To provide a manufacturing method and an inspection method of a fiber-reinforced product.

Abstract

To provide a method for easily manufacturing a fiber-reinforced product by which a defect can be satisfactorily detected even when the fiber-reinforced product has a complicated appearance design in which reinforcing fibers are dispersed in an in-plane direction.SOLUTION: A method of manufacturing a fiber reinforced product comprising discontinuous reinforcing fibers dispersed in an in-plane direction and a resin, the method comprising: Step 101. Applying light by an illumination means and imaging the surface of the fibre reinforced product by an imaging means. Step 201: A step of storing the surface image picked up by the image pickup means in the storage means. Step 501. Detecting the surface imperfection, if any, by the learning-type surface imperfection detection means. The learning-type surface defect detection means is means for detecting a surface defect on the surface of the fiber-reinforced product by artificial intelligence on the basis of learning of surface image data of the fiber-reinforced product in a state where the surface defect is absent and surface image data of the fiber-reinforced product in a state where the surface defect is present.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method for manufacturing a fiber-reinforced product that includes a step of inspecting surface defects of the fiber-reinforced product as part of the manufacturing process, and to a method for inspecting surface defects of a fiber-reinforced product. [Background technology]

[0002] In recent years, fiber-reinforced products using reinforcing fibers have been widely adopted for sports and leisure applications such as fishing rods and golf shafts, as well as for aircraft, automobiles, and industrial equipment. Since even the slightest defect in a fiber-reinforced product can be the starting point for failure or reduce the strength of the product, a high level of precision is required for determining whether the product meets these inspection standards. While these inspections were generally performed by human eyes, automation technologies have been increasingly introduced with the aim of reducing labor, improving accuracy, and preventing defects from being overlooked. Various technologies have been proposed to shorten inspection time and improve inspection accuracy.

[0003] Patent Document 1 provides a non-destructive inspection method for detecting defects with high accuracy by measuring the time change in the surface temperature distribution of a C / C composite material during thermal transients.The heat rays of a flash lamp are irradiated, an image is input using an infrared thermograph device, and the time change in the surface temperature distribution is measured to detect defects.

[0004] Patent Document 2 provides a method for inspecting prepreg defects, shortening the inspection time and improving the inspection accuracy when manufacturing prepreg using release paper. In particular, it provides a prepreg defect inspection system that identifies the location and type of defects on the side where the release paper is attached and transmits this information to the subsequent fabric inspection process.

[0005] Patent Document 3 relates to a defect inspection method for the surface of prepreg impregnated with carbon fibers and matrix resin. In particular, it provides an optical means for inspecting the surface of prepreg aligned in one direction with high accuracy and reliability. The rectangularity ratio, maximum diameter angle, and unevenness of the defect candidate area are calculated, crack defects are detected, and then a threshold is calculated from an image in which the crack defect detection area is masked, and fluff defects are detected. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Publication No. 10-096705 [Patent Document 2] Japanese Patent Application Laid-Open No. 2012-184929 [Patent Document 3] Japanese Patent Application Laid-Open No. 2010-85166 Summary of the Invention [Problem to be solved by the invention]

[0007] However, the invention described in Patent Document 1 uses two heat sources for measurement, and an infrared thermograph captures an image of the thermal radiation energy from the surface of the carbon fiber-reinforced carbon composite. In this case, (1) a certain amount of space is required around the product under inspection, (2) if the resin contained in the fiber-reinforced product is a thermoplastic resin, there is a risk that the resin in the fiber-reinforced product will melt when heated, and (3) furthermore, when using a heat source, it takes time to raise the temperature of the fiber-reinforced product, increasing the labor required for inspection.

[0008] The invention described in Patent Document 2 detects defects by comparing with a preset threshold value, so it can only be used when the appearance of the inspected product is uniform and unchanging (the appearance design is consistent), such as prepregs made with continuous fibers. On the other hand, in the case of fiber-reinforced products with dispersed discontinuous fibers, the surface design of the inspected product is too complex to set a predetermined threshold value. For fiber-reinforced products with dispersed discontinuous fibers, the surface design of each product is different.

[0009] In Patent Document 3, defects are detected in three dimensions, and in order to determine the degree of unevenness, a dedicated camera is required, a relatively long imaging distance must be secured, and a certain amount of space is required. Furthermore, three-dimensional inspection often takes a long time, and there are cases where the takt time of the process cannot be achieved.

[0010] The present invention provides a method for easily manufacturing a fiber-reinforced product, which does not require a relatively large design space and can effectively detect defects even in fiber-reinforced products with complex exterior designs in which reinforcing fibers are dispersed in the in-plane direction. [Means for solving the problem]

[0011] As a result of extensive investigations, the present inventors have found that the above problems can be solved by the following means, and have arrived at the present invention. 1. A method for producing a fiber-reinforced product comprising in-plane dispersed discontinuous reinforcing fibers and a resin, comprising the steps of: Step 101: A step of irradiating light by an illumination means and imaging the surface of the fiber-reinforced product by an imaging means. Step 201: A step of storing the surface image captured by the imaging means in storage means. Step 501: If there is a surface defect, the surface defect is detected by a learning surface defect detection means. However, a learning-type surface defect detection means is a means in which artificial intelligence detects surface defects on the surface of a fiber-reinforced product based on learning surface image data of the fiber-reinforced product when it is free of the surface defect and surface image data of the fiber-reinforced product when it has the surface defect. 2. A method for producing a fiber-reinforced product as described in 1 above, comprising the following steps 301 and 401 between steps 201 and 501: Step 301: A step of reading out the surface image of the fiber reinforced product stored in the storage means, processing the image, and outputting pixel values. Step 401: A step of extracting points A and B, which have a difference in pixel value of 128 or more and are adjacent to each other. 3. The method for producing a fiber-reinforced product according to claim 2, wherein in step 101, the surface of the fiber-reinforced product is imaged using an imaging device located at an imaging distance of 100 mm or more and 800 mm or less from the fiber-reinforced product. 4. The method for manufacturing a fiber-reinforced product according to any one of 2 and 3, wherein the imaging device and the lighting means are each one. 5. The method for producing a fiber-reinforced product according to any one of 2 to 4 above, wherein the image processing in step 301 replaces the function of fixational eye movement with an electronic circuit. 6. The method for producing a fiber-reinforced product according to 5 above, wherein the fixational eye movement is tremor. 7. The method for manufacturing a fiber-reinforced product described in any one of 2 to 6 above, wherein the surface image is a grayscale image. 8. The method for producing a fiber-reinforced product according to any one of 2 to 6, wherein the surface image is a red-green-blue (RGB) image. 9. A method for producing a fiber-reinforced product according to any one of 2 to 8 above, wherein in step 401, the pixel value of point A is 200 or more and the pixel value of point B is 50 or less. 10. The method for manufacturing a fiber-reinforced product described in any one of 2 to 9, wherein the imaging device and the lighting means are each fixed. 11. The light reaching the imaging device contains more diffusely reflected light than specularly reflected light. 11. A method for producing a fiber-reinforced product according to any one of 2 to 10 above. 12. A method for inspecting a fiber-reinforced product comprising discontinuous reinforcing fibers dispersed in the in-plane direction and a resin, using the following steps: Step 101: A step of irradiating light by an illumination means and imaging the surface of the fiber-reinforced product by an imaging means. Step 201: A step of storing the surface image captured by the imaging means in storage means. Step 301: A step of reading out the surface image of the fiber reinforced product stored in the storage means, processing the image, and outputting pixel values. Step 401: A step of extracting points A and B, which have a difference in pixel value of 128 or more and are adjacent to each other. Step 501: If there is a surface defect, the surface defect is detected by a learning surface defect detection means. However, a learning-type surface defect detection means is a means in which artificial intelligence detects surface defects on the surface of a fiber-reinforced product based on learning surface image data of the fiber-reinforced product when it is free of the surface defect and surface image data of the fiber-reinforced product when it has the surface defect. 13. A method for producing a fiber-reinforced product as described in 1 above, comprising the following step 302 between step 201 and step 501: Step 302: A step of reading out the surface image of the fiber reinforced product stored in the storage means, detecting the black pattern, and outputting the area of ​​the black pattern 14. A method for producing a fiber-reinforced product as described in 1 above, comprising the following step 303 between step 201 and step 501: Step 303: A step of reading out the surface image of the fiber-reinforced product stored in the storage means, detecting the black line pattern, and outputting the area of ​​the black line and the degree of bending. 15. The fiber reinforced product is a flat plate that is continuously manufactured in the MD direction, A method for producing a fiber-reinforced product as described in item 1, comprising the following step 304 between step 201 and step 501. Step 304: A step of reading out the surface image of the fiber reinforced product stored in the storage means, detecting the black line pattern, and outputting the angle between the MD direction of the fiber reinforced product and the direction of the black line pattern. 16. A method for producing a fiber-reinforced product according to any one of claims 13 to 15, wherein the light reaching the imaging device is specularly reflected light. 17. A method for manufacturing a fiber-reinforced product according to any one of 1 to 11 or 13 to 16, wherein a plurality of fiber-reinforced products are continuously manufactured, and the fiber patterns observed in each fiber-reinforced product are different from each other. 18. A method for producing a fiber-reinforced product described in any one of 1 to 11 or 13 to 17, wherein the thickness of the thinnest part of the resin between the surface of the fiber-reinforced product and the discontinuous reinforcing fibers present inside the fiber-reinforced product is less than 100 μm. [Effects of the Invention]

