Multimodal evaluation method for the degree of dispersion of the fiber distribution of a robotically injected part

Abstract

This application relates to a multimodal evaluation method for the dispersion of fiber distribution in robot injection molded parts. The method includes: acquiring the visible light color difference value of the injection molded sample to be tested; activating a multimodal scanning sequence in response to the color difference value being lower than the optical recognition limit; determining the detection mode of the first modal detection based on the concentration information of conductive fibers; performing the first modal detection and the second modal detection to acquire a first feature quantity characterizing the total dispersion of fibers and a second feature quantity characterizing the dispersion of conductive fibers, respectively; calculating the independent dispersion index of non-conductive fibers using a decoupling model based on the first and second feature quantities, and determining the distribution uniformity evaluation results of the two types of fibers respectively. This application has the advantages of effectively separating the distribution state of mixed fibers, overcoming the optical blind zone caused by dark conductive fibers, and improving the detection accuracy under complex working conditions.

Description

Technical Field

[0001] This application relates to the field of robot manufacturing, and in particular to a multimodal evaluation method for the discreteness of fiber distribution in robot injection molded parts. Background Technology

[0002] In the injection molding process of robot structural components, fiber reinforcement materials are often incorporated into the polymer matrix to improve the mechanical properties of the product. The uniformity of fiber distribution in the matrix directly affects key indicators such as strength, modulus, and fatigue life of the product. If fibers agglomerate in local areas, a sudden gradient in the mechanical properties of the agglomerated area will occur compared to the surrounding area. Under alternating loads, the product is prone to crack initiation from the agglomeration boundary, leading to premature failure of the structural component. Therefore, evaluating the uniformity of fiber distribution during injection molding is a necessary step to ensure batch consistency of robot structural components.

[0003] Currently, the most widely used method for evaluating fiber distribution in industry is the visible light colorimetric method. This method utilizes the difference in optical reflectivity between fibers and the polymer matrix. A colorimeter is used to collect colorimetric values ​​at various measurement points on the surface of the injection-molded part, and the spatial fluctuation of these colorimetric values ​​indirectly characterizes the uniformity of fiber distribution. When there is sufficient contrast between the fiber color and the matrix color, areas of fiber agglomeration will exhibit localized colorimetric anomalies that can be identified by the colorimeter. This method allows for evaluation with relatively low equipment costs and fast measurement speed.

[0004] However, when injection-molded parts contain dark-colored conductive fibers such as carbon fiber, the colorimetric method faces fundamental physical limitations. Carbon fiber itself is black, and as the mass fraction of carbon fiber in the matrix increases, the overall lightness of the product surface is suppressed to an extremely narrow low-value range. Within this range, the lightness difference between areas with uniform fiber distribution and areas with local fiber aggregation is less than the effective resolution of the colorimeter, and the spatial gradient of colorimetric values ​​tends to flatten, making it impossible to extract meaningful information on distribution uniformity. In other words, the evaluation capability of the colorimetric method depends on sufficient optical contrast between the fiber and the matrix, and dark-colored conductive fibers precisely eliminate this contrast basis, rendering the colorimetric method ineffective for such products.

[0005] Microscopic morphology methods, such as scanning electron microscopy cross-sectional analysis and microfocus CT tomography, can directly observe the spatial position of fibers at the micrometer scale. However, these methods have extremely small fields of view, and the time required for area-by-area imaging at the injection molding part level far exceeds the industrial production cycle. Furthermore, CT methods usually require destructive sampling of the product, which cannot meet the needs of online or near-line evaluation. Summary of the Invention

[0006] In order to effectively evaluate the uniformity of fiber distribution within an industrially acceptable timescale, this application provides a multimodal evaluation method for the dispersion of fiber distribution in robot injection molded parts.

[0007] This application provides a multimodal evaluation method for the dispersion of fiber distribution in robot injection molded parts, which adopts the following technical solution: A multimodal evaluation method for the dispersion of fiber distribution in robot injection molded parts includes: S1. Obtain the visible light color difference value of the surface of the injection molded sample containing conductive and non-conductive fibers; in response to the visible light color difference value being lower than the preset optical recognition lower limit, activate the multimodal scanning sequence; S2. Obtain the concentration information of the conductive fiber in the injection molded sample to be tested, and determine the detection method of the first mode detection in the multimodal scanning sequence based on the concentration information; S3. Perform the first modal detection and the second modal detection on the injection molded sample to be tested through the multimodal scanning sequence. Obtain a first feature quantity characterizing the total dispersion of the conductive fibers and the non-conductive fibers through the first modal detection, and obtain a second feature quantity characterizing the dispersion of the conductive fibers through the second modal detection. S4. Based on the first feature quantity and the second feature quantity, calculate the independent discrete index of the non-conductive fiber using a decoupling model; S5. Based on the independent discrete index and the second characteristic quantity, determine the distribution uniformity evaluation results of the non-conductive fiber and the conductive fiber, respectively.

[0008] Optionally, the visible light color difference value is the brightness value L* in the CIEL*a*b* color space; the preset optical recognition lower limit is when the brightness value L* is less than a preset brightness threshold.

[0009] Optionally, in response to the concentration information indicating that the mass fraction of the conductive fiber is less than or equal to a preset concentration threshold, the detection method for the first modal detection is terahertz time-domain spectroscopy detection, and obtaining the first feature quantity in S3 includes the following sub-steps: S301. A terahertz pulse is emitted to the injection molded sample to be tested, and the backscattered echo of the injection molded sample to be tested is obtained; S302. Extract the spatial distribution variance of the backscattered echo intensity within the target area of ​​the injection-molded sample to be tested, and use the spatial distribution variance as the first characteristic quantity.

