Inspection methods and devices for storage media, molded body areas

Abstract

This invention provides an inspection device that can infer the mechanical properties of a molded body without measuring its mechanical properties, thus aiding in the evaluation of the molded body. The inspection device includes a storage unit for storing a mechanical property inference model. This model infers the mechanical property characteristics of a second molded body region by generating characteristics based on the mechanical property information of a first molded body region reinforced with reinforcing fibers and known non-destructive inspection information, and using the non-destructive inspection information of the non-destructive inspection information as input. The model then obtains the non-destructive inspection information of the second molded body region, inputs this non-destructive inspection information into the mechanical property inference model, obtains the mechanical property information of the second molded body region from the model, and outputs information based on this mechanical property information.

Description

Technical Field

[0001] The present invention relates to an inspection program, an inspection method, and an inspection device for a formed body region after fiber reinforcement. Background Art

[0002] A formed body after fiber reinforcement using carbon fiber can reinforce the vulnerability of a matrix resin with a high-strength fiber. Therefore, it is widely adopted as a material with excellent light weight and high characteristics.

[0003] Conventionally, in the production process of a fiber-reinforced composite material, a non-destructive inspection is performed to check for defective products during the manufacture of the composite material. For example, in Patent Document 1, in the process of impregnating carbon fiber with a thermoplastic resin, the following inspection is performed. First, a directional ultrasonic transmitter and a wave receiver are opposed to each other at a fixed distance from the object to be inspected (a composite material in which a thermoplastic resin is impregnated with carbon fiber). Then, ultrasonic waves are emitted from one ultrasonic transmitter, and the ultrasonic waves transmitted through the object to be inspected are received by the opposed wave receiver, and the propagation time of the ultrasonic waves is measured using a signal processing circuit, thereby detecting internal defects of the object to be inspected in a non-contact manner. The data of the inspection using ultrasonic waves here is converted into an image, and based on this image, it is possible to determine whether the object to be inspected is qualified or not.

[0004] In Patent Documents 2 and 3, a device for automatically performing high-precision retrieval in order to efficiently sort dark red fish meat, feathers, etc. in the production process of processed foods is disclosed.

[0005] Prior Art Documents

[0006] Patent Documents

[0007] Patent Document 1: Japanese Unexamined Patent Application Publication No. 2019-158459

[0008] Patent Document 2: International Publication No. 2019 / 151393

[0009] Patent Document 3: International Publication No. 2019 / 151394 Summary of the Invention

[0010] Technical Problem to be Solved by the Invention

[0011] In recent years, there have been situations such as soaring raw material prices and labor costs, and it has become a problem to suppress production costs while maintaining high quality. There is a demand for low-cost and high-precision inspection of a formed body region, which is a region obtained by forming a fiber-reinforced composite material using a mold (forming mold).

[0012] The non-destructive testing of composite materials described in Patent Document 1 relies on visual inspection of the obtained images. Therefore, it is difficult to obtain a detailed understanding of the composite material's condition. In particular, when determining the pass / fail status of images visually, it is difficult to establish objective evaluation criteria and to determine what criteria to use to calculate the material's pass / fail status.

[0013] Furthermore, the food inspection systems described in Patent Documents 2 and 3 are devices used by humans to locate hard bones, which differs from the technology used to inspect shaped objects. These food inspection systems simply allow neural networks to perform the judgments that humans could make simply by observing images and measuring the quality of objects.

[0014] The purpose of this invention is to provide an inspection procedure, inspection method, and inspection device for a molded body region, which can infer the mechanical properties of a molded body region obtained by molding a composite material using a mold without measuring its mechanical properties, thus aiding in the evaluation of the molded body region.

[0015] Technical means for solving problems

[0016] The above objectives can be achieved through the following methods.

[0017] An inspection procedure for a molded body region according to one aspect of the present invention causes a processor to perform the following steps: generating a mechanical property prediction model by performing machine learning on the mechanical property information and non-destructive inspection information of a fiber-reinforced first molded body region, wherein the mechanical property prediction model uses non-destructive inspection information of a fiber-reinforced second molded body region with unknown mechanical property information as input to predict the mechanical property information of the second molded body region; obtaining the non-destructive inspection information of the second molded body region; and inputting the non-destructive inspection information of the second molded body region into the mechanical property prediction model, obtaining the mechanical property information of the second molded body region from the mechanical property prediction model, and outputting based on the mechanical property information, wherein the first molded body region and the second molded body region are respectively obtained by forming a plate-shaped composite material with a molding die, wherein the projected area of ​​the composite material is set as S1, the projected area of ​​the portion in the molding die cavity corresponding to the first molded body region and the second molded body region is set as S2, and the value obtained by calculating (S1 / S2)×100% is used as the fill rate, wherein the first molded body region and the second molded body region are respectively formed by forming the composite material with a fill rate of 10% or more and 500% or less.

[0018] An inspection method for a molded body region according to one aspect of the present invention comprises: a step of generating a mechanical property prediction model by performing machine learning on the mechanical property information and non-destructive inspection information of a fiber-reinforced first molded body region, thereby generating a model that predicts the mechanical property information of a fiber-reinforced second molded body region by taking non-destructive inspection information of a second molded body region whose mechanical property information is unknown as input; a step of obtaining the non-destructive inspection information of the second molded body region; and a step of inputting the non-destructive inspection information of the second molded body region into the mechanical property prediction model, obtaining the mechanical property information of the second molded body region from the mechanical property prediction model, and outputting based on the mechanical property information, wherein the first molded body region and the second molded body region are respectively obtained by forming a plate-shaped composite material using a molding die, the projected area of ​​the composite material is set as S1, the projected area of ​​the portion of the molding die cavity corresponding to the first molded body region and the second molded body region is set as S2, the value obtained by calculating (S1 / S2)×100% is used as the filling rate, and the first molded body region and the second molded body region are respectively formed by forming the composite material with a filling rate of 10% or more and 500% or less.

[0019] An inspection apparatus for a molded body region according to one aspect of the present invention includes a processor capable of accessing a model storage unit. The model storage unit stores a mechanical property prediction model generated by machine learning based on mechanical property information and non-destructive inspection information of a fiber-reinforced first molded body region. The mechanical property prediction model uses non-destructive inspection information of a fiber-reinforced second molded body region (whose mechanical property information is unknown) as input to predict the mechanical property information of the second molded body region. The first and second molded body regions are respectively obtained by forming a plate-shaped composite material using a molding die. The projected area of ​​the composite material is defined as S1. The projected area of ​​the portion in the cavity of the forming mold corresponding to the first forming body region and the second forming body region is set as S2. The value obtained by calculating (S1 / S2)×100% is used as the filling rate. The first forming body region and the second forming body region are formed into the composite material with a filling rate of more than 10% and less than 500%. The processor obtains the non-destructive inspection information of the second forming body region, inputs the non-destructive inspection information into the mechanical property prediction model, obtains the mechanical property information of the second forming body region from the mechanical property prediction model, and outputs based on the mechanical property information.

[0020] Invention Effects

[0021] According to the present invention, the mechanical properties of a molded body region can be inferred solely from non-destructive inspection information without measuring mechanical properties, which is helpful for evaluating the molded body region. According to the present invention, mechanical property information that cannot be inferred by humans using non-destructive inspection information can be inferred instantaneously with high accuracy. Therefore, waste during the production of molded bodies containing molded body regions can be reduced, and high-quality molded bodies can be provided at low cost. Attached Figure Description

[0022] Figure 1 This is a diagram illustrating an example of the structure of an inspection system.

[0023] Figure 2 This is a flowchart for learning and processing.

[0024] Figure 3 This is a diagram representing an example of a neural network that outputs three response values.

