Automatic sorting method and system for wear resistance grade of finished arrow belt

Abstract

The application discloses a garter finished product wear resistance grade automatic sorting method and system, and relates to the technical field of garter finished product sorting. The method comprises the following steps: performing wear resistance characteristic analysis on a garter finished product set to obtain garter wear resistance grade labels; performing abnormality detection on the garter finished product set to generate garter composite labels; performing control decision on a plurality of automatic sorting devices to determine a garter sorting control scheme; performing sorting risk prediction on the garter sorting control scheme to obtain a garter sorting risk sequence; establishing a sorting control adjustment first group; performing sorting comprehensive risk variation optimization to generate a sorting control optimization result, and performing automatic sorting on the garter finished product set. The application solves the technical problems that garter sorting in the prior art relies on manual operation, is inefficient, and is prone to sorting errors, resulting in insufficient sorting accuracy and safety, and achieves the technical effects of realizing automatic sorting of garter finished products according to wear resistance grades and improving the accuracy and safety of sorting.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of arrow belt finished product sorting, in particular to an arrow belt finished product wear resistance grade automatic sorting method and system. BACKGROUND

[0002] In the field of arrow belt finished product production, wear resistance grade sorting is an important link to ensure product quality and subsequent use effect. The traditional sorting method relies on manual operation, which is not only low in efficiency, but also prone to sorting errors due to human judgment errors in the sorting process, and it is difficult to effectively control the risks such as misplacement, mistaken grabbing and damage in the sorting process, which cannot meet the requirements of sorting precision and safety in large-scale production.

[0003] The existing technology has the technical problems that arrow belt sorting relies on manual operation, which is low in efficiency and prone to sorting errors, resulting in insufficient sorting accuracy and safety. SUMMARY

[0004] The present application provides an arrow belt finished product wear resistance grade automatic sorting method and system to solve the technical problems that arrow belt sorting in the prior art relies on manual operation, which is low in efficiency and prone to sorting errors, resulting in insufficient sorting accuracy and safety.

[0005] In view of the above problems, the present application provides an arrow belt finished product wear resistance grade automatic sorting method and system.

[0006] In a first aspect of the present application, an arrow belt finished product wear resistance grade automatic sorting method is provided, which comprises:

[0007] The wear resistance grade confidence evaluation model is used to analyze the wear resistance characteristics of the arrow belt finished product set, and the wear resistance grade labels of each arrow belt are obtained. The arrow belt finished product set is subjected to abnormality detection, and the composite labels of each arrow belt are generated in combination with the wear resistance grade labels of each arrow belt. Based on the arrow belt finished product set, control decisions are made for a plurality of automatic sorting devices according to the composite labels of each arrow belt, and an arrow belt sorting control scheme is determined. The arrow belt sorting risk analysis channel is introduced to predict the sorting risk of the arrow belt sorting control scheme, and an arrow belt sorting risk sequence is obtained. Based on the arrow belt sorting risk sequence, the arrow belt sorting control scheme is adjusted according to the arrow belt sorting risk constraint, and a sorting control adjustment first group is established. The sorting control adjustment first group is subjected to sorting comprehensive risk variation optimization to generate a sorting control optimization result, and the arrow belt finished product set is subjected to automatic sorting in combination with the plurality of automatic sorting devices.

[0008] In a second aspect of the present application, an arrow belt finished product wear resistance grade automatic sorting system is provided, which comprises:

[0009] The wear-resistant grade label acquisition module is configured to obtain wear-resistant grade labels of each arrow belt by performing wear-resistant feature analysis on the arrow belt product set through a wear-resistant grade confidence evaluation model; the anomaly detection module is configured to perform anomaly detection on the arrow belt product set, and generate a composite label of each arrow belt in combination with the wear-resistant grade labels of the arrow belt; the control decision module is configured to perform control decision on a plurality of automatic sorting devices based on the arrow belt product set and according to the composite label of each arrow belt, and determine an arrow belt sorting control scheme; the sorting risk prediction module is configured to introduce an arrow belt sorting risk analysis channel to perform sorting risk prediction on the arrow belt sorting control scheme, and obtain an arrow belt sorting risk sequence; the sorting control module is configured to adjust the arrow belt sorting control scheme according to arrow belt sorting risk constraints based on the arrow belt sorting risk sequence, and establish a first group of sorting control adjustment; and the sorting control optimization result generation module is configured to perform sorting comprehensive risk variation optimization on the first group of sorting control adjustment, generate a sorting control optimization result, and perform automatic sorting on the arrow belt product set in combination with the plurality of automatic sorting devices.

[0010] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0011] The wear-resistant feature of the arrow belt product set is analyzed through the wear-resistant grade confidence evaluation model to obtain wear-resistant grade labels of each arrow belt; anomaly detection is performed on the arrow belt product set to generate a composite label of each arrow belt; control decision is performed on a plurality of automatic sorting devices based on the arrow belt product set to determine an arrow belt sorting control scheme; the arrow belt sorting risk analysis channel is introduced to perform sorting risk prediction on the arrow belt sorting control scheme to obtain an arrow belt sorting risk sequence; the arrow belt sorting control scheme is adjusted according to arrow belt sorting risk constraints based on the arrow belt sorting risk sequence to establish a first group of sorting control adjustment; and sorting comprehensive risk variation optimization is performed on the first group of sorting control adjustment to generate a sorting control optimization result, and the arrow belt product set is automatically sorted. The technical effect of realizing automatic sorting of arrow belt products according to wear-resistant grade is achieved, and the accuracy and safety of sorting are improved. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0013] Figure 1 The arrow belt product wear-resistant grade automatic sorting method flowchart provided by the embodiment of the present application;

[0014] Figure 2The application provides an automatic sorting system for wear-resistant grades of finished arrow belts.

[0015] The application provides an automatic sorting system for wear-resistant grades of finished arrow belts. DETAILED DESCRIPTION

[0016] The application provides an automatic sorting system for wear-resistant grades of finished arrow belts.

[0017] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0018] Embodiment one, as shown in the application provides an automatic sorting method for wear-resistant grades of finished arrow belts, which comprises the following steps. Figure 1

[0019] Step S100: wear-resistant feature analysis of the finished arrow belt set is performed by using a wear-resistant grade confidence evaluation model, and wear-resistant grade labels of each arrow belt are obtained.

