strength, shear strength, etc.

[0013] The information on the elastic modulus may be the elastic modulus itself (e.g., tensile elastic modulus or flexural elastic modulus) or a rank (hereinafter, referred to as a mechanical property rank) of the elastic modulus (e.g., tensile elastic modulus or flexural elastic modulus). The information on the elastic modulus may further include any one of information indicating that the elastic modulus (or its rank) corresponds to a defective product, information indicating that the elastic modulus (or its rank) corresponds to a non-defective product, and information indicating that the elastic modulus (or its rank) was difficult to estimate.

[0014] The information on the breaking strength may be the breaking strength itself (e.g., tensile strength or yield strength), or may be a rank (hereinafter, referred to as a mechanical property rank) of the breaking strength (e.g., tensile strength or yield strength). The information on the breaking strength may further include any one of information indicating that the breaking strength (or its rank) corresponds to a defective product, information indicating that the breaking strength (or its rank) corresponds to a non-defective product, and information indicating that the breaking strength (or its rank) was difficult to estimate.

[0015] A computer included in the inspection system acquires non-destructive inspection information of the second composite material, inputs this non-destructive inspection information into a mechanical property estimation model that has been generated in advance and stored in a model storage unit, estimates the mechanical property information of the second composite material using this mechanical property estimation model, and outputs based on the estimation results. Examples of the output method include displaying information (e.g., mechanical property rank, information on whether the product is good or bad, information on whether it is difficult to estimate, etc.) on a display unit, playing the information as a message from a speaker, printing the information on a printer, etc.

[0016] In the present invention, the non-destructive inspection information is preferably vibration characteristic information. The vibration characteristic is information on the composite material itself, such as displacement and acceleration, obtained by vibrating the composite material in a fixed or unfixed state. The non-destructive inspection information may be numerical data or an inspection image using a wavelet image. In the case of an image, for example, the wavelet image in FIG. 8 can be exemplified.

[0017] The mechanical property estimation model is a model that outputs mechanical property information using non-destructive testing information as an input, and is generated by performing machine learning (including supervised learning or unsupervised deep learning) on ​​data of a fiber-reinforced composite material (first composite material) whose mechanical property information and non-destructive testing information data are known. For example, a neural network or a support vector machine is used as the mechanical property estimation model.

[0018] The computer of the inspection system constitutes an inspection device. This computer may include a processor, a storage unit consisting of a device capable of storing information such as a hard disk device or a solid state drive (SSD), a random access memory (RAM) and a read only memory (ROM). This processor executes an inspection program stored in the ROM to perform processes such as acquiring non-destructive inspection information of the composite material of the object to be inspected, inputting the acquired non-destructive inspection information into a mechanical property estimation model, acquiring mechanical property information from the mechanical property estimation model, and outputting based on the acquired mechanical property information. The storage medium for storing the inspection program is not limited to the ROM, and any known storage medium can be used, but it is preferable that the storage medium is a non-transient storage medium, and a hard disk device or a solid state drive (SSD) may be used. The storage medium may also be referred to as a recording medium. The inspection program may also be stored in a server (cloud) on a network, and the processor may download the program from the server and execute the program.

[0019] Non-destructive inspection information of a composite material is usually used to determine whether defects, voids, or foreign matter exist inside the composite material, and if so, what degree of existence thereof. However, even if many defects, voids, or foreign matter exist inside the composite material, the mechanical properties may be good depending on the distribution state of the defects, voids, or foreign matter. In such a case, if the non-destructive inspection information is visually confirmed and it is determined that the product is defective because there are many defects, voids, or foreign matter, the composite material that should have been a good product will have to be discarded, resulting in a decrease in production efficiency. On the other hand, the opposite is also possible. In other words, even if the non-destructive inspection information is visually confirmed and it is determined that the product is good because there are few defects, voids, or foreign matter, the mechanical properties may be in a state corresponding to a defective product.

[0020] As a result of verification based on the above viewpoints, the inventors found that there is a correlation between nondestructive inspection information and mechanical property information, and succeeded in estimating the mechanical property information of a composite material with high accuracy from the nondestructive inspection information of the composite material by machine learning a large amount of actual measurement data of nondestructive inspection information and mechanical property information into a model such as a neural network or a support vector machine. It has not been considered in the past to obtain mechanical property information from nondestructive inspection information. For this reason, it has not been easy for a person skilled in the art to construct a machine learning model that inputs nondestructive inspection information and outputs mechanical property information. A detailed example of the inspection system will be described below. Note that, in the following, an example will be described in which the mechanical property estimation model is a neural network.

