An artificial intelligence-based inspection detection management system

Abstract

The application is suitable for the technical field of product detection, and provides an artificial intelligence-based inspection and detection management system, which comprises: a product classification module, which is used for receiving medicine category to-be-inspected product information and cosmetic category to-be-inspected product information, performing classification labeling based on a preset industry classification standard, and obtaining a primary classification result set; a clustering optimization module, an embedded quantum genetic simulated annealing clustering algorithm unit of which performs quantum genetic operation and simulated annealing optimization through iteration, and outputs a target clustering result meeting a fitness function threshold value; a task construction module, which constructs a multi-target optimization task allocation model according to the target clustering result; and a task allocation module, which generates a detection task sequence based on an optimal solution set output by the target task allocation model, adjusts a task queue according to real-time state data of an inspection device, and sends an execution instruction to the corresponding inspection device. The application improves the classification precision of products by improving the clustering algorithm, and effectively improves the inspection capacity and detection efficiency by using the allocation model.

Description

Technical Field

[0001] This invention relates to the field of product testing technology, and in particular to an artificial intelligence-based inspection and testing management system. Background Technology

[0002] In the testing and inspection processes of industries such as pharmaceuticals and cosmetics, the variety of products is increasing, and testing standards are becoming more stringent. Currently, the product testing and inspection field generally adopts a model combining traditional automated equipment with manual sampling, relying on technologies such as optical inspection, mechanical sensing, and image recognition for basic product quality screening. While current image classification models can complete the testing and inspection process, their detection capabilities are limited by insufficient algorithm generalization and low levels of intelligence, which can easily lead to misjudgments. Furthermore, due to the diversity of product types, relying on fixed parameter thresholds and limited sample libraries makes it difficult to dynamically adapt to complex scenarios, resulting in low detection efficiency when facing complex situations.

[0003] In view of this, an artificial intelligence-based inspection and testing management system is proposed. Summary of the Invention

[0004] This invention provides an artificial intelligence-based inspection and testing management system to solve the problems of low product inspection capabilities and low testing efficiency.

[0005] This invention provides an artificial intelligence-based inspection and testing management system, comprising: The product classification module is used to receive information on pharmaceutical products and cosmetic products to be inspected, classify and label them based on preset industry classification standards, and obtain a set of primary classification results. The clustering optimization module has a built-in quantum genetic simulated annealing clustering algorithm unit. The quantum genetic simulated annealing clustering algorithm unit outputs the target clustering result that satisfies the fitness function threshold by iteratively executing quantum genetic operations and simulated annealing optimization. The task construction module constructs a multi-objective optimization task allocation model based on the target clustering results; wherein the target parameters include maximizing detection efficiency, balancing equipment load, and minimizing testing costs, and the constraints include detection time constraints, maximum equipment throughput constraints, and reagent and consumable cost constraints. The task allocation module generates a sequence of detection tasks based on the optimal solution set output by the multi-objective optimized task allocation model, adjusts the task queue according to the real-time status data of the inspection equipment, and sends execution instructions to the corresponding inspection equipment.

[0006] Furthermore, the quantum genetic simulated annealing clustering algorithm unit iteratively executes quantum genetic operations and simulated annealing optimization to output target clustering results that satisfy the fitness function threshold, including: Based on the products to be inspected under each industry category in the primary classification result set, an initial population containing M candidate clustering schemes is generated, wherein each candidate clustering scheme consists of N cluster centers, and the position coordinates of the cluster centers are represented by qubit encoding. Calculate the feature clustering degree of similar products and the feature discrimination degree between different categories of products in each candidate clustering scheme, and calculate the fitness value of each candidate clustering scheme based on the feature clustering degree and feature discrimination degree; Quantum rotation gate and quantum NOT gate operations are performed on the candidate clustering schemes encoded by qubits. The correlation update between different candidate schemes is realized through the quantum entanglement mechanism to generate a new population. The new population is subjected to a neighborhood search at a preset initial annealing temperature. Based on preset criteria, inferior solutions are accepted or rejected, and the annealing temperature is reduced according to an exponential decay strategy. When the number of iterations reaches a preset threshold, the current optimal candidate clustering scheme is output as the target clustering result.