[0012] According to the manufacturing method and inspection method of the present invention, artificial intelligence can detect defects on the surface of a fiber-reinforced product based on learning of surface image data, so the defect detection process can be significantly reduced.

[0013] Another advantage of the present invention is that the imaging distance is short, so a large design space is not required. Another advantage of the present invention is that when a two-dimensional image is acquired using one imaging device and one light source, it is possible to detect unevenness (surface defects) even with an imaging device that can only obtain pixel value information. Furthermore, using 2D images for image processing shortens image processing time, minimizing the impact on takt time in the manufacturing and inspection processes. [Brief explanation of the drawings]

[0014] [Figure 1] Schematic diagram of fiber-reinforced plastic in which the reinforcing fibers are discontinuous and randomly arranged in two dimensions. [Figure 2] 10 is a schematic diagram showing how light is applied by an illumination means and the surface of the fiber-reinforced product is imaged by an imaging means from an imaging device. FIG. [Figure 3] (a) In step 301, the surface image of the fiber reinforced product stored in the storage means is read out (before image processing). (b) In step 301, the surface image of the fiber reinforced product is read out and image processed. [Figure 4] (a) The surface image of the fiber-reinforced product was read and processed. (b) Only area A with a pixel value of 200 or more was extracted from (a). (c) The pixels around area A were extracted, and the area adjacent to area B with a pixel value of 50 or less was extracted. [Figure 5] (a) By shifting the position of the fiber-reinforced product 2 to 3 mm toward the imaging means, the light from the lighting means is diffusely reflected by the fiber-reinforced product and input to the imaging means. (b) The light from the lighting means is reflected by the fiber-reinforced product and input directly to the imaging means. [Figure 6]The surface image of the fiber-reinforced fiber product was read and processed using Technos' 5000 series-S (application name: Analyzer). [Figure 7] The black pattern was detected and the area of ​​the black pattern was output to detect the defect. A defect caused by sticking to the belt was observed when a fiber-reinforced product was made using a double belt press. [Figure 8] The black line pattern is detected, and the area and degree of curvature of the black line are output to detect defects. Miscutting of the reinforcing fibers has increased the fiber length, causing the reinforcing fibers to become twisted. [Figure 9] The black line pattern is detected, and the MD direction and angle of the fiber reinforced product are output to detect defects. These are scratches. DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments of the present invention will be described, but the present invention is not limited to these.

[0016] [Reinforced fiber] In this specification, the reinforcing fiber is preferably at least one selected from the group consisting of carbon fiber, aramid fiber, glass fiber, polyester fiber, nylon fiber, polypropylene fiber, and polyethylene fiber, and more preferably carbon fiber or glass fiber.

[0017] [Reinforced fiber: carbon fiber] 1. Carbon fiber in general When carbon fibers are used, 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, etc. are generally known, and any of these carbon fibers can be suitably used in the present invention. Among these, polyacrylonitrile (PAN)-based carbon fibers are preferred in the present invention because of their excellent tensile strength. As a PAN-based carbon fiber, for example, TENAX (registered trademark) STS40-24KS (average fiber diameter 7 μm) carbon fiber manufactured by Teijin Limited can be used.

[0018] 2. Carbon fiber sizing agent The carbon fiber used in the present invention may have a sizing agent attached to its surface. When using carbon fiber with a sizing agent attached, the type of sizing agent can be appropriately selected depending on the type of carbon fiber and the type of resin used in the X material or Y material, and is not particularly limited.

[0019] [Reinforced fiber: glass fiber] The case where the reinforcing fiber used in the present invention is glass fiber will be described. 1. Glass fiber in general The glass fiber used in the present invention may be any glass fiber generally referred to as glass fiber. The glass composition is not particularly limited, and may include A-glass, C-glass, E-glass, etc., and may contain components such as TiO2, SO3, and P2O5 in some cases. For example, Nitto Boseki's E-glass RS240QR-483 (count: 2400 g / 1000 m) glass fiber can be used as the glass fiber.

[0020] 2.Glass fiber sizing agent The glass fiber used in the present invention may have a sizing agent attached to its surface. When using glass fiber with a sizing agent attached, the type of sizing agent can be appropriately selected depending on the type of glass fiber and the type of resin, and is not particularly limited. Glass fiber that has been pre-treated with a conventionally known coupling agent such as an organosilane compound, an organotitanium compound, an organoborane compound, or an epoxy compound can be preferably used.

[0021] [Fiber reinforced products] In the present invention, the fiber-reinforced product may be a composite material or a molded body. A composite material is a material for producing a molded body, and the composite material is preferably press-molded (also called compression molding) to form a molded body. When the fiber-reinforced product is a composite material, a flat plate shape is preferable. On the other hand, when the fiber-reinforced product is a molded body, the molded body is shaped and has a three-dimensional shape. When a thermoplastic matrix resin is used for cold pressing, the shape of the reinforcing fibers is largely maintained before and after molding, so by analyzing the shape of the reinforcing fibers contained in the molded body, the shape of the reinforcing fibers in the composite material can be determined.

[0022] [Reinforced fiber: fiber length] The reinforcing fibers contained in the fiber-reinforced product of the present invention are discontinuous fibers. When discontinuous fibers are used, formability is improved compared to fiber-reinforced products using only continuous fibers, making it easier to create complex molded articles. The fiber-reinforced product of the present invention preferably contains discontinuous reinforcing fibers having a weight-average fiber length Lw of 1 mm or more and 100 mm or less. When discontinuous reinforcing fibers are used, formability is improved compared to fiber-reinforced plastics using only continuous fibers, making it easier to create complex molded articles. Furthermore, by using discontinuous reinforcing fibers, no matter what direction stress is applied to the molded article from, it is unlikely that a direction will result in extremely weak mechanical properties.