[0021] [Reinforced fiber] The type of reinforcing fiber used in the present invention can be appropriately selected depending on the application of the composite material a (second composite material whose mechanical property information is unknown) to be inspected, and is not particularly limited. Either inorganic fiber or organic fiber can be suitably used as the reinforcing fiber. Examples of the inorganic fibers include carbon fibers, activated carbon fibers, graphite fibers, glass fibers, tungsten carbide fibers, silicon carbide fibers, ceramic fibers, alumina fibers, natural mineral fibers (such as basalt fibers), boron fibers, boron nitride fibers, boron carbide fibers, and metal fibers.

[0022] [Carbon fiber] When carbon fibers are used as the fibers, generally known carbon fibers include polyacrylonitrile (PAN)-based carbon fibers, petroleum / coal pitch-based carbon fibers, rayon-based carbon fibers, cellulose-based carbon fibers, lignin-based carbon fibers, phenol-based carbon fibers, and vapor-grown carbon fibers, and any of these carbon fibers can be suitably used in the present invention.

[0023] [Mold of reinforcing fiber] In the present invention, the form of the reinforcing fiber is not particularly limited, but the following describes the continuous fiber that the present inventors have used as a specific example, although the present invention is not limited to the continuous fiber. The continuous fiber means a reinforcing fiber in which the reinforcing fiber bundles are arranged in a continuous state without cutting the reinforcing fiber into short fibers. In order to obtain a composite material a having excellent mechanical properties, it is preferable to use a continuous reinforcing fiber. More specifically, the continuous fiber is preferably a fiber having a length of 1 m or more, and is used by impregnating the resin by hand layup or the like after processing into a woven fabric such as a woven fabric or knitted fabric, or by impregnating the continuous fiber with uncured resin to form a prepreg.

[0024] [Composite material a] Composite material A is reinforced with reinforcing fibers. An example of an embodiment carried out by the present inventors will be described below, but the present invention is not limited to the composite material A described below. 1. Molded body The composite material A is preferably a molded body after molding, and may be a molded body using a thermoplastic resin or a molded body using a thermosetting prepreg. A prepreg is a molding intermediate material in which a substrate made of carbon fibers such as continuous carbon fibers arranged in one direction and formed into a sheet, or a substrate made of carbon fibers such as carbon fiber fabric is impregnated with a thermosetting resin, or the substrate is partially impregnated with a thermosetting resin and the remaining portion is placed on at least one surface. 2. Unidirectional material The composite material a is preferably a unidirectional material. A unidirectional material is one in which continuous reinforcing fibers with a length of 100 mm or more are arranged in one direction inside the composite material a. The unidirectional material may be one in which a plurality of continuous reinforcing fibers are laminated. In particular, when the composite material a is a unidirectional material and is a molded body using a thermosetting prepreg, the influence of fiber orientation on the mechanical properties is small. For this reason, the accuracy of estimating mechanical property information using the model described below can be improved.

[0025] [Preferred composite material] The composite material contains reinforcing fibers and a matrix resin as essential components, and contains other components as optional components, and has a porosity Vr of 10% or less, which is calculated by the following formulas (A) and (B). Vr = (t2-t1) / t2 × 100 Equation (A) t1 = (Wf / Df + Wm / Dm + Wz / Dz) ÷ unit area (mm 2 ) ... Formula (B) t1: Theoretical thickness of the composite material mm t2: Actual thickness of composite material mm Df: Density of reinforcing fiber (mg / mm 3 ) Dm: density of matrix resin (mg / mm 3 ) Dz: Density of other components (mg / mm 3 ) Wf: Mass ratio of reinforcing fibers (%) Wm: Mass ratio of matrix resin (%) Wz: Mass ratio of other components (%) The porosity Vr is more preferably 5% or less, and further preferably 3% or less. If the porosity is within this range, the accuracy of the mechanical property prediction of the present invention is improved.

[0026] [Production of composite material a] For example, composite material a can be prepared as follows. 1.Material Reinforcement fiber: Carbon fiber "Tenax (registered trademark)" STS40-24K (tensile strength 4,300 MPa, tensile modulus 240 GPa, number of filaments 24,000, fineness 1,600 tex, elongation 1.8%, density 1.78 g / cm 3 (Teijin Limited) Base resin: Thermosetting resin composition with epoxy resin as the main component 2. Preparation of unidirectional prepreg The unidirectional prepreg was produced by the hot melt method as follows. First, the above-mentioned thermosetting resin composition was applied onto a release paper using a coater to produce a resin film. Next, the above-mentioned carbon fiber bundle was sent out from the creel, passed through a comb to align the pitch between the carbon fiber bundles, and then spread through a fiber spreading bar to produce a fiber weight per unit area of ​​100 g / m. 2 Sheet The carbon fibers were aligned in one direction so as to form a unidirectional prepreg. The resin film was then placed on both sides of the carbon fibers, heated and pressurized to impregnate the carbon fibers with the thermosetting resin composition, and the carbon fibers were wound up with a winder to produce a unidirectional prepreg. The resin content of the resulting unidirectional prepreg was 30 wt.%.

