Embedded AI-driven leather defect laser marking and process optimization closed-loop system

Abstract

The invention discloses an embedded AI-driven leather defect laser marking and process optimization closed-loop system, and relates to the technical field of leather production and processing, the system comprises the following components: a data preprocessing module, an embedded AI defect detection module, a laser marking device and a production control system; leather images are analyzed in real time through the embedded AI defect detection module, defects on leather can be quickly and accurately recognized, and the defect positions are accurately marked through the laser marking device, so that the defect detection and marking process is more efficient, errors caused by manual operation are reduced, and the detection efficiency is improved. The overall efficiency of leather production is improved, meanwhile, the production control system dynamically adjusts the technological parameters according to the detection result, leather defects can be effectively reduced, and therefore the quality of leather products is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of leather production and processing, and specifically to a closed-loop system for laser marking of leather defects driven by embedded AI and process optimization. Background Art

[0002] With the rapid development of the leather industry, the demand for leather products is increasing day by day, and at the same time, the quality requirements for leather are getting higher and higher. In the process of leather production, the detection and marking of leather defects are key links to ensure product quality. Traditionally, this link mainly relies on manual operation. However, manual detection is not only inefficient but also easily affected by human factors, resulting in difficulties in ensuring the accuracy and consistency of detection results.

[0003] Traditional technologies have deficiencies. First of all, manual detection is time-consuming and laborious and cannot meet the needs of large-scale production. Secondly, manual detection is easily affected by subjective factors, resulting in unstable detection results. Moreover, traditional technologies lack an intelligent dynamic adjustment mechanism and cannot optimize production process parameters in real time according to detection results, thus restricting the further improvement of leather product quality. In addition, traditional technologies also have obvious shortcomings in quality traceability and it is difficult to achieve comprehensive recording and traceability of production data.

[0004] In summary, the traditional leather defect detection and marking technologies are no longer able to meet the development needs of the modern leather industry. Therefore, in order to improve the efficiency of leather production and product quality, it is particularly important to develop a closed-loop system for laser marking of leather defects driven by embedded AI and process optimization. Summary of the Invention

[0005] The purpose of the present invention is to make up for the deficiencies of the existing technology and provide a closed-loop system for laser marking of leather defects driven by embedded AI, which can achieve rapid and accurate detection and marking of leather defects and dynamic optimization of process parameters through advanced embedded AI technology and laser marking technology, providing strong support for the high-quality development of the leather industry.

[0006] In order to solve the above technical problems, the present invention provides the following technical solution: a closed-loop system for laser marking of leather defects driven by embedded AI, which system includes the following components: a data preprocessing module, an embedded AI defect detection module, a laser marking device, and a production control system;

[0007] The data preprocessing module: is used for preprocessing the leather images obtained by the image acquisition device;

[0008] The embedded AI defect detection module: performs real-time analysis on the preprocessed leather images based on deep learning algorithms to identify leather defects;

[0009] The laser marking device: According to the defect position information fed back by the embedded AI defect detection module, accurately mark the defect position.

[0010] The production control system: Receive the detection results of the embedded AI defect detection module, dynamically adjust the leather processing process parameters, and establish a quality traceability system to store leather production process data.

[0011] Furthermore, the data preprocessing module uses an adaptive multi-scale image enhancement algorithm. This algorithm is based on the local features of the image and determines the enhancement parameters through the following formula:

[0012]

[0013] where E(x, y) is the pixel value of the enhanced image, (x, y) is the image pixel coordinate, I(x, y) is the pixel value of the original image, is the average pixel value of the local region of the image, σ(x, y) is the standard deviation of the local region of the image, and α, β, γ are adaptive parameters determined by analyzing the texture complexity and contrast at different scales of the image. First, divide the image into multiple sub-regions of different scales, calculate the texture complexity T and contrast C of each sub-region, and according to the formula Determine the parameter values. ∈ is a very small constant to prevent the denominator from being zero, realizing targeted enhancement of different feature image regions, improving the image quality, and providing a more accurate data basis for subsequent defect detection.

[0014] Even further, the embedded AI defect detection module uses an improved deep residual attention network. This network introduces an attention mechanism on the basis of the traditional residual network and calculates the attention weight through the following formula:

[0015] A = σ(w2·δ(W1·GAP(F)))

[0016] where A is the attention weight, F is the feature map, GAP is the global average pooling operation, W1, W2 are learnable weight matrices, δ is the ReLU activation function, σ is the Sigmoid activation function. Weight the feature map through the attention weight to highlight the features related to leather defects and suppress irrelevant information. The weights of the network are trained with a large number of leather defect samples, using the cross-entropy loss function where yi is the true label, is the predicted label, and use the Adam optimizer to update the weights, continuously adjust the network parameters to improve the accuracy and recall rate of defect recognition.