[0020] Specifically, the nth finished arrow belt (n is a positive integer) is extracted from the finished arrow belt set, and wear-resistant feature detection is performed to obtain nth wear-resistant detection data; the data is input into the wear-resistant grade confidence evaluation model to obtain nth wear-resistant grade evaluation results and a nth grade evaluation confidence coefficient; then, it is determined whether the confidence coefficient is greater than or equal to a grade evaluation confidence threshold value, if yes, the nth wear-resistant grade label is generated by combining the nth wear-resistant grade evaluation results and the nth grade evaluation confidence coefficient, and if no, the nth arrow belt re-inspection signal is generated, so as to complete the wear-resistant grade label marking of all arrow belts in the finished arrow belt set.

[0021] Step S200: abnormality detection is performed on the finished arrow belt set, and each arrow belt composite label is generated by combining the wear-resistant grade labels of the arrow belts.

[0022] ​Specifically, when performing anomaly detection on the arrow belt product set, the production monitoring parameters of the nth arrow belt product are collected to obtain nth arrow belt production data, and the data is input into a plurality of arrow belt anomaly detection models supervised and trained by the arrow belt production sample set and the anomaly detection sample set to obtain a plurality of anomaly detection results and generate the nth arrow belt anomaly label accordingly; subsequently, the anomaly label is combined with the corresponding nth arrow belt wear resistance grade label to form the nth arrow belt composite label, and the same method is used to process all arrows in the arrow belt product set, and finally the arrow belt composite labels are generated.

[0023] Step S300: Based on the arrow belt product set, control decisions are made on a plurality of automatic sorting devices according to the arrow belt composite labels to determine an arrow belt sorting control scheme.

[0024] Specifically, the arrow belt composite labels corresponding to the arrow belt product set are read through a data interface, and the wear resistance grade interval and anomaly type code contained in each type of label are parsed; the capability matrix database of the automatic sorting device is called to match the device group that can handle the corresponding grade and anomaly type, the optimal transfer route between the device groups is calculated through a path planning algorithm, and a control instruction set containing device ID, sorting order, and clamping force parameter is generated; at the same time, based on the priority field in the composite label (such as the P1 level of the high wear resistance no anomaly label), a sorting scheduling priority queue is configured, the control instruction set is sent to the PLC controller of each automatic sorting device in the order of the queue through an industrial bus, and finally an arrow belt sorting control scheme containing real-time scheduling logic and execution parameters is formed to ensure that arrow belts with different composite labels can be accurately distributed to the corresponding processing units.

[0025] Step S400: Introducing an arrow belt sorting risk analysis channel to predict the sorting risk of the arrow belt sorting control scheme to obtain an arrow belt sorting risk sequence.

[0026] Specifically, the plurality of automatic sorting devices are three-dimensionally reconstructed to construct a sorting integrated model; based on the model, the arrow belt product set is virtually sorted according to the arrow belt sorting control scheme to obtain a virtual sorting data set; since the arrow belt sorting risk analysis channel includes a sorting misplacement risk analysis model, a superimposed misgrabbing risk analysis model, and a sorting damage risk analysis model, the virtual sorting data set is input into these three models to correspondingly obtain a sorting misplacement risk coefficient, a superimposed misgrabbing risk coefficient, and a sorting damage risk coefficient, and the three risk coefficients are combined to form the arrow belt sorting risk sequence.

[0027] Step S500: Based on the arrow belt sorting risk sequence, the arrow belt sorting control scheme is adjusted according to the arrow belt sorting risk constraint optimization to establish a sorting control adjustment first group.

[0028] Specifically, after obtaining the arrow sorting risk sequence, the arrow sorting control scheme is optimized and adjusted based on the arrow sorting risk constraints (including sorting misplacement risk constraints, stacking misgrabbing risk constraints, and sorting damage risk constraints), and a process of establishing a first group of sorting control adjustment is performed. First, it is determined whether the arrow sorting risk sequence meets the arrow sorting risk constraints. If not, the original sorting control scheme is adjusted to generate multiple alternative sorting control adjustment schemes, forming a set of sorting control adjustment schemes. Then, the sorting risk of each scheme in the set is predicted through the arrow sorting risk analysis channel to obtain the sorting risk sequence corresponding to each scheme. Then, the sorting risk sequence of each scheme is compared with the arrow sorting risk constraints, and the schemes that meet the risk constraint requirements are selected according to the comparison results, and finally these schemes are integrated to establish the first group of sorting control adjustment.

[0029] Step S600: performing sorting comprehensive risk variation optimization on the first group of sorting control adjustment to generate a sorting control optimization result, and combining the multiple automatic sorting devices to perform automatic sorting on the arrow product set.

[0030] Specifically, the first group of sorting control adjustment is subjected to sorting comprehensive risk variation optimization to generate a sorting control optimization result, and the multiple automatic sorting devices are combined to complete the automatic sorting of the arrow product set. The specific process is as follows: first, configure weights according to multiple indexes such as sorting misplacement risk, stacking misgrabbing risk, and sorting damage risk, and establish a sorting comprehensive risk analysis model; use the model to select schemes with comprehensive risk less than a threshold value from the first group to form a second group of sorting control adjustment; after mutation processing of the second group, combine the arrow sorting risk analysis channel and the comprehensive risk analysis model for optimization to obtain a variation optimization group of sorting control; compare the second group with the variation optimization group to determine the sorting control optimization result with the goal of minimizing the comprehensive risk; and finally, control the multiple automatic sorting devices to implement automatic sorting on the arrow product set according to the result.

[0031] In one possible implementation, step S100 further includes:

[0032] Step S110: extracting an nth arrow product from the arrow product set, and performing wear-resistant feature detection on the nth arrow product to obtain nth wear-resistant detection data, n being a positive integer.

[0033] Step S120: inputting the nth wear-resistant detection data into the wear-resistant grade confidence evaluation model to obtain an nth wear-resistant grade evaluation result and an nth grade evaluation confidence coefficient.

[0034] Step S130: determining whether the nth grade evaluation confidence coefficient is greater than or equal to a grade evaluation confidence threshold.

[0035] Step S140: If the nth grade evaluation confidence coefficient is greater than or equal to the grade evaluation confidence threshold, a nth arrow belt wear grade label is generated according to the nth wear grade evaluation result and the nth grade evaluation confidence coefficient.