[0023] The weight-average fiber length Lw of the reinforcing fibers is preferably 3 mm or more and 80 mm or less, more preferably 5 mm or more and 60 mm or less, and even more preferably 10 mm or more and 40 mm or less. If the weight-average fiber length Lw of the reinforcing fibers is 100 mm or less, the fluidity of the composite material is less likely to decrease when the composite material is produced by press molding, and it is easy to produce the composite material in the desired shape. Furthermore, if the weight-average fiber length Lw is 1 mm or more, the mechanical strength of the resulting molded product is less likely to decrease, which is preferable. In a molded product produced by injection molding, the weight-average fiber length of the reinforcing fibers is about 0.1 to 0.3 mm. Therefore, when the weight-average fiber length of the reinforcing fibers is set to 1 mm or more and 100 mm or less, it is preferable to produce the molded product by press molding.

[0024] In the present invention, discontinuous reinforcing fibers having different fiber lengths may be used in combination. In other words, the discontinuous reinforcing fibers used in the present invention may have a single peak in the weight-average fiber length distribution, or may have multiple peaks.

[0025] [Reinforcing fiber: Measurement method for number average fiber length Ln and weight average fiber length Lw] Generally, if the fiber length of each reinforcing fiber is Li, the number average fiber length Ln and weight average fiber length Lw can be calculated by the following formulas (1) and (2). The units of the number average fiber length Ln and weight average fiber length Lw are mm. Ln=ΣLi / I equation (1) Lw=(ΣLi 2 ) / (ΣLi)...Equation (2) Here, "I" indicates the number of reinforcing fibers measured.

[0026] When the fiber length is constant, the number-average fiber length and the weight-average fiber length are the same value. Reinforcing fibers can be extracted from molded products, for example, by heating them at 500°C for about 1 hour and removing the resin in a furnace.

[0027] The average fiber length can be calculated, for example, by measuring the fiber lengths of 100 fibers randomly extracted from a fiber-reinforced product to the nearest 1 mm using a vernier caliper or the like, and then calculating the average fiber length based on formula (1).

[0028] If the dispersion contains short fibers that cannot be measured with a caliper, remove the resin, then place the resulting reinforcing fibers in water containing a surfactant and thoroughly stir using ultrasonic vibrations. Samples for evaluation can be obtained by randomly sampling the stirred dispersion with a measuring spoon, and measuring the lengths of 3,000 fibers using a Nireco Luzex AP image analyzer. The measured fiber lengths can be used to calculate the number-average fiber length Ln and weight-average fiber length Lw using the same formulas (1) and (2) described above.

[0029] [Reinforced fiber: fiber volume ratio] There is no particular limitation on the fiber volume fraction (Vf) of the reinforcing fibers contained in the fiber-reinforced product, but it is preferably 20 to 70%, more preferably 25 to 60%, and even more preferably 30 to 55%. The fiber volume fraction (Vf, unit: volume %) is the ratio of the volume of the reinforcing fibers to the total volume including not only the reinforcing fibers and resin but also other additives. There is no limitation on the analysis of the reinforcing fiber volume fraction, but it is recommended to measure it as follows.

[0030] A sample is cut from the fiber-reinforced product, and the resin is burned off in a furnace at 500°C for 1 hour. The mass of the sample is then weighed before and after treatment to calculate the mass of the reinforcing fiber, resin, and other additives. Next, the volume ratio of the reinforcing fiber to the resin is calculated using the specific gravity of each component. Vf = 100 x reinforcing fiber volume / (reinforcing fiber volume + resin volume + other additive volume)

[0031] [Reinforced fiber: fiber bundle] When the reinforcing fibers are discontinuous fibers and used in the form of opened fiber bundles, the reinforcing fibers contained in the fiber-reinforced plastic may be solely in the form of monofilaments, solely in the form of fiber bundles, or a mixture of both. When the reinforcing fibers used in the present invention are in the form of fiber bundles, the number of single fibers (also referred to as monofilaments) constituting each fiber bundle is not particularly limited, but is typically within the range of 1,000 to 100,000. When carbon fibers are used as reinforcing fibers, they are generally in the form of fiber bundles consisting of several thousand to several tens of thousands of single fibers. If carbon fibers are used as they are, the entangled portions of the fiber bundles may become locally thick, making it difficult to obtain a thin-walled impact absorbing member. For this reason, when carbon fibers are used as reinforcing fibers, the fiber bundles are usually widened or opened before use.

[0032] [Reinforced fibers: Dispersed in the in-plane direction] The reinforcing fibers contained in the fiber-reinforced product of the present invention are discontinuous fibers and dispersed in the in-plane direction. When the fiber-reinforced product is a molded article, it is preferable to disperse the reinforcing fibers contained in the composite material in the in-plane direction in order to achieve dispersion in the in-plane direction in the molded article. Dispersion of the reinforcing fibers in the in-plane direction means that the reinforcing fibers are dispersed so that their fiber axes are oriented in the in-plane direction. It is preferable that the angle between the fiber axes of the reinforcing fibers and the in-plane direction is 45° or less. 1. In-plane direction The in-plane direction is an arbitrary direction of a parallel plane perpendicular to the thickness direction of the fiber-reinforced product (molded body or composite material). The composite material is preferably a plate-shaped material.

[0033] 2. Random distribution in two dimensions It is preferable that the reinforcing fibers contained in a fiber-reinforced product are randomly dispersed in two dimensions within the plane. In regions where the composite material is press-molded without flow, the shape of the reinforcing fibers is largely maintained before and after molding. Therefore, it is preferable that the reinforcing fibers contained in the non-flowing regions of a molded product formed from a composite material are also randomly dispersed in two dimensions within the plane.

[0034] Here, "randomly dispersed in two dimensions within the plane" refers to a state in which the reinforcing fibers are not oriented in a specific direction within the plane of the fiber-reinforced product (composite material or molded article), but are oriented randomly, and are arranged within the sheet plane without any specific directionality overall. A fiber-reinforced product obtained using discontinuous fibers that are randomly dispersed in two dimensions is a substantially isotropic fiber-reinforced product (composite material or molded article) that does not have anisotropy within the plane.

[0035] The degree of two-dimensional random orientation is evaluated by determining the ratio of the tensile modulus in two mutually perpendicular directions. If the ratio (Eδ) obtained by dividing the larger of the tensile modulus values ​​measured in any direction of a fiber-reinforced product (composite material or molded product) by the smaller of the measured values ​​is 5 or less, more preferably 2 or less, and even more preferably 1.5 or less, the reinforcing fibers can be evaluated as being two-dimensionally randomly dispersed.

[0036] When the resin contained in the molded product is a thermoplastic resin and the molded product has a three-dimensional shape including curved surfaces, the two-dimensional random dispersion in the in-plane direction can be evaluated by heating the molded product above its softening temperature, returning it to a flat plate shape, and then solidifying it. After that, test specimens can be cut out and the tensile modulus measured to confirm the state of random dispersion in the two-dimensional direction.

[0037] [resin] The resin contained in the fiber reinforced product may be thermosetting or thermoplastic. 1.Thermoplastic resin When the resin used is a thermoplastic resin, the type is not particularly limited, and a resin having a desired softening point or melting point can be appropriately selected and used. As the thermoplastic resin, one having a softening point in the range of 180°C to 350°C is usually used, but is not limited thereto.

[0038] Examples of thermoplastic resins include polyolefin resins, polystyrene resins, polyamide resins, polyester resins, polyacetal resins (polyoxymethylene resins), polycarbonate resins, (meth)acrylic resins, polyarylate resins, polyphenylene ether resins, polyimide resins, polyethernitrile resins, phenoxy resins, polyphenylene sulfide resins, polysulfone resins, polyketone resins, polyether ketone resins, thermoplastic urethane resins, fluorine-based resins, and thermoplastic polybenzimidazole resins.