[0010] Optionally, in response to the concentration information indicating that the mass fraction of the conductive fiber is greater than a preset concentration threshold, the detection method of the first modal detection is transient thermal wave mapping detection, and obtaining the first feature quantity in S3 includes the following sub-steps: S311. The injection-molded sample to be tested is subjected to pulsed thermal excitation; S312. Obtain the transient temperature decay curve of the surface of the injection molded sample to be tested; S313. Extract the spatial distribution features of the thermal response based on the transient temperature decay curve, and use the variance of the spatial distribution features of the thermal response as the first feature quantity.

[0011] Optionally, extracting the spatial distribution features of the thermal response in S313 includes: taking the second derivative of the transient temperature decay curve in the time dimension to obtain the peak arrival time matrix on the logarithmic time scale, and using the peak arrival time matrix as the spatial distribution features of the thermal response.

[0012] Optionally, the second modal detection is eddy current impedance detection, and obtaining the second characteristic quantity in step S3 includes the following sub-steps: S32. Obtain the impedance response value output by the eddy current sensor for the target area of ​​the injection molded sample to be tested; S33. Extract the maximum spatial gradient of the impedance response value within the target region, and use the maximum spatial gradient as the second feature quantity.

[0013] Optionally, step S32 includes the following sub-steps: S321. Obtain the initial impedance response value output by the eddy current sensor; S322. Obtain the real-time lift-off distance between the eddy current sensor and the surface of the injection molded sample to be tested; S323. Based on the real-time lift-off distance and eddy current attenuation coefficient, the initial impedance response value is exponentially attenuated to obtain the impedance response value.

[0014] Optionally, the decoupling model is a weighted difference model, and the independent discrete index is equal to the product of the first feature and the first calibration weight coefficient minus the product of the second feature and the second calibration weight coefficient.

[0015] Optionally, the following steps are included before S3 is executed: S6. Obtain injection mold flow analysis data for the target robot injection molded part; S7. Based on the shear rate distribution in the injection mold flow analysis data, extract several coordinate points whose shear rate gradient is greater than a preset gradient threshold as target feature nodes; The multimodal scanning sequence in S3 performs the first modality detection and the second modality detection based on the targeted feature node.

[0016] Optionally, the step S7 may be followed by the following steps: S71. Obtain the real-time process parameters of the current batch of injection molding machines; S72. Calculate the deviation between the real-time process parameters and the basic process parameters used to generate the injection mold flow analysis data; S73. In response to the deviation being greater than a preset process fluctuation threshold, the coordinates of the target feature node are expanded outward according to a preset diffusion radius to generate a compensation sampling grid; The multimodal scanning sequence in S3 performs the first modal detection and the second modal detection on the compensated sampling grid.

[0017] In summary, this application includes at least one of the following beneficial technical effects: 1. By combining visible light color difference recognition with multimodal scanning sequences, and introducing independent detection channels based on different physical responses into the multimodal scanning sequences, independent evaluation of hybrid fibers becomes possible. Since eddy current impedance only responds to the alternating electromagnetic induction of conductive fibers, while the first modal detection responds to both the dielectric and thermophysical properties of conductive and non-conductive fibers, a decoupling model is used to perform weighted differential calculations on the first and second characteristic quantities. This mathematically strips away the distribution state of non-conductive fibers, which was originally obscured by the black appearance and conductive shielding effect of conductive fibers, outputting an independent discrete index for non-conductive fibers and eliminating the physical blind spot of single optical detection under low brightness conditions.

[0018] 2. By acquiring the concentration information of conductive fibers and dynamically switching the detection mode of the first mode, the electromagnetic skin effect barrier caused by conductive fibers at high concentrations is overcome. In response to the conductive fiber mass fraction exceeding a preset concentration threshold, transient thermal wave mapping detection is used instead of terahertz time-domain spectroscopy detection. Utilizing the difference in thermal diffusivity between the fiber and the polymer matrix, the fiber distribution state is transformed into transient temperature decay characteristics. Furthermore, by calculating the second derivative of the transient temperature decay curve and extracting the peak arrival time matrix on a logarithmic time scale, the absolute radiation temperature amplitude affected by release agents or surface roughness is transformed into a time-dimensional feature independent of surface emissivity, eliminating spatial artifacts in the infrared thermometry process and improving the reliability of feature extraction in high-concentration conductive fiber scenarios.

[0019] 3. By introducing injection mold flow analysis data for the target robot injection molded parts and combining it with real-time process parameters for deviation calculation, the problem of mismatch between multimodal area array scanning time and injection molding production cycle time is solved. Target feature nodes are extracted based on shear rate gradient, transforming global blind scanning into local targeted scanning for areas with drastic flow field changes. In response to the deviation between real-time process parameters and basic process parameters exceeding the process fluctuation threshold, a compensation sampling grid is generated by expanding the coordinates of the target feature nodes according to the diffusion radius. This allows the detection system to adaptively cover fiber agglomeration areas that have undergone physical displacement due to fluctuations in parameters such as holding pressure or melt temperature, reducing overall scanning time while avoiding missed detection of local defects. Attached Figure Description

[0020] Figure 1 A flowchart illustrating a multimodal evaluation method for the dispersion of fiber distribution in robot injection molded parts according to an embodiment of the present invention is shown.