[0025] Figure 4 It is a diagram representing the operations and processing between units of a neural network.

[0026] Figure 5 This is a flowchart of the speculative processing.

[0027] Figure 6 It is the distribution of the recognition surface and response values ​​when RBF is used in the activation function.

[0028] Figure 7 This refers to the distribution of the recognition surface and response values ​​when the Sigmoid function is used in the activation function.

[0029] Figure 8 This is a schematic diagram representing an image obtained through ultrasonic flaw detection.

[0030] Figure 9 This is a side view schematically representing an example of a forming die used for stamping composite material MX.

[0031] Figure 10 View from the movable mold side Figure 9 A top view of the fixed mold in the forming mold shown.

[0032] Figure 11 It means through Figure 9 The diagram shows an example of a method for forming a single composite material M using a forming die.

[0033] Figure 12 It is observed along direction D1. Figure 11 The diagram shows a fixed mold and a planar schematic of the composite material M.

[0034] Figure 13 It means through Figure 9 The diagram shows the state in which the forming mold shapes the composite material M.

[0035] Figure 14 This is a schematic diagram illustrating an example of pre-shaping a composite material M by heating it and then placing it in a fixed mold.

[0036] Figure 15 This is a schematic diagram illustrating an example of pre-shaped composite materials being arranged on a fixed mold in a non-overlapping manner before being formed.

[0037] Figure 16 It means from Figure 15 The diagram shows the state of the composite material being formed.

[0038] Figure 17 This is a schematic diagram illustrating the relationship between filler content and flow distance of composite materials.

[0039] Symbol Explanation

[0040] 11: Image Storage Unit

[0041] 12: Processing Department

[0042] 13: Input Data Generation Department

[0043] 14: Learning about data storage department

[0044] 15: Academic Department

[0045] 16: Model Storage Department

[0046] 17: Speculation Department

[0047] 18: Display Section

[0048] 19: Operations Department

[0049] 301: Neural Networks

[0050] 302: Input Layer

[0051] 303: Hidden Layer

[0052] 304: Output layer

[0053] 311, 312, 313: Units Detailed Implementation

[0054] Hereinafter, an inspection system including an inspection device as one embodiment of the present invention will be described, but the present invention is not limited thereto.

[0055] [Overview of the Inspection System]

[0056] The inspection system of this embodiment takes the area (second molded body area) obtained by molding a composite material of a specified shape (e.g., plate-like) after fiber reinforcement with unknown mechanical property information by a mold as the object to be inspected. The mechanical property information of the second molded body area is inferred instead of measured.

[0057] The mechanical property information of the molded body region is information representing the mechanical properties of the molded body region, such as information related to the strength or elasticity of the molded body region. Examples of mechanical property information include information related to breaking strength such as tensile strength or flexural strength, and information related to elastic modulus such as fracture strength, compressive strength, or shear strength.

[0058] Information related to the modulus of elasticity can be the modulus of elasticity (e.g., tensile modulus of elasticity or flexural modulus of elasticity) itself, or, in the case of classifying the modulus of elasticity (e.g., tensile modulus of elasticity or flexural modulus of elasticity), its grade (hereinafter referred to as mechanical property grade). Information related to the modulus of elasticity may further include any of the following: information indicating that the modulus of elasticity (or its grade) conforms to nonconforming products; information indicating that the modulus of elasticity (or its grade) conforms to conforming products; or information indicating that the modulus of elasticity (or its grade) is difficult to predict.

[0059] Information relating to breaking strength may be the breaking strength (e.g., tensile strength or yield strength) itself, or, in the case of classifying the breaking strength (e.g., tensile strength or yield strength), its grade (hereinafter referred to as mechanical property grade). Information regarding breaking strength may further include any of the following: information indicating that the breaking strength (or its grade) conforms to nonconforming products; information indicating that the breaking strength (or its grade) conforms to conforming products; or information indicating that the breaking strength (or its grade) is difficult to predict.

[0060] The inspection system includes a computer that acquires non-destructive inspection information of the second molded body region, inputs this non-destructive inspection information into a pre-generated mechanical property prediction model stored in a model storage unit, and uses this mechanical property prediction model to predict the mechanical property information of the second molded body region, and outputs the result based on the prediction. Examples of output methods include displaying the information (e.g., mechanical property grade, whether it is a qualified or unqualified product, information that is difficult to predict, etc.) on a display unit, broadcasting the information as a message through a speaker, or printing the information using a printer.

[0061] Non-destructive inspection information refers to information obtained by non-destructively inspecting the internal state of a shaped body region using methods such as radiation, infrared radiation, or ultrasound. Non-destructive inspection information can be images obtained using vibration or sound, or it can be numerical data. In the case of images, examples include radiation images, infrared images, or ultrasound images. Furthermore, in one of the embodiments described later, an ultrasound image will be used as an example, but the present invention is not limited to this.

[0062] The mechanical property prediction model is a model that generates mechanical property information by using machine learning (including supervised learning or unsupervised deep learning) on ​​the fiber-reinforced molded body region (first molded body region), which has known mechanical property information and non-destructive inspection information, as input and outputs mechanical property information. Mechanical property prediction models may use, for example, neural networks or support vector machines.

[0063] The inspection system comprises a computer-based inspection device. This computer includes a processor, a storage unit consisting of a hard disk drive or an SSD (Solid State Drive) capable of storing information, RAM (Random Access Memory), and ROM (Read Only Memory). The processor executes an inspection program stored in the ROM, thereby performing processing such as acquiring non-destructive inspection information of the shaped area of ​​the inspected object, inputting the acquired non-destructive inspection information into a mechanical property prediction model, acquiring mechanical property information from the mechanical property prediction model, and outputting information based on the acquired mechanical property information.

[0064] Non-destructive inspection information of the molded body area is typically used to determine the presence of defects, voids, or foreign objects within that area, and, if present, their degree. However, even with a large number of defects, voids, or foreign objects present within the molded body area, there are cases where the mechanical properties are good, depending on their distribution. In such cases, visually confirming the non-destructive inspection information might lead to a rejection due to the abundance of defects, voids, or foreign objects, resulting in the discarding of the molded body area that should have been acceptable and a decrease in production efficiency. Conversely, the opposite situation may also exist. That is, even if visually confirming the non-destructive inspection information results in a rejection due to a small number of defects, voids, or foreign objects, the mechanical properties may sometimes be equivalent to those of an unacceptable product.

[0065] Based on the above-mentioned viewpoints, the inventors verified that non-destructive inspection information and mechanical property information are correlated. By using models such as neural networks or support vector machines to perform machine learning on multiple measured data of non-destructive inspection information and mechanical property information, the inventors successfully inferred the mechanical property information of the formed body region with high accuracy based on the non-destructive inspection information. Previously, it was not considered to derive mechanical property information from non-destructive inspection information. Therefore, constructing a machine learning model that uses non-destructive inspection information as input to output mechanical property information is not easy for those skilled in the art.

[0066] The following provides a detailed example of the inspection system. Additionally, the following explains an example where the mechanical characteristic prediction model is a neural network.

[0067] [Reinforced Fibers]

[0068] The type of reinforcing fiber used in this invention can be appropriately selected according to the intended use of the inspected object, i.e., shaped body region a (corresponding to a second shaped body region with unknown mechanical properties), and is not particularly limited. Either inorganic or organic fibers can be suitably used as the reinforcing fiber.

[0069] Examples of such inorganic fibers include carbon fiber, activated carbon fiber, graphite fiber, glass fiber, tungsten carbide fiber, silicon carbide fiber, ceramic fiber, alumina fiber, natural mineral fiber (basalt fiber, etc.), boron fiber, boron nitride fiber, boron carbide fiber, and metal fiber.