[0039] The molded article of the present invention may use one type of thermoplastic resin or two or more types of thermoplastic resins. Examples of the use of two or more types of thermoplastic resins in combination include, but are not limited to, the use of thermoplastic resins having different softening points or melting points or the use of thermoplastic resins having different average molecular weights. When a thermoplastic resin is used, it is more preferable to use a polyolefin resin, and even more preferable to use a polypropylene resin.

[0040] 2.Thermosetting resin The resin may be a thermosetting resin. When a thermosetting resin is used, it is preferably an unsaturated polyester resin, a vinyl ester resin, an epoxy resin, or a phenol resin. One type of resin may be used alone, or two or more types may be used in combination. Furthermore, when a thermosetting resin is used as the resin of the present invention, it is preferable to use a sheet molding compound (sometimes called SMC) containing reinforcing fibers. Because of its high moldability, sheet molding compounds can be easily molded into even complex shapes. Sheet molding compounds have higher fluidity and formability than continuous fibers, making it easy to create ribs and bosses.

[0041] [Other agents] The resin may contain additives such as various fibrous or non-fibrous fillers such as organic or inorganic fibers, flame retardants, UV-resistant agents, stabilizers, release agents, pigments, softeners, plasticizers, surfactants, etc. When a thermosetting resin is used, it may also contain thickeners, curing agents, polymerization initiators, polymerization inhibitors, etc. One type of additive may be used alone, or two or more types may be used in combination.

[0042] [Method of manufacturing molded body: Press molding] 1. Hot press molding and cold press molding When the fiber-reinforced product of the present invention is a molded body, the molded body can be produced by a press molding method (sometimes called compression molding), and as the press molding, molding methods such as hot press molding and cold press molding can be used. By press molding, various shapes can be given to the molded body.

[0043] 2.Cold press molding When the fiber-reinforced product of the present invention is a molded article, and when a thermoplastic resin is used as the resin, press molding using cold press is preferred. In cold press molding, for example, a fiber-reinforced product heated to a first predetermined temperature is placed in a mold set to a second predetermined temperature, and then pressurized and cooled. Specifically, if the thermoplastic resin contained in the fiber reinforced product is crystalline, the first predetermined temperature is equal to or higher than the melting point, and the second predetermined temperature is lower than the melting point. If the thermoplastic resin is amorphous, the first predetermined temperature is equal to or higher than the glass transition temperature, and the second predetermined temperature is lower than the glass transition temperature. That is, the cold pressing method includes at least the following steps A-1) to A-2). Step A-1) A step of heating the thermoplastic resin to a temperature above the melting point but below the decomposition temperature if the thermoplastic resin is crystalline, or above the glass transition temperature but below the decomposition temperature if the thermoplastic resin is amorphous. Step A-2) The composite material heated in step A-1) is placed in a mold whose temperature is adjusted to below the melting point if the thermoplastic resin is crystalline, or below the glass transition temperature if the thermoplastic resin is amorphous, and then pressurized. By performing these steps, the molding of the composite material can be completed (a press-molded body can be produced).

[0044] The above steps must be performed in the order described above, but other steps may be included between each step, such as a shaping step, which is performed before step A-2), in which a shaping mold different from the mold used in step A-2) is used to pre-shape the mixture into the shape of the cavity of the mold. The shape of the fiber reinforced product may be a shape developed by computer inverse molding analysis from the three-dimensional shape of the press-molded product to be manufactured.

[0045] 3.Hot press molding In the hot press molding method, for example, a fiber-reinforced product is placed in a mold, pressure is applied while the temperature of the mold is raised to a first predetermined temperature, and the mold is cooled to a second predetermined temperature. Specifically, if the thermoplastic resin constituting the fiber-reinforced product is crystalline, the first predetermined temperature is equal to or higher than the melting point, and the second predetermined temperature is lower than the melting point. If the thermoplastic resin contained in the fiber-reinforced product is amorphous, the first predetermined temperature is equal to or higher than the glass transition temperature, and the second predetermined temperature is lower than the glass transition temperature. Hot press molding preferably includes at least the following steps B-1) to B-4). B-1) A step of placing the composite material in a mold (second mold, lower mold). B-2) A process of applying pressure while heating the mold to a temperature above the melting point and below the thermal decomposition temperature of the thermoplastic resin if the thermoplastic resin is crystalline, or to a temperature above the glass transition temperature and below the thermal decomposition temperature of the thermoplastic resin if the thermoplastic resin is amorphous (first pressing process). B-3) A process of applying pressure in one or more stages so that the pressure in the final stage is 1.2 to 100 times the pressure in the first pressing process (second pressing process). B-4) A step of adjusting the mold temperature to below the melting point if the thermoplastic resin is crystalline, or below the glass transition temperature if the thermoplastic resin is amorphous. By carrying out these steps, a molded body can be produced.

[0046] 4. Commonalities between cold press molding and hot press molding Steps A-2) and B-3) are steps in which pressure is applied to the composite material to obtain a molded product of the desired shape. The molding pressure is not particularly limited, but is preferably as low as possible within a range that allows the desired structure shape to be obtained. Specifically, a pressure of less than 30 MPa relative to the mold cavity projected area is preferred, more preferably 20 MPa or less, and even more preferably 10 MPa or less. A molding pressure of less than 30 MPa is preferred because it does not require capital investment or maintenance costs for a press. Naturally, various processes may be inserted between the above compression molding steps, such as vacuum compression molding, in which compression molding is performed under vacuum.

[0047] [First embodiment] As a first embodiment, a method for detecting surface defects using the following steps 301 and 401 will be described below. Step 301: A step of reading out the surface image of the fiber reinforced product stored in the storage means, processing the image, and outputting pixel values. Step 401: A step of extracting points A and B, which have a difference in pixel value of 128 or more and are adjacent to each other. 1. Surface of fiber-reinforced products (fiber pattern) The pattern of reinforcing fibers observed on the surface of a fiber-reinforced product is shown, for example, in Figure 1. In the fiber-reinforced plastic depicted in Figure 1, the reinforcing fibers are discontinuous and randomly dispersed in two dimensions. This type of fiber morphology causes the appearance design of each individual product to differ, even among products of the same brand. More specifically, differences arise in the characteristics of the arranged fiber bundles, such as the position, orientation, thickness, and length, for each individual product. In other words, the design observed on the surface of the fiber-reinforced product of the present invention can be said to be a fingerprint that differs from product to product.

[0048] As such, when detecting defects in fiber-reinforced products, which have different appearance designs (i.e., different observed fiber patterns), it is impossible to set a predetermined threshold. For example, in the invention described in Patent Document 2, defects are detected by comparing with a preset threshold, but this can only be used when the appearance of the inspected product is consistent, such as in prepregs using continuous fibers. In fiber-reinforced products with dispersed discontinuous fibers, as in the present invention, the surface design is too complex, making it impossible to uniquely determine the surface design of the product (a predetermined threshold cannot be set).

[0049] In the present invention, the thickness of the thinnest part of the resin between the surface of the fiber-reinforced product and the fibers present inside the fiber-reinforced product is preferably less than 100 μm. If the fibers are not present in a fluffy form on the surface of the composite material, the thickness of the resin layer may be 0 μm when observed in a cross-sectional photograph, but it is preferable for the resin layer to have a certain thickness, and it is preferably maintained at 0.01 μm or more. Here, the thickness of the resin layer present between the reinforcing fibers and the surface of the composite material refers to the distance between the fiber bundle or the fiber isolated from the fiber bundle that is closest to the surface and the surface of the composite material.

