Knitted fabric defect real-time identification and production regulation and control system based on AI visual inspection

Abstract

The invention discloses a knitted fabric defect real-time identification and production regulation and control system based on AI visual inspection, and relates to the field of knitted fabric production, and the system comprises an acquisition module which is used for adjusting the posture and operation parameters of a camera to enable the camera to adapt to the width of a knitted fabric and the operation speed of production equipment, collecting a surface image in the production process of the knitted fabric; the equalization module is used for carrying out noise filtering and illumination equalization processing on the acquired fabric surface image; according to the method, the width of the knitted fabric and the running speed of production equipment are precisely adapted, the image acquisition quality is optimized, yarn texture details are clearly presented through targeted noise filtering and illumination equalization processing, and the defect position is precisely positioned and the type is judged by means of a yarn track capturing and fitting algorithm, so that the defect detection accuracy is improved. And the severity of the flaws is quantified based on multi-dimensional indexes, so that the evaluation result is more scientific.

Description

Technical Field

[0001] This invention relates to the field of knitted fabric production technology, specifically to a real-time defect identification and production control system for knitted fabrics based on AI visual inspection. Background Technology

[0002] Knitted fabric production uses yarn as raw material. The yarn is formed into loops and interlocked with each other through a knitting machine to complete warp knitting or weft knitting.

[0003] Patent application No. 202411764182.X discloses a machine vision-based method and system for detecting defects in knitted fabrics. This application aims to address the problem that "currently, many studies have proposed defect detection methods based on edge detection, image segmentation, and image recognition technologies. However, these methods usually rely on the analysis of static images and still have limitations in detecting complex shapes and small defects."

[0004] However, most existing technologies focus on defect identification or intelligent control of knitted fabrics as design improvement directions, but there is no technology that combines the two and has the functions of both.

[0005] To address this, we propose a real-time defect identification and production control system for knitted fabrics based on AI visual inspection. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a real-time defect identification and production control system for knitted fabrics based on AI visual inspection, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a real-time defect identification and production control system for knitted fabrics based on AI visual inspection, comprising: The acquisition module adjusts the camera's posture and operating parameters to adapt the camera to the width of the knitted fabric and the operating speed of the production equipment, acquiring surface images of the knitted fabric during the production process. The equalization module performs noise filtering and illumination equalization on the acquired fabric surface images. The recognition module captures the yarn trajectory in the fabric area of ​​the image after noise filtering and illumination equalization, determining the location and type of defects in the knitted fabric based on the trajectory offset. The integration module receives the defect location and type output from the recognition module and generates a production report containing the defect location, type, and severity. The analysis module iterates through the production report, uploads fabric production indicators, and quantifies the matching degree between defect location, type, and equipment parameters by constructing a parameter correlation model, determining the items and directions for adjustment among the current knitting production equipment parameters such as needle pitch, tension, and yarn feed speed. The execution module receives the analysis results from the analysis module, generates corresponding adjustment signals based on the analysis results, and transmits the signals to the corresponding execution unit of the equipment in real time to complete equipment control. The acquisition module is interconnected with a balancing module via a wireless network. The balancing module is interconnected with a recognition module via a wireless network. The recognition module is interconnected with an integration module via a wireless network. The integration module is interconnected with an analysis module via a wireless network. The analysis module is interconnected with an execution module via a wireless network.

[0008] Furthermore, the adjustment of the camera's posture in the acquisition module includes the adaptation adjustment of the field of view: ; In the formula: To adjust the rear field of view; For the width of the knitted fabric; The vertical distance between the camera lens and the surface of the knitted fabric; The adjustment of the camera's operating parameters in the acquisition module includes the adaptation adjustment of the acquisition frame rate: ; In the formula: To adjust the capture frame rate; The fabric production feed speed on the production equipment; The preset yarn texture sampling coefficient; This is the actual length coefficient corresponding to the pixels in a single frame of image, with the unit being length / pixel; This represents the minimum resolution pixel spacing of the yarn texture in a single frame image.

[0009] Furthermore, when the equalization module performs noise filtering on the acquired fabric surface image, it employs adaptive median filtering based on the yarn texture direction, with the size of the filtering window... ; In the formula: The preset texture adaptation coefficient; The mean gradient of the yarn texture direction within the neighborhood of the current pixel; The standard deviation of the texture gradient within the neighborhood of the current pixel; During illumination equalization, the image is divided into several sub-regions based on the yarn texture area, and the illumination compensation value for each sub-region is... ,in This represents the average grayscale value of the entire image. This represents the average gray value of the i-th sub-region. This represents the texture complexity coefficient of the sub-region. This represents the basic compensation threshold.