[0021] Figure 2 A flowchart illustrating two parallel detection channels in S3 of one embodiment of the present invention is shown. Detailed Implementation

[0022] The present application will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the scope of the application.

[0023] This application discloses a multimodal evaluation method for the discreteness of fiber distribution in robot injection molded parts. Before describing the embodiments of this application in detail, some terms will be explained first.

[0024] Conductive fibers refer to fiber-reinforced materials that can induce eddy currents in an alternating electromagnetic field. Examples of conductive fibers include chopped carbon fibers, long carbon fibers, nickel-plated carbon fibers, and stainless steel fibers. Non-conductive fibers refer to fiber-reinforced materials that do not generate eddy current responses in an alternating electromagnetic field. Examples of non-conductive fibers include E-glass fibers, S-glass fibers, aramid fibers, and basalt fibers. In the embodiments of this application, the polymer matrix of the injection-molded sample to be tested contains both conductive and non-conductive fibers. The injection-molded sample to be tested can be a robot structural part directly taken from the injection molding production line, or a standardized sheet prepared under laboratory conditions using a stepped mold.

[0025] The aggregation mechanisms of the two types of fibers in the matrix differ. Conductive fibers, due to their inherent conductivity, are prone to localized aggregation during injection molding due to electrostatic adsorption; the aggregation of non-conductive fibers is mainly influenced by the melt shear field and the fiber aspect ratio. Therefore, even if the overall fiber distribution index is within the acceptable range, there may be instances where one type of fiber is severely skewed and masked by the complementary distribution of the other. The method in this application aims to obtain independent evaluation results of the distribution uniformity of each type of fiber.

[0026] Figure 1 A general flowchart of a multimodal evaluation method according to some embodiments of this application is shown. Figure 1 As shown, the method includes steps S1 to S5. In step S1, the visible light color difference value of the surface of the injection-molded sample to be tested is obtained, and whether to activate the multimodal scanning sequence is determined based on whether the visible light color difference value is lower than the optical recognition lower limit. In step S2, the concentration information of conductive fibers in the injection-molded sample to be tested is obtained, and the detection mode of the first modal detection is determined based on the concentration information. In step S3, the first modal detection and the second modal detection are performed on the injection-molded sample to be tested, and the first feature quantity and the second feature quantity are obtained respectively. In step S4, the first feature quantity and the second feature quantity are calculated using a decoupling model to calculate the independent dispersion index of the non-conductive fibers. In step S5, the distribution uniformity evaluation results of the two types of fibers are output based on the independent dispersion index and the second feature quantity. The implementation method of the method is described in detail below with reference to each step.

[0027] Visible light color difference is a colorimetric parameter output by a colorimeter after optical measurements of the surface of an injection-molded sample. The colorimeter emits visible light onto the sample surface and receives the reflected light, acquiring the colorimetric coordinates of each measurement point on the sample surface through photoelectric conversion. When the fiber concentration in the sample is low and there is sufficient contrast between the fiber color and the matrix color, fiber agglomeration areas will exhibit identifiable colorimetric anomalies in the colorimeter's output. Color difference methods can effectively assess fiber distribution uniformity. The lower limit of optical recognition is the critical color difference value at which the spatial gradient resolution of the colorimeter degrades drastically. Below this value, the colorimetric difference between areas of uniform fiber distribution and areas of localized fiber agglomeration is less than the colorimeter's resolution, and color difference methods cannot output meaningful information about distribution uniformity.

[0028] For example, sample A has a visible light color difference value of 45, which is higher than the optical recognition limit. The color difference method can still distinguish between areas with uniform fiber distribution and areas of aggregation. In this case, the multimodal scanning sequence is not activated, and the color difference method is used directly for evaluation. Sample B has a visible light color difference value of 22, which is lower than the optical recognition limit. The color difference method fails, and the multimodal scanning sequence is activated to replace the color difference method for subsequent evaluation. Multimodal detection requires significantly more equipment complexity and scanning time than the color difference method. This graded triggering mechanism allows samples where the color difference method is still effective to be exempt from multimodal scanning, thereby reducing the overall consumption of detection resources.

[0029] The specific value of the lower limit of optical recognition can be determined through calibration experiments. The calibration method is as follows: take several sets of standard pieces with known fiber distribution (including uniform standard pieces and agglomerated standard pieces), measure them one by one on a colorimeter, record the trend of the color space gradient output by the colorimeter as a function of the color difference value, and determine the color difference value corresponding to the gradient resolution being reduced to the preset minimum acceptable level as the lower limit of optical recognition.

[0030] In some embodiments, the visible light color difference value is the lightness value L* in the CIE Lab color space, and the optical recognition lower limit is when the lightness value L* is less than a preset lightness threshold. In the CIE Lab color space, the L* axis represents lightness, ranging from 0 (pure black) to 100 (pure white), and the a* and b* axes represent the red-green and yellow-blue phase directions, respectively. The black color of carbon fiber mainly lowers the lightness of the sample surface rather than changing the hue; therefore, L* is the color channel most sensitive to carbon fiber concentration. For example, the lightness value L* of a sample containing 5% carbon fiber by mass is approximately 55, and the color difference method can still effectively distinguish uniform areas from agglomerated areas at this lightness level; the lightness value L* of a sample containing 12% carbon fiber by mass is approximately 25, which is lower than the preset lightness threshold of 30, causing the color difference method to fail and activating the multimodal scanning sequence. It should be understood that this application is not limited to the CIE Lab color space; similar determinations can also be made using the V channel (lightness channel) in the HSV color space, but the CIE Lab space is the most commonly used in industrial colorimeters.