[0007] Furthermore, the calculation of the feature clustering degree of similar products and the feature discrimination degree between different categories of products in each candidate clustering scheme, and the calculation of the fitness value of each candidate clustering scheme based on the feature clustering degree and feature discrimination degree, includes: The formula for calculating the feature clustering degree is: in: This represents the total number of categories in the current candidate clustering scheme. Let be the number of dimensions of the product feature vector. For the first Among the candidate solutions, the first one is... Class 1 The characteristic values ​​of each product For the first Among the candidate solutions, the first one is... The cluster centers of the class are in the th order. Coordinates on the dimensional feature This is the noise suppression coefficient. The variance of the characteristic measurement error; The formula for calculating feature discrimination is: in: and For the first Class and First The cluster centers of the class are in the th order. Coordinates on the dimensional feature For the first Confidence weights of dimensional features The confidence compensation coefficient is... This is the probability of category confusion based on historical data; The formula for calculating fitness value is: in: , For preset weighting coefficients, As an intra-class similarity correction factor, To prevent extremely small positive numbers from being divided by zero, The decay factor is related to the number of iterations. The entropy penalty coefficient is... For the first Shannon entropy of each candidate scheme.

[0008] Furthermore, the process of performing quantum rotation gate and quantum NOT gate operations on the candidate clustering schemes encoded with qubits, and updating the correlation between different candidate schemes through quantum entanglement mechanism to generate a new population, includes: The candidate clustering scheme with the highest fitness value is selected from the current population as the parent scheme. Quantum crossover operation is performed on the qubit encoding of the cluster center in the parent scheme to generate a set of child schemes with quantum entanglement. A quantum NOT gate operation is performed on the cluster center qubits in the set of sub-schemes, and the number and frequency of qubit phase flips are dynamically adjusted according to the current iteration progress. The mutated offspring schemes are mixed with the historical best schemes, and a temporary retention scheme is selected based on a probability acceptance mechanism controlled by temperature parameters, and the temperature parameters are gradually reduced. When the fitness improvement rate of the temporary retention scheme is continuously lower than a preset threshold, a joint quantum state measurement is performed on the associated candidate schemes, and the probability amplitude distribution of the qubits is updated synchronously to form a final new population.

[0009] Furthermore, maximizing the detection efficiency includes: in: The total number of testing devices. The number of detection tasks to be assigned. For the total testing time window, Let be a binary decision variable, representing the task. Assign to device , For the task Number of tests, For equipment Execute the task The baseline time for a single test, For equipment Idle time penalty factor The efficiency attenuation coefficient, This represents the task timeout variance.

[0010] Furthermore, the device load balancing includes: in: Peak load penalty weight, For equipment The composite load rate, Average load factor For equipment Maximum theoretical throughput For the task In the equipment The amount of reagent resources consumed. For equipment The maximum capacity of the reagent storage compartment.

[0011] Furthermore, minimizing the inspection cost includes: in: For equipment Fixed usage costs For equipment Time cost coefficient, This is the reagent consumption coefficient. The unit price is the price per reagent. Energy transfer factor This refers to industrial electricity prices.

[0012] Furthermore, the constraints include detection time constraints, maximum equipment throughput constraints, and reagent and consumable cost constraints, including: Detection time limit constraints: in: To be allocated to equipment The total number of tasks For equipment Execute the The single detection time for each task For equipment superior Number of tests for each task For the task The final deadline, The current system time; Maximum throughput constraint of the device: in: For equipment Maximum theoretical throughput For the total testing time window, For the task In the equipment Total time spent on; Reagent and consumable cost constraints: in: For the task In the equipment The nominal reagent consumption is listed above. The reagent consumption rate, For equipment The maximum capacity of the reagent storage compartment.

[0013] Furthermore, the step of generating a detection task sequence based on the optimal solution set output by the multi-objective optimization task allocation model, adjusting the task queue according to the real-time status data of the inspection equipment, and sending execution instructions to the corresponding inspection equipment includes: Based on the optimal solution set output by the multi-objective optimization task allocation model, the initial detection task sequence is generated by sorting the results according to a preset comprehensive score of urgency and cost. The system obtains real-time operating load rate, reagent balance and fault alarm information of each testing equipment through the equipment status monitoring interface, and calculates equipment availability index. When the availability metric of any device is detected to be below the first threshold, a dynamic adjustment strategy is triggered. Spatiotemporal conflict detection was performed on the adjusted task queue to verify the non-overlap between task execution periods and equipment maintenance cycles, and to confirm that the total reagent consumption did not exceed the replenishment warning line. The verified task queue is converted into a set of device control instructions, which are then distributed to the target verification device for execution according to a preset communication protocol, and the task status database is updated synchronously.