[0010] Furthermore, when the recognition module captures the yarn trajectory of the fabric area in the image, it calculates the gray-level gradient value of each pixel in the image in the direction of yarn texture extension, extracts the local peak points of the gray-level gradient value in each yarn texture area as initial feature points, and then performs gradient consistency verification of the initial feature points in a 3×3 neighborhood, and removes noise points in the neighborhood whose gradient direction deviation exceeds a preset angle threshold, thereby obtaining an effective feature point set of the yarn trajectory. The number N of valid feature points is counted, according to Determine the order m of the Bézier curve, where m is a positive integer; Using the coordinates of the effective feature points as constraint points, the m-th order Bézier curve is fitted using the least squares method to obtain the actual yarn trajectory. ; Finally, calculate the mean Euclidean distance between the fitted Bézier curve and each effective feature point. If the mean value is greater than the preset fitting error threshold, extract 2 to 3 new feature points on the side of the feature point with the largest distance, and repeat the above steps until the value does not exceed the preset fitting error threshold. The trajectory offset ; In the formula: This represents the x-axis coordinate value corresponding to the actual yarn trajectory at parameter t; This represents the y-axis coordinate value of the actual yarn trajectory at parameter t; This represents the x-axis coordinate value of the standard yarn trajectory at parameter t. This represents the y-axis coordinate value of the standard yarn trajectory at parameter t. This is the parameter range corresponding to the yarn segment; when If the offset exceeds a preset threshold, the corresponding area is determined to have defects.

[0011] Furthermore, when the identification module determines the type of defect in the knitted fabric based on the trajectory offset, it follows the following rules: Within the same yarn segment When the mean is not equal to zero, the fluctuation value ; like If the defect falls within the first preset range, the defect type is determined to be loose yarn. like Within the second preset range and If the average value is greater than the preset offset threshold, the defect type is determined to be yarn misweaving; like If a sudden change occurs and the pixel grayscale variance of the corresponding area is greater than the preset variance threshold, the defect type is determined to be yarn breakage.

[0012] Furthermore, the severity of the situation in the production report is quantified using the following formula: ; In the formula: The severity of the defect; This is the defect area weighting coefficient; This represents the actual area of ​​the defective region. The area of ​​the fabric region corresponding to the currently acquired image; This is the quantified value corresponding to the defect type; This refers to the weighting coefficient for the defect type. This represents the weighting coefficient for the defect location. This represents the influence coefficient of the defect location.

[0013] Furthermore, during the operation phase of the analysis module, a multivariate coupled correlation model is constructed using the needle pitch, tension, and yarn feed speed of the knitting production equipment as input variables and the mean of the defect severity as the output variable. ; In the formula: This represents the average severity of the defects. , , These are the quadratic coefficient of needle pitch, the coupling coefficient between tension and yarn feed speed, and the ratio coefficient between needle pitch and tension. The needle spacing; For tension; For yarn feeding speed; For model residuals; When quantifying the matching degree using an association model, the matching degree... ,in This indicates the maximum permissible value for the severity of the defect. When M is lower than the preset matching threshold, the device parameters that need to be adjusted are determined. The adjustment direction is determined by calculating the average value of each parameter to be adjusted in relation to the severity of the defect. The partial derivative of a parameter is such that if the partial derivative is positive, the parameter is adjusted in the direction of decreasing; if the partial derivative is negative, the parameter is adjusted in the direction of increasing.

[0014] Furthermore, the adjustment signal generated in the execution module includes the adjustment amplitude and the adjustment response time: ; In the formula: To adjust the amplitude and response time; The target matching degree is preset; This represents the current matching degree. The actual values ​​of the current equipment parameters to be adjusted; The gain coefficient is used for parameter adjustment; The upper limit of the parameter adjustment rate of the device execution unit, and when When the response time is less than the minimum response time of the device execution unit, the minimum response time of the device execution unit shall be used as the reference. application.

[0015] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention provides a real-time defect identification and production control system for knitted fabrics based on AI visual inspection. During operation, the system optimizes image acquisition quality by precisely adapting to the width of the knitted fabric and the operating speed of the production equipment. Through targeted noise filtering and illumination equalization, it clearly presents the details of the yarn texture. With the help of yarn trajectory capture and fitting algorithms, it accurately locates the defect location and determines the type. Based on multi-dimensional indicators, it quantifies the severity of the defect, making the evaluation results more scientific. By constructing a parameter correlation model, the matching degree between defects and equipment parameters is quantified, the adjustment direction and amplitude are clarified, and real-time precise control of production equipment is achieved. The closed-loop optimization mechanism continuously corrects parameters to ensure that the severity of defects is within a safe range, effectively improving the production quality and yield of knitted fabrics, reducing resource consumption, increasing production efficiency, adapting to the production needs of knitted fabrics of different grades and textures, and enhancing the stability and controllability of the production process. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0017] Figure 1 This is a schematic diagram of a system for real-time identification and production control of defects in knitted fabrics based on AI visual inspection. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] The present invention will be further described below with reference to embodiments. Example:

[0020] This embodiment presents a real-time defect identification and production control system for knitted fabrics based on AI visual inspection, such as... Figure 1 As shown, it includes: The acquisition module is used to adjust the camera's posture and operating parameters to adapt the camera to the width of the knitted fabric and the operating speed of the production equipment, and to acquire surface images of the knitted fabric during the production process. The camera pose adjustment in the acquisition module includes the adaptation adjustment of the field of view: ; In the formula: To adjust the rear field of view; For the width of the knitted fabric; The vertical distance between the camera lens and the surface of the knitted fabric; The above formula combines geometric imaging logic, with the width of the knitted fabric and the vertical distance between the camera and the fabric as the core parameters. It accurately calculates the field of view through trigonometric function relationships to ensure that the camera's shooting range can fully cover the width of the fabric. This avoids the loss of fabric edge information due to insufficient shooting range and also eliminates the waste of resources caused by range redundancy. It achieves dynamic adaptation of the field of view to knitted fabrics of different widths and can adapt to various production scenarios without manual intervention. The camera posture adjustment in the acquisition module includes focal length adjustment: by automatically adjusting the focal length, the two side boundaries of the knitted fabric are aligned with the two side boundaries of the knitted fabric area image in the surface image. Adjusting the camera's operating parameters in the acquisition module includes adapting the acquisition frame rate: ; In the formula: To adjust the capture frame rate; The fabric production feed speed on the production equipment; The preset yarn texture sampling coefficient; This is the actual length coefficient corresponding to the pixels in a single frame of image, with the unit being length / pixel; The minimum resolvable pixel spacing of the yarn texture in a single frame image; The above formula constructs the frame rate calculation logic around key factors such as production feed speed, yarn texture sampling requirements, actual pixel length and minimum resolution pixel spacing. It flexibly controls the number of frames collected per unit length of yarn through preset yarn texture sampling coefficients, and at the same time uses pixel correlation coefficients to ensure sampling resolution, so that even if the production speed changes, the yarn texture information can be completely captured, providing sufficient and high-quality image data support for subsequent defect identification. in, Used to characterize the number of frames that need to be collected per unit length of yarn; The equalization module is used to perform noise filtering and illumination equalization processing on the acquired fabric surface images. When the equalization module performs noise filtering on the acquired fabric surface image, it uses adaptive median filtering based on the yarn texture direction. The size of the filtering window is... ; The above formula combines the preset texture adaptation coefficient, the mean gradient of the yarn texture direction in the neighborhood of the current pixel, and the standard deviation of the texture gradient to dynamically determine the window size of the adaptive median filter. The texture adaptation coefficient is flexibly selected according to the consistency of the yarn texture direction, so that the window size accurately matches the texture complexity. While effectively filtering out image noise, it retains the detailed features of the yarn texture to the greatest extent and avoids the texture blurring or noise residue problems caused by traditional fixed window filtering. In the formula: The preset texture adaptation coefficient; The mean gradient of the yarn texture direction within the neighborhood of the current pixel; The standard deviation of the texture gradient within the neighborhood of the current pixel; During illumination equalization, the image is divided into several sub-regions based on the yarn texture area, and the illumination compensation value for each sub-region is... ,in This represents the average grayscale value of the entire image. This represents the average gray value of the i-th sub-region. This represents the texture complexity coefficient of the sub-region. Indicates the basic compensation threshold; The above formula is based on the difference between the average gray value of the entire image and the average gray value of the sub-region. It combines the texture complexity coefficient of the sub-region and the basic compensation threshold to calculate the illumination compensation amount. The texture complexity coefficient is adjusted according to the yarn texture interlacing density and detail richness of the sub-region. The basic compensation threshold is adapted to the overall image gray value based on the standard reference gray value. This enables fine-grained illumination adjustment for sub-regions with different texture complexities, effectively balancing the overall image illumination and eliminating the interference of uneven illumination on yarn trajectory capture. in, The preset value range is [0.5, 2]. The value is larger when the yarn texture direction consistency is lower, and smaller when the yarn texture direction consistency is higher. The preset value range is [0.3, 1.5]. The value is larger when the interlacing density of the yarn texture in the sub-region is higher and the details are richer, and the value is smaller when the interlacing density of the yarn texture in the sub-region is lower and the details are simpler. ∈[10, 30], taking the standard reference gray level corresponding to the knitted fabric as the benchmark, when the overall average gray level of the image is lower than the standard reference gray level, the larger value in this range is taken, and when the overall average gray level of the image is higher than the standard reference gray level, the smaller value in this range is taken. The recognition module is used to capture the yarn trajectory of the fabric area in the image after noise filtering and illumination equalization, and to determine the location and type of defects in the knitted fabric by the trajectory offset. When the recognition module captures the yarn trajectory of the fabric area in the image, it calculates the gray-level gradient value of each pixel in the image in the direction of yarn texture extension, extracts the local peak points of gray-level gradient values ​​in each yarn texture area as initial feature points, and then performs gradient consistency verification on the initial feature points in a 3×3 neighborhood, and removes noise points in the neighborhood whose gradient direction deviation exceeds a preset angle threshold, thus obtaining the effective feature point set of the yarn trajectory. The number N of valid feature points is counted, according to Determine the order m of the Bézier curve, where m is a positive integer; Using the coordinates of the effective feature points as constraint points, the m-th order Bézier curve is fitted using the least squares method to obtain the actual yarn trajectory. t∈[0,1] are curve parameters; Finally, calculate the mean Euclidean distance between the fitted Bézier curve and each effective feature point. If the mean value is greater than the preset fitting error threshold, extract 2 to 3 new feature points on the side of the feature point with the largest distance, and repeat the above steps until the value does not exceed the preset fitting error threshold. trajectory offset ; In the formula: Let x be the x-axis coordinate value of the actual yarn trajectory at parameter t, where t∈[0,1], and t is the parameter of the Bézier curve. This represents the y-axis coordinate value of the actual yarn trajectory at parameter t; The x-axis coordinate value at parameter t is the standard yarn trajectory (i.e., the flawless ideal yarn trajectory corresponding to this knitted fabric, which is a Bézier curve of the same order as the actual yarn trajectory); This represents the y-axis coordinate value of the standard yarn trajectory at parameter t. This is the parameter range corresponding to the yarn segment; The above formula calculates the Euclidean distance integral between the actual yarn trajectory and the standard flawless yarn trajectory within the corresponding parameter range, comprehensively quantifies the overall offset of the yarn trajectory, abandons the deviation judgment of a single pixel or local line segment, fully considers the continuity characteristics of the yarn trajectory, accurately reflects the cumulative offset of the yarn during the extension process, and provides an objective and comprehensive quantitative basis for the location and type determination of defects. when If the offset exceeds a preset threshold, the corresponding area is determined to have defects. When the identification module determines the type of defect in knitted fabric based on trajectory offset, it follows the following rules: Within the same yarn segment When the mean is not equal to zero, the fluctuation value ; like If the defect falls within the first preset range, the defect type is determined to be loose yarn. like Within the second preset range and If the average value is greater than the preset offset threshold, the defect type is determined to be yarn misweaving; like If a sudden change occurs and the pixel grayscale variance of the corresponding area is greater than the preset variance threshold, the defect type is determined to be yarn breakage. Among them, mutation determination means that the mutation value is greater than the preset mutation threshold, and the numerical range of the second preset interval is greater than that of the first preset interval. The integration module receives the defect location and defect type output by the identification module and generates a production report containing the defect location, defect type, and severity. The severity of a production report is quantified using the following formula: ; In the formula: The severity of the defect; This is the defect area weighting coefficient; This represents the actual area of ​​the defective region. The area of ​​the