[0031] In step S2, the concentration information of conductive fibers in the injection-molded sample to be tested is obtained, and the detection method of the first mode detection in the multimodal scanning sequence is determined based on the concentration information. The concentration information can be read from the nominal mass fraction in the injection molding formula database, or the fiber content can be calculated by weighing the sample and combining it with the density conversion relationship.

[0032] At different concentration ranges, the physical response sensitivity of the same detection principle to fiber distribution information may degrade or even fail. Specifically, when the mass fraction of conductive fibers is below a certain threshold, the conductive fibers have not yet formed a continuous conductive permeation network in the matrix, and electromagnetic waves can penetrate the sample and generate effective scattered signals. In this case, a detection method based on electromagnetic wave penetration can be selected. When the mass fraction of conductive fibers exceeds this threshold, the conductive fiber network forms a skin effect layer on the sample surface, and the penetration depth of electromagnetic waves decreases sharply. The detection method based on electromagnetic wave penetration can no longer obtain fiber distribution information inside the sample, and it is necessary to switch to another detection method that does not rely on electromagnetic wave penetration. Therefore, step S2 selects between the two detection methods based on the concentration information, so that the first modal detection can output a first feature quantity with clear physical meaning in a wide concentration range. The output of step S2 (the result of the detection method selection) is passed to step S3, and step S3 calls the corresponding detection hardware and signal processing flow according to the selection result.

[0033] Step S3 comprises two parallel detection channels. The first modal detection channel responds to all fibers (conductive and non-conductive fibers) in the injection-molded sample under test, and the first characteristic quantity output represents the total spatial dispersion of the two types of fibers. The second modal detection channel utilizes the eddy current effect to respond only to the conductive fibers, and the second characteristic quantity output represents the spatial dispersion of the conductive fibers. The outputs of both channels are simultaneously transmitted to step S4.

[0034] The following sections describe two implementation paths for first modality detection and implementation methods for second modality detection.

[0035] In some embodiments, in response to a concentration information indicating that the mass fraction of the conductive fiber is less than or equal to a preset concentration threshold, the detection method for the first mode detection is terahertz time-domain spectroscopy. The specific value of the preset concentration threshold depends on the type of conductive fiber and the dielectric properties of the matrix material. For example, for a composite system of chopped carbon fibers and a polyamide matrix, the preset concentration threshold can be set to 15% by mass fraction.

[0036] The basic principle of terahertz time-domain spectroscopy is as follows: When a terahertz pulse is incident on the injection-molded sample to be tested, the difference in dielectric constant between the fiber and the polymer matrix causes the terahertz wave to be scattered at the fiber-matrix interface, generating backscattered echoes. Regions with dense fibers have more scattering interfaces, resulting in higher intensity backscattered echoes; regions with sparse fibers have fewer scattering interfaces, leading to lower intensity backscattered echoes. Therefore, the spatial distribution of echo intensity can reflect the spatial distribution of fiber concentration.

[0037] Figure 2 A flowchart illustrating the acquisition of a first feature quantity by terahertz time-domain spectral detection according to some embodiments of this application is shown. Figure 2As shown, in response to the concentration information indicating that the mass fraction of the conductive fiber is less than or equal to a preset concentration threshold, obtaining the first feature quantity includes steps S301 and S302.

[0038] In step S301, a terahertz pulse is emitted to the injection molded sample to be tested, and the backscattered echo of the sample is obtained. The center frequency of the terahertz pulse can be selected in the range of 0.1THz to 3THz, and there is no limitation here.

[0039] In step S302, the spatial distribution variance of the backscattered echo intensity within the target area of ​​the injection-molded sample is extracted, and this spatial distribution variance is used as the first characteristic quantity. For example, echo intensity values ​​are collected at 100 measuring points (10×10) within a 40mm×40mm target area. When the fiber distribution is uniform, the echo intensity at each measuring point is concentrated around a certain mean, and the spatial distribution variance is small, for example, 0.03. When the fiber is locally agglomerated, the echo intensity at the measuring points within the agglomerated area is abnormally high, and the spatial distribution variance is large, for example, 0.41. The larger the variance, the more uneven the spatial distribution of the fibers.

[0040] In some embodiments, in response to a concentration information indicating that the mass fraction of the conductive fiber is greater than a preset concentration threshold, the detection method of the first mode detection is transient thermal wave mapping detection.

[0041] When the mass fraction of conductive fibers exceeds a preset concentration threshold, the conductive fiber network forms a skin effect layer on the sample surface. Under this condition, the penetration depth of terahertz waves shrinks to tens of micrometers above the surface, and the backscattered echo no longer carries information about the fiber distribution inside the sample, rendering terahertz time-domain spectroscopy detection ineffective. Transient thermal wave mapping detection uses the thermal diffusion process instead of electromagnetic wave penetration as the information carrier, bypassing the problem of limited electromagnetic wave penetration.

[0042] The physical principle of transient thermal imaging detection is as follows: a pulsed heat source applies a transient heat flow to the surface of the injection-molded sample, and the heat diffuses along the depth direction of the sample. The thermal conductivity of fibers is higher than that of the polymer matrix, and the equivalent thermal diffusivity of densely fiberd regions is greater than that of sparsely fiberd regions. Therefore, the surface temperature of densely fiberd regions decays faster after pulse excitation, while the surface temperature of sparsely fiberd regions decays more slowly. By acquiring temperature decay curves at various locations on the sample surface using an infrared thermal imager, fiber distribution information can be extracted from the spatial differences in decay rates.