[0027] 3. Preparation of composite material A 11 sheets of unidirectional prepreg are stacked by hand in the 0° direction to form a laminated structure

[0011] T The prepreg laminate was placed in a bag film, which was then heated in an autoclave at 130°C for 120 minutes to harden the laminate and produce a CFRP molded product (composite material a, which is a unidirectional carbon fiber reinforced thermosetting resin composite material) having a thickness of 1 mm.

[0028] [Measurement of tensile modulus and tensile strength] As a specific example of the breaking strength or elastic modulus of the present invention, the present inventors measured the tensile modulus and tensile strength of composite material a as described below. The CFRP molded body was processed into a test piece shape (length 250 mm × width 15 mm) using a water jet, and a tab made of a glass fiber reinforced resin-based composite material was attached. A tensile test in the 0° direction was performed at a test speed of 2 mm / min using a universal testing machine in accordance with the ASTM D3039 method, and the tensile modulus and tensile strength of the CFRP molded body (composite material a) were calculated.

[0029] [Vibration characteristics information] In the present invention, the non-destructive inspection information is preferably vibration characteristic information. There is no particular limitation on the vibration inspection method used to obtain the vibration characteristic information, and any inspection method that detects internal defects, voids, or foreign objects in the molded body region without destroying the molded body region may be used. The vibration characteristic information may also be obtained by using a finite element method (FEA). The vibration characteristic information is preferably obtained by converting information obtained by the vibration inspection or the finite element method into an image, and it is particularly preferable that the converted image is a wavelet image. Specific wavelet images are shown in Figure 8 (a) (b) (c) (d). To obtain these images, vibration measurements were performed using impulse excitation under free-free boundary conditions. A hole of Φ2 mm was drilled in the compact, and nylon string (22-8231, made by Takagi Rope Industry) was passed through the hole and the compact was hung from a beam to achieve free-free boundary conditions. An acceleration pickup sensor (PCB PIEZOTRONICS 356A01) was installed in the compact area, and vibration was applied using an impulse hammer (GK-3100, made by Ono Sokki). A real-time acoustic vibration analysis system (DS-3000, made by Ono Sokki) was used to measure and analyze the data. The sampling frequency was set to 2000 Hz. Figure 8 (a) (b) shows wavelet images of the compact area without defects, and (c) (d) shows wavelet images of the compact area with defects. It was confirmed that 803 and 804 in Figures 8(c) and (d) have a spherical pattern, which is clearly different from 801 and 802 in Figures 8(a) and (b).

[0030] [Vibration characteristics: vibration damping] The image preferably includes information on vibration damping. Vibration damping is a vibration phenomenon in which the amplitude decreases over time in vibration time history data. For example, looking at the patterns 801 and 802 in Figures 8(a) and (b), it can be seen that the pattern becomes thinner from left to right on the paper in Figure 8, and the vibration damping is greater. In this case, Figure 8(a) has greater vibration damping than (b). In the molded body region where the measurement data in Figure 8 was measured, the greater the vibration damping, the smaller the mechanical properties. Note that greater vibration damping does not necessarily mean smaller mechanical properties. Figure 8 is merely an example.

[0031] [Acoustic characteristics] The non-destructive inspection information in the present invention may be acoustic property information. [frequency] 1. Sampling Frequency Generally, the sampling frequency refers to the frequency at which samples are taken per unit time in sampling, which is a process required to convert analog waveforms such as audio into digital data. In order to correctly sample a waveform, it is necessary to sample at a frequency that is at least twice the bandwidth of the frequency components of the waveform. In the present invention, the nondestructive testing information is vibration characteristic information or acoustic characteristic information, and the sampling frequency for obtaining the nondestructive testing information of the second composite material is greater than 0 Hz. The lower limit is preferably 50 Hz or more, more preferably 250 Hz or more, even more preferably 5,000 Hz or more, and even more preferably 10,000 Hz or more, whereas the upper limit is preferably 50,000 Hz or less, more preferably 40,000 Hz or less, and even more preferably 30,000 Hz or less. From another point of view, the sampling frequency when acquiring the vibration characteristics of the second composite material is preferably at least twice the natural frequency of the first mode described below, and more preferably at least twice the natural frequency of the second mode. 2. Natural frequency of the first mode When the nondestructive testing information is vibration characteristic information, the natural frequencies of the first mode of the first composite material and the second composite material are preferably between 0 Hz and 1000 Hz. Within this range, for example, when the molded body is assembled in an automobile, it does not resonate with vibrations from the outside or the engine room, improving comfort. More preferably, the natural frequencies of the first mode of the first composite material and the second composite material are between 0 Hz and 500 Hz. On the other hand, when the nondestructive inspection information is acoustic property information, the natural frequencies of the primary modes of the first composite material and the second composite material are preferably greater than 0 Hz and less than or equal to 20,000 Hz, and more preferably greater than 0 Hz and less than or equal to 10,000 Hz.