fabric region corresponding to the currently acquired image; This is the quantified value corresponding to the defect type; This refers to the weighting coefficient for the defect type. This represents the weighting coefficient for the defect location. The influence coefficient of the defect location; The above formula integrates three core factors: the actual area ratio of the defective area, the risk level of the defect type, and the impact of the defect location. It achieves multi-dimensional weighted quantification, where the weight coefficients are flexibly adjusted according to the fabric category, defect risk level, and core location. The type quantification value distinguishes between high-risk and low-risk defects, and the location impact coefficient clarifies the difference between the core area and the edge area, thereby reflecting the degree of impact of different defects on product quality. in, , , All are positive numbers, and their sum is 1. ∈[0.2, 0.5], the larger the value is when the knitted fabric is a high-end fine product, and the smaller the value is when the fabric is a rough texture product; ∈[0.3, 0.6], the larger the value is when the defect type belongs to the high-risk category that directly affects the strength of the fabric, and the smaller the value is when the defect type belongs to the low-risk category that only slightly affects the appearance; ∈[0.1, 0.3], the value is larger when the defect is located in the core area of ​​the fabric, and smaller when the defect is located in the edge area of ​​the fabric (non-core area). The values ​​are integers from 1 to 5. High-risk defects such as broken yarn and mis-weaving that directly affect the functionality of the fabric correspond to values ​​of 4 to 5, while low-risk defects such as loose yarn that only slightly affect the appearance correspond to values ​​of 1 to 2. The value ranges from 0.5 to 1. When the defect is located in the core usable area of ​​the fabric, such as the front of the main body of the garment, the value is 1. When it is located in the non-core area of ​​the fabric edge, such as the side seam or hem of the garment, the value is 0.5. The analysis module is used to traverse production reports, upload fabric production indicators, and quantify the matching degree between defect location, defect type and equipment parameters by building a parameter correlation model. It determines the items and directions for adjustment among the current knitting production equipment parameters such as needle pitch, tension and yarn feed speed. During the analysis module's operation phase, a multivariate coupled correlation model is constructed using the needle pitch, tension, and yarn feed speed of the knitting production equipment as input variables and the mean of defect severity as the output variable. ; In the formula: This represents the average severity of the defects. , , These are the quadratic coefficient of needle pitch, the coupling coefficient between tension and yarn feed speed, and the ratio coefficient between needle pitch and tension. The needle spacing; For tension; For yarn feeding speed; For model residuals; The above formula fully considers the complex relationship between the three key parameters of needle pitch, tension, and yarn feed speed in knitting production equipment and the average severity of defects. It introduces a quadratic term for needle pitch, a coupling term for tension and yarn feed speed, and a ratio term for needle pitch and tension. By fitting actual production data, the correlation coefficients are obtained, which accurately depict the comprehensive impact of the interaction of multiple parameters on defects and provide a model support for adjusting equipment parameters that fits the actual production scenario. When quantifying the matching degree using an association model, the matching degree... ,in This indicates the maximum permissible value for the severity of the defect. When M is lower than the preset matching threshold, the device parameters that need to be adjusted are determined. The method for determining the adjustment direction is as follows: calculate the average value of each parameter to be adjusted in relation to the severity of the defect. The partial derivatives of a parameter are such that if the partial derivative is positive, the parameter is adjusted in the direction of decreasing; if the partial derivative is negative, the parameter is adjusted in the direction of increasing. in, ∈[0.02,0.15] represents the average severity of defects in knitted fabrics collected from several groups with different stitch lengths. With the square of the stitch length as the independent variable, As the dependent variable, the correlation coefficient corresponding to the quadratic term of the needle spacing is obtained by fitting using the least squares method. ∈[0.008, 0.05] represents the average severity of defects in knitted fabrics obtained by collecting data under several combinations of different tension and yarn feed speeds. With the product of tension and yarn feed speed as the independent variable, The dependent variable is the correlation coefficient corresponding to the coupling term obtained by fitting multiple linear regression. ∈[0.1, 0.8] represents the average severity of defects in knitted fabrics collected under several different stitch lengths and tensions. The ratio of needle pitch to tension is used as the independent variable. The correlation coefficients corresponding to the ratio terms obtained by fitting the least squares method with the dependent variable as the variable. ∈[0, ], The preset residual threshold is set to a larger value when the actual coupling relationship between equipment parameters and the severity of defects deviates more significantly from the model fitting relationship, and a smaller value when the actual coupling relationship fits the model fitting relationship better. The execution module is used to receive the analysis results from the analysis module, generate corresponding adjustment signals based on the analysis results, and transmit the signals to the corresponding execution unit of the equipment in real time to complete the equipment control. The adjustment signal generated in the execution module includes the adjustment amplitude and the adjustment response time: ; In the formula: To adjust the amplitude and response time; The target matching degree is preset; This represents the current matching degree. The actual values ​​of the current equipment parameters to be adjusted; The gain coefficient is used for parameter adjustment; The upper limit of the parameter adjustment rate of the device execution unit. Furthermore, when the response time is less than the minimum response time of the device execution unit, the minimum response time of the device execution unit shall be used as the reference time. application; In the above formula, the adjustment amplitude calculation is based on the difference between the target matching degree and the current matching degree. Combined with the actual values ​​of the current equipment parameters and the gain coefficient, the parameter adjustment amount is quantitatively calculated to ensure that the adjustment amplitude and the matching degree difference are accurately matched. The adjustment response time calculation combines the parameter adjustment amount and the upper limit of the adjustment rate of the equipment execution unit to reasonably control the action rate of the execution unit and avoid production anomalies caused by parameter mutations. At the same time, when the calculated response time is less than the minimum response time of the equipment, the minimum response time is adopted, which takes into account both the accuracy of adjustment and the safety and stability of equipment operation. The analysis module determines the equipment parameters to be adjusted, namely needle pitch, tension, and yarn feed speed, and uses the current actual values ​​of these parameters as... Substituting into the above adjustment amplitude formula, the specific adjustment amplitude of the corresponding parameter is calculated; Based on the adjustment direction output by the analysis module, the corresponding equipment parameters are quantitatively adjusted: If the adjustment direction is to increase, the adjusted parameter value is the actual value of the current parameter plus the corresponding parameter's dedicated adjustment amplitude; If the adjustment direction is to decrease, the adjusted parameter value is the current parameter's actual value minus the corresponding parameter's dedicated adjustment amplitude. The adjusted equipment parameter values ​​must be within the preset safe operating range of the parameter. If the adjusted parameter value exceeds the safe operating range, it will be corrected to the corresponding boundary value of the safe operating range. Otherwise, the operation will be judged as abnormal and the equipment will be driven to stop. The execution module encodes the adjusted parameter value into a control signal that can be recognized by the equipment execution unit (needle pitch adjustment motor, tension adjustment valve, yarn feed roller speed controller), and synchronously controls the action rate of the execution unit in combination with the adjustment response time. After the execution module completes the equipment adjustment, the acquisition module acquires the fabric surface image of the corresponding production position after the equipment adjustment within a preset time period. After being processed by the equalization module and the recognition module, the integration module generates a defect verification report after adjustment. If the defect severity in the verification report is still higher than the preset verification threshold, the analysis module re-optimizes the parameter association model based on the new production report, and the execution module generates a secondary adjustment signal until the defect severity is within the preset safety range. The acquisition module interacts with the balancing module via a wireless network. The balancing module interacts with the identification module via a wireless network. The identification module interacts with the integration module via a wireless network. The integration module interacts with the analysis module via a wireless network. The analysis module interacts with the execution module via a wireless network.