[0017] Further, after receiving the defect position information, the laser marking device adopts a dynamic focus spot adjustment algorithm. According to the undulation of the leather surface and the defect depth information, the laser focus position is calculated by the following formula:

[0018] Z = Z0 + k·d

[0019] Where Z is the adjusted laser focus position, Z0 is the initial focus position, d is the defect depth, and k is a coefficient related to the leather material, which is determined by experimental measurement of different leather materials. At the same time, according to the size and shape of the defect, the laser power P and pulse width t are adjusted. The formula P0 and t0 are the initial power and pulse width, S is the defect area, l is the longest side length of the defect, and l0 is the reference length, to ensure clear and accurate marking in different defect situations without damaging other parts of the leather.

[0020] Further, when adjusting the process parameters, the production control system adopts an adaptive parameter adjustment algorithm based on Bayesian optimization. This algorithm builds a probability model between the process parameters and leather defects, and continuously iteratively optimizes the process parameters. The objective function f(x) is defined as the defect occurrence rate, x is the process parameter vector. According to prior knowledge and initial experimental data, a Gaussian process model GP(x) is constructed, and the expected improvement value EI(x) is calculated to select the next process parameter point to be tested:

[0021]

[0022] Where f min is the currently known minimum defect occurrence rate, μ(x) and σ(x) are the mean and standard deviation of the Gaussian process model at x respectively, Φ and are the cumulative distribution function and probability density function of the standard normal distribution respectively. By continuously selecting the point with the largest EI(x) for experiments and updating the Gaussian process model, the optimal process parameters are gradually approximated, effectively reducing the generation of leather defects.

[0023] Further, the quality traceability system uses blockchain technology for data storage. The raw material information, production time, defect detection results, and process parameter adjustment record data of each piece of leather are hashed to generate a hash value, which is stored in the blocks of the blockchain. Each block contains the hash value of the previous block, forming a chain structure to ensure the immutability and traceability of the data.

[0024] Furthermore, the system further includes an environmental monitoring module for real-time monitoring of the temperature, humidity, and light parameters of the leather production environment, collecting environmental data through sensors, and processing the data using the Kalman filtering algorithm to improve the accuracy and stability of the data. The Kalman filtering algorithm performs state estimation and update through the following formulas:

[0025]

[0026] P k|k-1 = A·P k-1|k-1 ·A T + Q

[0027] K k = P k|k-1 ·H T ·(H·P k|k-1 ·H T + R) -1

[0028]

[0029] P k|k = (I - K k ·H)·P k|k-1

[0030] Wherein, is the prior state estimate at time k, is the posterior state estimate at time k - 1, A is the state transition matrix, B is the control matrix, u k is the control input, P k|k-1 is the prior covariance estimate at time k, P k -1|k - 1 is the posterior covariance estimate at time k - 1, Q is the process noise covariance, K k is the Kalman gain, H is the observation matrix, z k is the observed value, R is the observation noise covariance, is the posterior state estimate at time k, P k|k is the posterior covariance estimate at time k. The production control system further optimizes the process parameters based on the environmental monitoring data and in combination with the leather defect situation to reduce the impact of environmental factors on leather quality.

[0031] Furthermore, during the training process of the embedded AI defect detection module, an adversarial learning strategy is adopted, introducing a generative adversarial network structure. The generator G attempts to generate realistic leather defect images, and the discriminator D discriminates between real defect images and the generated defect images. The adversarial training is carried out through the following formulas:

[0032]

[0033] Among them, x is the real leather defect image, z is the random noise vector, and p data (x) is the real data distribution, and p z (z) is the noise distribution. During the training process, the generator and the discriminator are alternately optimized. The generator continuously learns to generate more realistic defect images to deceive the discriminator, while the discriminator continuously improves its discrimination ability. This adversarial learning strategy enables the embedded AI defect detection module to learn richer defect features and improve the recognition ability for complex defects.

[0034] Furthermore, the system also has a fault diagnosis and early warning function. For each module in the system, a fault prediction model is established, and a multi-classification fault diagnosis algorithm based on support vector machines is adopted. By monitoring and analyzing various index data during the operation of the module, these index data are used as feature vectors and input into the SVM model for training and classification. When the model predicts that a certain module may have a fault, an early warning message is sent in a timely manner to remind the maintenance personnel to check and maintain, ensuring the stable operation of the system.