[0032] [Finite element method] The mechanical property information or non-destructive testing information of the first composite material is preferably obtained by the finite element method. Here, the finite element method (FEM) is one of the numerical analysis techniques, and can numerically obtain an approximate solution of a differential equation that is difficult to solve analytically. By creating an analytical model in which material parameters identified in advance by comparing actual measurements with analysis are applied to the shape of the first composite material, it is possible to obtain such information by the finite element method without actually measuring the mechanical properties or non-destructive testing (preferably vibration characteristic or acoustic characteristic information).

[0033] [Inspection system] Hereinafter, data converted into a format that can be input to the input layer of the neural network will be referred to as input data. In the inspection system, non-destructive testing information (hereinafter referred to as non-destructive testing information sample) of a sample of the first composite material a (hereinafter referred to as composite material sample b) and mechanical property information (hereinafter referred to as mechanical property information sample) obtained by actual measurement of the composite material sample b are acquired as second input data, and the neural network is trained using this second input data. After the neural network has completed training, the non-destructive testing information of the composite material a is input to the neural network as the first input data, and the mechanical property information of the composite material a is estimated based on the reaction value in the output layer from the neural network. Composite material a may be sorted into good and bad products based on the estimated mechanical property information.

[0034] In order to perform effective learning and highly accurate inference, the inspection system can use non-destructive inspection information (preferably vibration inspection images or acoustic characteristic images) that is optimized for the learning and inference processes. For example, various image processing may be performed on the vibration inspection images so that they can be easily detected. Note that the vibration inspection images are one type of vibration characteristic information.

[0035] [Inspection equipment] The inspection device 1 performs image processing, generation of input data, learning of a neural network, estimation of mechanical property information using a neural network, etc. The inspection device 1 is an information processing device such as a computer that includes one or more processors composed of a CPU (Central Processing Unit) etc., a storage unit, and a communication unit, and on which an OS (Operating System) and applications run. The inspection device 1 may be a physical computer, or may be realized by a virtual computer (Virtual Machine: VM), a container, or a combination of these. More specifically, the structure of the processor is an electric circuit that combines circuit elements such as semiconductor elements.

[0036] The inspection device 1 includes an image storage unit 11 that stores nondestructive inspection information and nondestructive inspection information samples, a processing unit 12 that processes nondestructive inspection information and nondestructive inspection information samples, an input data generation unit 13, a learning data storage unit 14, a learning unit 15, a model storage unit 16, an estimation unit 17, a display unit 18, and an operation unit 19. The processing unit 12, the input data generation unit 13, the learning unit 15, and the estimation unit 17 are each a functional block realized by the processor of the inspection device 1 executing a program. The program includes an inspection program for composite materials.

[0037] The image storage unit 11 is preferably a storage area for storing vibration inspection images (or acoustic characteristic images). The image storage unit 11 may be a volatile memory such as SRAM or DRAM, or a non-volatile memory such as NAND, MRAM, or FRAM (registered trademark).

[0038] The processing unit 12 preferably performs image processing on the vibration inspection image (or the acoustic characteristic image) and stores the image after the image processing in the image storage unit 11. Examples of the image processing include generating an image in which the brightness of each color of red, green, and blue (RGB) is extracted from each pixel in the image, generating an image in which the brightness of green (G) is subtracted from the brightness of red (R) from the brightness of red (R) in each pixel, and generating an image in which only the red component is extracted after conversion to the HSV color space, but other types of image processing may also be performed. The processing unit 12 may also perform other operations such as enlarging, reducing, cropping, removing noise, rotating, inverting, changing color depth, adjusting contrast, adjusting brightness, adjusting sharpness, and correcting colors of the image.

[0039] The input data generating unit 13 generates input data to be input to the input layer of the neural network from the nondestructive inspection information or nondestructive inspection information samples stored in the image storage unit 11. For example, when performing learning described below using a vibration inspection image, it is preferable to cut out a desired portion from the vibration inspection image or remove unnecessary portions to obtain the second input data.

[0040] When the inspection device 1 is executing a learning process, the input data generation unit 13 stores the input data in the learning data storage unit 14. When the inspection device 1 is inspecting the composite material a, the input data is transferred to the estimation unit 17. When performing the learning process, the input data generating unit 13 may generate input data using, for example, an image (non-destructive inspection information sample) captured by an external device or system.