[0021] In this embodiment, the acquisition module adjusts the camera's posture and operating parameters to adapt the camera to the width of the knitted fabric and the operating speed of the production equipment, acquiring surface images of the knitted fabric during the production process. The equalization module then performs noise filtering and illumination equalization on the acquired fabric surface images. The recognition module further captures the yarn trajectory of the fabric area in the image after noise filtering and illumination equalization, determining the location and type of defects in the knitted fabric based on the trajectory offset. The integration module receives the defect location and type output by the recognition module and generates a production report containing the defect location, type, and severity. The analysis module then iterates through the production report, uploads the fabric production indicators, and quantifies the matching degree between defect location, defect type, and equipment parameters by constructing a parameter correlation model. This determines the items and directions for adjustment among the current knitting production equipment parameters such as needle pitch, tension, and yarn feed speed. Finally, the execution module receives the analysis results from the analysis module, generates corresponding adjustment signals based on the analysis results, and transmits the signals to the corresponding execution unit of the equipment in real time to complete equipment control.

[0022] In the above embodiments, the system can adapt to the acquisition of images of the width and production speed of the knitted fabric in real time. After noise reduction and illumination equalization processing, it can accurately capture the yarn trajectory, determine the defects, quantify their severity, and adjust the production equipment parameters in conjunction with the process. Through closed-loop optimization and continuous correction, it can significantly improve the accuracy and real-time performance of defect identification, reduce the defect rate, optimize production efficiency, stabilize fabric quality, reduce manual intervention costs, and achieve intelligent production control.