[0070] [Carbon fiber]

[0071] When using carbon fiber as the fiber, the carbon fibers generally known include polyacrylonitrile (PAN) carbon fibers, petroleum / coal tar carbon fibers, rayon carbon fibers, cellulose carbon fibers, lignin carbon fibers, phenolic carbon fibers, and vapor-grown carbon fibers. In this invention, any one of these carbon fibers can be preferred.

[0072] [Reinforcing the morphology of fibers]

[0073] In this invention, the morphology of the reinforcing fiber is not particularly limited. Hereinafter, continuous fibers, as specific examples, will be described. However, this invention is not limited to continuous fibers.

[0074] Continuous fiber refers to reinforcing fiber that is pulled together in a continuous state without cutting the reinforcing fiber into short fibers. Continuous reinforcing fibers are preferred for obtaining the molded body region a with excellent mechanical properties. More specifically, continuous fibers are preferably fibers with a length of 1 m or more, so that they can be used as impregnation materials for resins processed into fabrics, woven fabrics, etc., by manual coating or other methods, or as prepregs made by impregnating uncured resin with continuous fibers.

[0075] [Formed body region a]

[0076] The molded body region a is a region reinforced by reinforcing fibers. Hereinafter, an example of an embodiment of the present invention will be described, but the present invention is not limited to the molded body region a described below.

[0077] 1. Molded body

[0078] The molded body region a is the molded body of the plate-shaped composite material after molding. It can be a molded body using thermoplastic resin or a molded body using thermosetting prepreg.

[0079] Prepreg is a material used to make molded articles. It is a material in which continuous carbon fibers are arranged in a sheet in one direction, a material in which a thermosetting resin is impregnated on a substrate made of carbon fibers such as carbon fiber fabric, or a molding intermediate material in which a portion of a thermosetting resin is impregnated and the remainder is disposed on at least one surface.

[0080] 2. Unidirectional materials

[0081] The molded body region a is preferably made of a unidirectional material. A unidirectional material refers to a material in which continuous reinforcing fibers, with a length of 100 mm or more, are aligned in one direction within the molded body region a. Alternatively, a material composed of multiple layers of continuous reinforcing fibers can also be used. In particular, when the molded body region a is made of a unidirectional material and a thermosetting prepreg is used, the fiber orientation has less influence on the mechanical properties. Therefore, the accuracy of the inference of mechanical property information based on the model described later can be improved.

[0082] [Preferred Molded Body Region]

[0083] The above-mentioned molded body region contains reinforcing fibers and matrix resin as necessary components, and other components as optional components. The porosity Vr of the molded body region, calculated by the following formulas (A) and (B), is preferably 10% or less.

[0084] Vr=(t2-t1) / t2×100%···Formula (A)

[0085] t1=(Wf / Df+Wm / Dm+Wz / Dz)÷unit area (mm²) 2 Formula (B)

[0086] t1: Theoretical thickness of the formed area (mm)

[0087] t2: Measured thickness of the formed area (mm)

[0088] Df: Density of reinforcing fiber (mg / mm²) 3 )

[0089] Dm: Density of the matrix resin (mg / mm³) 3 )

[0090] Dz: Density of other components (mg / mm³) 3 )

[0091] Wf: Mass percentage of reinforcing fibers (%)

[0092] Wm: Mass percentage of the matrix resin

[0093] Wz: Mass percentage of other ingredients (%)

[0094] The porosity (Vr) is more preferably 5% or less, and even more preferably 3% or less. If the porosity is within this range, the accuracy of the mechanical property prediction of the present invention is improved.

[0095] [Manufacturing of the formed body region a]

[0096] For example, the shaped body region a can be prepared as follows.

[0097] 1. Materials

[0098] • Reinforcing fiber: Carbon fiber "Tenex (registered trademark)" STS 40-24K (tensile strength 4300MPa, tensile modulus 240GPa, filament count 24000, fineness 1600tex, elongation 1.8%, density 1.78g / cm³) 3 , Teijin Co., Ltd.)

[0099] • Base resin: A thermosetting resin composition with epoxy resin as the main component.

[0100] 2. Preparation of unidirectional prepreg

[0101] The unidirectional prepreg is produced by the following hot-melt method. First, the above-mentioned thermosetting resin composition is coated onto release paper using a coating machine to create a resin film. Next, the above-mentioned carbon fiber bundles are fed from the bobbin holder and passed through a comb to ensure uniform spacing between the carbon fiber bundles. Then, they are widened by a fiber-opening rod to achieve a fiber weight of 100 g / m². 2The carbon fibers are arranged in a sheet-like manner in one direction. Then, the resin film is overlapped from both sides of the carbon fibers, heated and pressurized to impregnate it with a thermosetting resin composition, and wound using a paper winding machine to produce a unidirectional prepreg. The resulting unidirectional prepreg has a resin content of 30 wt.%.

[0102] 3. Creation of the formed body region a

[0103] Eleven sheets of unidirectional prepreg are manually stacked in the 0° direction to obtain a laminated structure

[011] . T The prepreg laminate was placed inside a bag film, positioned in a molding die, and heated in an autoclave at 130°C for 120 minutes to cure, producing a 1 mm thick CFRP molded body (unidirectional carbon fiber reinforced thermosetting resin molded body, i.e., molded body region a). The fill rate shown below during autoclave molding was 100%.

[0104] [Determination of tensile modulus of elasticity and tensile strength]

[0105] As a specific example of the breaking strength or elastic modulus of the present invention, the inventors measured the tensile elastic modulus and tensile strength of the molded body region a as follows.

[0106] The CFRP molded body was processed into a test piece shape (250 mm in length × 15 mm in width) using water jet, and a label made of glass fiber reinforced resin matrix composite was bonded to it. A tensile test was performed in the 0° direction using a universal testing machine at a test speed of 2 mm / min according to ASTM D3039, and the tensile modulus of elasticity and tensile strength of the CFRP molded body (molded body region a) were calculated.

[0107] [Non-destructive inspection information]

[0108] The non-destructive testing (NDT) method used in generating NDT information is not particularly limited, as long as it is an inspection method that detects internal defects, voids, or foreign objects in the shaped body region a without damaging it. NDT methods include using radiation, ultrasound, or infrared light. NDT information preferably uses information that transforms inspection data into an image, and particularly preferably, the transformed image is an ultrasonic flaw detection image. Alternatively, images obtained using vibration or sound inspection can also be used. There are no particular limitations on the method of image conversion of the inspection data. In many cases, when purchasing an ultrasonic inspection device, image conversion software is also provided along with a computer terminal for data processing.

[0109] [Ultrasonic flaw detection image]

[0110] Generally, a representative method for non-destructive testing of materials is the ultrasonic testing method. A directional ultrasonic transmitter and receiver are positioned opposite each other at a fixed distance on either side of the shaped body region a to be inspected. Pulse-modulated ultrasonic waves are emitted from the transmitter and received by the receiver. The propagation time (echo intensity) of the ultrasonic waves is measured by a signal processing circuit. This propagation time varies depending on whether the shaped body region a contains voids or foreign objects. Therefore, the internal state of the shaped body region a can be inspected non-contactly using ultrasonic waves.

[0111] Next, a specific example of a method for acquiring ultrasonic flaw detection images as non-destructive inspection information will be described. Here, an ultrasonic flaw detection device (SDS-3600: manufactured by Claude Clayman, Japan) was used to measure the CFRP molded body as the molded body region a, and an ultrasonic flaw detection image of the CFRP molded body was obtained, which evaluated the internal defects of the CFRP molded body.