[0041] The learning data storage unit 14 is a storage area for storing a plurality of input data used for learning the neural network. The input data stored in the learning data storage unit 14 is used as learning data for the learning unit 15. The input data (second input data) used as learning data is stored in association with a mechanical property information sample obtained by measuring the composite material sample b from which the input data was obtained. This mechanical property information sample may include, in addition to at least one of the mechanical property rank and the mechanical property value (e.g., elastic modulus) of the composite material sample b, information indicating that the mechanical property value corresponds to a non-defective product, information indicating that the mechanical property value corresponds to a defective product, information indicating that it is difficult to estimate the mechanical property value, and the like.

[0042] For example, the correspondence of the mechanical property information sample to the second input data obtained from the composite material sample b (hereinafter, this correspondence will also be referred to as labeling) can be achieved by the user directly inputting the mechanical property value (e.g., elastic modulus) of the composite material sample b by operating the operation unit 19. After the input, the inspection device 1 classifies the mechanical property value into mechanical property ranks. For example, the tensile elastic modulus can be classified into the following mechanical property ranks. Mechanical properties rank 1: Tensile modulus of composite material is 30 GPa or more Mechanical properties rank 2: Tensile modulus of composite material is 25-30GPa Mechanical properties rank 3: Tensile modulus of composite material is 25 GPa or less

[0043] The mechanical property rank may be output in units of 3 GPa or 1 GPa instead of 5 GPa as described above. If the mechanical property information sample corresponding to the second input data obtained from the composite material sample b is known, the mechanical property rank may be automatically labeled by a program, script, or the like, instead of by a user operation. The mechanical property rank may be labeled before or after the nondestructive inspection information sample obtained from the composite material sample b is converted into the second input data.

[0044] The learning unit 15 uses the input data (second input data) stored in the learning data storage unit 14 to learn the neural network. The learning unit 15 stores the learned neural network in the model storage unit 16. The learning unit 15 can learn, for example, a three-layered neural network including an input layer, a hidden layer, and an output layer. By learning a three-layered neural network, it is possible to ensure real-time response performance during inspection of the composite material a. There is no particular limit to the number of units included in each of the input layer, hidden layer, and output layer. The number of units included in each layer can be determined based on the desired response performance, the target of estimation, the discrimination performance, etc.

[0045] Note that the three-layer neural network is just one example, and does not preclude the use of a multi-layer neural network with more layers. When using a multi-layer neural network, various types of neural networks, such as a convolutional neural network, can be used.

[0046] The model storage unit 16 is a storage area that stores the neural network learned by the learning unit 15. The model storage unit 16 may store a plurality of neural networks according to the type of the composite material a to be inspected. Since the model storage unit 16 is set so as to be accessible by the estimation unit 17, the estimation unit 17 can use the neural network stored in the model storage unit 16 to inspect the composite material a (estimate mechanical property information). The model storage unit 16 may be a volatile memory such as RAM or DRAM, or a non-volatile memory such as NAND, MRAM, or FRAM (registered trademark). The model storage unit 16 may be located in a place accessible by the processor of the inspection device 1, and may not be built into the inspection device 1. For example, the model storage unit 16 may be a storage external to the inspection device 1, or may be a network storage connected to a network accessible from the inspection device 1.

[0047] The estimation unit 17 estimates the mechanical property information of the composite material a using the neural network stored in the model storage unit 16. The estimation unit 17 estimates the mechanical property rank of the composite material a based on the reaction values ​​output from the units of the output layer. Examples of the units of the output layer include units of mechanical property rank 1, units of mechanical property rank 2, units of mechanical property rank 3, and units that are difficult to estimate, but other types of units may be prepared. For example, those with a low estimated mechanical property rank may contain a large amount of foreign matter. The mechanical property rank of the composite material a may be estimated using the difference or ratio of the reaction values ​​of multiple units.

[0048] The display unit 18 is a display that displays images and texts. The display unit 18 may display a captured image, an image after image processing, and the result of estimation by the estimation unit 17.

[0049] The operation unit 19 is a device that provides a means for a user to operate the inspection device 1. The operation unit 19 is, for example, a keyboard, a mouse, a button, a switch, a voice recognition device, or the like, but is not limited to these.

[0050] [Learning process] Before the inspection device 1 estimates the mechanical property rank of the composite material a, it is necessary to train the neural network using a nondestructive inspection information sample and a mechanical property information sample of a composite material sample b of the same type as the composite material a. Figure 2 is a flowchart of the training process.