[0043] The sub-step of obtaining the first characteristic quantity in response to the concentration information indicating that the mass fraction of the conductive fiber is greater than a preset concentration threshold includes S311-S313.

[0044] In step S311, the injection-molded sample to be tested is subjected to pulsed thermal excitation. The pulsed heat source can be a xenon lamp flash, a laser pulse, or a halogen lamp stepped heating; there are no restrictions here.

[0045] In step S312, the transient temperature decay curve of the surface of the injection-molded sample to be tested is obtained. The infrared thermal imager continuously acquires the temperature field sequence of the sample surface within a preset time window after pulsed thermal excitation, and extracts the temperature change curve of each pixel location over time, which is the transient temperature decay curve. For example, on a sample with a carbon fiber mass fraction of 20%, after flash pulse excitation, the surface temperature of the densely fiberd area decays to 40% of the initial temperature rise within 0.5 seconds, while the surface temperature of the sparsely fiberd area decays to only 70% of the initial temperature rise within 0.5 seconds.

[0046] In step S313, the spatial distribution features of the thermal response are extracted based on the transient temperature decay curve, and the variance of the spatial distribution features is used as the first feature quantity. The spatial distribution features of the thermal response can be a thermal diffusivity matrix or a peak arrival time matrix (further explained below). After extracting the spatial distribution features of the thermal response, the variance of the feature at each location within the target area is calculated. The larger the variance, the more uneven the fiber distribution.

[0047] Furthermore, in some embodiments, the method for extracting the spatial distribution features of the thermal response in step S313 is as follows: the second derivative of the transient temperature decay curve in the time dimension is calculated to obtain the peak arrival time matrix in the logarithmic time scale, and the peak arrival time matrix is ​​used as the spatial distribution features of the thermal response.

[0048] Infrared thermal imagers measure radiated temperature, not actual temperature. Radiated temperature equals actual temperature multiplied by surface emissivity. Factors such as residual release agent and differences in surface roughness cause fluctuations in emissivity at different locations on the sample surface, resulting in spatial artifacts unrelated to fiber distribution superimposed on the radiated temperature field. If the thermal diffusivity is directly extracted from the radiated temperature, the thermal diffusivity will include artifact contributions, reducing the reliability of the evaluation results.

[0049] Let T(t) be the true temperature decay curve at a certain measuring point, and let ε be the surface emissivity of that point, where ε is a position-dependent constant that does not change with time within the measurement time window. Let ε·T(t) be the radiation temperature collected by the infrared thermal imager at that point. Taking the first derivative of ε·T(t) with respect to time gives ε·T'(t), and taking the second derivative gives ε·T''(t). The time corresponding to the extreme points or zero-crossing points of T''(t) is then determined. hour, The value of depends only on the shape of the curve T(t) itself, and is not affected by the multiplicative constant ε. In other words, the peak arrival time It is an emissivity-independent quantity. For example, at the same measuring point, under two conditions of emissivity ε=0.85 and ε=0.92 (corresponding to two measurements with different amounts of residual release agent), the absolute amplitudes of the radiation temperature decay curves are different, but the extreme points of the second derivatives are at the same time. All values ​​are 0.38 seconds. Therefore, after constructing a matrix from the peak arrival times at various locations on the sample surface, each element of the matrix reflects only the local thermal diffusion characteristics at the corresponding location, and is not affected by the surface condition. It can be used as a reliable spatial distribution characteristic of the thermal response.

[0050] The above describes one implementation of step S313, namely, extracting the peak arrival time matrix using the second derivative. In other embodiments, if the spatial uniformity of the emissivity on the sample surface can be guaranteed (e.g., by coating the sample surface with a uniform high-emissivity coating before measurement), the thermal diffusivity at each location can be directly fitted from the transient temperature decay curve, and the thermal diffusivity matrix can be used as the spatial distribution feature of the thermal response. Both methods belong to the implementation of extracting the spatial distribution feature of the thermal response based on the transient temperature decay curve.

[0051] In some embodiments, the second modal detection is eddy current impedance detection. The basic principle of eddy current impedance detection is as follows: the excitation coil of the eddy current sensor generates an alternating magnetic field. After the alternating magnetic field penetrates the injection-molded sample under test, eddy currents are induced in the conductive fibers. The counter-magnetic field generated by the eddy currents acts on the sensor coil, changing the impedance of the coil. In areas with dense conductive fibers, the induced eddy current intensity is large, and the impedance change is large; in areas with sparse conductive fibers, the induced eddy current intensity is small, and the impedance change is small. Non-conductive fibers (such as glass fibers) do not induce eddy currents; therefore, the eddy current impedance response only reflects the local concentration and distribution of conductive fibers and is not affected by non-conductive fibers.

[0052] In the detection channel of the second modality, the second feature quantity is obtained through steps S32 and S33.

[0053] In step S32, the impedance response value output by the eddy current sensor for the target area of ​​the injection molded sample is acquired. The eddy current sensor scans the target area point by point, and outputs an impedance response value for each measuring point.