[0112] More specifically, after configuring the ultrasonic probes, ultrasonic inspection is performed on the shaped body area a, which is the object to be inspected, using ultrasonic waves at a frequency of 600 kHz. The two probes are configured such that the transmitting probe is 30 mm away from the upper surface of the object to be inspected, the receiving probe is 30 mm away from the lower surface of the object to be inspected, and the shafts of the receiving probe and the transmitting probe are aligned vertically.

[0113] In the imaging of the electrical signal converted from the ultrasonic waves received by the receiving probe, a C-scan is used to display the position (2D) of the object under inspection as a Cartesian coordinate by modulating the intensity of the received echo at a certain depth in the ultrasonic probe. In the resulting C-scan image, differences in the propagation behavior of the ultrasonic waves are represented by variations in color and intensity.

[0114] The specific C-scan image obtained is shown below. Figure 8 .in addition, Figure 8 The C-scan images shown can change color according to the propagation behavior of ultrasound and echo intensity, but the images used in the learning described below and the images used in the inference of mechanical property information of the inspected object are all acquired under the same conditions.

[0115] [Inspection System]

[0116] Hereinafter, data transformed into a form that can be input to the input layer of a neural network will be described as input data. In the inspection system, non-destructive inspection information (hereinafter referred to as non-destructive inspection information sample) of a sample of molded body region a (hereinafter referred to as molded body region sample b, equivalent to the first molded body region) and mechanical characteristic information actually measured from molded body region sample b (hereinafter referred to as mechanical characteristic information sample) are acquired and used as second input data. This second input data is then used for neural network learning. After the neural network has completed its learning, the non-destructive inspection information of molded body region a is input into the neural network as the first input data. Based on the response values ​​from the output layer of the neural network, the mechanical characteristic information of molded body region a is inferred. Alternatively, based on the inferred mechanical characteristic information, qualified and unqualified products of molded body region a can be classified.

[0117] To enable effective learning and high-precision inference, the inspection system can utilize optimized non-destructive inspection information (preferably ultrasonic flaw detection images) in both learning and inference processing. For example, various image processing techniques can be applied to ultrasonic flaw detection images to facilitate their detection.

[0118] [Inspection device]

[0119] The inspection device 1 performs image processing, input data generation, neural network learning, and inference using mechanical characteristic information from the neural network. The inspection device 1 is an information processing device such as a computer, equipped with one or more processors (including a CPU), storage, and communication units, and operated by an OS (operating system) and application programs. The inspection device 1 can be a physical computer or implemented using a virtual machine (VM), a container, or a combination thereof. More specifically, the processor's structure is a circuit composed of semiconductor elements and other circuit components.

[0120] The inspection apparatus 1 includes: an image storage unit 11 that stores non-destructive inspection information and non-destructive inspection information samples; a processing unit 12 that processes the non-destructive inspection information and non-destructive 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 functional blocks implemented by a program executed by the processor of the inspection apparatus 1. This program includes an inspection program for the shaped body area.

[0121] The image storage unit 11 is preferably a storage area for storing ultrasonic flaw detection images. The image storage unit 11 can be a volatile memory such as SRAM or DRAM, or a non-volatile memory such as NAND, MRAM, or FRAM (registered trademark).

[0122] The processing unit 12 preferably performs image processing on the ultrasonic flaw detection image and saves the processed image in the image storage unit 11. Examples of image processing include generating an image by extracting the brightness of each of the red, green, and blue (RGB) colors from the pixels in the image, generating an image by subtracting the brightness of the green (G) color from the brightness of the red (R) color in each pixel, and generating an image by extracting only the red component after transforming to the HSV color space. However, other types of image processing can also be performed.

[0123] The processing unit 12 can also perform image enlargement, reduction, cropping, noise removal, rotation, inversion, color depth change, contrast adjustment, brightness adjustment, sharpness adjustment, color correction, etc.

[0124] The input data generation unit 13 generates input data for the input layer of the neural network based on the non-destructive inspection information or non-destructive inspection information samples stored in the image storage unit 11. For example, when using ultrasonic flaw detection images for learning (described later), it is preferable to cut out the desired area from the ultrasonic flaw detection image or remove redundant areas to create the second input data.

[0125] When the inspection device 1 performs learning processing, the input data generation unit 13 saves the input data in the learning data storage unit 14. When the inspection device 1 inspects the shaped body region a, the input data is transmitted to the estimation unit 17.

[0126] Furthermore, during learning processing, the input data generation unit 13 may use images (non-destructive inspection information samples) captured by external devices or systems to generate input data.

[0127] The learning data storage unit 14 is a storage area that stores multiple input data 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. For the input data used as learning data (second input data), a corresponding mechanical property information sample obtained by measuring the molded body region sample b, which is the source of the input data, is established and stored. In addition to including at least one of the mechanical property level and mechanical property value (e.g., elastic modulus) of the molded body region sample b, the mechanical property information sample also includes information indicating that the mechanical property value is equivalent to a qualified product, information indicating that the mechanical property value is equivalent to a non-qualified product, and information indicating that the mechanical property value is difficult to predict.

[0128] For example, the correspondence between the mechanical property information sample and the second input data obtained from the molded body region sample b (hereinafter referred to as "the correspondence") can be directly input by the user through operation of the operation unit 19, allowing the mechanical property value (e.g., elastic modulus) of the molded body region sample b to be entered directly. After input, the inspection device 1 classifies the mechanical property value into mechanical property grades. For example, the tensile elastic modulus can be classified into the following mechanical property grades.

[0129] Mechanical property grade 1: The tensile modulus of elasticity in the formed part region is above 30 GPa.

[0130] Mechanical property grade 2: The tensile modulus of elasticity in the formed part region is 25-30 GPa.

[0131] Mechanical property grade 3: The tensile modulus of elasticity in the formed part region is below 25 GPa.

[0132] The mechanical property level can also be output in units of 3 GPa or 1 GPa, instead of every 5 GPa as described above. Furthermore, if the mechanical property information sample corresponding to the second input data obtained from the molded body region sample b is known, the mechanical property level can be automatically labeled using programs, scripts, etc., without user intervention. The labeling of the mechanical property level can be performed either before or after the transformation of the non-destructive inspection information sample obtained from the molded body region sample b into the second input data.

[0133] 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 then stores the learned neural network in the model storage unit 16. For example, the learning unit 15 can learn a three-layer neural network consisting of an input layer, a hidden layer, and an output layer. By learning this three-layer neural network, real-time response performance when inspecting the shaped object region a can be ensured. There is no particular limitation on the number of units included in each of the input, hidden, and output layers. The number of units included in each layer can be determined based on the required response performance, the object to be inferred, the recognition performance, etc.

[0134] Furthermore, while a 3-layer neural network is one example, it doesn't preclude the use of multi-layer neural networks with more layers. When using multi-layer neural networks, various other neural networks, such as convolutional neural networks, can be employed.

[0135] Model storage unit 16 is a storage area that stores the neural network learned by learning unit 15. Multiple neural networks can be stored in model storage unit 16 depending on the type of the shaped body region a to be inspected. Model storage unit 16 is configured to be referenced by estimation unit 17, so estimation unit 17 can use the neural networks stored in model storage unit 16 to inspect the shaped body region a (estimate mechanical characteristic information). Model storage unit 16 can be volatile memory such as RAM or DRAM, or non-volatile memory such as NAND, MRAM, or FRAM (registered trademark). Furthermore, model storage unit 16 can be located anywhere accessible to the processor of inspection device 1, and may not be built into inspection device 1. For example, model storage unit 16 can be an external memory connected to inspection device 1, or a network memory connected to a network accessible from inspection device 1.