[0051] First, the processor of the inspection device 1 acquires a non-destructive information sample for each of the multiple composite material samples b (step S201). The non-destructive inspection information samples here include those with high and low mechanical property ranks. When a unit that outputs a difficult-to-estimate reaction value is provided in the output layer of the neural network, a non-destructive inspection information sample whose mechanical property information is difficult to estimate may be prepared. Examples of non-destructive inspection information samples whose mechanical property information is difficult to estimate include, when the non-destructive inspection information sample is an image, an image in which the composite material sample b is not sufficiently captured, an image in which the brightness adjustment by lighting or exposure is inappropriate, and the like in which the composite material sample b is not clearly captured.

[0052] The processor of the inspection device 1 generates second input data from each acquired nondestructive inspection information sample (step S201). Next, the processor of the inspection device 1 acquires a mechanical property information sample of each of the multiple composite material samples b, and stores the acquired mechanical property information sample in association with each second input data (step S203). Note that step S203 may be performed before step S202. In this case, even after each nondestructive inspection information sample is converted into the second input data, the mechanical property information sample associated with the nondestructive inspection information sample may be inherited by the second input data.

[0053] Next, the processor of the inspection device 1 starts learning by the neural network based on the second input data (step S204).

[0054] Figure 3 shows an example of a neural network that outputs three reaction values. Neural network 301 in Figure 3 is a neural network having three layers: input layer 302, hidden layer 303, and output layer 304. Output layer 304 includes units 311, 312, and 313 that estimate the mechanical property rank. Although there are three units 311, 312, and 313 in Figure 3, the number can be increased or decreased as appropriate depending on the rank of the mechanical property. In the neural network, a value input to the input layer is propagated to the hidden layer and the output layer, and a reaction value of the output layer is obtained. In step S204, when second input data is input to the neural network, each parameter and structure of the neural network, such as the number of hidden layers 303, the number of units included in each of the input layer 302 and the hidden layer 303, and the coupling coefficient between the units included in each of the input layer 302 and the hidden layer 303, are adjusted so that a mechanical property information sample corresponding to the second input data or information similar thereto is output from the neural network with a high probability. In this way, a mechanical property estimation model is generated and stored in the model storage unit 16.

[0055] Figure 4 shows the computational processing between units in a neural network. Figure 4 shows units in the (m-1)th layer and units in the mth layer. For the sake of explanation, only some of the units in the neural network are shown in Figure 4. The unit numbers in the (m-1)th layer are k = 1, 2, 3, etc. The unit numbers in the mth layer are j = 1, 2, 3, etc. The reaction value of unit number k in the m-1th layer is ak m-1 Then, the reaction value a of unit number j of the mth layer is j m can be calculated using the following formula (2).

[0056]

number

[0057] Here, W jk m is the weight, which indicates the strength of the connection between units. j m is the bias. f( ) is the activation function. From equation (2), we can see that the response value of any unit in the mth layer is the output value when the response values ​​of all units (k = 1, 2, 3, ) in the m-1th layer are weighted together and input as a variable of the activation function. Next, an example of the activation function will be explained. The following formula (3) is a normal distribution function.

[0058]

number

[0059] Here, μ is the mean value, which indicates the center position of the bell-shaped peak drawn by the normal distribution function. σ is the standard deviation, which indicates the width of the peak. Since the value of equation (3) depends only on the distance from the center of the peak, the Gaussian function (normal distribution function) can be said to be a type of radial basis function (RBF). The Gaussian function (normal distribution function) is just one example, and other RBFs may also be used. The following equation (4) is the sigmoid function. The sigmoid function asymptotically approaches 1.0 in the limit of x → ∞. It also asymptotically approaches 0.0 in the limit of x → -∞. In other words, the sigmoid function takes values ​​in the range of (0.0, 1.0).

[0060]

number

[0061] Note that this does not preclude the use of functions other than the Gaussian function or the sigmoid function as the activation function. For example, the present inventors used Relu in the convolution layer and softmax in the output layer.

[0062] In neural network training, when input data is input to the input layer, the weights W, which are the strengths of connections between units, are adjusted to obtain the correct output. jk The correct output (the response value of the unit in the output layer) expected when input data labeled with a certain mechanical property rank is input to a neural network is also called a teacher signal. For example, when input data labeled with a mechanical property rank of 311 is input to neural network 301, the teacher signal will have a response value of 1 for unit 311, a response value of 0 for unit 312, and a response value of 0 for unit 313. When input data labeled with a mechanical property rank of 312 is input to neural network 301, the teacher signal will have a response value of 0 for unit 311, a response value of 1 for unit 312, and a response value of 0 for unit 313. For example, the weight W jk The adjustment of can be performed using the backpropagation method. In the backpropagation method, the weights W are adjusted in order from the output layer side so that the deviation between the output of the neural network 310 and the teacher signal becomes small. jk The following equation (5) shows the improved backpropagation method.