[0023] Referring to the system in the above embodiments, the following is an application example of the system: XX knitted fabric manufacturer produces fabrics with a width of 1.8 meters. The fabric feed speed of the production equipment is 0.8 meters per second. This system is needed to achieve real-time defect identification and production control.

[0024] After the system starts, the acquisition module first performs parameter adaptation. Based on the fabric width and the vertical distance from the camera lens to the fabric surface, the adjusted field of view is calculated to be 35 degrees. Through automatic focus adjustment, the two sides of the fabric are precisely aligned with the boundaries of the fabric area in the acquired image. Combining the preset yarn texture sampling coefficient, the actual length coefficient corresponding to the pixels in a single frame image, and the minimum resolution pixel spacing of the yarn texture in a single frame image, the acquisition frame rate is determined to be 30 frames per second to ensure clear capture of fabric surface details.

[0025] After acquiring the fabric surface image, the equalization module begins processing. An adaptive median filter based on the yarn texture direction is used. The filter window size is determined to be 5×5 based on the mean gradient of the yarn texture direction and the standard deviation of the texture gradient within the current pixel's neighborhood, effectively filtering out image noise. Subsequently, the image is divided into several sub-regions according to the yarn texture area, and the average gray value of the entire image is calculated. Combining the average gray value of each sub-region, the texture complexity coefficient of the sub-region, and the basic compensation threshold, the illumination compensation values ​​for three sub-regions are determined to be 22, 19, and 25, respectively, completing the illumination equalization process and ensuring uniform image brightness.

[0026] The recognition module captures the yarn trajectory in the processed image. It calculates the grayscale gradient value of each pixel along the yarn texture extension direction, extracts local peak points as initial feature points, and obtains 12 valid feature points after 3×3 neighborhood gradient consistency verification. Based on the number of valid feature points, the Bézier curve order is determined to be 10. The actual yarn trajectory is fitted using the coordinates of the valid feature points as constraint points. The trajectory offset from the standard yarn trajectory is calculated to be 0.8 mm, which is greater than a preset offset threshold. Further analysis reveals that the fluctuation value of the trajectory offset is within the second preset interval and the mean is greater than the preset offset threshold, indicating the presence of yarn misweaving defects in this area.

[0027] After receiving defect information, the integration module quantifies the severity of the defects. Combining the defect area weighting coefficient of 0.4 (due to the high-end and fine fabric), the ratio of the actual area of ​​the defect area to the area of ​​the corresponding fabric area in the currently acquired image, the quantification value of 4 for mis-woven yarns, the defect type weighting coefficient of 0.5, the influence coefficient of 1 for the defect location (the core area of ​​the fabric), and the defect location weighting coefficient of 0.1, the defect severity is calculated to be 3.8.