[0014] Furthermore, the step of triggering a dynamic adjustment strategy when the availability index of any device is detected to be lower than a first threshold includes: The affected tasks are split into sub-task sets and redistributed to adjacent available devices based on the remaining device capacity. For tasks that cannot be split, activate the priority preemption mechanism to pause low-priority tasks and release resources.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: This invention utilizes a quantum genetic simulated annealing clustering algorithm to perform secondary classification on the basis of the initial product classification, effectively improving the granularity of product categories. It also leverages the parallel search characteristics of quantum computing to improve the clustering speed of the dataset. Through a multi-objective optimization model, it dynamically balances detection efficiency, equipment load, and cost consumption, combined with a real-time equipment status feedback mechanism, thereby increasing detection throughput in pharmaceutical or cosmetic testing scenarios. Ultimately, it effectively improves industrial-grade detection capabilities and efficiency in complex testing scenarios. Attached Figure Description

[0016] Figure 1 This is a structural block diagram of an embodiment of an artificial intelligence-based inspection and testing management system according to the present invention; Figure 2 This is a schematic diagram of the implementation process corresponding to the clustering optimization module in this invention; Figure 3 This is a flowchart illustrating an embodiment of the task allocation module in this invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Example 1 Please see Figure 1The AI-based testing and inspection management system provided in this application includes: a product classification module 101, used to receive information on pharmaceutical products and cosmetic products to be tested, classify and label them based on preset industry classification standards, and obtain a primary classification result set; a clustering optimization module 102, which has a built-in quantum genetic simulated annealing clustering algorithm unit, which outputs a target clustering result that meets the fitness function threshold by iteratively executing quantum genetic operations and simulated annealing optimization; a task construction module 103, which constructs a multi-objective optimization task allocation model based on the target clustering result; wherein the objective parameters include maximizing detection efficiency, balancing equipment load, and minimizing testing costs, and the constraints include detection time constraints, maximum equipment throughput constraints, and reagent and consumable cost constraints; and a task allocation module 104, which generates a testing task sequence based on the optimal solution set output by the multi-objective optimization task allocation model, adjusts the task queue according to the real-time status data of the testing equipment, and sends execution instructions to the corresponding testing equipment.

[0019] Specifically, the product classification module 101 primarily receives structured data on pharmaceuticals and cosmetics through API interfaces, file uploads, and IoT devices. This data includes key information such as dosage forms and active ingredients for pharmaceuticals, and INCI ingredients and functional labels for cosmetics. The received data undergoes feature cleaning, rule matching, and semantic analysis. Preset industry classification standards include those such as the dosage form system in the Chinese Pharmacopoeia and EU cosmetic ingredient regulations. After labeling, a primary classification result set containing classification levels, key features, and quality labels is output through confidence assessment and anomaly detection mechanisms, providing structured input for subsequent intelligent optimization.

[0020] The clustering optimization module 102 performs secondary classification based on the preliminary classification results obtained from the product classification module. It employs a quantum computing fusion optimization algorithm to deeply mine the feature space of the primary classification results, generating more precise cluster grouping granularity and improving the ability to identify complex product features. The task construction module 103 establishes a multi-objective dynamic optimization model based on the clustering results, comprehensively balancing detection efficiency, resource utilization, and cost control to generate a globally optimal task allocation strategy. The task allocation module 104 dynamically adjusts the detection task queue based on real-time equipment status feedback, achieving adaptive optimization and precise resource scheduling during task execution. Through the combined effects of industry standard pre-screening, quantum intelligent clustering, multi-objective dynamic decision-making, and real-time feedback control, the accuracy of pharmaceutical and cosmetic product testing and the execution efficiency of large-scale testing tasks are significantly improved.

[0021] Example 2 Please see Figure 2 In this invention, the quantum genetic simulated annealing clustering algorithm unit iteratively executes quantum genetic operations and simulated annealing optimization to output the target clustering result that satisfies the fitness function threshold, including the following steps: S21. Based on the products to be inspected under each industry category in the primary classification result set, generate an initial population containing M candidate clustering schemes, where each candidate clustering scheme consists of N cluster centers, and the position coordinates of the cluster centers are represented by qubit encoding; Pharmaceutical products are categorized into dosage forms, and cosmetic products into functional categories, forming industry subsets. For example, dosage forms can be divided into tablets, capsules, and injections, while functional categories can be categorized into skincare, makeup, and cleansing. The feature vector of each cluster center is converted into a probability amplitude distribution of qubits, utilizing the quantum superposition property to simultaneously represent multiple possible cluster combinations. The M value is dynamically adjusted based on the product type; for pharmaceutical testing, the default M=50, and for cosmetic testing, M=30, ensuring coverage of typical classification scenarios.