[0036] Specifically, the arrow belt product set is sequentially transmitted to the detection area by the automatic conveying line, the positioning sensor triggers the mechanical arm to grab the nth arrow belt product and fix it on the detection platform; the surface hardness tester is started to detect the hardness of the key wear area of the arrow belt, and the friction tester is started to simulate the friction process under the actual working condition of the arrow belt, and the wear amount and the friction coefficient change curve in unit time are recorded; at the same time, the high-resolution microscope is used to collect the surface texture image of the arrow belt, and the feature parameters such as the number and depth of the wear spots are extracted; the hardness data, wear amount data and texture feature parameters are integrated into structured nth wear detection data and stored in the local database for subsequent calling.

[0037] The nth wear detection data is input into the wear grade confidence evaluation model after normalization processing, and the model adopts the algorithm architecture based on gradient boosting decision tree. The model first performs hierarchical fitting on the wear characteristics (such as surface hardness, wear amount, texture characteristics, etc.) in the detection data through multiple decision trees, each tree iteratively optimizes the error of the previous model, and gradually improves the discrimination ability of the wear grade; at the same time, the out-of-bag data is used to evaluate the confidence of the prediction result of each tree, and the nth wear grade evaluation result (such as first, second, third, etc.) and the corresponding nth grade evaluation confidence coefficient are calculated by combining the voting weight in the decision tree integration process, so as to quantify the reliability of the evaluation result.

[0038] The nth grade evaluation confidence coefficient is compared with the set grade evaluation confidence threshold to determine whether the reliability of the evaluation result meets the requirements, which provides a basis for determining whether to generate a formal label or trigger a recheck.

[0039] If the nth grade evaluation confidence coefficient is greater than or equal to the grade evaluation confidence threshold, it means that the evaluation result has sufficient reliability, and the nth arrow belt wear grade label containing the wear grade of the arrow belt and the evaluation reliability information is generated by combining the nth wear grade evaluation result and the corresponding confidence coefficient, and the wear grade labeling of the arrow belt is completed.

[0040] In one possible implementation manner, step S200 further includes:

[0041] Step S210: Collect the production monitoring parameters of the nth arrow belt product to obtain the nth arrow belt production data.

[0042] Step S220: Supervise and train a plurality of learners with the arrow belt production sample set as input information and the arrow belt anomaly detection sample set as output information to generate a plurality of arrow belt anomaly detection models.

[0043] Step S230: inputting the nth arrow belt production data into the plurality of arrow belt anomaly detection models to obtain a plurality of arrow belt anomaly detection results.

[0044] Step S240: generating an nth arrow belt anomaly label according to the plurality of arrow belt anomaly detection results, and combining the nth arrow belt wear resistance grade label to obtain an nth arrow belt composite label.

[0045] Specifically, various sensors such as temperature sensors, pressure sensors, and speed sensors deployed on the production line are used to collect real-time key monitoring parameters of the nth arrow belt product during the production process. These parameters include raw material ratio, temperature change in the processing link, equipment operating pressure, production time, and other data closely related to the production process. After collecting, filtering, and standardizing these parameters, the nth arrow belt production data that can fully reflect the production status of the arrow belt is integrated, providing complete and accurate basic information for subsequent anomaly detection.

[0046] First, the arrow belt production sample set is preprocessed, including data cleaning to remove noise, standardization to make different parameters in the same order of magnitude, and converting the anomaly types (such as size deviation, material defects, etc.) in the arrow belt anomaly detection sample set into numerical labels. Random forest, gradient boosting tree, and multilayer perceptron are selected as multiple learners. The processed arrow belt production sample set and corresponding anomaly detection sample set are divided into training set and validation set in the ratio of 8:2. Each learner is iteratively trained through the training set, and the model hyperparameters (such as the number of trees for random forest and the learning rate for gradient boosting tree) are adjusted in a cross-validation manner. The model performance is evaluated using the validation set, and training is stopped when the accuracy and recall rate of the model on the validation set reach the preset threshold. Finally, multiple arrow belt anomaly detection models with performance meeting the standard are generated.

[0047] First, the nth arrow belt production data is preprocessed in the same way as the training sample set, such as data standardization and missing value filling, to ensure that its format and feature distribution are consistent with the model requirements. Then, the processed production data is input into the multiple arrow belt anomaly detection models that have been trained. Each model will independently analyze the production data based on its learned anomaly recognition rules and output corresponding anomaly detection results. These results include information such as whether there is an anomaly, the probability value of the anomaly, and the suspected anomaly category, providing multi-dimensional judgment basis for subsequent generation of anomaly labels.

[0048] The multiple arrow belt anomaly detection results are first fused, a weighted fusion algorithm based on model performance is used, a weight positively correlated with the detection accuracy of each model is given to the model, a comprehensive score of each anomaly type is calculated by weighting, the anomaly type with the highest score is determined as the final determination result, if the scores of all anomaly types are lower than a set threshold, it is determined as "no anomaly", and the nth arrow belt anomaly label is generated; then the data association module is used to structurally integrate the anomaly label and the nth arrow belt wear resistance grade label, form the nth arrow belt composite label containing the "wear resistance grade-anomaly type" field, and store it in the database for subsequent sorting control decision calling.

[0049] In a possible implementation manner, the step S400 further includes:

[0050] Step S410: performing three-dimensional reconstruction according to the multiple automatic sorting devices to obtain a sorting integrated model.

[0051] Step S420: based on the sorting integrated model, performing virtual sorting on the arrow belt finished product set according to the arrow belt sorting control scheme to obtain a virtual sorting data set.

[0052] Step S430: the arrow belt sorting risk analysis channel includes a sorting misplacement risk analysis model, a superposition misgrab risk analysis model and a sorting damage risk analysis model.

[0053] Step S440: inputting the virtual sorting data set into the sorting misplacement risk analysis model to obtain a sorting misplacement risk coefficient.

[0054] Step S450: inputting the virtual sorting data set into the superposition misgrab risk analysis model to obtain a superposition misgrab risk coefficient.

[0055] Step S460: inputting the virtual sorting data set into the sorting damage risk analysis model to obtain a sorting damage risk coefficient, combining the sorting misplacement risk coefficient and the superposition misgrab risk coefficient to generate the arrow belt sorting risk sequence.