[0035] Compared with the prior art, the embedded AI-driven leather defect laser marking and process optimization closed-loop system has the following beneficial effects:

[0036] First, the system can quickly and accurately identify the defects on the leather through the embedded AI defect detection module's real-time analysis of the leather image, and accurately mark the defect positions through the laser marking device. This not only makes the defect detection and marking process more efficient, but also reduces the error of manual operation, improves the overall efficiency of leather production. At the same time, the production control system dynamically adjusts the process parameters according to the detection results, which can effectively reduce the generation of leather defects, thus improving the quality of leather products.

[0037] Second, the system has established a quality traceability system and uses blockchain technology for data storage to ensure that the raw material information, production time, defect detection results, and process parameter adjustment record data of each piece of leather are tamper-proof and traceable. This helps the enterprise to achieve comprehensive monitoring and management of the leather production process. Once a problem occurs, the cause can be quickly located and traced. In addition, the system also has environmental monitoring, fault diagnosis and early warning functions, which can monitor the production environment parameters in real time, predict and warn system faults, providing strong support for the intelligent management of the enterprise.

[0038] Other advantages, objectives and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0040] Figure 1 It is a flowchart for realizing the functions of an embedded AI-driven leather defect laser marking and process optimization closed-loop system;

[0041] Figure 2 It is a flowchart of the environmental monitoring module and the fault diagnosis and early warning module. Detailed implementation manners

[0042] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, in conjunction with the drawings and preferred embodiments, detail the specific implementation manners, structures, features and their effects of the present invention as follows.

[0043] Embodiment 1

[0044] This embodiment describes a small leather products processing factory focusing on the production of leather shoes uppers. The products are mainly supplied to several small shoe enterprises. In a highly competitive market environment, product quality has become the key for the processing factory to survive. In the past, due to the low efficiency and unstable accuracy of manual inspection of leather defects, the defective rate was relatively high, which not only increased the production cost but also affected customer satisfaction. To solve these problems, the processing factory introduced an embedded AI-driven leather defect laser marking and process optimization closed-loop system.

[0045] In the production workshop, the image acquisition device is installed above the leather feeding conveyor belt and can stably obtain images of the leather shoes uppers. The data preprocessing module then starts the adaptive multi-scale image enhancement algorithm. For example, when processing an image of a leather shoes upper, it is divided into multiple sub-regions of different scales. The texture complexity T of one sub-region is 0.3, the maximum texture complexity max(T) of this batch of images is 0.5, the sub-region contrast C is 0.25, the maximum contrast max(C) is 0.4, the sub-region standard deviation σ(x, y) is 0.1, and the extremely small constant ∈ is 10 -6 , according to the formula It is calculated that Then substitute these parameters into the enhancement formula to perform enhancement processing on the image. After processing, the originally blurred texture and potential defects on the leather surface become clearer, providing a better data basis for subsequent defect detection.

[0046] The improved deep residual attention network within the embedded AI defect detection module starts to work. Suppose at a certain moment, the network obtains a preprocessed leather feature map F. First, a global average pooling operation GAP is performed on F, and then it is multiplied by the learnable weight matrix W1, processed through the ReLU activation function δ, multiplied by the weight matrix W2, and finally the attention weight A is calculated through the Sigmoid activation function σ. During the training phase, the processing factory collected a large number of upper leather samples containing various defects such as scratches, holes, and color spots, and used the cross-entropy loss function and continuously adjusted the network weights with the help of the Adam optimizer. After long-term training, the network can accurately identify various leather defects. For example, when a fine scratch is detected on the upper leather, the system can quickly and accurately locate and mark the defect position.

[0047] Once a defect is detected, the laser marking device immediately receives the defect position information. If the depth d of a certain defect is 0.5 mm, the initial focusing position Z0 is 10 mm, and the correlation coefficient k related to the leather material is 0.2, according to the formula Z = Z0 + k·d, the laser focusing position Z = 10 + 0.2×0.5 = 10.1 mm can be calculated. If the defect area S = 2 mm 2 , and the initial power P0 = 5 W, then the laser power If the longest side length l of the defect is 3 mm, the reference length l0 is 2 mm, and the initial pulse width t0 is 2 ms, then the pulse width The laser marking device accurately marks the defect position based on these adjusted parameters, and the marked leather is convenient for subsequent sorting and processing.