[0022] S22. Calculate the feature clustering degree of similar products and the feature discrimination degree between different categories of products in each candidate clustering scheme, and calculate the fitness value of each candidate clustering scheme based on the feature clustering degree and feature discrimination degree; The formula for calculating the feature clustering degree is: in: This represents the total number of categories in the current candidate clustering scheme. Let be the number of dimensions of the product feature vector. For the first Among the candidate solutions, the first one is... Class 1 The characteristic values ​​of each product For the first Among the candidate solutions, the first one is... The cluster centers of the class are in the th order. Coordinates on the dimensional feature This is the noise suppression coefficient. The variance of the characteristic measurement error; The formula for calculating feature discrimination is: in: and For the first Class and First The cluster centers of the class are in the th order. Coordinates on the dimensional feature For the first Confidence weights of dimensional features The confidence compensation coefficient is... This is the probability of category confusion based on historical data; The formula for calculating fitness value is: in: , For preset weighting coefficients, As an intra-class similarity correction factor, To prevent extremely small positive numbers from being divided by zero, The decay factor is related to the number of iterations. The entropy penalty coefficient is... For the first Shannon entropy of each candidate scheme.

[0023] S23. Perform quantum rotation gate and quantum NOT gate operations on the candidate clustering schemes after qubit encoding, and realize the correlation update between different candidate schemes through quantum entanglement mechanism to generate a new population; S231. Select the candidate clustering scheme with the highest fitness value from the current population as the parent scheme, perform quantum crossover operation on the qubit encoding of the cluster center in the parent scheme, and generate a set of child schemes with quantum entanglement properties. S232. Perform a quantum NOT gate operation on the cluster center qubits in the sub-scheme set, and dynamically adjust the number and frequency of qubit phase flips according to the current iteration progress; S233. Mix the mutated offspring scheme with the historical best scheme, select a temporary retention scheme based on the probability acceptance mechanism controlled by the temperature parameter, and gradually reduce the temperature parameter; S234. When the fitness improvement rate of the temporary retention scheme is continuously lower than the preset threshold, perform joint quantum state measurement on the associated candidate schemes, synchronously update the probability amplitude distribution of the qubits, and form a final new population.

[0024] Specifically, firstly, the candidate scheme with the highest fitness value in the current population is selected as the parent scheme. Here, the highest fitness value can be the top 10%. A quantum crossover operation is performed on the cluster center of its qubit encoding—the qubit representing the key quality attribute in the parent scheme is selected as the control bit, and an entangled channel is established with the target bit through a controlled NOT gate (CNOT), generating a set of offspring schemes that combine the advantages of the parent scheme. Secondly, dynamic mutation is applied to the qubit sequence of the offspring schemes. According to the iteration progress, proportionally (e.g., 30% in the early stage and 5% in the later stage), the feature dimension is randomly selected, and the probability amplitude phase of the selected qubit is flipped through the quantum NOT gate to break the local optimum. Finally, the mutated offspring are mixed with the historical best solution. Based on the temperature decay strategy, the initial temperature T0=1000 and the decay coefficient γ=0.92 are set. The acceptance probability calculated by the current temperature and fitness difference is used to accept inferior solutions with decreased fitness. When the improvement rate of the best solution is <0.5% for 3 consecutive iterations, the Belki joint measurement is performed on the associated solution to collapse the entangled state into explicit cluster center coordinates, forming a new population that integrates global search and local optimization.

[0025] S24. Perform a neighborhood search on the new population at a preset initial annealing temperature, accept or reject inferior solutions based on preset criteria, and reduce the annealing temperature according to an exponential decay strategy. The purpose of this step is to refine the genetically optimized population. The domain search is initiated at an initial temperature T0=1000, allowing for the temporary acceptance of suboptimal solutions with decreased fitness, thus enhancing global search capabilities. For pharmaceutical high-temperature sterilization testing, a rapid annealing mode (temperature decay coefficient γ=0.85) is set, while slow annealing (γ=0.93) is used for cosmetic stability testing. The acceptance probability is calculated based on the difference between the current temperature and fitness. Suboptimal solutions are retained when the acceptance probability is greater than or equal to 0.3. Finally, for key quality attributes, such as drug impurity content, an acceptance limit is set at a maximum fitness difference of 50 to ensure that core indicators do not deteriorate.