[0056] Specifically, the appearance structure, size parameters of each component, relative position relationship and stroke range of the movement shaft of the multiple automatic sorting devices are comprehensively scanned by using a three-dimensional scanning device to obtain a large amount of three-dimensional point cloud data; then the data are denoised, registered and fused by using a point cloud preprocessing software to eliminate the redundant information and errors in the data, and the model details are further optimized and adjusted in combination with the design drawings of the devices to finally construct the sorting integrated model which can accurately reflect the overall layout and collaborative workspace of the multiple automatic sorting devices.

[0057] After obtaining the sorting integrated model reconstructed by the three-dimensional reconstruction of the multiple automatic sorting devices, the model is taken as a virtual operation platform, and the simulation sorting operation is performed on all arrow products concentrated in the arrow product according to the determined arrow sorting control scheme. The virtual sorting process can completely reproduce the device actions, arrow transmission path, sorting logic and other key links that may be involved in the actual sorting. By collecting and arranging various data generated during the simulation process, such as the flow information of the arrow between devices, the parameter record of the device action, the interaction state of the arrow and the device, and the like, a virtual sorting data set is finally formed, which provides comprehensive and detailed simulation data support for subsequent risk analysis.

[0058] The arrow sorting risk analysis channel is a comprehensive analysis channel specially used for analyzing various risks in the arrow sorting process. It includes three core models, namely, a sorting misposition risk analysis model, a superimposed mistaken grabbing risk analysis model, and a sorting damage risk analysis model. The sorting misposition risk analysis model is mainly used to analyze the risk of position deviation of the arrow in the sorting process and the failure to accurately enter the target sorting area. The superimposed mistaken grabbing risk analysis model focuses on evaluating the risk of mistaken grabbing of multiple arrows by the sorting device when the arrows are superimposed. The sorting damage risk analysis model focuses on the risk of damage to the arrow in the sorting process due to collision and extrusion with the device. These three models cooperate with each other to build a comprehensive risk analysis system for arrow sorting from different dimensions.

[0059] The construction of the sorting misposition risk analysis model uses the gradient boosting decision tree algorithm. First, the real-time position coordinates of the arrow, the running trajectory parameters of the sorting device, the arrow transmission speed, the device response delay, and other feature variables related to sorting misposition are extracted from the virtual sorting data set to form the model input feature set. Then, the actual misposition events and corresponding feature data in the historical sorting process are used as training samples, where the positive samples are the cases of misposition, and the negative samples are the cases of normal sorting. The GBDT model is trained, and multiple decision trees are constructed through continuous iteration. Each new tree is fitted to the prediction residual of the previous model, and the model's ability to identify sorting misposition risk is gradually optimized. After training, the virtual sorting data set to be analyzed is input into the model, and the model will perform comprehensive calculations based on the importance weights of each feature variable. The output is the risk probability value of the arrow's position deviation and failure to accurately enter the target area during the sorting process. This value is the sorting misposition risk coefficient, which quantitatively represents the level of sorting misposition risk.

[0060] The construction of the superimposed mis-grab risk analysis model adopts a support vector machine (SVM) algorithm: first, the stacking thickness of the arrow tape, the spacing between adjacent arrow tapes, the grabbing force of the sorting device, the friction coefficient of the surface of the arrow tape and other characteristic variables related to the superimposed mis-grab are extracted from the virtual sorting data set to construct the input feature space of the model; then, the case data of the superimposed mis-grab occurring in the historical sorting is taken as the positive sample, and the case data of the normal grabbing is taken as the negative sample, the SVM model is trained, the two types of samples are distinguished by finding the optimal hyperplane, and the kernel function is introduced to map the low-dimensional feature space to the high-dimensional space to solve the non-linear relationship between the characteristic variables and improve the recognition accuracy of the model to the superimposed mis-grab risk; after the training is completed, the virtual sorting data set to be analyzed is input into the model, the model calculates based on the relative position of the characteristic variables and the optimal hyperplane, and outputs the probability value of the arrow tape being mis-grabbed due to superimposition in the sorting process. The value is the superimposed mis-grab risk coefficient, which quantitatively represents the degree of the superimposed mis-grab risk.

[0061] The construction of the sorting damage risk analysis model adopts a convolutional neural network (CNN) algorithm: first, the pressure distribution image of the contact point between the arrow tape and the sorting device, the surface deformation data of the arrow tape, the running acceleration parameter of the device, the contact time and other characteristic information related to the sorting damage are extracted from the virtual sorting data set, which is converted into a multi-dimensional feature matrix suitable for CNN processing; then, the cases in which the arrow tape is damaged (marked as positive samples) and the cases in which the arrow tape is not damaged (marked as negative samples) in the historical sorting are taken as training data, the CNN model is trained, the local key features in the feature matrix are extracted through the convolution layer, the feature dimension is compressed through the pooling layer, and then the feature integration and risk classification learning are performed through the full connection layer. The network parameters are constantly optimized to minimize the prediction error; after the training is completed, the virtual sorting data set to be analyzed is input into the model, the model evaluates the possibility of damage to the arrow tape due to collision, extrusion and other reasons in the sorting process, and outputs the corresponding probability value as the sorting damage risk coefficient. Then, the sorting damage risk coefficient obtained through the above CNN model and the sorting misalignment risk coefficient and the superimposed mis-grab risk coefficient obtained before are integrated and arranged in a predetermined order to form an arrow tape sorting risk sequence containing the three coefficients. The sequence comprehensively reflects the quantification degree of various risks faced by the arrow tape in the sorting process.

[0062] In a possible implementation manner, the step S500 further includes:

[0063] Step S510: If the arrow tape sorting risk sequence does not satisfy the arrow tape sorting risk constraint, adjusting the arrow tape sorting control scheme to obtain a set of sorting control adjustment schemes.

[0064] Step S520: According to the arrow belt sorting risk analysis channel, the sorting risk of each scheme in the sorting control adjustment scheme set is predicted to obtain the sorting risk sequence of each scheme.

[0065] Step S530: Compare the sorting risk sequence of each scheme with the arrow belt sorting risk constraint to obtain the risk comparison result of each scheme.

[0066] Step S540: According to the risk comparison result of each scheme, the sorting control adjustment scheme set is screened to generate the first group of sorting control adjustment schemes.