[0048] According to the detection results, the production control system adopts an adaptive parameter adjustment algorithm based on Bayesian optimization. The system sets the tanning time and temperature process parameters as the adjustment objects, and continuously iteratively optimizes the process parameters by establishing a probability model between the process parameters and leather defects. For example, after a period of production and data accumulation, the system finds that when the tanning temperature is slightly reduced within a certain range and the tanning time is extended at the same time, the defect incidence rate significantly decreases. At the same time, blockchain technology is used to record the raw material information, production time, defect detection results, and process parameter adjustment records of each piece of upper leather. In this way, once a quality problem occurs, the processing factory can quickly trace back to the source of the problem through the blockchain and take timely measures for improvement.

[0049] Embodiment 2

[0050] This embodiment describes that in a professional sofa leather manufacturing factory, to ensure the production of high-quality sofa leather products, the entire production process comprehensively applies an embedded AI-driven closed-loop system for leather defect laser marking and process optimization.

[0051] At the starting position where the leather enters the production line, multiple high-definition image acquisition devices are ingeniously installed above and on both sides of the conveyor belt to collect leather images in all directions. These image acquisition devices can work at high resolution and high frame rate to ensure that every detail on the leather surface is captured. The original leather images collected often have problems such as uneven illumination and blurred texture, so they will be immediately transmitted to the data preprocessing module. Since the sofa leather is large in area and rich in diverse textures, the preprocessing module uses an adaptive multi-scale image enhancement algorithm. Based on the local features of the image, the leather image is divided into multiple sub-regions of different scales. For example, for a piece of cowhide leather with unique texture, in the divided sub-regions, the algorithm will calculate the texture complexity and contrast of each sub-region respectively, and determine the adaptive parameters α, β, γ according to the formula. Through the adjustment of these parameters, potential defects in the image, such as small holes, color spots, and fine cracks left by mosquito bites, are significantly enhanced, providing clearer and more distinguishable image data for subsequent defect detection.

[0052] The preprocessed image is quickly sent to the embedded AI defect detection module. This module is based on an improved deep residual attention network to perform real-time analysis on the image. The attention mechanism in the network plays a key role. It assigns attention weights to different feature maps through a complex calculation process. When faced with a large piece of leather for making a sofa cushion, the network can accurately identify multiple small hole defects in it. Once a defect is detected, the detection module will quickly feedback the defect position information to the laser marking device. Since the sofa leather is usually thick, after receiving the information, the laser marking device will adopt a dynamic focused spot adjustment algorithm. According to the natural undulation of the leather surface and the depth information of the small holes, it calculates the laser focusing position through a specific formula to ensure that the mark can accurately cover the defect position and reach an appropriate depth. At the same time, according to the size and distribution shape of the holes, the device will adjust the laser power and pulse width accordingly. For larger holes, the laser power is appropriately increased and the pulse width is extended to ensure that the mark is clearly visible; for smaller holes, the power is reduced and the pulse width is shortened to avoid excessive damage to the leather. After completing the parameter adjustment, the laser marking device makes accurate marks at the defect positions, providing clear identification for subsequent processing.

[0053] The production control system receives the detection results of the embedded AI defect detection module in real time, and dynamically adjusts the leather production process based on these results. If a batch of leather is found to have more color spot defects after the dyeing process, the production control system will start the adaptive parameter adjustment algorithm based on Bayesian optimization. The system will establish a probability model between the dyeing process parameters (such as dyeing temperature, time, dye concentration and stirring speed) and the leather color spot defects. Through continuous iterative optimization, the dyeing temperature, time and dye concentration parameters are adjusted. For example, after many experiments and model calculations, it was found that appropriately lowering the dyeing temperature, extending the dyeing time and fine-tuning the dye concentration ratio can significantly reduce the occurrence of color spot defects. While adjusting the process parameters, the quality traceability system begins to play a role. Every piece of sofa leather from the raw material procurement link At the beginning, its supplier information, leather type and grade, and purchase date data are recorded in detail. During the production process, the production time of each process, operator information, process parameter settings, defect detection results and processing data are collected in real time. These data are hashed to generate hash values and stored in a quality traceability system based on blockchain technology. Each data block contains the hash value of the previous data block, forming an unalterable chain structure. In this way, when consumers find that there are quality problems with the leather after purchasing a sofa, the company can use the quality traceability system to quickly and accurately locate which link of the production and which batch of raw materials the problem occurred in, and even trace it back to the specific operator, so as to take timely measures to solve the problem. At the same time, it also provides strong data support for companies to improve their production processes.