[0026] S25. When the number of iterations reaches the preset threshold, output the current optimal candidate clustering scheme as the target clustering result.

[0027] Iteration thresholds: For pharmaceutical testing, a maximum of 100 iterations is set to match the time window requirements of pharmaceutical batch processing; for cosmetic testing, a maximum of 80 iterations is set to accommodate rapid production changeovers. Early termination is triggered when the optimal fitness improvement rate is less than 0.5% for five consecutive iterations, with this threshold relaxed to 1% for special dosage forms to ensure sufficient optimization of complex features. Finally, iteration terminates when these conditions are met, and the optimal candidate clustering scheme output at this point is taken as the optimal clustering result.

[0028] The above embodiments improve existing clustering algorithms by using the improved clustering algorithm to classify product categories more finely. Based on the finely classified products, the accuracy of subsequent detection task allocation is effectively improved.

[0029] Example 3 The calculation formulas for the objective function and various constraints in the multi-objective optimization task allocation model of this invention include the following: Maximizing detection efficiency includes: in: The total number of testing devices. The number of detection tasks to be assigned. For the total testing time window, Let be a binary decision variable, representing the task. Assign to device , For the task Number of tests, For equipment Execute the task The baseline time for a single test, For equipment Idle time penalty factor The efficiency attenuation coefficient, This represents the task timeout variance.

[0030] Equipment load balancing includes: in: Peak load penalty weight, For equipment The composite load rate, Average load factor For equipment Maximum theoretical throughput For the task In the equipment The amount of reagent resources consumed. For equipment The maximum capacity of the reagent storage compartment.

[0031] Minimizing inspection costs includes: in: For equipment Fixed usage costs For equipment Time cost coefficient, This is the reagent consumption coefficient. The unit price is the price per reagent. Energy transfer factor This refers to industrial electricity prices.

[0032] Detection time limit constraints: in: To be allocated to equipment The total number of tasks For equipment Execute the The single detection time for each task For equipment superior Number of tests for each task For the task The final deadline, The current system time; Maximum throughput constraint of the device: in: For equipment Maximum theoretical throughput For the total testing time window, For the task In the equipment Total time spent on; Reagent and consumable cost constraints: in: For the task In the equipment The nominal reagent consumption is listed above. The reagent consumption rate, For equipment The maximum capacity of the reagent storage compartment.

[0033] Specifically, the task construction module establishes a multi-objective optimization mathematical model to map the clustering results of the objectives into a detection task allocation scheme. First, based on the clustered product feature set, it sets objective functions for detection efficiency, equipment load balancing, and testing cost, while simultaneously determining constraints such as detection time limits, equipment throughput, and reagent capacity. Then, it iteratively searches the multi-objective Pareto front solution set using the NSGA-II algorithm, combined with Lagrange relaxation to handle constraint conflicts, ultimately generating an optimal task allocation scheme that incorporates efficiency, load, and cost. For example, in a drug pooled testing scenario, high-priority antibiotic testing tasks are automatically assigned to high-performance chromatographs, while routine excipient testing tasks are scheduled to low-cost general-purpose equipment.

[0034] Example 4 Please see Figure 3 This invention generates a detection task sequence based on the optimal solution set output by a multi-objective optimization task allocation model, adjusts the task queue according to the real-time status data of the inspection equipment, and sends execution instructions to the corresponding inspection equipment, including the following steps: S31. Based on the optimal solution set output by the multi-objective optimization task allocation model, sort it according to the preset comprehensive score of urgency and cost to generate an initial detection task sequence; S32. Real-time acquisition of operating load rate, reagent balance and fault alarm information of each testing equipment through the equipment status monitoring interface, and calculation of equipment availability index; Specifically, the task allocation module, based on the Pareto optimal solution set output by the aforementioned multi-objective optimization model, first performs a comprehensive scoring and sorting of the solution set according to preset rules to generate an initial task sequence; for example, if the preset rule prioritizes the urgency of expiration date for drug testing tasks and emphasizes cost weight for cosmetic tasks, then antibiotic testing tasks will be preferentially allocated to the high-performance liquid chromatograph. Equipment operating data is collected in real time through industrial IoT protocols to calculate equipment availability indicators.