[0054] In step S33, the maximum spatial gradient of the impedance response value within the target region is extracted and used as the second feature. The maximum spatial gradient reflects the magnitude of the change in conductive fiber concentration at the location where the spatial variation is most drastic. For example, within the same 40mm × 40mm target region, the eddy current sensor scans point by point to acquire the spatial distribution map of the impedance response value. When the conductive fibers are uniformly distributed, the impedance value changes gently in space, and the maximum spatial gradient is small, for example, 0.12Ω / mm. When the conductive fibers are locally aggregated, the impedance value at the aggregation boundary shows a steep jump, and the maximum spatial gradient is large, for example, 1.85Ω / mm. The larger the gradient, the more drastic the spatial concentration change of the conductive fibers, i.e., the more severe the aggregation.

[0055] Furthermore, in some embodiments, the process of obtaining the impedance response value in step S32 includes steps S321-S323.

[0056] The surface of stepped sheet metal or actual injection molded parts exhibits variations in wall thickness and surface undulations, causing the distance between the eddy current sensor and the sample surface (lift-off distance) to be non-constant during the scanning process. Eddy current impedance is extremely sensitive to the lift-off distance; for every 0.1 mm increase in lift-off distance, the impedance response may decrease by several percentage points. If the effect of the lift-off distance is not compensated for, a systematic bias independent of fiber distribution will be introduced.

[0057] In step S321, the initial impedance response value output by the eddy current sensor is obtained. The initial impedance response value is the raw measurement value directly output by the eddy current sensor under the current lift-off distance conditions, without lift-off distance compensation.

[0058] In step S322, the real-time lift-off distance between the eddy current sensor and the surface of the injection-molded sample to be tested is obtained. The laser displacement sensor and the eddy current sensor are mounted on the same scanning probe, and the real-time lift-off distance is output simultaneously at each measuring point.

[0059] In step S323, based on the real-time lift-off distance and the eddy current attenuation coefficient, exponential attenuation compensation is performed on the initial impedance response value to obtain the impedance response value. The compensation method is to multiply the initial impedance response value by a natural exponential function value, with the product of the eddy current attenuation coefficient and the real-time lift-off distance as the exponent. The eddy current attenuation coefficient is related to the excitation frequency of the eddy current sensor. For example, the real-time lift-off distance of measuring point P1 is 0.5 mm, the initial impedance response value is 3.2 Ω, and the eddy current attenuation coefficient is 2.1 mm. -1Therefore, the compensated impedance response value is 3.2 × exp(2.1 × 0.5) ≈ 9.14 Ω. The real-time lift-off distance of measuring point P2 is 0.2 mm, and the initial impedance response value is 6.8 Ω. Therefore, the compensated impedance response value is 6.8 × exp(2.1 × 0.2) ≈ 10.35 Ω. In multi-frequency eddy current scenarios, each excitation frequency corresponds to a different eddy current attenuation coefficient, and compensation needs to be performed frequency-by-frequency.

[0060] The above describes one implementation of step S32, which involves synchronously measuring the real-time lift-off distance using a laser displacement sensor and performing exponential attenuation compensation. In other embodiments, a mechanical clamp can be used to fix the distance between the sample surface and the eddy current sensor to a constant value, thus eliminating the need for lift-off distance compensation. Both methods pertain to obtaining the impedance response value output by the eddy current sensor for the target region.

[0061] In step S4, the independent discrete index of the non-conductive fiber is calculated using a decoupling model based on the first and second characteristic quantities.

[0062] The first feature quantity includes the combined contribution of both conductive and non-conductive fibers to the detection signal, while the second feature quantity includes only the contribution of conductive fibers. If the first feature quantity is directly used to evaluate the uniformity of the non-conductive fiber distribution, the evaluation result will be affected by the distribution state of the conductive fibers. The decoupling model separates the contribution of conductive fibers from the first feature quantity, and the output is an independent discrete index that reflects only the distribution state of the non-conductive fibers.

[0063] In some embodiments, the decoupling model is a weighted difference model, where the independent discrete index is equal to the product of the first feature quantity and the first calibration weight coefficient minus the product of the second feature quantity and the second calibration weight coefficient. The first and second calibration weight coefficients are determined through calibration experiments: single-fiber standard parts containing only conductive fibers and single-fiber standard parts containing only non-conductive fibers are prepared respectively; first modal detection and second modal detection are performed on the two sets of standard parts respectively; and the weight coefficients are calculated back based on the response sensitivity coefficients of each detection mode to the single fiber type.

[0064] For example, if a sample has a first characteristic quantity V1=0.41, a second characteristic quantity V2=1.85Ω / mm, a first calibration weighting coefficient w1=1.0, and a second calibration weighting coefficient w2=0.18mm / Ω, then the independent dispersion index D of the non-conductive fiber is D=w1×V1-w2×V2=1.0×0.41-0.18×1.85=0.41-0.333=0.077.

[0065] The above describes one implementation of the decoupling model, namely the weighted difference model. In other embodiments, the decoupling model can also employ a matrix inversion model. The matrix inversion model constructs a two-dimensional observation vector from the first and second eigenvalues. By multiplying the observation vector by the inverse of a pre-calibrated second-order response matrix (each element of which represents the response coefficient of each detection mode to each fiber), the discrete indexes of both conductive and non-conductive fibers are simultaneously solved. Compared to the weighted difference model, the matrix inversion model can handle cases where both detection modes exhibit cross-responses to both fibers, making it suitable for multi-component systems with more complex fiber types. Both methods belong to the implementation of calculating the independent discrete indexes of non-conductive fibers using a decoupling model based on the first and second eigenvalues.

[0066] In step S5, the distribution uniformity evaluation results of non-conductive fibers and conductive fibers are determined based on the independent discrete index and the second characteristic quantity, respectively.