[0136] The estimation unit 17 uses the neural network stored in the model storage unit 16 to estimate the mechanical property information of the molded body region a. The estimation unit 17 estimates the mechanical property level of the molded body region a based on the response values ​​output from the units in the output layer. Examples of units in the output layer include units with mechanical property level 1, units with mechanical property level 2, units with mechanical property level 3, and units that are difficult to estimate, but other types of units can also be prepared. For example, if the estimated mechanical property level is low, it is possible that many foreign objects are mixed in. The difference or ratio of the response values ​​of multiple units can also be used to estimate the mechanical property level of the molded body region a.

[0137] Display unit 18 is a display for showing images and text. It can also display captured images, processed images, and prediction results based on prediction unit 17.

[0138] The operation unit 19 is a device that provides an operating unit for the user to operate the inspection device 1. The operation unit 19 may be, for example, a keyboard, mouse, button, switch, voice recognition device, etc., but is not limited to these.

[0139] [Learning Processing]

[0140] Before inferring the mechanical property level of the molded body region a based on the inspection device 1, it is necessary to use non-destructive inspection information samples and mechanical property information samples of the same type as molded body region a to learn the neural network. Figure 2 This is a flowchart for learning and processing.

[0141] First, the processor of the inspection device 1 acquires non-destructive information samples for each of the multiple shaped body region samples b (step S201). These non-destructive inspection information samples include samples with high mechanical property levels and samples with low mechanical property levels. When a unit is provided that outputs a difficult-to-predict response value to the output layer of the neural network, non-destructive inspection information samples with difficult-to-predict mechanical property information can also be prepared. Examples of difficult-to-predict non-destructive inspection information samples include images where the shaped body region sample b is not fully displayed, or images where brightness adjustment is inappropriate due to lighting or exposure, resulting in unclear display of the shaped body region sample b.

[0142] The processor of inspection device 1 generates second input data from the acquired non-destructive inspection information samples (step S202). Next, the processor of inspection device 1 acquires mechanical characteristic information samples of each of the plurality of shaped body region samples b, and stores the acquired mechanical characteristic information samples in correspondence with each second input data (step S203).

[0143] Alternatively, step S203 can be performed before step S202. In this case, after each non-destructive inspection information sample is transformed into second input data, the mechanical characteristic information sample corresponding to the non-destructive inspection information sample is also integrated with the second input data.

[0144] Next, the processor of the inspection device 1 begins neural network-based learning based on the second input data (step S204).

[0145] Figure 3 This is an example of a neural network that outputs three response values. Figure 3 The neural network 301 is a three-layer neural network with an input layer 302, a hidden layer 303, and an output layer 304. The output layer 304 includes units 311, 312, and 313 for inferring the level of mechanical characteristics. Units 311, 312, and 313... Figure 3 There are 3 in total, but the number can be increased or decreased appropriately according to the level of mechanical properties.

[0146] In a neural network, the value input to the input layer propagates to the hidden layer and the output layer, resulting in the response value of the output layer. In step S204, when the second input data is input into the neural network, the parameters and structure of the neural network, such as the number of hidden layers 303, the number of units contained in each of the input layer 302 and the hidden layer 303, and the association coefficients between the units contained in each of the input layer 302 and the hidden layer 303, are adjusted to enable the neural network to output mechanical characteristic information samples corresponding to or closely related to the second input data with a high probability. This generates a mechanical characteristic prediction model and stores it in the model storage unit 16.

[0147] Figure 4 This represents the operations and processing between units in a neural network. Figure 4 The cells of layer m-1 and layer m are shown. For illustration, in... Figure 4 Only a portion of the units in the neural network are shown. The unit numbers in the (m-1)th layer are k = 1, 2, 3... The unit numbers in the mth layer are j = 1, 2, 3...

[0148] If the reaction value of the unit number k in the (m-1)th layer is set as a k m-1 Then the reaction value a of unit number j in the m-th layer j m The following formula (2) is used to find it.

[0149] [Number 1]

[0150]

[0151] Here, W jk m This is the weight, representing the strength of the coupling between units. j m is the bias. f(······) is the activation function. According to equation (2), the response value of any unit in the m-th layer is the output value when the response values ​​of all units (k=1, 2, 3...) in the (m-1)-th layer are weighted and summed, and used as the input variable of the activation function.

[0152] Next, an example of an activation function will be given. Equation (3) below is the normal distribution function.

[0153] [Number 2]

[0154]

[0155] Here, μ is the mean, representing the center of the bell-shaped peak described by the normal distribution function. σ is the standard deviation, representing the width of the peak. The value of equation (3) depends only on the distance from the center of the peak, therefore it can be said that the Gaussian function (normal distribution function) is a type of radial basis function (RBF). The Gaussian function (normal distribution function) is one example; other RBFs can also be used.

[0156] Equation (4) below is the Sigmoid function. The Sigmoid function gradually approaches 1.0 as x approaches infinity. In addition, it gradually approaches 0.0 as x approaches -∞. That is, the Sigmoid function takes values ​​in the range (0.0, 1.0).

[0157] [Number 3]

[0158]

[0159] Furthermore, the use of activation functions other than Gaussian and Sigmoid functions is not prohibited. For example, the inventors used ReLU in the convolutional layer and softmax in the output layer.

[0160] In neural network learning, after input data is fed into the input layer, the strength of the coupling between units, i.e., the weights W, is determined. jk The adjustment is made to obtain the correct output. When input data labeled with a certain mechanical characteristic level is input into a neural network, the expected correct output (the response value of the unit in the output layer) is also called the teacher signal.

[0161] For example, if input data labeled with a mechanical characteristic level of 311 is input into neural network 301, then in the teacher signal, the response value of unit 311 is 1, the response value of unit 312 is 0, and the response value of unit 313 is 0. If input data labeled with a mechanical characteristic level of 312 is input into neural network 301, then in the teacher signal, the response value of unit 311 is 0, the response value of unit 312 is 1, and the response value of unit 313 is 0.

[0162] For example, weight W jk The adjustments can be performed using the back propagation method (error back propagation method). In the back propagation method, the weights W are adjusted sequentially from the output layer side. jk This is to reduce the deviation between the output of the neural network 310 and the teacher signal. Equation (5) below represents the improved backpropagation method.

[0163] [Number 4]

[0164] W jk (t+1)=W jk (t)+ΔW jk (t)

[0165] ΔW jk (t)=-ηδ k O j +αΔW jk (t-1)+βΔW jk (t-2) (5)

[0166] Furthermore, when using a Gaussian function as the activation function, it's not just the weight W... jkFurthermore, σ and μ in equation (3) are used as parameters in the improved backpropagation method and adjusted accordingly. By adjusting the values ​​of parameters σ and μ, the learning convergence of the neural network is assisted. Equation (6) below represents the adjustment of the value of parameter σ.

[0167] [Number 5]

[0168] σ jk (t+1)=σ jk (t)+Δσ jk (t)

[0169] Δσ jk (t)=-ηδ k O j +αΔσ jk (t-1)+βΔσ jk (t-2) (6)

[0170] Equation (7) below represents the adjustment of the value of parameter μ.

[0171] [Number 6]

[0172] μ jk (t+1)=μ jk (t)+Δμ jk (t)

[0173] Δμ jk (t)=-ηδ k O j +αΔμ jk (t-1)+βΔμ jk (t-2) (7)

[0174] Here, t is the number of learning iterations, η is the learning constant, and δ is the learning frequency. k It is the generalized error, O j ΔW is the response value for unit number j, α is the insensitivity constant, and β is the vibrational constant. jk , Δσ jk , Δμ jk Indicates weight W jk The correction values ​​for σ and μ respectively.