[0054] The environmental monitoring module plays an important role in the sofa leather manufacturing workshop. Multiple sensors are distributed and installed at different positions in the workshop to monitor the temperature, humidity, and light parameters in the workshop in real time. Since the production environment has an undeniable impact on the quality of leather, for example, too high temperature may cause the leather to dry and become brittle, prone to cracks, and too high humidity may lead to mildew and deterioration of the leather. The environmental data collected by the sensors will be transmitted to the environmental monitoring module, and the module uses the Kalman filter algorithm to process the data to improve the accuracy and stability of the data. When it is detected that the workshop temperature exceeds the appropriate range, the production control system will combine the environmental data and the leather crack defect situation to automatically adjust the humidity of the leather storage environment. For example, by starting the humidification or dehumidification equipment in the workshop to adjust the humidity to an appropriate level, or during the processing, adding a leather moisturizing treatment link, such as spraying an appropriate amount of special moisturizer on the leather surface, to ensure the flexibility and quality of the leather. In addition, the system also has a powerful fault diagnosis and early warning function. Special fault prediction models are established for each module in the system, including the image acquisition device, the embedded AI defect detection module, the laser marking device, and the production control system. The multi-classification fault diagnosis algorithm based on support vector machine is used to monitor and analyze the index data during the operation of the module. For example, when it is detected that the frame rate of the image acquisition device fluctuates abnormally, the laser power of the laser marking device is unstable, or the parameter adjustment of the production control system appears abnormal, these index data will be used as feature vectors and input into the SVM model for training and classification. Once the model predicts that a certain module may have a fault, the system will send out an early warning message in time, reminding the maintenance personnel to check and maintain through the alarm device in the workshop and the mobile terminal of the staff. The maintenance personnel can quickly locate the faulty module according to the early warning message and take corresponding repair measures to avoid production interruption caused by equipment failure, thus ensuring the continuity and stability of sofa leather manufacturing production and improving production efficiency and product quality.

[0055] The above is only a preferred embodiment of the present invention, and it is not intended to limit the present invention in any form. Although the present invention has been disclosed above with the preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to be equivalent embodiments with equivalent changes within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent change, and modification made to the above embodiment according to the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An embedded AI-driven closed-loop system for laser marking and process optimization of leather defects, characterized in that, The system includes the following components: a data preprocessing module, an embedded AI defect detection module, a laser marking device, and a production control system; The data preprocessing module: is used to preprocess the leather images obtained by the image acquisition device; The embedded AI defect detection module: performs real-time analysis on the preprocessed leather images based on deep learning algorithms to identify leather defects; The laser marking device: according to the defect position information fed back by the embedded AI defect detection module, precisely marks the defect positions; The production control system: receives the detection results of the embedded AI defect detection module, dynamically adjusts the leather processing process parameters, and establishes a quality traceability system to store the leather production process data.

2. The embedded AI-driven closed-loop system for laser marking and process optimization of leather defects according to claim 1, wherein The data preprocessing module adopts an adaptive multi-scale image enhancement algorithm, which is based on the local features of the image and determines the enhancement parameters through the following formula: Among them, E(x, y) is the pixel value of the enhanced image, (x, y) is the pixel coordinate of the image, and I(x, y) is the pixel value of the original image. is the average pixel value of the local area of the image, σ(x, y) is the standard deviation of the local area of the image, and α, β, and γ are adaptive parameters. First, the image is divided into multiple sub-regions of different scales, and the texture complexity T and contrast C of each sub-region are calculated. According to the formula to determine the parameter values, where ∈ is a very small constant to prevent the denominator from being zero.

3. The embedded AI-driven closed-loop system for laser marking and process optimization of leather defects according to claim 1, characterized in that, The embedded AI defect detection module adopts an improved deep residual attention network, which introduces an attention mechanism on the basis of the traditional residual network and calculates the attention weight through the following formula: A = σ(w2·δ(W1·GAP(F))) Among them, A is the attention weight, F is the feature map, GAP is the global average pooling operation, W1 and W2 are learnable weight matrices, δ is the ReLU activation function, σ is the Sigmoid activation function. The feature map is weighted by the attention weight to highlight the features related to leather defects and suppress irrelevant information. The weights of the network are trained with a large number of leather defect samples, and the cross-entropy loss function is adopted. where yi is the true label, is the predicted label, and the Adam optimizer is used to update the weights to continuously adjust the network parameters.