[0028] The analysis module iterates through production reports, constructs a multivariate coupled correlation model, and calculates the current matching degree to be 0.6, which is lower than the preset target matching degree of 0.85. By calculating the partial derivatives of each equipment parameter with respect to the mean of defect severity, it is determined that the needle pitch and yarn feed speed need to be adjusted, with the needle pitch to be adjusted by decreasing it and the yarn feed speed to be adjusted by increasing it.

[0029] The execution module calculates the adjustment amplitude and response time based on the analysis results: the current actual needle pitch is 0.3 mm, the adjustment amplitude is 0.02 mm, and the response time is 0.3 seconds; the current actual yarn feed speed is 1.2 m / s, the adjustment amplitude is 0.15 m / s, and the response time is 0.3 seconds. Following the adjustment direction, the adjusted needle pitch is 0.28 mm, and the yarn feed speed is 1.35 m / s, both within the preset safe operating range. The execution module encodes the adjusted parameters into control signals and transmits them to the corresponding execution unit, controlling the action rate according to the response time to complete the adjustment.

[0030] After the adjustment is completed, the acquisition module collects an image of the fabric surface at the production location within 5 minutes. The subsequent module processes the image to generate a defect verification report after adjustment. The report shows that the defect severity has been reduced to 1.2, which is within the preset safe range.

[0031] In summary, the above embodiments, during operation, optimize image acquisition quality by precisely adapting to the width of the knitted fabric and the operating speed of the production equipment. Through targeted noise filtering and illumination equalization, the details of the yarn texture are clearly presented. By using yarn trajectory capture and fitting algorithms, the location and type of defects are accurately located. The severity of defects is quantified based on multi-dimensional indicators, making the evaluation results more scientific. Furthermore, by constructing a parameter correlation model, the matching degree between defects and equipment parameters is quantified, clarifying the adjustment direction and amplitude, and achieving real-time precise control of the production equipment. The closed-loop optimization mechanism continuously corrects parameters to ensure that the severity of defects is within a safe range. This effectively improves the production quality and yield of knitted fabrics, reduces resource consumption, increases production efficiency, adapts to the production needs of knitted fabrics of different grades and textures, and enhances the stability and controllability of the production process.

[0032] 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 will 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.

Claims

1. A real-time defect identification and production control system for knitted fabrics based on AI visual inspection, characterized in that, include: The acquisition module is used to adjust the camera's posture and operating parameters to adapt the camera to the width of the knitted fabric and the operating speed of the production equipment, and to acquire surface images of the knitted fabric during the production process. The equalization module is used to perform noise filtering and illumination equalization processing on the acquired fabric surface images. The recognition module is used to capture the yarn trajectory of the fabric area in the image after noise filtering and illumination equalization, and to determine the location and type of defects in the knitted fabric by the trajectory offset. The integration module receives the defect location and defect type output by the identification module and generates a production report containing the defect location, defect type, and severity. The analysis module is used to traverse production reports, upload fabric production indicators, and quantify the matching degree between defect location, defect type and equipment parameters by building a parameter correlation model. It determines the items and directions for adjustment among the current knitting production equipment parameters such as needle pitch, tension and yarn feed speed. The execution module receives the analysis results from the analysis module, generates corresponding adjustment signals based on the analysis results, and transmits the signals to the corresponding execution unit of the equipment in real time to complete the equipment control.

2. The real-time defect identification and production control system for knitted fabrics based on AI visual inspection according to claim 1, characterized in that, The adjustment of the camera posture in the acquisition module includes the adaptation adjustment of the field of view: ; In the formula: To adjust the rear field of view; For the width of the knitted fabric; The vertical distance between the camera lens and the surface of the knitted fabric; The adjustment of the camera's operating parameters in the acquisition module includes the adaptation adjustment of the acquisition frame rate: ; In the formula: To adjust the capture frame rate; The fabric production feed speed on the production equipment; The preset yarn texture sampling coefficient; This is the actual length coefficient corresponding to the pixels in a single frame of image, with the unit being length / pixel; This represents the minimum resolution pixel spacing of the yarn texture in a single frame image.

3. The real-time defect identification and production control system for knitted fabrics based on AI visual inspection according to claim 1, characterized in that, The equalization module performs noise filtering on the acquired fabric surface image using an adaptive median filter based on the yarn texture direction, with the filter window size... ; In the formula: The preset texture adaptation coefficient; The mean gradient of the yarn texture direction within the neighborhood of the current pixel; The standard deviation of the texture gradient within the neighborhood of the current pixel; During illumination equalization, the image is divided into several sub-regions based on the yarn texture area, and the illumination compensation value for each sub-region is... ,in This represents the average grayscale value of the entire image. This represents the average gray value of the i-th sub-region. This represents the texture complexity coefficient of the sub-region. This represents the basic compensation threshold.