[0175] Here, we take the modified backpropagation method as an example to analyze the weight W. jk The parameter adjustment process has been explained, but the general backpropagation method can also be used instead. The following descriptions, focusing solely on the backpropagation method, include both the modified backpropagation method and the general backpropagation method.

[0176] Weight W based on backpropagation method jkThe number of parameter adjustments can be once or multiple times, without particular limitation. Typically, the decision to perform weighted W based on the backpropagation method is made based on the inferred accuracy of the mechanical characteristic level using the test data. jk Repeated adjustment of parameters. If the weight W is repeatedly adjusted... jk Adjusting the parameters can sometimes improve the accuracy of predicting the mechanical property level.

[0177] By using the method described above, the weight W can be determined in step S205. jk The values ​​of parameters σ and μ. To determine the weight W jk The values ​​of parameters σ and μ enable inference processing using neural networks.

[0178] Figure 5 This is a flowchart illustrating the inference action of the inspection device 1, which operates according to the inspection procedure for the formed body region, regarding mechanical characteristic information. The processor of the inspection device 1 acquires non-destructive inspection information of the formed body region a (preferably by capturing an ultrasonic flaw detection image) (step S501). When the ultrasonic flaw detection image is used as non-destructive inspection information, there may also be a step of image processing of the ultrasonic flaw detection image between step S501 and step S502.

[0179] Next, the processor of the inspection device 1 generates first input data based on the non-destructive inspection information (step S502). The first input data has N elements equal to the number of units in the input layer of the neural network, and is in a form that can be input into the neural network.

[0180] Next, the processor of inspection device 1 inputs the first input data into the neural network (step S503). The first input data is passed in the order of input layer, hidden layer, and output layer. The processor of inspection device 1 makes a prediction of the mechanical characteristic level based on the response value in the output layer of the neural network (step S504).

[0181] Inference processing using neural networks is equivalent to finding the location of the first input data within the recognition space. Figure 6 An example of the recognition space is shown when a Gaussian function is used as the activation function. If an RBF such as a Gaussian function is used as the activation function, the recognition surface that divides the recognition space into regions for each level of mechanical characteristic becomes a closed surface. Furthermore, for each category of mechanical characteristic level, by adding an index in the height direction, it is possible to localize the regions involved in each category within the recognition space.

[0182] Figure 7This illustrates an example of the recognition space when the sigmoid function is used as the activation function. With the sigmoid function as the activation function, the recognition surface becomes an open surface. Furthermore, the learning process of the neural network described above is equivalent to learning the recognition surface within the recognition space. Figure 6 , Figure 7 Only mechanical property grades 311 and 312 are shown in the region, but there may be a distribution of more than three mechanical property grades.

[0183] As described above, if the inspection system of this embodiment is used, the mechanical properties of the shaped body region a can be inferred from the non-destructive inspection information (preferably an ultrasonic flaw detection image) of the shaped body region a. In the inventions described in Patent Document 2 (International Publication No. 2019 / 151393) or Patent Document 3 (International Publication No. 2019 / 151394), the only difference is that a neural network replaces the judgment that a human can make simply by seeing an image and measuring the object. That is, in these inventions, the object of inspection is a photograph of food, so a person can easily determine whether there are foreign objects or the like in the food.

[0184] On the other hand, mechanical property information is numerical or based on a rating, while non-destructive inspection information (preferably ultrasonic testing images) is information that visualizes or quantifies the internal state of a formed body region. That is, even if a skilled worker observes non-destructive inspection information, they cannot infer mechanical property information from it. For example, it is obvious that no matter how hard a human tries, they cannot infer mechanical property information from... Figure 8 Mechanical characteristic information can be inferred from the ultrasonic flaw detection images (a) to (d). Using the inspection device 1 of this embodiment, mechanical characteristic information that a skilled worker could not predict could be instantly inferred without actual measurement.

[0185] Next, an example of the mechanical properties of the molded body region obtained by molding a sheet-like composite material using a molding die will be described. Hereinafter, the sheet-like composite material before molding using a molding die will be collectively referred to as composite material MX. In the following description, composite material MX can be exemplified as composite material M, composite material Ms, composite material M2, and composite material M3. Composite material MX preferably includes discontinuous fibers. Composite material MX is particularly preferably a sheet molding compound impregnated with a thermosetting resin as the matrix resin in a chopped fiber bundle pad.

[0186] [Forming method for composite material MX]

[0187] There are no particular limitations on the forming method of composite material MX. Stamping (compression forming), autoclave forming, vacuum forming, etc. can be used, but stamping is preferred.

[0188] [Molding Die]

[0189] Figure 9 This is a side view schematically illustrating an example of a forming die used in the stamping of composite material MX. Figure 10 Viewed from the 30th side of the movable mold Figure 9 A top view of the fixed mold 20 in the forming mold shown. Figure 9 The molding die shown includes a fixed die 20 and a movable die 30 that is freely movable relative to the fixed die 20. A recess 22 is formed on the upper surface 21 of the movable die 30 side of the fixed die 20. The recess 22 is a region defined by a bottom surface 22A and a pair of side surfaces 22B connecting the bottom surface 22A and the upper surface 21. The movable die 30 is configured to move freely in a direction D including a direction D1 approaching the fixed die 20 and a direction D2 away from the fixed die 20. Direction D1 is the direction in which pressure is applied to the composite material MX when molding the composite material MX. In addition, if the molding method is vacuum forming, direction D1 becomes the attraction direction.

[0190] When the movable mold 30 is closest to the fixed mold 20 (the movable mold 30 is located at the position indicated by the single-dot dashed line in the figure), a forming mold cavity (space SP shown by the diagonal line in the figure) is formed between the movable mold 30 and the fixed mold 20.

[0191] A molding die cavity refers to the space in which the molded body is formed. In molded bodies obtained by forming composite material MX using a molding die, sometimes unwanted parts such as the ends are trimmed. In this case, Figure 9 The space within the space SP shown, which forms the shaped body that is trimmed and ultimately remains, is called the molding die cavity. For example, suppose that in the process of passing through... Figure 9 The molding die shown is used to shape the composite material M to obtain the product. Figure 13 The molded body MD1 shown is a case where the portion indicated by the dashed line is trimmed to produce the final product. In this case, the portion of space SP other than the area indicated by the dashed line becomes the molding die cavity.

[0192] Figure 11 It means to utilize Figure 9 The diagram illustrates an example of a method for forming a single composite material M using a molding die. The plate-shaped composite material M is positioned on the upper surface 21 of the fixed die 20, covering the recess 22. Figure 12 Observing from direction D1 Figure 11 The diagram shows a top view of the fixed mold 20 and the composite material M. If the movable mold 30 is moved from... Figure 11 If the state shown moves in direction D1, the composite material M flows due to the pressure from the movable mold 30 and deforms along the shape of the space SP, as... Figure 13As shown, the molded body MD1 is obtained. In Figures 11-13 In the example, the entire molded body region obtained by molding the composite material M constitutes the molded body MD1.

[0193] In the case where the forming method is cold stamping, the composite material M is pre-shaped by heating before stamping. For example... Figure 14 As shown, the pre-shaped composite material M is positioned on the upper surface and recess of the fixed mold 20. Then, if the movable mold 30 is moved from... Figure 14 If the state shown moves in direction D1, the pre-shaped composite material M flows due to the pressure from the movable mold 30 and deforms along the shape of the space SP, as... Figure 13 As shown, the shaped body MD1 is obtained.