4. The embedded AI-driven closed-loop system for laser marking and process optimization of leather defects according to claim 1, characterized in that, After receiving the defect position information, the laser marking device adopts a dynamic focus spot adjustment algorithm, and calculates the laser focus position through the following formula according to the undulation of the leather surface and the defect depth information: Z = Z0 + k·d Among them, Z is the adjusted laser focusing position, Z0 is the initial focusing position, d is the defect depth, k is a coefficient related to the leather material. At the same time, according to the size and shape of the defect, the laser power P and the pulse width t are adjusted, and the formula P0 and t0 are the initial power and pulse width, S is the defect area, l is the longest side length of the defect, and l0 is the reference length.

5. The embedded AI-driven closed-loop system for laser marking and process optimization of leather defects according to claim 1, wherein When adjusting the process parameters, the production control system adopts an adaptive parameter adjustment algorithm based on Bayesian optimization. This algorithm continuously iteratively optimizes the process parameters by establishing a probability model between the process parameters and the leather defects. The objective function f(x) is defined as the defect occurrence rate, x is the process parameter vector, a Gaussian process model GP(x) is constructed according to prior knowledge and initial experimental data, and the expected improvement value EI(x) is calculated to select the next process parameter point to be tested: where fmi n is the currently known minimum defect occurrence rate, μ(x) and σ(x) are the mean and standard deviation of the Gaussian process model at x respectively, Φ and are the cumulative distribution function and probability density function of the standard normal distribution respectively. By continuously selecting the point with the largest EI(x) for experiments, the Gaussian process model is updated to gradually approach the optimal process parameters.

6. The embedded AI-driven closed-loop system for laser marking and process optimization of leather defects according to claim 1, wherein The quality traceability system uses blockchain technology for data storage. The raw material information, production time, defect detection results, and process parameter adjustment record data of each piece of leather are hashed to generate a hash value, which is stored in the blocks of the blockchain. Each block contains the hash value of the previous block, forming a chain structure.

7. The embedded AI-driven closed-loop system for laser marking and process optimization of leather defects according to claim 1, characterized in that, The system further includes an environmental monitoring module, which is used to monitor the temperature, humidity, and light parameters of the leather production environment in real time, collect environmental data through sensors, and process the data using the Kalman filter algorithm. The Kalman filter algorithm performs state estimation and update through the following formula: P k|k-1 = A·P k-1|k-1 ·A T + Q K k = P k|k-1 · H T · (H · P k|k-1 · H T + R) -1 P k|k = (I - K k · H) · P k|k-1 Among them, is the prior state estimate at time k, is the posterior state estimate at time k-1, A is the state transition matrix, B is the control matrix, and u k is the control input, and P k|k-1 is the prior covariance estimate at time k, and P k is the posterior covariance estimate at time k-1|k-1, Q is the process noise covariance, and K k is the Kalman gain, H is the observation matrix, and z k is the observed value, and R is the observation noise covariance. is the posterior state estimate at time k, and P k|k is the posterior covariance estimate at time k. The production control system further optimizes the process parameters according to the environmental monitoring data and combined with the leather defect situation to reduce the impact of environmental factors on leather quality.

8. The embedded AI-driven closed-loop system for laser marking and process optimization of leather defects according to claim 1, characterized in that, During the training process, the embedded AI defect detection module adopts an adversarial learning strategy, introduces a generative adversarial network structure. The generator G attempts to generate realistic leather defect images, and the discriminator D discriminates between real defect images and generated defect images, and performs adversarial training through the following formula: Among them, x is the real leather defect image, z is the random noise vector, and p data (x) is the real data distribution, and p z (z) is the noise distribution. During the training process, the generator and the discriminator are alternately optimized. The generator continuously learns to generate more realistic defect images to deceive the discriminator, while the discriminator continuously improves its discrimination ability.

9. The closed-loop system for laser marking and process optimization of leather defects driven by embedded AI according to claim 1, characterized in that, The system also has a fault diagnosis and early warning function. For each module in the system, a fault prediction model is established, and a multi-classification fault diagnosis algorithm based on support vector machines is adopted. By monitoring and analyzing the index data during the operation of the module, these index data are used as feature vectors and input into the SVM model for training and classification. When the model predicts that a certain module may have a fault, an early warning message is sent in time to remind the maintenance personnel to check and maintain, ensuring the stable operation of the system.

Images