[0050] Furthermore, the thickness of the resin layer at the thinnest part between the surface of the composite material and the discontinuous reinforcing fibers present inside the composite material is preferably in the range of 0.5 μm or more and less than 100 μm.

[0051] When the thickness of the surface resin layer is 100 μm or less, the pattern of the reinforcing fibers becomes more visible, improving the design of the "complex pattern," and making it easy to create an exterior design with a different pattern for each fiber-reinforced product.

[0052] 2. Process 101. Step 101 in the present invention is a step of irradiating the surface of the fiber-reinforced product with light from an illumination means and imaging the surface with an imaging means. 2.1 Lighting means Although there are no particular limitations on the lighting means, examples include line scan lighting, dome lighting (diffused light), ring lighting, and multi-angle lighting (combining ring lights shining from various directions), line scan lighting or dome lighting, which can provide uniform illuminance to the imaging location, are preferred, and dome lighting is particularly preferred. Line scan lighting may be used to illuminate from only one direction, or imaging may be performed from the position of specular reflection, or imaging may be performed from a position slightly shifted from the position of specular reflection.

[0053] 2.2 Diffuse Reflection It is preferable that the light that reaches the imaging device contains more diffusely reflected light than specularly reflected light. Compared to specularly reflected light, diffusely reflected light does not produce an image that is too bright, making defects easier to recognize. This is because when the proportion of specular reflection is high compared to diffuse reflection, the "gloss (glare)" and "shininess" of the surface of the fiber-reinforced product are captured in large quantities. In other words, when there is a lot of diffusely reflected light, it is easier to capture the shadow area 204, making defects easier to recognize. For example, in Figure 5(b), light from the lighting means is reflected by the fiber-reinforced product and is directly input to the imaging means. On the other hand, in Figure 5(a), the position of the fiber-reinforced product is shifted 2 to 3 mm toward the imaging means (x in Figure 5(a)). As a result, the light from the lighting means is diffusely reflected by the fiber-reinforced product and input to the imaging means.

[0054] 2.3 Imaging distance In the present invention, the imaging distance is preferably 100 mm or more and 800 mm or less. By making it 100 mm or more, the imaging width can be secured and the autofocus function can be easily used. Furthermore, by capturing an image of a relatively wide range, there is no need to take multiple photos. On the other hand, by setting the lens to 800mm or less, it is possible to maintain good image resolution. In other words, even with an inexpensive lens, it is possible to increase the resolution. Also, by setting the lens to closer to 800mm or less, it becomes easier to secure space in the design of the equipment. Because the space between the imaging device and the fiber-reinforced product is narrow, the design space behind the imaging device can be expanded. In this expanded design space, it becomes possible to introduce robots for transporting the fiber-reinforced product, for example. The imaging distance is more preferably 200 mm or more and 700 mm or less, and even more preferably 300 mm or more and 600 mm or less.

[0055] 2.4 Imaging device The imaging device for capturing images is not particularly limited, but may be, for example, a digital camera equipped with a lens, which captures the identification information of the captured fiber-reinforced plastic as electronic data of the digital image and sends the image data to a recording means. For example, the imaging device may be set to face the fiber-reinforced product directly so that it can capture the entire identification information. To reduce extraneous light reflected or glare from the surface of the fiber-reinforced product, the position of the product, the ambient light source, and the camera may be adjusted, or a polarizing filter that linearly polarizes the light entering the lens may be installed at the tip of the imaging device's lens, thereby enabling accurate capture of the surface image of the fiber-reinforced product. Any known computer may be used.

[0056] 2.5 Movable imaging and lighting devices In the present invention, the imaging device and the lighting device may be moved simultaneously to capture images continuously. Even if the imaging device and the lighting device are moved simultaneously, it is possible to extract adjacent points A and B from the obtained image in step 401.

[0057] 2.6 Fixing of imaging device and lighting means In the present invention, it is preferable that the imaging device is fixed. In other words, it is preferable to move the fiber-reinforced product rather than moving the imaging device to photograph the fiber-reinforced product. As mentioned above, it is possible to move the imaging device to photograph, but in the present invention, it becomes necessary to move the lighting device at the same time. It is preferable to fix the lighting device and the imaging device separately, as this makes it less likely to shake when photographing, compared to moving the lighting device and imaging device at the same time.

[0058] 2.7 Number of imaging devices and lighting means It is preferable to have one imaging device and one lighting means. Generally, images are taken from two directions using one light source, or one image is taken from two lights. In other words, it is common to take two or more photos of a defect. When extracting adjacent points A and B with pixel values ​​of 128 or greater, the takt time can be shortened by using one imaging device and one lighting means. In this case, it is necessary to actively create a shadow of the defect, so it is preferable to have one lighting source. If lighting is applied from two directions, the shadow of the defect (e.g., 204 in Figure 2) will become faint.

[0059] 2.8 Image capture area The range of the fiber reinforced product is not particularly limited, and although it depends on the size of the fiber reinforced product, it is preferable to take images of an area of ​​400 mm x 400 mm continuously.

[0060] 2.9 Image Composition In step 401, when extracting adjacent portions A and B, it is preferable not to combine the captured images. In other words, it is preferable to either delete one of the overlapping areas during shooting, or capture each image independently. When images are combined, the shadow of the defect tends to be faded. If the images are not combined, the defect will be clearly visible as a difference in pixel value, but combining images with different brightness may result in a change in pixel value.

[0061] 2.10 Imaging defects Imaging a defect when extracting the defect using the difference in pixel values ​​will be described with reference to FIG. 2. When light 211 emitted from illumination means 206 strikes defect 205, a reflective area 203 and a shadow area 204 are generated. When this is imaged by imaging means 202, the pixel value of reflective area 203 is high and the pixel value of shadow area 204 is low. The present invention detects defects using this difference in pixel values. After image processing, reflective area 203 becomes area A, and after image processing, shadow area 204 becomes area B. In this way, the present invention can detect defect 205 by distinguishing between reflective area 203, which has a high pixel value, and shadow area 204, which has a relatively low pixel value, both of which are adjacent to defect 205. In other words, an example of a defect according to the present invention is a blister such as 205 in FIG. 2.

[0062] 2.11 Disadvantages of concave shapes In step 401, when extracting adjacent portions A and B, it is possible to detect not only blister defects such as 205 in Figure 2, but also concave defects. Concave defects include scratches and dents, but even these concave defects generate reflective and shadow areas due to the lighting means. Defects can be detected by recognizing the difference in pixel values ​​between these reflective and shadow areas.

[0063] 3. Process 201. The image captured by the imaging means is stored in a storage means, preferably in a database, preferably in association with product information and manufacturing conditions of the fiber-reinforced product, and preferably in a storage unit of a computer.

[0064] 4. Process 301. 4.1 Step 301 is a step of reading out the surface image of the fiber reinforced product stored in the storage means, processing the image, and outputting the pixel values. It is preferable that the image processing replaces the function of fixational eye movement with an electronic circuit, and it is more preferable that the fixational eye movement is tremor. Figure 3(a) is a captured image, and Figure 3(b) is the image after processing, specifically, after tremor processing and averaging. In Figure 3(a), 301 is the area where light is reflected, which corresponds to the reflection area 203 in Figure 2. Similarly, 302 in Figure 3(a) is a shadow, which corresponds to the shadow area 204 in Figure 2. On the other hand, Figure 3(b) shows the result after image processing, with 303 in Figure 3 being the light reflection after image processing and 304 in Figure 3 being the enhanced shadow after image processing.