[0194] When a molded body is manufactured using a molding die, the composite material MX that becomes the raw material for the molded body can be one or more. For example, such as Figure 11 and Figure 12 As shown by the dashed line, sometimes other plate-shaped composite materials Ms are further configured on the composite material M before molding. Additionally, as... Figure 15 As shown, there are also cases where multiple pre-shaped plate-like composite materials (composite material M2 and composite material M3) are arranged on the fixed mold 20 in a non-overlapping manner before forming. When the movable mold 30 is moved from... Figure 15 As the state shown moves in direction D1, the pre-shaped composite materials M2 and M3 flow due to pressure from the movable mold 30, deforming along the shape of space SP. Then, as... Figure 16 As shown, a molded body MD2 is obtained by forming a molded body region MA2 obtained from molding composite material M2 and a molded body region MA3 obtained from molding composite material M3. Figure 15 and Figure 16 In the example, the molded body region MA2 obtained by molding composite material M2 and the molded body region MA3 obtained by molding composite material M3 each constitute a part of the molded body MD2.

[0195] [Projected area S1 of the composite material MX constituting the molded body]

[0196] When the molded body is composed of a single composite material MX, the planar area of ​​the single composite material MX (the composite material in its pre-forming state in the case of cold stamping) when viewed in the thickness direction is defined as the projected area S1. Figure 11 In the example, the planar area of ​​composite material M is called the projected area S1.

[0197] Assume the molded body is composed of multiple composite material MX, and these multiple composite material MX are formed in an overlapping state. In this case, the molded body can be regarded as being composed of a single molded body region. Moreover, the planar area of ​​the multiple composite material MX in their overlapping state (the composite material before pre-forming in the case of cold stamping) when viewed in their respective thickness directions is defined as the projected area S1. Figure 12 In the example, the entire composite material Ms overlaps with composite material M. Therefore, the projected area S1 is the same as the planar area of ​​composite material M. Figure 12 For example, consider the case where only a portion of composite material Ms is configured to overlap with composite material M. In this case, the sum of the planar area of ​​composite material M and the planar area of ​​the region of composite material Ms that does not overlap with composite material M becomes the projected area S1.

[0198] Assume the formed body is composed of multiple composite material MXs arranged in a non-overlapping manner before forming. In this case, the planar area of ​​each composite material MX (in the case of cold stamping, the composite material before pre-forming) when viewed in its respective thickness direction is defined as the projected area S1. Figure 15 In the example, the planar area of ​​composite material M2 before pre-forming, when viewed in the thickness direction, is called the projected area S1. Similarly, the planar area of ​​composite material M3 before pre-forming, when viewed in the thickness direction, is called the projected area S1.

[0199] [The projected area S2 of the portion of the mold cavity corresponding to the area of ​​the molded body]

[0200] Consider a molded body formed by a molding die cavity, consisting of a single molded body region obtained by molding a single or overlapping composite material MX. In this case, the molding die cavity as a whole becomes the portion corresponding to this single molded body region. The planar area of ​​this portion when viewed from direction D1 is defined as the projected area S2. Figure 13 In the example, the planar area of ​​space SP when viewed from direction D1 is called the projected area S2.

[0201] Assume that the molded body formed by the molding die cavity consists of multiple molded body regions obtained by molding multiple composite materials MX arranged in a non-overlapping state. In this case, the portion of the molding die cavity containing each molded body region is the portion corresponding to each molded body region. The planar area of ​​each portion when viewed from direction D1 is defined as the projected area S2.

[0202] exist Figure 16In the example, the planar area of ​​the portion of space SP containing the shaped body region MA2 when viewed from direction D1 is called the projected area S2. Similarly, the planar area of ​​the portion of space SP containing the shaped body region MA3 when viewed from direction D1 is also called the projected area S2.

[0203] Furthermore, when multiple composite material MXs are separately arranged and cold-stamped to form a molded body, each composite material MX flows during forming and welds at the boundary. Therefore, by observing the completed molded body, it is easy to identify the "molded body area obtained by forming each composite material MX".

[0204] [Fill Rate]

[0205] For a molded body manufactured by a molding die, the value obtained by the following formula (C) is defined as the fill rate.

[0206] Fill rate [%] = 100 × {(projected area S1 of the composite material MX constituting the molded body) / (projected area S2 of the part of the mold cavity corresponding to the area of ​​the molded body)} ·····(C)

[0207] In addition, such as Figure 16 As shown in the example, when the molded body includes multiple molded body regions, two fill rates are obtained for one molded body by substituting the projected area S1 and the projected area S2 of each molded body region into equation (C).

[0208] That is, the first filling rate is calculated by setting the planar area of ​​composite material M2 as the projected area S1 and the planar area of ​​the portion of the molding die cavity containing the molded body region MA2 as the projected area S2. Similarly, the second filling rate is calculated by setting the planar area of ​​composite material M3 as the projected area S1 and the planar area of ​​the portion of the molding die cavity containing the molded body region MA3 as the projected area S2.

[0209] A higher fill rate value means a shorter flow distance for the composite material MX during molding. For example, to use... Figure 9 The forming mold shown only Figure 16 The molded area MA2 shown is an example of product manufacturing. In this case, as... Figure 17 As shown, if the projected area S1 of composite material M2 is made larger than that of composite material M2, then... Figure 15 The example shown is small, and the filling rate is reduced, so the flow distance of composite material M2 is greater than that of composite material M2. Figure 15 The example shown is long.

[0210] When a composite material MX includes discontinuous fibers, and the molded body region obtained by molding the composite material MX is reinforced by the discontinuous fibers, if the composite material MX is flowed to generate the molded body region, the fibers will orient themselves during the flow. Therefore, when the flow distance of the composite material MX is short, the control of flow and fiber orientation during molding becomes easier, and the mechanical properties of the molded body region remain stable during mass production.

[0211] By generating the aforementioned mechanical property prediction model using learning data obtained under conditions where the mechanical property quality of the molded body region is as free from deviation as possible, the accuracy of mechanical property prediction can be improved. Specifically, the first molded body region, which serves as the measurement source of the learning data (non-destructive inspection information and mechanical property information) for learning the mechanical property prediction model, and the second molded body region, which serves as the object of mechanical property prediction using the mechanical property prediction model, are preferably materials obtained by molding composite material MX with a filling rate of 10% or more and 500% or less, respectively.

[0212] As the structures of the first and second formed body regions respectively, for example, can be adopted Figure 13 The molded body MD1 shown Figure 16 The shaped body region MA2 shown Figure 16 The molded area MA3 shown is Figure 16 The molded body MD2 is shown. Furthermore, the first molded body region and the second molded body region are respectively... Figure 16 In the case of the molded body MD2 shown, either the fill rate calculated for composite material M2 and molded body region MA2 or the fill rate calculated for composite material M3 and molded body region MA3 is 10% or more and 500% or less.

[0213] If the filler content is less than 10%, the flow distance of the composite material MX increases. Therefore, it becomes difficult to stabilize the mechanical properties of both the first and second molded body regions. Thus, a lower limit of the filler content is set at 10%. The upper limit of the filler content can be determined based on the intended use of the product manufactured using the molding die. Setting this upper limit to 500% allows for the selection of products suitable for a wide range of applications. Furthermore, to further stabilize the mechanical properties of both the first and second molded body regions, a filler content of 50% or more is preferred, more preferably 70% or more, and even more preferably 80% or more. It should be noted that when the reinforcing fiber is discontinuous and it is desirable to increase mechanical strength in a specific direction, a filler content of 10% or more but less than 50% can easily lead to anisotropy in the molded body region, which is therefore preferable.

[0214] [Isotropy of the formed body region]

[0215] When the flow of the composite material MX during molding is minimal, the fiber orientation of the composite material MX and its molded body region is approximately similar. In particular, when the composite material MX is an isotropic substrate reinforced with discontinuous fibers, isotropy can be ensured in the molded body region if the flow during molding is minimal (in other words, if the filler content is high). Therefore, the isotropy of the first molded body region and the second molded body region is preferably 1.5 or less, more preferably 1.3 or less.