[0067] Specifically, when any one or more of the sorting misplacement risk coefficient, the superimposed misgrab risk coefficient, and the sorting damage risk coefficient contained in the arrow belt sorting risk sequence exceeds the corresponding arrow belt sorting risk constraint (i.e., does not meet the preset risk threshold requirement), the original arrow belt sorting control scheme needs to be adjusted. In the adjustment process, the key operating parameters of the automatic sorting device are adjusted in multiple dimensions and multiple combinations, such as changing the operating speed of the sorting device, adjusting the force parameter of the grabbing mechanism, optimizing the transmission path planning of the arrow belt, etc. Through different combinations of these parameters, multiple sorting control adjustment schemes with differences are formed, and then integrated to form the sorting control adjustment scheme set, providing a basis for subsequent risk assessment and screening.

[0068] According to the arrow belt sorting risk analysis channel, the sorting risk of each scheme in the sorting control adjustment scheme set is predicted: for each adjustment scheme, first, based on the sorting integrated model reconstructed by multiple automatic sorting devices, the arrow belt finished product set is virtually sorted according to the scheme to obtain the corresponding virtual sorting data set; then the virtual sorting data set is input into the sorting misplacement risk analysis model, the superimposed misgrab risk analysis model and the sorting damage risk analysis model respectively to obtain the sorting misplacement risk coefficient, the superimposed misgrab risk coefficient and the sorting damage risk coefficient corresponding to the scheme; finally, the three coefficients are combined to form the sorting risk sequence of the scheme. By executing the above operation on all schemes in the scheme set one by one, the sorting risk sequence of each scheme is finally obtained.

[0069] The sorting risk sequence (including the sorting misplacement risk coefficient, the superimposed misgrab risk coefficient, and the sorting damage risk coefficient) of each scheme in the sorting control adjustment scheme set is compared with the arrow belt sorting risk constraint (corresponding to the sorting misplacement risk constraint, the superimposed misgrab risk constraint, and the sorting damage risk constraint) item by item, and the three risk coefficients of each scheme are checked whether they are less than or equal to the corresponding risk constraint threshold respectively. At the same time, the difference or deviation between each risk coefficient and the corresponding constraint threshold is recorded to form the risk comparison result of each scheme, which clearly presents whether each adjustment scheme meets the risk control requirement and the specific degree of meeting the requirement.

[0070] According to the risk ratio results of each scheme, the sorting control adjustment scheme set is optimized and screened: those adjustment schemes in which the sorting misplacement risk coefficient, the stacking misgrab risk coefficient, and the sorting damage risk coefficient all meet the corresponding arrow belt sorting risk constraints (i.e., each coefficient does not exceed the preset threshold) are preferentially selected; for the case where individual risk coefficients in some schemes are close to but do not exceed the constraint threshold, and the overall risk level is relatively low, it is also included in the screening range. Through such a screening process, the schemes that meet the requirements are determined from the sorting control adjustment scheme set to collectively constitute the sorting control adjustment first group.

[0071] In a possible implementation manner, step S600 further includes:

[0072] Step S610: According to the sorting risk multi-index, a weight configuration is performed to establish a sorting comprehensive risk analysis model, the sorting risk multi-index including a sorting misplacement risk, a stacking misgrab risk, and a sorting damage risk.

[0073] Step S620: According to the sorting comprehensive risk analysis model, a sorting comprehensive risk optimization is performed on the sorting control adjustment first group to establish a sorting control adjustment second group smaller than a sorting comprehensive risk threshold.

[0074] Step S630: According to the arrow belt sorting risk analysis channel and the sorting comprehensive risk analysis model, a variation optimization is performed on the sorting control adjustment second group to establish a sorting control variation optimization group.

[0075] Step S640: According to the sorting control adjustment second group and the sorting control variation optimization group, a sorting comprehensive risk minimization optimization is performed to obtain the sorting control optimization result.

[0076] Specifically, for the three sorting risk multi-indexes of the sorting misplacement risk, the stacking misgrab risk, and the sorting damage risk, an analytic hierarchy process can be used for weight configuration: first, a judgment matrix is constructed, and the relative importance values are determined by comparing the influence degrees (such as the occurrence probability and the severity of loss caused) of each risk index in the arrow belt sorting process; then, the characteristic vector of the judgment matrix is calculated to obtain the weight values of each index, so as to ensure that the weight distribution conforms to the actual risk influence priority. Based on the weight configuration, a sorting comprehensive risk analysis model is established, the model obtains a comprehensive risk value by multiplying the sorting misplacement risk coefficient, the stacking misgrab risk coefficient, and the sorting damage risk coefficient by the corresponding weight and then summing, so as to realize the quantitative analysis of the overall risk of the sorting scheme.

[0077] The established sorting comprehensive risk analysis model is used to perform comprehensive risk evaluation on each sorting control adjustment scheme in the first group of sorting control adjustments. The sorting misplacement risk coefficient, the stacking mis-grabbing risk coefficient, and the sorting damage risk coefficient corresponding to each scheme are substituted into the model, and the sorting comprehensive risk value of each scheme is calculated by combining the pre-configured weights of the risk indicators. Then, the sorting comprehensive risk values of the schemes are compared with the preset sorting comprehensive risk threshold, and the schemes with a sorting comprehensive risk value less than the threshold are selected to form a second group of sorting control adjustments, thereby further narrowing the range based on the first group and retaining the alternative schemes with a lower comprehensive risk level.

[0078] The genetic algorithm is used for mutation optimization. First, the schemes in the second group of sorting control adjustments are encoded as chromosomes, and the chromosome genes (corresponding to the sorting device running speed, grabbing force, etc.) are randomly mutated by setting a mutation probability (such as a mutation rate of 5% to 10% based on the sensitivity of the scheme parameters) to generate a primary population after mutation. Then, the risk of each scheme in the primary population after mutation is predicted using the arrow belt sorting risk analysis channel, and the individuals with a risk sequence satisfying the constraints are selected as an intermediate population. The comprehensive risk values of the intermediate population are calculated as the fitness by the sorting comprehensive risk analysis model, and the individuals with a fitness higher than a threshold (i.e., a comprehensive risk value lower than the sorting comprehensive risk threshold) are retained. The mutation and selection are repeated for multiple rounds, and finally a sorting control mutation optimization group is formed.

[0079] All the sorting control schemes in the second group of sorting control adjustments and the sorting control mutation optimization group are summarized, and the comprehensive risk values of each scheme are calculated by the sorting comprehensive risk analysis model. The scheme with the smallest comprehensive risk value is selected from the comparison of the comprehensive risk values, which is the sorting control optimization result. This ensures that the finally determined sorting control scheme minimizes the sorting comprehensive risk while meeting various risk constraints, and provides an optimal control basis for automatic sorting of the arrow belt finished product set.