[0067] The distribution uniformity assessment result can be a binary judgment of pass or fail, or a continuous uniformity score. Taking binary judgment as an example: the independent dispersion index is compared with a preset non-conductive fiber uniformity threshold, and the second characteristic quantity is compared with a preset conductive fiber uniformity threshold. For example, if the non-conductive fiber uniformity threshold is set to 0.15, and the independent dispersion index D = 0.077, which is less than 0.15, the distribution uniformity of the non-conductive fiber is judged to be passable. If the conductive fiber uniformity threshold is set to 2.0 Ω / mm, and the second characteristic quantity V2 = 1.85 Ω / mm, which is less than 2.0 Ω / mm, the distribution uniformity of the conductive fiber is judged to be passable.

[0068] The uniformity threshold is calibrated as follows: Take several sets of standard pieces with known fiber distribution states (including uniform standard pieces and agglomerated standard pieces), perform a complete multimodal evaluation process on each standard piece to obtain the independent discrete index and the second feature quantity, and use the ROC curve to determine the discrimination threshold that maximizes the classification accuracy.

[0069] In some embodiments, the distribution uniformity assessment results may also be in the form of continuous scores. For example, the independent dispersion index can be mapped to a score range of 0 to 100, where 0 points correspond to the least uniform distribution and 100 points correspond to perfect uniformity. The mapping function can be a linear function or a nonlinear function fitted based on experimental data of standard parts; there is no limitation here.

[0070] In some embodiments, steps S6 and S7 are included before step S3, for determining the target region to be scanned before the multimodal scanning sequence is executed.

[0071] Performing multimodal point-by-point scanning on the entire surface of an injection-molded part takes far longer than the injection molding production cycle time. For example, a 200mm × 150mm injection-molded part surface divided at 1mm intervals can yield 30,000 potential measurement points, and performing multimodal scanning on all points could take approximately 6 hours. However, the typical injection molding production cycle time is tens of seconds to several minutes, making a full inspection solution impractical for production lines. Targeted sparse sampling, by pre-locating high-risk areas for fiber aggregation using mold flow analysis data, scans only a small number of nodes within these high-risk areas, significantly reducing scanning time.

[0072] In step S6, injection mold flow analysis data for the target robot injection molded part is obtained. The injection mold flow analysis data is obtained by simulating the mold geometry and process parameters before injection molding using commercial mold flow analysis software (such as Moldflow), including the spatial distribution of the melt flow velocity field, temperature field, pressure field, and shear rate in the cavity.

[0073] In step S7, based on the shear rate distribution in the injection molding flow analysis data, several coordinate points with shear rate gradients greater than a preset gradient threshold are extracted as targeted feature nodes. During injection molding, fiber agglomeration is most likely to occur in regions with large melt shear rate gradients. At locations with large shear rate gradients, the velocity difference between the high-shear and low-shear sides of the melt leads to asymmetric fiber migration speeds perpendicular to the flow direction. Fibers converge and accumulate from the high-shear region to the low-shear region, forming localized agglomerations.

[0074] For example, in the mold flow analysis results of the 200mm×150mm injection molded part surface mentioned above, the shear rate gradient is greater than 500s. -1 There are 47 coordinate points per mm. Using these 47 coordinate points as target feature nodes, the multimodal scanning sequence performs first and second modal detection based solely on these 47 target feature nodes. The number of scan points is reduced from 30,000 to 47, and the scanning time is shortened from approximately 6 hours to approximately 35 seconds, which can match the injection molding production cycle.

[0075] Furthermore, in some embodiments, step S7 is followed by steps S71, S72 and S73, which are used to compensate for the position of the target feature node when the actual process parameters deviate from the nominal parameters used in the mold flow analysis.

[0076] Mold flow analysis uses nominal process parameters (basic process parameters), but parameters such as holding pressure, melt temperature, injection speed, and mold temperature during actual injection molding vary from batch to batch. These fluctuations in process parameters cause changes in the melt flow field, resulting in actual agglomeration locations deviating from the positions predicted by the mold flow analysis. If the targeted feature nodes are not compensated for, the agglomeration areas may drift outside the coverage area of ​​the targeted nodes, leading to missed detections.

[0077] In step S71, the real-time process parameters of the current batch of injection molding machine are obtained, including holding pressure, melt temperature, injection speed and mold temperature.

[0078] In step S72, the deviation between the real-time process parameters and the base process parameters used to generate injection mold flow analysis data is calculated. The deviation can be calculated using either the independent comparison threshold method or the normalized Euclidean distance method. The independent comparison threshold method calculates the absolute deviation between the real-time value and the base value for each process parameter; if the absolute deviation of any parameter exceeds its own tolerance, the deviation is considered excessive. The normalized Euclidean distance method normalizes the real-time and base values ​​of all process parameters to the [0,1] interval and calculates the Euclidean distance between the normalized vectors as the deviation. For example, if the base process parameters are a holding pressure of 80 MPa and a melt temperature of 280°C, and the real-time process parameters are a holding pressure of 73 MPa and a melt temperature of 285°C, the deviation calculated using the normalized Euclidean distance method is 0.12.

[0079] In step S73, in response to a deviation exceeding a preset process fluctuation threshold, the coordinates of the target feature nodes are expanded outward according to a preset diffusion radius to generate a compensation sampling grid. For example, if the preset process fluctuation threshold is 0.08, and the deviation of 0.12 is greater than 0.08, compensation is triggered. The diffusion radius is set to 3 mm, and annular regions with a radius of 3 mm are expanded outward for each of the 47 target feature nodes. All expanded regions are merged and discretized into a sampling grid, resulting in a compensation sampling grid containing approximately 120 measurement points. The multimodal scanning sequence performs first and second modal detection on the compensation sampling grid.