4. The real-time defect identification and production control system for knitted fabrics based on AI visual inspection according to claim 1, characterized in that, When the recognition module captures the yarn trajectory of the fabric area in the image, it calculates the gray-level gradient value of each pixel in the image in the direction of yarn texture extension, extracts the local peak points of the gray-level gradient value in each yarn texture area as initial feature points, and then performs gradient consistency verification on the initial feature points in a 3×3 neighborhood, and removes noise points in the neighborhood whose gradient direction deviation exceeds a preset angle threshold, thus obtaining an effective feature point set of the yarn trajectory. The number N of valid feature points is counted, according to Determine the order m of the Bézier curve, where m is a positive integer; Using the coordinates of the effective feature points as constraint points, the m-th order Bézier curve is fitted using the least squares method to obtain the actual yarn trajectory. ; Finally, calculate the mean Euclidean distance between the fitted Bézier curve and each effective feature point. If the mean value is greater than the preset fitting error threshold, extract 2 to 3 new feature points on the side of the feature point with the largest distance, and repeat the above steps until the value does not exceed the preset fitting error threshold. The trajectory offset ; In the formula: This represents the x-axis coordinate value corresponding to the actual yarn trajectory at parameter t; This represents the y-axis coordinate value of the actual yarn trajectory at parameter t; This represents the x-axis coordinate value of the standard yarn trajectory at parameter t. This represents the y-axis coordinate value of the standard yarn trajectory at parameter t. This is the parameter range corresponding to the yarn segment; when If the offset exceeds a preset threshold, the corresponding area is determined to have defects.

5. The real-time defect identification and production control system for knitted fabrics based on AI visual inspection according to claim 1, characterized in that, When the identification module determines the type of defect in knitted fabric based on trajectory offset, it follows the following: Within the same yarn segment When the mean is not equal to zero, the fluctuation value ; like If the defect falls within the first preset range, the defect type is determined to be loose yarn. like Within the second preset range and If the average value is greater than the preset offset threshold, the defect type is determined to be yarn misweaving; like If a sudden change occurs and the pixel grayscale variance of the corresponding area is greater than the preset variance threshold, the defect type is determined to be yarn breakage.

6. The real-time defect identification and production control system for knitted fabrics based on AI visual inspection according to claim 1, characterized in that, The severity of the situation in the production report is quantified using the following formula: ; In the formula: The severity of the defect; This is the defect area weighting coefficient; This represents the actual area of ​​the defective region. The area of ​​the fabric region corresponding to the currently acquired image; This is the quantified value corresponding to the defect type; This refers to the weighting coefficient for the defect type. This represents the weighting coefficient for the defect location. This represents the influence coefficient of the defect location.

7. The real-time defect identification and production control system for knitted fabrics based on AI visual inspection according to claim 1, characterized in that, During the operation of the analysis module, a multivariate coupled correlation model is constructed using the needle pitch, tension, and yarn feeding speed of the knitting production equipment as input variables and the mean of the defect severity as the output variable. ; In the formula: This represents the average severity of the defects. , , These are the quadratic coefficient of needle pitch, the coupling coefficient between tension and yarn feed speed, and the ratio coefficient between needle pitch and tension. The needle spacing; For tension; For yarn feeding speed; For model residuals; When quantifying the matching degree using an association model, the matching degree... ,in This indicates the maximum permissible value for the severity of the defect. When M is lower than the preset matching threshold, the device parameters that need to be adjusted are determined. The adjustment direction is determined by calculating the average value of each parameter to be adjusted in relation to the severity of the defect. The partial derivative of a parameter is such that if the partial derivative is positive, the parameter is adjusted in the direction of decreasing; if the partial derivative is negative, the parameter is adjusted in the direction of increasing.

8. The real-time defect identification and production control system for knitted fabrics based on AI visual inspection according to claim 1, characterized in that, The adjustment signal generated in the execution module includes the adjustment amplitude and the adjustment response time. ; In the formula: To adjust the amplitude and response time; The target matching degree is preset; This represents the current matching degree. The actual values ​​of the current equipment parameters to be adjusted; The gain coefficient is used for parameter adjustment; The upper limit of the parameter adjustment rate of the device execution unit, and when When the response time is less than the minimum response time of the device execution unit, the minimum response time of the device execution unit shall be used as the reference. application.

9. A real-time defect identification and production control system for knitted fabrics based on AI visual inspection as described in claim 1, characterized in that, The acquisition module is interconnected with a balancing module via a wireless network. The balancing module is interconnected with a recognition module via a wireless network. The recognition module is interconnected with an integration module via a wireless network. The integration module is interconnected with an analysis module via a wireless network. The analysis module is interconnected with an execution module via a wireless network.

Images