[0065] 4.2 Tremor Tremor (microtremor) is a type of fixational eye movement in the human eye, where the eyeball vibrates vertically at 10 μm approximately 80 times per second. It is said that if a person with 1.0 visual acuity stops fixational eye movement, their visual acuity will drop to 0.1. This behavior is believed to enable detection of subtle defects that would be undetectable without movement. This allows humans to see objects with high accuracy. Accuracy can be improved by converting the function of fixational eye movement, known as tremor, into an electronic circuit and processing it as if the object being inspected is vibrating slightly. The process of converting fixational eye movement into an electronic circuit to eliminate oversights and increase accuracy is also called tremor sensing. For specific surface treatment methods, see Patents 04932819, 04250076, and 03739965. For more specific examples, use Technos Co., Ltd.'s Super 5000 7K model. This technology (neuro-visual sensor "Brain-Neuro") won the Electrical and Electronic Components Award at the Japan Manufacturing Parts Awards in 2015. (URL: https: / / award.cho-monodzukuri.jp / award2015 / 2015electric-electron / ) In addition to tremors, fixational eye movements also include microsaccades and drifts.

[0066] 4.3 Averaging Filter When processing images, it is preferable to apply an averaging filter to reduce noise and emphasize features such as edges. An averaging filter is a filter that blurs and smooths an image. An example of image noise is when differences in pixel values ​​become noticeable, making uniform areas appear rough. In such cases, blurring the image can make the roughness less noticeable.

[0067] 5. Process 401. Step 401 is a step of extracting adjacent portions A and B whose pixel value difference is 128 or more. Preferably, the pixel value difference is 128 or more and 255 or less, more preferably 150 or more and 255 or less, even more preferably 200 or more and 255 or less, and even more preferably 230 or more and 255 or less. Furthermore, it is preferable that the pixel value of location A is 200 or more, and the pixel value of location B is 50 or less. More preferably, the pixel value of location A is 240 or more and 255 or less, and the pixel value of location B is 0 or more and 10 or less.

[0068] 5.1 Defect extraction The images obtained by the image processing in step 301 are exemplified in FIG. 3(b) and FIG. 4(a). 5.1.1 Figure 4(b) From Figure 4(a), areas with pixel values ​​of 200 or more are extracted and designated as area A. Figure 4(b) shows an example of an image with only area A extracted. Figure 4(b) shows areas with pixel values ​​of 200 or more extracted from Figure 4(a). 5.1.2 Figure 4(c) FIG. 4(c) shows an observation of the area surrounding the area shown in FIG. 4(b), where the pixel value is high. When observing FIG. 4(c), if a point A is extracted (FIG. 4(b)), and the diameter of the maximum circumscribing circle of point A is α mm, the area observed is expanded to a diameter of 1.2 × α mm. Then, within the observed area, a point B with a pixel value of 50 or less is extracted. For example, in FIG. 4(c), point A (402 in FIG. 4) has a pixel value of 200 or more, and point B (401 in FIG. 4) has a pixel value of 50 or less. In this case, FIG. 4(c) is a candidate for a defect (blister). In the present invention, a point where the difference in pixel value between point A and point B is 128 or more is extracted as a defect candidate. In other words, a point where the difference in pixel value is less than 128 is not extracted as a defect candidate.

[0069] 5.2 Definition of "adjacent" Although portions A and B are adjacent to each other, "adjacent" here refers not only to the case where portions A and B are in contact, but also to the case where they are not in contact. In other words, "adjacent" refers to the state where portions A and B are adjacent to each other, and includes the state where the two components are not in contact. This is because, depending on the shape of the defect, the boundary between portions A and B may or may not be in contact. More specifically, the distance between the reflective region 203 and the shadow region 204 is preferably 0 mm or more and 100 mm or less, more preferably 5 mm or more and 50 mm or less, and even more preferably 5 mm or more and 20 mm or less.

[0070] 5.3 Pixel value information 5.3.1 Grayscale images It is preferable that the captured surface image contains only pixel value information (grayscale image). An image created using pixel value information is called a grayscale image. An image calibrated using a collection of pixel values ​​is a grayscale image. It is preferable that the surface image is a grayscale image that expresses shades of black and white. A grayscale image represents one pixel using 8 bits and contains only brightness information, not color information.

[0071] Generally, monochrome images express brightness in 256 levels from 0 to 255, with 0 being black (pure black among blacks), 255 being white (pure white among whites), and the intermediate gray, a mixture of equal parts black and white, being quantified as 127. This is because the range between the minimum and maximum energy detectable by the sensor is divided into 256 levels (8 bits), which are then converted into integer values ​​using a sliding formula. All you have to do is look at the pixel values ​​of the image and detect points A and B.

[0072] 5.3.2 Red-Green-Blue (RGB) Images The captured surface image may be a red-green-blue (RGB) image. When a red-green-blue (RGB) image is captured, it can be filtered using one of the red, green, or blue pixels to use 256 levels of grayscale information. The filtered 256 levels of grayscale information can be used as the surface pixel value to detect points A and B.

[0073] 6. Process 501. Step 501 is a step of detecting the surface defect, if any, by a learning type surface defect detection means. As described above, in step 401, points A and B are extracted, and defect candidates are extracted. However, at this stage, even things that are not defects are recognized as defects and extracted. For example, in the manufacturing process of fiber-reinforced products, when a product is crushed between rollers, a part where the roller surface is transferred will be extracted as a defect because a difference in pixel value occurs (the pattern will be recognized as a defect). Therefore, a learning-type surface defect detection means is used. Here, the learning-type surface defect detection means in the present invention is a means in which artificial intelligence detects surface defects on the surface of a fiber-reinforced product based on learning of surface image data of a fiber-reinforced product without the surface defect and surface image data of a fiber-reinforced product with the surface defect.

[0074] The basic operation of the learning-type surface defect detection means is that the artificial intelligence learns from normal data without surface defects among the surface image data of fiber-reinforced products, and the artificial intelligence detects surface defects by finding surface image data of the fiber-reinforced product that differs from the learned normal data. AI with a machine learning model is preferred as this learning-type surface defect detection means.

[0075] Specifically, the AI ​​learns using normal data without surface defects from surface image data of fiber-reinforced products. This allows the AI ​​to quickly identify normal data and similar data. However, when the AI ​​encounters abnormal data that differs from normal data, it struggles to make a judgment and outputs a predetermined abnormality level. By having the AI ​​learn the results of surface defects detected by the on-demand surface defect detection means, even if the results are small, the AI ​​learns to indicate a high abnormality level for data determined to be surface defects by the on-demand surface defect detection means, even if the abnormality level is the same. By outputting this predetermined abnormality level, the AI ​​is deemed to have detected a surface defect, making it possible to detect surface defects. Once the AI ​​is able to detect surface defects, it is possible to stop learning using the results of surface defects detected by the on-demand surface defect detection means, further improving the AI's judgment speed.

[0076] Furthermore, the artificial intelligence may be configured to detect surface defects on the surface of the bar material based on the results of surface defect detection by the on-demand surface defect detection means as well as the results of human judgment. By learning human judgment, it becomes possible to detect surface defects that are difficult to judge, such as those that can only be judged by humans.

[0077] According to a surface inspection device having such a configuration and operation, the surface defect detection means comprises an on-demand surface defect detection means that detects defects on the surface of a fiber-reinforced product from the acquired surface image data of the fiber-reinforced product, and a learning surface defect detection means that uses artificial intelligence to detect surface defects on the surface of a fiber-reinforced product based on the surface image data of the fiber-reinforced product when there are no surface defects and learning from the results of surface defects detected by the on-demand surface defect detection means, making it possible to accurately detect many types of surface defects one by one.