[0080] It should be understood that when the deviation does not exceed the process fluctuation threshold, no compensation sampling grid is generated, and the multimodal scanning sequence still performs detection based on the original target feature nodes. The specific value of the diffusion radius can be set according to the historical statistical relationship between the process parameter deviation and the agglomeration position drift, and is not limited here.

[0081] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0082] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0083] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for multi-modal evaluation of the degree of fiber distribution dispersion of a robotic injection molded part, characterized in that, include: S1. Obtain the visible light color difference value of the surface of the injection molded sample containing conductive and non-conductive fibers; In response to the visible light color difference value being lower than a preset optical recognition lower limit, a multimodal scanning sequence is activated; S2. Obtain the concentration information of the conductive fiber in the injection molded sample to be tested, and determine the detection method of the first mode detection in the multimodal scanning sequence based on the concentration information; S3. Perform the first modal detection and the second modal detection on the injection molded sample to be tested through the multimodal scanning sequence. Obtain a first feature quantity characterizing the total dispersion of the conductive fibers and the non-conductive fibers through the first modal detection, and obtain a second feature quantity characterizing the dispersion of the conductive fibers through the second modal detection. S4. Based on the first feature quantity and the second feature quantity, calculate the independent discrete index of the non-conductive fiber using a decoupling model; S5. Based on the independent discrete index and the second characteristic quantity, determine the distribution uniformity evaluation results of the non-conductive fiber and the conductive fiber, respectively.

2. The multi-modal assessment method of claim 1, wherein, The visible light color difference value is the lightness value L* in the CIE Lab color space; the preset optical recognition lower limit is when the lightness value L* is less than the preset lightness threshold.

3. The multimodal evaluation method according to claim 2, characterized in that, In response to the concentration information indicating that the mass fraction of the conductive fiber is less than or equal to a preset concentration threshold, the detection method of the first modal detection is terahertz time-domain spectroscopy detection, and obtaining the first feature quantity in S3 includes the following sub-steps: S301. A terahertz pulse is emitted to the injection molded sample to be tested, and the backscattered echo of the injection molded sample to be tested is obtained; S302. Extract the spatial distribution variance of the backscattered echo intensity within the target area of ​​the injection-molded sample to be tested, and use the spatial distribution variance as the first characteristic quantity.

4. The multimodal evaluation method according to claim 3, characterized in that, In response to the concentration information indicating that the mass fraction of the conductive fiber is greater than a preset concentration threshold, the detection method of the first modal detection is transient thermal wave mapping detection, and obtaining the first feature quantity in S3 includes the following sub-steps: S311. The injection-molded sample to be tested is subjected to pulsed thermal excitation; S312. Obtain the transient temperature decay curve of the surface of the injection molded sample to be tested; S313. Extract the spatial distribution features of the thermal response based on the transient temperature decay curve, and use the variance of the spatial distribution features of the thermal response as the first feature quantity.

5. The multimodal evaluation method according to claim 4, characterized in that, Extracting the spatial distribution features of the thermal response in S313 includes: taking the second derivative of the transient temperature decay curve in the time dimension to obtain the peak arrival time matrix in the logarithmic time scale, and using the peak arrival time matrix as the spatial distribution features of the thermal response.

6. The multimodal evaluation method according to claim 5, characterized in that, The second modal detection is eddy current impedance detection, and obtaining the second characteristic quantity in step S3 includes the following sub-steps: S32. Obtain the impedance response value output by the eddy current sensor for the target area of ​​the injection molded sample to be tested; S33. Extract the maximum spatial gradient of the impedance response value within the target region, and use the maximum spatial gradient as the second feature quantity.

7. The multimodal evaluation method according to claim 6, characterized in that, S32 includes the following sub-steps: S321. Obtain the initial impedance response value output by the eddy current sensor; S322. Obtain the real-time lift-off distance between the eddy current sensor and the surface of the injection molded sample to be tested; S323. Based on the real-time lift-off distance and eddy current attenuation coefficient, the initial impedance response value is exponentially attenuated to obtain the impedance response value.

8. The multimodal evaluation method according to claim 7, characterized in that, The decoupling model is a weighted difference model, and the independent discrete index is equal to the product of the first feature and the first calibration weight coefficient minus the product of the second feature and the second calibration weight coefficient.

9. The multimodal evaluation method according to claim 8, characterized in that, The following steps are included before S3 is executed: S6. Obtain injection mold flow analysis data for the target robot injection molded part; S7. Based on the shear rate distribution in the injection mold flow analysis data, extract several coordinate points whose shear rate gradient is greater than a preset gradient threshold as target feature nodes; The multimodal scanning sequence in S3 performs the first modality detection and the second modality detection based on the targeted feature node.

10. The multimodal evaluation method according to claim 9, characterized in that, The step following S7 is: S71. Obtain the real-time process parameters of the current batch of injection molding machines; S72. Calculate the deviation between the real-time process parameters and the basic process parameters used to generate the injection mold flow analysis data; S73. In response to the deviation being greater than a preset process fluctuation threshold, the coordinates of the target feature node are expanded outward according to a preset diffusion radius to generate a compensation sampling grid; The multimodal scanning sequence in S3 performs the first modal detection and the second modal detection on the compensated sampling grid.

Images