[0216] [Coefficient of variation (CV) of weight per unit area of ​​the formed body region]

[0217] Preferably, the coefficient of variation (CV) of the weight per unit area for both the first and second molded body regions is 10% or less. If the coefficient of variation (CV) of the weight per unit area is 10% or less, the characteristics of both the first and second molded body regions become more uniform. Therefore, it becomes easier to predict the mechanical properties of the mechanical property prediction model.

[0218] [Reinforcing fibers contained in the molded area]

[0219] When the fiber length of the reinforcing fiber bundle is set to Li, the diameter of the monofilament of the reinforcing fiber constituting the reinforcing fiber bundle is set to Di, and the number of monofilaments contained in the reinforcing fiber bundle is set to Ni, it is preferable that the first forming body region and the second forming body region respectively include areas where Li is 1 mm or more and 100 mm or less, and Li / (Ni×Di) 2 ) is 8.0×10 1 Above and 3.3×10 3 The following is a reinforcing fiber bundle A.

[0220] The volume percentage of the reinforcing fiber bundle A relative to the total reinforcing fibers contained in the molded body is preferably 50 to 100 vol%, more preferably 70 to 90 vol%.

[0221] Various embodiments have been described above with reference to the accompanying drawings, but it is self-evident that the present invention is not limited to such examples. Those skilled in the art will conceive of various modifications and variations within the scope of the claims, and it should be understood that they naturally fall within the technical scope of the present invention. Furthermore, the structural elements in the above embodiments can be arbitrarily combined without departing from the spirit of the invention.

[0222] Furthermore, this application is based on Japanese patent application filed on July 8, 2020 (Japanese Patent Application No. 2020-118107), the contents of which are incorporated herein by reference.

Claims

1. A storage medium, characterized in that, The procedure for inspecting the storage area of ​​the shaped object. The inspection procedure for the shaped body region causes the processor to perform the following steps: The step of generating a mechanical property prediction model by performing machine learning on the mechanical property information and non-destructive inspection information of the fiber-reinforced first molded body region, wherein the mechanical property prediction model uses the non-destructive inspection information of the fiber-reinforced second molded body region, whose mechanical property information is unknown, as input to predict the mechanical property information of the second molded body region. The steps for obtaining non-destructive inspection information of the second shaped body region; as well as The steps include inputting non-destructive inspection information of the second formed body region into the mechanical property prediction model, obtaining mechanical property information of the second formed body region from the mechanical property prediction model, and outputting based on the mechanical property information. The first molded body region and the second molded body region are respectively obtained by forming plate-shaped composite materials using molding dies. The first molded body region and the second molded body region are respectively reinforced by discontinuous fibers. Let the projected area of ​​the composite material be S1, and let the projected area of ​​the portion in the molding cavity of the molding die corresponding to each of the first molded body region and the second molded body region be S2. Let the value obtained by calculating (S1 / S2)×100% be the filling rate. The first molded body region and the second molded body region are respectively molded into the composite material with a fill rate of 10% or more and 500% or less.

2. The storage medium according to claim 1, characterized in that, The first molded body region and the second molded body region are respectively molded into the composite material with a fill rate of 50% or more.

3. The storage medium according to claim 1 or 2, characterized in that, The isotropy of the first molded body region and the second molded body region is 1.5 or less.

4. The storage medium according to claim 1 or 2, characterized in that, The variation coefficient of the unit area weight of the first molded body region and the second molded body region is less than 10%.

5. The storage medium according to claim 1 or 2, characterized in that, In a case where a fiber length of the reinforcing fiber bundle is set to Li, a filament diameter of the reinforcing fiber constituting the reinforcing fiber bundle is set to Di, and a fiber number of the filament contained in the reinforcing fiber bundle is set to Ni, the reinforcing fiber of each of the first molded body region and the second molded body region includes the reinforcing fiber bundle in which Li is 1 mm or more and 100 mm or less, and Li / (Ni x Di) is 8.0 x 10 2 or more and 3.3 x 10 1 or less. 3 ​ 6. The storage medium according to claim 1 or 2, characterized in that, The first molded body region and the second molded body region are respectively obtained by stamping plate-shaped composite materials using a molding die.

7. The storage medium according to claim 1 or 2, characterized in that, The composite material is a sheet molding compound made by impregnating a thermosetting resin, which serves as the matrix resin, into a chopped fiber bundle pad.

8. The storage medium according to claim 1 or 2, characterized in that, The non-destructive inspection information is image or numerical data.

9. The storage medium according to claim 1 or 2, characterized in that, The mechanical property information is related to the elastic modulus or breaking strength of the formed body region.

10. The storage medium according to claim 9, characterized in that, Information related to the elastic modulus includes the elastic modulus itself, or, if the elastic modulus is graded, its level. Information related to the destructive strength includes the destructive strength itself, or, if the destructive strength has been graded, its level.

11. The storage medium according to claim 10, characterized in that, Information related to the elastic modulus or breaking strength includes at least one of the following: information indicating that the elastic modulus or breaking strength is difficult to predict, information indicating that the elastic modulus or breaking strength is equivalent to a defective product, and information indicating that the elastic modulus or breaking strength is equivalent to a qualified product.

12. A method for inspecting a region of a shaped body, characterized in that, have: The step of generating a mechanical property prediction model by performing machine learning on the mechanical property information and non-destructive inspection information of the fiber-reinforced first molded body region, wherein the mechanical property prediction model uses the non-destructive inspection information of the fiber-reinforced second molded body region, whose mechanical property information is unknown, as input to predict the mechanical property information of the second molded body region. The steps for obtaining non-destructive inspection information of the second shaped body region; as well as The steps include inputting non-destructive inspection information of the second formed body region into the mechanical property prediction model, obtaining mechanical property information of the second formed body region from the mechanical property prediction model, and outputting based on the mechanical property information. The first molded body region and the second molded body region are respectively obtained by forming plate-shaped composite materials using molding dies. The first molded body region and the second molded body region are respectively reinforced by discontinuous fibers. Let the projected area of ​​the composite material be S1, and let the projected area of ​​the portion in the molding cavity of the molding die corresponding to each of the first molded body region and the second molded body region be S2. Let the value obtained by calculating (S1 / S2)×100% be the filling rate. The first molded body region and the second molded body region are respectively molded into the composite material with a fill rate of 10% or more and 500% or less.

13. An inspection device for a shaped body region, characterized in that, The device includes a processor capable of accessing a model storage unit that stores mechanical property prediction models generated through machine learning based on mechanical property information and non-destructive inspection information of the fiber-reinforced first molded body region. The mechanical property prediction model uses non-destructive inspection information of the fiber-reinforced second molded body region, whose mechanical property information is unknown, as input to predict the mechanical property information of the second molded body region. The first molded body region and the second molded body region are respectively obtained by forming plate-shaped composite materials using molding dies. The first molded body region and the second molded body region are respectively reinforced by discontinuous fibers. Let the projected area of ​​the composite material be S1, and let the projected area of ​​the portion in the molding cavity of the molding die corresponding to each of the first molded body region and the second molded body region be S2. Let the value obtained by calculating (S1 / S2)×100% be the filling rate. The first molded body region and the second molded body region are respectively formed by molding the composite material with a fill rate of 10% or more and 500% or less. The processor acquires non-destructive inspection information of the second formed body region, inputs the non-destructive inspection information into the mechanical property prediction model, obtains mechanical property information of the second formed body region from the mechanical property prediction model, and outputs based on the mechanical property information.

Images