[0080] In one possible implementation, step S630 further includes:

[0081] Step S631: Mutating the second group of sorting control adjustments to obtain a first group of sorting control mutations.

[0082] Step S632: Based on the arrow belt sorting risk analysis channel, performing sorting risk optimization on the first group of sorting control mutations according to the arrow belt sorting risk constraints to generate a second group of sorting control mutations.

[0083] Step S633: Based on the sorting comprehensive risk analysis model, performing sorting comprehensive risk optimization on the second group of sorting control mutations according to the sorting comprehensive risk threshold to obtain a sorting control mutation optimization group.

[0084] Specifically, the genetic algorithm is adopted to mutate the second group of sorting controls to obtain a first group of sorting control mutations: first, each sorting control scheme in the second group is coded into a chromosome, and the genes on the chromosome correspond to the operating parameters (such as speed, force, path nodes, etc.) of the sorting device; then, according to a preset mutation probability, some genes on the chromosome are randomly mutated, that is, the corresponding parameters are randomly adjusted within a reasonable range, and in this way, a plurality of new schemes after mutation are generated from each original scheme, and all these mutated schemes together constitute the first group of sorting control mutations.

[0085] Relying on the arrow belt sorting risk analysis channel composed of the sorting misplacement risk analysis model, the superimposed misgrab risk analysis model and the sorting damage risk analysis model, the sorting risk of each scheme in the first group of sorting control mutations is predicted: first, based on the automatic sorting device parameters corresponding to each scheme, a sorting integrated model is constructed, virtual sorting is performed to obtain a virtual sorting data set, and then the data set is input into the three risk analysis models respectively to obtain the sorting misplacement risk coefficient, the superimposed misgrab risk coefficient and the sorting damage risk coefficient of each scheme, forming a complete sorting risk sequence; then, the sorting risk sequence of each scheme is compared with the arrow belt sorting risk constraints (including the constraint thresholds of the three risks) one by one, and the schemes whose risk coefficients all satisfy the corresponding constraint conditions are selected, and these schemes are integrated to generate the second group of sorting control mutations.

[0086] Using the established sorting comprehensive risk analysis model, the comprehensive risk of each scheme in the second group of sorting control mutations is evaluated, the sorting misplacement risk coefficient, the superimposed misgrab risk coefficient and the sorting damage risk coefficient of each scheme are substituted into the model, and the comprehensive risk value of each scheme is calculated by combining the preset weights of each risk index; then, the comprehensive risk values of each scheme are compared with the preset sorting comprehensive risk threshold, and the schemes whose comprehensive risk values are less than the threshold are selected, and these schemes are integrated to form the second group of sorting control mutations.

[0087] In one possible implementation, step S140 further includes:

[0088] Step S141: If the nth level evaluation confidence coefficient is less than the level evaluation confidence threshold, an nth arrow belt recheck signal is generated.

[0089] Specifically, when it is determined that the grade evaluation confidence coefficient of the nth arrow strip product is less than the preset grade evaluation confidence threshold, a re-inspection mechanism is triggered, and the nth arrow strip re-inspection signal containing the unique identifier (such as the number, production batch, etc.) of the arrow strip is automatically generated by the signal generation module. The signal is sent to the terminal device (such as a computer display screen, a handheld terminal) of the re-inspection workstation and a prompt sound is emitted at the same time, and the state of the arrow strip is marked as “to be re-inspected” in the database, so that the re-inspection personnel can quickly locate and perform secondary wear-resistant feature detection, thereby ensuring the accuracy of subsequent sorting.

[0090] In one possible implementation manner, the step S500 further includes:

[0091] The arrow strip sorting risk constraint includes a sorting misposition risk constraint, a superimposed misgrab risk constraint, and a sorting damage risk constraint.

[0092] Specifically, the arrow strip sorting risk constraint is a series of standards for evaluating and regulating the acceptable degree of risk in the arrow strip sorting process, and specifically includes three aspects: the sorting misposition risk constraint, which limits the upper limit of the risk coefficient of the arrow strip deviating from the preset track or target area during sorting; the superimposed misgrab risk constraint, which limits the maximum value of the risk coefficient of multiple arrow strips being simultaneously misgrabbed or superimposed grabbing; and the sorting damage risk constraint, which specifies that the risk coefficient of physical damage of the arrow strip caused by collision, extrusion, etc. during sorting operation cannot exceed a threshold. The three types of constraints together constitute the basic control standard of the arrow strip sorting risk, ensuring the safety and accuracy of the sorting process.

[0093] In the second embodiment, based on the same inventive concept as the arrow strip product wear-resistant grade automatic sorting method in the foregoing embodiments, as shown in the following table, the present application provides an arrow strip product wear-resistant grade automatic sorting system, and the system and method embodiments in the present application are based on the same inventive concept. The system includes: Figure 2

[0094] The wear-resistant grade label acquisition module 10 is configured to analyze the wear-resistant features of the arrow strip product set by using a wear-resistant grade confidence evaluation model to obtain wear-resistant grade labels of the arrow strip products.

[0095] The anomaly detection module 20 is configured to perform anomaly detection on the arrow strip product set, and generate composite labels of the arrow strip products in combination with the wear-resistant grade labels of the arrow strip products.

[0096] The control decision module 30 is configured to perform control decision on a plurality of automatic sorting devices based on the arrow strip product set and according to the composite labels of the arrow strip products, and determine an arrow strip sorting control scheme.

[0097] ​The sorting risk prediction module 40 is configured to introduce the arrow belt sorting risk analysis channel to the arrow belt sorting control scheme to obtain an arrow belt sorting risk sequence.

[0098] The sorting control module 50 is configured to adjust the arrow belt sorting control scheme according to the arrow belt sorting risk constraint to establish a sorting control adjustment first group.

[0099] The sorting control optimization result generation module 60 is configured to perform sorting comprehensive risk variation optimization on the sorting control adjustment first group to generate a sorting control optimization result, and combine the multiple automatic sorting devices to automatically sort the arrow belt finished product set.