[0078] [Second embodiment] 1. Overview A second embodiment of the present invention is a method for producing a fiber-reinforced product containing discontinuous reinforcing fibers dispersed in the in-plane direction and a resin, which includes step 302 between step 201 and step 501. Step 101: A step of irradiating light by an illumination means and imaging the surface of the fiber-reinforced product by an imaging means. Step 201: A step of storing the surface image captured by the imaging means in storage means. Step 302: A step of reading out the surface image of the fiber reinforced product stored in the storage means, detecting the black pattern, and outputting the area of ​​the black pattern Step 501: A step of detecting the surface defects by a learning surface defect detection means. However, a learning-type surface defect detection means is a means in which artificial intelligence detects surface defects on the surface of a fiber-reinforced product based on learning surface image data of the fiber-reinforced product when it is free of the surface defect and surface image data of the fiber-reinforced product when it has the surface defect. In the second embodiment, steps 101, 201, and 501 are the same as those in the first embodiment.

[0079] 2. Process 302 Step 302 is a step of reading out the surface image of the fiber reinforced product stored in the storage means, detecting the black pattern, and outputting the area of ​​the black pattern. 2 If the area of ​​the black pattern is more than 1.0cm, it should be recognized as a defect. 2 In cases where the above applies, it is preferable to have it recognized as a defect. In Figure 7, the area of ​​the black pattern is 1 cm 2 The following is an example of a defect 701 equivalent to the above. Defect 701 is an example of a defect that occurs when a fiber-reinforced product is manufactured using a double belt press (for example, the manufacturing method described in International Publication No. 2016 / 002470). After the heated resin is impregnated into the fibers, insufficient cooling may cause the resin to stick to the side of the double belt press, and this part becomes defect 701 in the fiber-reinforced product. 3. In the second embodiment, it is preferable that the light that reaches the imaging device is specularly reflected light.

[0080] [Third embodiment] 1. Overview A third embodiment of the present invention is a method for producing a fiber-reinforced product containing discontinuous reinforcing fibers dispersed in the in-plane direction and a resin, which includes step 303 between step 201 and step 501. Step 101: A step of irradiating light by an illumination means and imaging the surface of the fiber-reinforced product by an imaging means. Step 201: A step of storing the surface image captured by the imaging means in storage means. Step 303: A step of reading out the surface image of the fiber-reinforced product stored in the storage means, detecting the black line pattern, and outputting the area of ​​the black line and the degree of bending. Step 501: A step of detecting the surface defects by a learning surface defect detection means. However, a learning-type surface defect detection means is a means in which artificial intelligence detects surface defects on the surface of a fiber-reinforced product based on learning surface image data of the fiber-reinforced product when it is free of the surface defect and surface image data of the fiber-reinforced product when it has the surface defect. In the third embodiment, steps 101, 201, and 501 are the same as those in the first embodiment.

[0081] 2.Process 303 Step 303 is a step of reading out the surface image of the fiber-reinforced product stored in the storage means, detecting the black line pattern, and outputting the area and curvature of the black line. The curvature radius R of the black line is preferably in the range of 30 mm to 5000 mm, more preferably in the range of 100 mm to 1000 mm. Within the output curvature radius, the area of ​​the black pattern is 300 mm 2 More than 5000mm 2 It is preferable to detect the following as defects: The area of ​​the black pattern is 500 mm within the output curvature radius. 2 More than 3000mm 2 It is more preferable to detect the following as defects: A defect detected in the third embodiment is shown in 801 in Fig. 8. When the fibers contained in a fiber-reinforced resin product are made from discontinuous fibers of about 20 mm, if a cutting error occurs in the reinforcing fibers, the fiber length will remain longer than the other fibers (more than 20 mm), resulting in a defect such as that shown in 801 in Fig. 8. 3. In the third embodiment, it is preferable that the light that reaches the imaging device is specularly reflected light.

[0082] [Fourth embodiment] 1. Overview A fourth embodiment of the present invention is a method for producing a fiber-reinforced product comprising discontinuous reinforcing fibers dispersed in the in-plane direction and a resin, which includes step 304 between step 201 and step 501. That is, the fiber reinforced product is a flat plate that is continuously manufactured in the MD direction, Step 101: A step of irradiating light by an illumination means and imaging the surface of the fiber-reinforced product by an imaging means. Step 201: A step of storing the surface image captured by the imaging means in storage means. Step 304: A step of reading out the surface image of the fiber reinforced product stored in the storage means, detecting the black line pattern, and outputting the angle between the MD direction of the fiber reinforced product and the direction of the black line pattern. Step 501: A step of detecting the surface defect, if any, by the learning type surface defect detection means. However, a learning-type surface defect detection means is a means in which artificial intelligence detects surface defects on the surface of a fiber-reinforced product based on learning surface image data of the fiber-reinforced product when it is free of the surface defect and surface image data of the fiber-reinforced product when it has the surface defect. In the fourth embodiment, steps 101, 201, and 501 are the same as those in the first embodiment.

[0083] 2.Process 305 Step 305 is a step of reading out the surface image of the fiber reinforced product stored in the storage means, detecting the black line pattern, and outputting the angle between the MD direction of the fiber reinforced product and the direction of the black line pattern. A black line that is 40 mm or longer and whose longitudinal direction is at an angle of ±5° or less to the MD direction is detected as a possible scratch, and more preferably a black line that is 50 mm or longer and whose longitudinal direction is at an angle of ±1° or less to the MD direction is detected as a possible scratch. 3. In the fourth embodiment, it is preferable that the light that reaches the imaging device is specularly reflected light. [Example]

[0084] 1. Evaluation (1) Material 1.Material 1.1 Carbon fiber Teijin's carbon fiber "Tenax" (registered trademark) STS40-24K (average fiber diameter 7 μm, 24,000 single fibers) 2.1 Thermoplastic resin Polyamide 6 (A1030 manufactured by Unitika Ltd., sometimes abbreviated as PA6). (1) Fiber volume fraction Vf contained in the fiber-reinforced product The fiber-reinforced product was placed in a furnace at 500°C for 1 hour to burn off the resin, and the mass of the reinforcing fiber and resin was calculated by weighing the sample before and after treatment. Next, the volume ratio of the reinforcing fiber to the resin was calculated using the specific gravity of each component. Note that the amount of other additives was so small that they could be ignored in terms of volume. Vf = 100 x reinforcing fiber volume / (reinforcing fiber volume + resin volume + other additive volume) 2.2 Preparation of fiber-reinforced products Teijin Limited's "Tenax" (registered trademark) STS40-24K carbon fiber (average fiber diameter 7 μm, 24,000 single fibers) cut to 20 mm in length was used as the carbon fiber, and Unitika Limited's polyamide 6 resin A1030 was used as the resin. A carbon fiber and polyamide 6 resin molding material with two-dimensionally randomly oriented carbon fibers was prepared based on the method described in U.S. Patent No. 10,006,677. The resulting molding material was heated at 270°C and 2.0 MPa for 5 minutes to obtain a flat fiber-reinforced product measuring 1250 mm wide x approximately 2000 mm long x 2.65 mm thick. The same process was repeated to prepare a total of 796 fiber-reinforced products. The fiber patterns of each fiber-reinforced product were different from one another, resulting in different surface designs. Analysis of the carbon fibers contained in this flat fiber-reinforced plastic revealed that the carbon fiber volume fraction (Vf) was 35%, the fiber length was constant, and the weight-average fiber length was 20 mm. The surface of the fiber-reinforced product was two-dimensionally randomly oriented, as shown in Figure 1. Of the 796 images prepared, 621 were used for training the learning-type defect detection means, and 109 were prepared as test data.