[0063]

number

[0064] When a Gaussian function is used as the activation function, the weight W jkIn addition, σ and μ in formula (3) are also adjusted as parameters in the improved backpropagation method. Adjusting the values ​​of parameters σ and μ helps the neural network converge in learning. The following formula (6) shows the adjustment process for the value of parameter σ.

[0065]

number

[0066]

number

[0067] Here, t is the number of times of learning, η is the learning constant, and δ k is the generalization error, O j is the reaction value of unit number j, α is the sensitivity constant, and β is the vibration constant. ΔW jk , Δσ jk , Δμ jk is the weight W jk , σ, and μ are shown.

[0068] Here, we use the improved backpropagation method as an example to find the weights W jk In the above, the parameter adjustment process has been described, but the general backpropagation method may be used instead. In the following, when the term "backpropagation method" is used, it is assumed that both the improved backpropagation method and the general backpropagation method are included. Weight W by pack propagation method jk The number of times that the parameters are adjusted may be one or more times, and is not particularly limited. In general, the weights W are calculated by the backpropagation method based on the prediction accuracy of the mechanical property rank when the test data is used. jk It is possible to determine whether or not to repeat the parameter adjustment. jk Repeated adjustment of parameters or parameters may improve the accuracy of estimating the mechanical property rank. By using the above method, in step S204, the weight W jk , the values ​​of the parameters σ and μ can be determined. jk Once the values ​​of the parameters σ and μ are determined, it becomes possible to perform estimation processing using a neural network.

[0069] 5 is a flow chart for explaining the operation of estimating mechanical property information by the inspection device 1 operating according to an inspection program for a composite material. The processor of the inspection device 1 acquires non-destructive inspection information of the composite material a (preferably takes a vibration inspection image or an acoustic characteristic image) (step S501). When the vibration inspection image (or the acoustic characteristic image) is used as the non-destructive inspection information, a step of performing image processing on the vibration inspection image (or the acoustic characteristic image) may be included between steps S501 and S502.

[0070] Next, the processor of the inspection device 1 generates first input data from the non-destructive inspection information (step S502). The first input data has N elements, which is equal to the number of units in the input layer of the neural network, and is in a format that can be input to the neural network.

[0071] Next, the processor of the inspection device 1 inputs the first input data to the neural network (step S503). The first input data is transmitted in the order of the input layer, hidden layer, and output layer. The processor of the inspection device 1 estimates the mechanical property rank based on the reaction value in the output layer of the neural network (step S504).

[0072] The estimation process using a neural network is equivalent to the process of finding the position of the first input data in the discrimination space. Figure 6 shows an example of the discrimination space when a Gaussian function is used as the activation function. When an RBF such as a Gaussian function is used as the activation function, the discrimination surface that divides the discrimination space into regions for each rank of mechanical physical properties becomes a closed surface. In addition, by adding a height index for each category of mechanical physical property rank, it is possible to localize the region related to each category in the discrimination space.

[0073] Fig. 7 shows an example of the discrimination space when a sigmoid function is used as the activation function. When the activation function is a sigmoid function, the discrimination surface becomes an open surface. The above-mentioned neural network learning process corresponds to a process of learning the discrimination surface in the discrimination space. Although only the mechanical property rank 311 and the mechanical property rank 312 are shown in the area in Figs. 6 and 7, there may be a distribution of three or more mechanical property ranks.

[0074] As described above, by using the inspection system of this embodiment, the mechanical property information of the composite material a can be estimated from the non-destructive inspection information (preferably a vibration inspection image or an acoustic property image) of the composite material a. In the inventions described in Patent Document 2 (WO 2019 / 151393) and Patent Document 3 (WO 2019 / 151394), the neural network merely replaces what a human can judge by looking at an image or a measured object. In other words, in these inventions, since the object to be inspected is a photographed food, a human can easily judge the presence or absence of a foreign object in the food. On the other hand, the mechanical property information is a numerical value or a rank equivalent thereto, and the non-destructive inspection information (preferably a vibration inspection image or an acoustic property image) is a visualization or quantification of the internal state of the composite material. In other words, even if a skilled worker looks at the non-destructive inspection information, he or she cannot infer the mechanical property information from it. For example, no matter how hard a person tries, it is clear that he or she cannot infer mechanical property information from the vibration inspection images of Figures 8(a), (b), (c), and (d). By using the inspection device 1 of this embodiment, mechanical property information that a skilled worker cannot infer no matter how hard he or she tries can be instantly inferred without actual measurement. [Explanation of symbols]

[0075] 11: Image storage section 12: Processing section 13: Input data generation unit 14: Learning data storage unit 15: Learning Department 16: Model memory section 17: Guessing part 18: Display section 19:Operation section 301: Neural Networks 302: Input layer 303: Hidden layer 304: Output layer 311, 312, 313 units 801, 802, 803, 804 Waveform showing vibration damping