[0100] Further, the system is also configured to implement the following functions:

[0101] According to the arrow belt finished product set, the nth arrow belt finished product is extracted, and wear-resistant feature detection is performed on the nth arrow belt finished product to obtain nth wear-resistant detection data, n is a positive integer; the nth wear-resistant detection data is input into the wear-resistant grade confidence evaluation model to obtain an nth wear-resistant grade evaluation result and an nth grade evaluation confidence coefficient; it is judged whether the nth grade evaluation confidence coefficient is greater than or equal to a grade evaluation confidence threshold; if the nth grade evaluation confidence coefficient is greater than or equal to the grade evaluation confidence threshold, according to the nth wear-resistant grade evaluation result and the nth grade evaluation confidence coefficient, an nth arrow belt wear-resistant grade label is generated.

[0102] Further, the system is also configured to implement the following functions:

[0103] The production monitoring parameters of the nth arrow belt finished product are collected to obtain nth arrow belt production data; the arrow belt production sample set is taken as input information, and the arrow belt abnormality detection sample set is taken as output information, and multiple learners are supervised and trained to generate multiple arrow belt abnormality detection models; the nth arrow belt production data is input into the multiple arrow belt abnormality detection models to obtain multiple arrow belt abnormality detection results; the nth arrow belt abnormality label is generated according to the multiple arrow belt abnormality detection results, and the nth arrow belt composite label is obtained in combination with the nth arrow belt wear-resistant grade label.

[0104] Further, the system is also configured to implement the following functions:

[0105] According to the plurality of automatic sorting devices, three-dimensional reconstruction is performed to obtain a sorting integrated model; based on the sorting integrated model, the arrow belt finished product set is virtually sorted according to the arrow belt sorting control scheme to obtain a virtual sorting data set; the arrow belt sorting risk analysis channel includes a sorting misposition risk analysis model, a superimposed mistaken grabbing risk analysis model, and a sorting damage risk analysis model; the virtual sorting data set is input into the sorting misposition risk analysis model to obtain a sorting misposition risk coefficient; the virtual sorting data set is input into the superimposed mistaken grabbing risk analysis model to obtain a superimposed mistaken grabbing risk coefficient; the virtual sorting data set is input into the sorting damage risk analysis model to obtain a sorting damage risk coefficient, and the sorting misposition risk coefficient and the superimposed mistaken grabbing risk coefficient are combined to generate the arrow belt sorting risk sequence.

[0106] Further, the system is also used to implement the following functions:

[0107] If the arrow belt sorting risk sequence does not satisfy the arrow belt sorting risk constraint, the arrow belt sorting control scheme is adjusted to obtain a sorting control adjustment scheme set; the sorting risk of each scheme in the sorting control adjustment scheme set is predicted according to the arrow belt sorting risk analysis channel to obtain a sorting risk sequence of each scheme; the sorting risk sequence of each scheme is compared with the arrow belt sorting risk constraint to obtain a risk comparison result of each scheme; and the sorting control adjustment scheme set is optimized and screened according to the risk comparison result of each scheme to generate a sorting control adjustment first group.

[0108] Further, the system is also used to implement the following functions:

[0109] According to the sorting risk multi-index, weight configuration is performed to establish a sorting comprehensive risk analysis model, and the sorting risk multi-index includes a sorting misposition risk, a superimposed mistaken grabbing risk, and a sorting damage risk; the sorting comprehensive risk analysis model is used to perform sorting comprehensive risk optimization on the sorting control adjustment first group to establish a sorting control adjustment second group smaller than a sorting comprehensive risk threshold value; the sorting control adjustment second group is subjected to variation optimization according to the arrow belt sorting risk analysis channel and the sorting comprehensive risk analysis model to establish a sorting control variation optimization group; and sorting comprehensive risk minimization optimization is performed according to the sorting control adjustment second group and the sorting control variation optimization group to obtain the sorting control optimization result.

[0110] Further, the system is also used to implement the following functions:

[0111] The sorting control adjustment second group is mutated to obtain a sorting control mutation first group; based on the arrow belt sorting risk analysis channel, the sorting control mutation first group is subjected to sorting risk optimization according to the arrow belt sorting risk constraint to generate a sorting control mutation second group; based on the sorting comprehensive risk analysis model, the sorting control mutation second group is subjected to sorting comprehensive risk optimization according to the sorting comprehensive risk threshold to obtain the sorting control mutation optimization group.

[0112] Further, the system is also used to implement the following functions:

[0113] If the nth grade evaluation confidence coefficient is less than the grade evaluation confidence threshold, an nth arrow belt re-inspection signal is generated.

[0114] Further, the system is also used to implement the following functions:

[0115] The arrow belt sorting risk constraint includes a sorting misposition risk constraint, a superposition mis-grab risk constraint and a sorting damage risk constraint.

[0116] It should be noted that the above sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above describes the specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0117] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0118] The present application and the drawings are only exemplary descriptions of the present application, and are considered to cover any and all modifications, changes, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art can make various modifications and changes to the present application without departing from the scope of the present application. Thus, if these modifications and changes of the present application belong to the scope of the present application and its equivalent technology, the present application intends to include these modifications and changes.