[0035] S33. When the availability index of any device is detected to be lower than the first threshold, a dynamic adjustment strategy is triggered; When the availability of a device is detected to be below a first threshold, dynamic adjustment is triggered. Optionally, the first threshold is set to a load rate > 85% or reagent balance < 15%. The adjustment strategy includes: S331. Split the affected tasks into sub-task sets and redistribute them to adjacent available devices according to the remaining device capacity; The affected tasks are broken down into sub-task sets based on the decomposability of the detection items. For example, if a chromatograph malfunction causes a backlog of 10 batches of drug tests, the tests can be performed in parallel on separate devices based on dissolution rate and content determination. The remaining capacity of neighboring devices is then redistributed, such as redistributing the remaining load space of 3 adjacent devices to 30%, 45%, and 25%, respectively.

[0036] S332. For tasks that cannot be split, initiate a priority preemption mechanism to pause low-priority tasks and release resources.

[0037] Prioritize indivisible tasks and pause low-priority tasks to release equipment resources. Here, low priority refers to the bottom 20% of all priorities. For example, if an indivisible task is cosmetic microbial aseptic testing that requires continuous incubation for 48 hours, then some low-priority tasks such as packaging integrity testing will be paused.

[0038] S34. Perform spatiotemporal conflict detection on the adjusted task queue to verify the non-overlap between the task execution period and the equipment maintenance cycle, and to ensure that the total reagent consumption does not exceed the replenishment warning line; S35. Convert the verified task queue into a set of device control instructions, distribute them to the target verification device for execution according to the preset communication protocol, and update the task status database synchronously.

[0039] After adjustments, spatiotemporal conflict verification is performed. For example, tasks are avoided during calibration and maintenance periods for certain mass spectrometers, ensuring that total reagent consumption does not exceed 80% of the safety stock level. Finally, verified queues are converted into equipment instructions, which are then sent via the MQTT protocol and the LIMS system task status is updated synchronously, achieving efficient dynamic scheduling. This effectively shortens the response time for drug testing task reallocation and reduces adjustment delays on cosmetic production lines.

[0040] This invention utilizes a quantum intelligent clustering algorithm to identify the complex features of pharmaceuticals and cosmetics, effectively improving classification accuracy. It also employs a multi-objective optimization model to allocate testing tasks, enhancing testing capabilities. Combined with real-time equipment status monitoring and dynamic scheduling mechanisms, it significantly improves testing efficiency. Furthermore, a load balancing strategy reduces equipment idle time, ensuring efficient resource utilization.

[0041] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0042] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An artificial intelligence-based inspection and testing management system, characterized in that, include: The product classification module is used to receive information on pharmaceutical products and cosmetic products to be inspected, classify and label them based on preset industry classification standards, and obtain a set of primary classification results. The clustering optimization module includes a built-in quantum genetic simulated annealing clustering algorithm unit. This algorithm unit iteratively executes quantum genetic operations and simulated annealing optimization to output target clustering results that satisfy the fitness function threshold. This includes: Based on the products to be inspected under each industry category in the primary classification result set, an initial population containing M candidate clustering schemes is generated, wherein each candidate clustering scheme consists of N cluster centers, and the position coordinates of the cluster centers are represented by qubit encoding. Calculate the feature clustering degree of similar products and the feature discrimination degree between different categories of products in each candidate clustering scheme, and calculate the fitness value of each candidate clustering scheme based on the feature clustering degree and feature discrimination degree; Quantum rotation gate and quantum NOT gate operations are performed on the candidate clustering schemes encoded by qubits. The correlation update between different candidate schemes is realized through the quantum entanglement mechanism to generate a new population. The new population is subjected to a neighborhood search at a preset initial annealing temperature. Based on preset criteria, inferior solutions are accepted or rejected, and the annealing temperature is reduced according to an exponential decay strategy. When the number of iterations reaches a preset threshold, the current optimal candidate clustering scheme is output as the target clustering result; The task construction module constructs a multi-objective optimization task allocation model based on the target clustering results; wherein the target parameters include maximizing detection efficiency, balancing equipment load, and minimizing testing costs, and the constraints include detection time constraints, maximum equipment throughput constraints, and reagent and consumable cost constraints. The task allocation module generates a detection task sequence based on the optimal solution set output by the multi-objective optimized task allocation model, adjusts the task queue according to the real-time status data of the inspection equipment, and sends execution instructions to the corresponding inspection equipment; including: Based on the optimal solution set output by the multi-objective optimization task allocation model, the initial detection task sequence is generated by sorting the results according to a preset comprehensive score of urgency and cost. The system obtains real-time operating load rate, reagent balance and fault alarm information of each testing equipment through the equipment status monitoring interface, and calculates equipment availability index. When the availability metric of any device is detected to be below the first threshold, a dynamic adjustment strategy is triggered. Spatiotemporal conflict detection was performed on the adjusted task queue to verify the non-overlap between task execution periods and equipment maintenance cycles, and to confirm that the total reagent consumption did not exceed the replenishment warning line. The verified task queue is converted into a set of device control instructions, which are then distributed to the target verification device for execution according to a preset communication protocol, and the task status database is updated synchronously.