[0085] (2) Preparation of learning-based surface defect detection method The presence or absence of defects (blisters) was manually detected from 796 fiber-reinforced products, and this was used as known image information. A surface defect prediction model was generated using machine learning based on this known image information, and this was used as a learning-based surface defect detection method.

[0086] [Example 1] 796 sheets of the fiber reinforced products prepared as described above were prepared, and an experiment was carried out in accordance with the first embodiment. (1) Process 101 The lighting device used was an LNSD-1800SW-DUM12TK (LED light) manufactured by CCS Inc., and was positioned 100 mm away from the position of the fiber-reinforced product at an incident angle of 20 degrees. The imaging device used was a 5000K-S manufactured by Technos Corporation, and was positioned 430 mm away from the position of the fiber-reinforced product (imaging distance was 430 mm). (2) Process 201 The surface images obtained by the imaging means were recorded on a 5000K-S manufactured by Technos Corporation. (3) Process 301. The surface image of the fiber-reinforced fiber product was read out and processed using a Technos 5000 Series-S (application name: Analyzer). The resulting image is shown in Figure 6. (4) Process 401. Areas A and B were extracted as adjacent areas with a pixel value difference of 230 or more. The pixel value of area A was set to 240 or more, and the pixel value of area B was set to less than 10. Some areas A and B were completely adjacent to each other, while others were separated by approximately 10 mm. An example of the portion A is shown in 601 in Fig. 6. It can be seen that the portion A is white in 601. An example of the portion B is shown at 602 in Fig. 6. It can be seen that the portion 602 is white. (5) Process 501. The image extracted in step 401 was judged using a learning type surface defect detection means to detect surface defects. (6) Manufacturing of continuous fiber-reinforced products Steps 101 to 501 were repeated for 100 fiber reinforced products to detect defects.

[0087] In process 101, it took 20 to 30 seconds to remove the fiber-reinforced product after imaging and place the fiber-reinforced product to be imaged, 2 seconds to image it, and 15 seconds to process the images. In other words, 772 defects were detected from 796 fiber-reinforced products measuring 1250 mm wide x approximately 2000 mm x average thickness of 2.65 mm, and this was completed in approximately 8 hours and 20 minutes. This was a time that would be impossible to complete manually. [Explanation of symbols]

[0088] 101: Fiber-reinforced products 102: Fiber bundle 211: Light 202: Imaging device 203, 301: Reflection area 204, 302: Shadow area 205: Defects (swelling, blisters) 206: Lighting means 303: Reflection of light after image processing 304: Enhanced shadow after image processing 401: Area B where pixel value is 50 or less 402: Point A where pixel value is 200 or more 601: Location A 602: Location B X: 2mm to 3mm (the distance the fiber-reinforced product is moved toward the imaging device) 701:Drawbacks caused by resin sticking to the side of the double belt press 801: Defects caused by incorrect cutting of reinforcement fibers 901: Scratch

Claims

1. A method for producing a fiber-reinforced product comprising discontinuous reinforcing fibers dispersed in an in-plane direction and a resin, the method comprising the steps of: Step 101. A step of irradiating light by an illumination means and imaging the surface of the fiber-reinforced product by an imaging means. Step 201: A step of storing the surface image captured by the imaging means in a storage means. Step 501. If there is a surface defect, the surface defect is detected by a learning surface defect detection means. However, a learning-type surface defect detection means is a means in which artificial intelligence detects surface defects on the surface of a fiber-reinforced product based on learning surface image data of the fiber-reinforced product when it is free of the surface defect and surface image data of the fiber-reinforced product when it has the surface defect.

2. 2. A method for producing a fiber-reinforced product according to claim 1, comprising, between step 201 and step 501, the following steps 301 and 401: Step 301: A step of reading out the surface image of the fiber reinforced product stored in the storage means, processing the image, and outputting pixel values. Step 401: A step of extracting adjacent portions A and B whose pixel value difference is 128 or more.

3. 3. The method for producing a fiber-reinforced product according to claim 2, wherein in step 101, an image of the surface of the fiber-reinforced product is taken using an imaging device located at an imaging distance of 100 mm to 800 mm from the fiber-reinforced product.

4. The method for manufacturing a fiber-reinforced product according to claim 2 or 3, wherein the number of the imaging device and the number of the lighting means are each one.

5. 5. The method for manufacturing a fiber-reinforced product according to claim 2, wherein the image processing in step 301 replaces the function of fixational eye movement with an electronic circuit.

6. The method for producing a fiber-reinforced product according to claim 5, wherein the involuntary eye movement is tremor.

7. The method for manufacturing a fiber-reinforced product according to any one of claims 2 to 6, wherein the surface image is a grayscale image.

8. The method for producing a fiber reinforced product according to any one of claims 2 to 6, wherein the surface image is a red-green-blue (RGB) image.

9. 9. The method for manufacturing a fiber-reinforced product according to claim 2, wherein in step 401, the pixel value of point A is 200 or more and the pixel value of point B is 50 or less.

10. The method for manufacturing a fiber-reinforced product according to claim 2 , wherein the imaging device and the lighting means are fixed.

11. The light that reaches the imaging device contains more diffusely reflected light than specularly reflected light. A method for producing a fiber-reinforced product according to any one of claims 2 to 10.

12. A method for inspecting a fiber-reinforced product comprising discontinuous reinforcing fibers dispersed in the in-plane direction and a resin, using the following steps: Step 101. A step of irradiating light by an illumination means and imaging the surface of the fiber-reinforced product by an imaging means. Step 201: A step of storing the surface image captured by the imaging means in a storage means. Step 301: A step of reading out the surface image of the fiber reinforced product stored in the storage means, processing the image, and outputting pixel values. Step 401: A step of extracting adjacent portions A and B whose pixel value difference is 128 or more. Step 501. If there is a surface defect, the surface defect is detected by a learning surface defect detection means. However, a learning-type surface defect detection means is a means in which artificial intelligence detects surface defects on the surface of a fiber-reinforced product based on learning surface image data of the fiber-reinforced product when it is free of the surface defect and surface image data of the fiber-reinforced product when it has the surface defect.

13. 2. A method for producing a fiber-reinforced product according to claim 1, comprising, between step 201 and step 501, step 302: Step 302: A step of reading out the surface image of the fiber-reinforced product stored in the storage means, detecting the black pattern, and outputting the area of ​​the black pattern.

14. 2. A method for producing a fiber-reinforced product according to claim 1, comprising, between step 201 and step 501, step 303: Step 303: A step of reading out the surface image of the fiber reinforced product stored in the storage means, detecting the black line pattern, and outputting the area of ​​the black line and the degree of bending.

15. The fiber reinforced product has a flat plate shape that is continuously manufactured in the MD direction, 2. A method for producing a fiber-reinforced product according to claim 1, comprising, between step 201 and step 501, step 304: Step 304: A step of reading out the surface image of the fiber reinforced product stored in the storage means, detecting a black line pattern, and outputting the angle between the MD direction of the fiber reinforced product and the direction of the black line pattern.

16. 16. A method for manufacturing a fiber reinforced product according to any one of claims 13 to 15, wherein the light reaching the imaging device is specularly reflected light.

17. 17. A method for producing a fiber-reinforced product according to any one of claims 1 to 11 or claims 13 to 16, wherein a plurality of fiber-reinforced products are continuously produced, and the fiber patterns observed in each fiber-reinforced product are different from each other.

18. 18. A method for producing a fiber-reinforced product according to any one of claims 1 to 11 or 13 to 17, wherein the thickness of the thinnest part of the resin between the surface of the fiber-reinforced product and the discontinuous reinforcing fibers present inside the fiber-reinforced product is less than 100 μm.

Images