Claims

1. A method for automatic sorting of finished girdle abrasion grades, characterized in that, The method includes: The wear resistance characteristics of the finished arrow belt set were analyzed using a wear resistance level confidence assessment model to obtain the wear resistance level label for each arrow belt; Anomaly detection is performed on the finished arrow belt assembly, and a composite label for each arrow belt is generated by combining the wear resistance grade labels of each arrow belt. Based on the finished arrow belt set, control decisions are made for multiple automatic sorting devices according to the composite labels of each arrow belt to determine the arrow belt sorting control scheme. A risk analysis channel for arrow belt sorting is introduced to predict the sorting risk of the arrow belt sorting control scheme and obtain the arrow belt sorting risk sequence. Based on the arrow belt sorting risk sequence, the arrow belt sorting control scheme is optimized and adjusted according to the arrow belt sorting risk constraints to establish the first group of sorting control adjustment; The sorting control adjustment group is optimized by performing comprehensive risk variation search to generate sorting control optimization results, and the arrow belt finished product set is automatically sorted by combining the multiple automatic sorting devices. Specifically, a risk analysis channel for arrow belt sorting is introduced to predict the sorting risk of the arrow belt sorting control scheme, obtaining an arrow belt sorting risk sequence, including: A sorting integrated model is obtained by performing three-dimensional reconstruction based on the multiple automated sorting devices. Based on the sorting integration model, the finished arrow belt set is virtually sorted according to the arrow belt sorting control scheme to obtain a virtual sorting dataset; The arrow belt sorting risk analysis channel includes a sorting misalignment risk analysis model, an overlay misgrabbing risk analysis model, and a sorting damage risk analysis model; Input the virtual sorting dataset into the sorting misalignment risk analysis model to obtain the sorting misalignment risk coefficient; Input the virtual sorting dataset into the superimposed mis-grabbing risk analysis model to obtain the superimposed mis-grabbing risk coefficient; The virtual sorting dataset is input into the sorting damage risk analysis model to obtain the sorting damage risk coefficient. The sorting misalignment risk coefficient and the superimposed misgrabbing risk coefficient are combined to generate the arrow belt sorting risk sequence. The sorting control adjustment group is optimized by performing comprehensive risk variation search to generate sorting control optimization results, including: A comprehensive risk analysis model for sorting is established by weighting multiple risk indicators. These multiple risk indicators include sorting misalignment risk, superimposed misgrabbing risk, and sorting damage risk. Based on the sorting comprehensive risk analysis model, the sorting control regulation group 1 is optimized for sorting comprehensive risk, and a sorting control regulation group 2 with a comprehensive risk threshold is established. Based on the arrow belt sorting risk analysis channel and the sorting comprehensive risk analysis model, the second group of sorting control adjustment is subjected to mutation optimization to establish a sorting control mutation optimization group. Based on the sorting control adjustment second group and the sorting control variation optimization group, the sorting comprehensive risk minimization optimization is performed to obtain the sorting control optimization result.

2. The method of claim 1, wherein the method further comprises: The abrasion resistance characteristics of the finished arrow belt set were analyzed using an abrasion resistance level confidence assessment model to obtain abrasion resistance level labels for each arrow belt, including: Extract the nth arrow belt finished product from the arrow belt finished product set, and perform wear resistance feature detection on the nth arrow belt finished product to obtain the nth wear resistance detection data, where n is a positive integer; input the nth abrasion detection data into the abrasion grade confidence evaluation model to obtain an nth abrasion grade evaluation result and an nth grade evaluation confidence coefficient; determine whether the nth grade evaluation confidence coefficient is greater than or equal to a grade evaluation confidence threshold value; if the nth grade evaluation confidence coefficient is greater than or equal to the grade evaluation confidence threshold value, generate an nth arrow belt abrasion grade label according to the nth abrasion grade evaluation result and the nth grade evaluation confidence coefficient.

3. The method of claim 1, wherein the method further comprises: perform anomaly detection on the arrow belt product set and generate a composite label for each arrow belt by combining the abrasion grade labels of the arrow belts, including: collect production monitoring parameters of an nth arrow belt product to obtain nth arrow belt production data; perform supervised training on multiple learners by taking the arrow belt production sample set as input information and the arrow belt anomaly detection sample set as output information to generate multiple arrow belt anomaly detection models; input the nth arrow belt production data into the multiple arrow belt anomaly detection models to obtain multiple arrow belt anomaly detection results; generate an nth arrow belt anomaly label according to the multiple arrow belt anomaly detection results and obtain an nth arrow belt composite label by combining the nth arrow belt abrasion grade label.

4. The method of claim 1, wherein the method further comprises: based on the arrow belt sorting risk sequence, adjust the arrow belt sorting control scheme according to arrow belt sorting risk constraints to establish a first group of sorting control adjustments, including: if the arrow belt sorting risk sequence does not satisfy the arrow belt sorting risk constraints, adjust the arrow belt sorting control scheme to obtain a set of sorting control adjustment schemes; perform sorting risk prediction on each scheme in the set of sorting control adjustment schemes according to the arrow belt sorting risk analysis channel to obtain a sorting risk sequence for each scheme; compare the sorting risk sequence of each scheme with the arrow belt sorting risk constraints to obtain a risk comparison result for each scheme; select the set of sorting control adjustment schemes according to the risk comparison result of each scheme to generate the first group of sorting control adjustments.

5. The method of claim 1, wherein the method further comprises: perform mutation optimization on the second group of sorting control adjustments according to the arrow belt sorting risk analysis channel and the sorting comprehensive risk analysis model to establish a second group of sorting control mutations, including: mutate the second group of sorting control adjustments to obtain a first group of sorting control mutations; perform sorting risk optimization on the first group of sorting control mutations according to the arrow belt sorting risk constraints based on the arrow belt sorting risk analysis channel to generate a second group of sorting control mutations; perform sorting comprehensive risk optimization on the second group of sorting control mutations according to the sorting comprehensive risk threshold value based on the sorting comprehensive risk analysis model to obtain the second group of sorting control mutations.

6. The method of claim 2, wherein the method further comprises: if the nth grade evaluation confidence coefficient is less than the grade evaluation confidence threshold value, generate an nth arrow belt re-inspection signal.

7. The method of claim 1, wherein the method further comprises: The arrow belt sorting risk constraints include a sorting misplacement risk constraint, a superimposed mis-grabbing risk constraint, and a sorting damage risk constraint.

8. An automatic sorting system for the wear grade of finished garters, characterized by, The system is used to implement the arrow belt product abrasion grade automatic sorting method according to any one of claims 1-7, and the system includes: an abrasion grade label acquisition module configured to analyze the abrasion characteristics of the arrow belt product set by an abrasion grade confidence evaluation model to obtain abrasion grade labels for each arrow belt; Anomaly detection module, for performing anomaly detection on the arrow belt product set, combining the arrow belt wear resistance grade labels to generate arrow belt composite labels; Control decision module, for performing control decision on a plurality of automatic sorting devices based on the arrow belt product set according to the arrow belt composite labels, and determining an arrow belt sorting control scheme; Sorting risk prediction module, for introducing an arrow belt sorting risk analysis channel to perform sorting risk prediction on the arrow belt sorting control scheme, and obtaining an arrow belt sorting risk sequence; Sorting control module, for adjusting the arrow belt sorting control scheme according to arrow belt sorting risk constraint optimization based on the arrow belt sorting risk sequence, and establishing a sorting control adjustment first group; Sorting control optimization result generation module, for performing sorting comprehensive risk variation optimization on the sorting control adjustment first group, generating a sorting control optimization result, and combining the plurality of automatic sorting devices to perform automatic sorting on the arrow belt product set.

Images