Claims

1. A mechanical property prediction model generated by machine learning based on mechanical property information and non-destructive testing information of a fiber-reinforced first composite material, the mechanical property information and non-destructive testing information of which are known, a processor that can access a model storage unit that stores a mechanical property estimation model that estimates mechanical property information of a fiber-reinforced second composite material by inputting non-destructive testing information of the second composite material, the mechanical property information of which is unknown; the processor acquires non-destructive testing information of the second composite material, inputs the non-destructive testing information into the mechanical property estimation model, acquires mechanical property information of the second composite material from the mechanical property estimation model, and performs output based on the mechanical property information; The non-destructive testing information is vibration characteristic information or acoustic characteristic information, and the sampling frequency when acquiring the non-destructive testing information of the second composite material is greater than 0 Hz.

2. 2. The inspection device according to claim 1, An inspection apparatus, wherein the non-destructive inspection information is image or numerical data.

3. 3. The inspection device according to claim 1, An inspection apparatus, wherein the mechanical property information is information regarding the elastic modulus or the breaking strength of a composite material.

4. 4. The inspection device according to claim 3, The information on the elastic modulus includes the elastic modulus or a rank when the elastic modulus is ranked, The information on the breaking strength includes the breaking strength or a rank when the breaking strength is classified into ranks.

5. 5. The inspection device according to claim 4, The information regarding the elastic modulus or breaking strength of an inspection device includes at least one of information indicating that the elastic modulus or breaking strength is difficult to predict, information indicating that the elastic modulus or breaking strength corresponds to a defective product, and information indicating that the elastic modulus or breaking strength corresponds to a good product.

6. The inspection device according to any one of claims 1 to 5, The inspection device, wherein the fiber is carbon fiber.

7. 7. The inspection device according to claim 1, An inspection device, wherein the first composite material and the second composite material are each a molded body using a thermosetting prepreg.

8. The inspection device according to any one of claims 1 to 7, The inspection device, wherein the first composite material and the second composite material are each a unidirectional material.

9. The inspection device according to any one of claims 1 to 8, An inspection device in which the mechanical property information or non-destructive inspection information of the first composite material includes at least one piece of information obtained by a finite element method.

10. 3. The inspection device according to claim 2, The image includes a waveform of a vibration decay.

11. The inspection device according to any one of claims 1 to 10, The non-destructive inspection information is vibration characteristic information.

12. 12. The inspection device according to claim 11, An inspection device in which the sampling frequency when acquiring the vibration characteristic information of the second composite material is 250 Hz or more and 50,000 Hz or less.

13. 13. The inspection device according to claim 11 or 12, An inspection device, wherein the natural frequency of the primary mode of the first composite material and the second composite material is greater than 0 Hz and less than or equal to 1000 Hz.

14. The inspection device according to any one of claims 1 to 10, The non-destructive inspection information is acoustic characteristic information.

15. 15. The inspection device according to claim 14, An inspection device, wherein the natural frequency of the primary mode of the first composite material and the second composite material is greater than 0 Hz and less than 20,000 Hz.

16. a step of generating a mechanical property estimation model for estimating the mechanical property information of a fiber-reinforced first composite material, the mechanical property information and the non-destructive testing information of which are known, by machine learning the mechanical property information and the non-destructive testing information of the second composite material, the mechanical property information of which is unknown, using the non-destructive testing information of a fiber-reinforced second composite material, the mechanical property information of which is unknown; obtaining non-destructive testing information of the second composite material; inputting the acquired non-destructive inspection information into the mechanical property estimation model, acquiring mechanical property information of the second composite material from the mechanical property estimation model, and performing output based on the mechanical property information; The non-destructive testing information is vibration characteristic information or acoustic characteristic information, and the sampling frequency for acquiring the non-destructive testing information of the second composite material is greater than 0 Hz.

17. a step of generating a mechanical property estimation model for estimating the mechanical property information of a fiber-reinforced first composite material, the mechanical property information and the non-destructive testing information of which are known, by machine learning the mechanical property information and the non-destructive testing information of the second composite material, the mechanical property information of which is unknown, using the non-destructive testing information of a fiber-reinforced second composite material, the mechanical property information of which is unknown; obtaining non-destructive testing information of the second composite material; inputting the acquired non-destructive inspection information into the mechanical property estimation model, acquiring mechanical property information of the second composite material from the mechanical property estimation model, and performing output based on the mechanical property information; A composite material inspection program in which the non-destructive inspection information is vibration characteristic information or acoustic characteristic information, and the sampling frequency when acquiring the non-destructive inspection information of the second composite material is greater than 0 Hz.

Images