2. The AI-based inspection and testing management system according to claim 1, characterized in that, The calculation of the feature clustering degree of similar products and the feature discrimination degree between different categories of products in each candidate clustering scheme, and the calculation of the fitness value of each candidate clustering scheme based on the feature clustering degree and feature discrimination degree, includes: The formula for calculating the feature clustering degree is: in: This represents the total number of categories in the current candidate clustering scheme. Let be the number of dimensions of the product feature vector. For the first Among the candidate solutions, the first one is... Class 1 The characteristic values ​​of each product For the first Among the candidate solutions, the first one is... The cluster centers of the class are in the th order. Coordinates on the dimensional feature This is the noise suppression coefficient. The variance of the characteristic measurement error; The formula for calculating feature discrimination is: in: and For the first Class and First The cluster centers of the class are in the th order. Coordinates on the dimensional feature For the first Confidence weights of dimensional features The confidence compensation coefficient is... This is the probability of category confusion based on historical data; The formula for calculating fitness value is: in: , For preset weighting coefficients, As an intra-class similarity correction factor, To prevent extremely small positive numbers from being divided by zero, The decay factor is related to the number of iterations. The entropy penalty coefficient is... For the first Shannon entropy of each candidate scheme.

3. The AI-based inspection and testing management system according to claim 1, characterized in that, The process of performing quantum rotation gate and quantum NOT gate operations on the candidate clustering schemes encoded with qubits, and updating the correlation between different candidate schemes through quantum entanglement mechanism to generate a new population, includes: The candidate clustering scheme with the highest fitness value is selected from the current population as the parent scheme. Quantum crossover operation is performed on the qubit encoding of the cluster center in the parent scheme to generate a set of child schemes with quantum entanglement. A quantum NOT gate operation is performed on the cluster center qubits in the set of sub-schemes, and the number and frequency of qubit phase flips are dynamically adjusted according to the current iteration progress. The mutated offspring schemes are mixed with the historical best schemes, and a temporary retention scheme is selected based on a probability acceptance mechanism controlled by temperature parameters, and the temperature parameters are gradually reduced. When the fitness improvement rate of the temporary retention scheme is continuously lower than a preset threshold, a joint quantum state measurement is performed on the associated candidate schemes, and the probability amplitude distribution of the qubits is updated synchronously to form a final new population.

4. The artificial intelligence-based inspection and testing management system according to claim 1, characterized in that, Maximizing detection efficiency includes: in: The total number of testing devices. The number of detection tasks to be assigned. For the total testing time window, Let be a binary decision variable, representing the task. Assign to device , For the task Number of tests, For equipment Execute the task The baseline time for a single test, For equipment Idle time penalty factor The efficiency attenuation coefficient, This represents the task timeout variance.

5. The artificial intelligence-based inspection and testing management system according to claim 1, characterized in that, The device load balancing includes: in: Peak load penalty weight, For equipment The composite load rate, Average load factor For equipment Maximum theoretical throughput For the task In the equipment The amount of reagent resources consumed. For equipment The maximum capacity of the reagent storage compartment.

6. The artificial intelligence-based inspection and testing management system according to claim 1, characterized in that, The minimization of inspection costs includes: in: For equipment Fixed usage costs For equipment Time cost coefficient, This is the reagent consumption coefficient. The unit price is the price per reagent. Energy transfer factor This refers to industrial electricity prices.

7. The artificial intelligence-based inspection and testing management system according to claim 1, characterized in that, The constraints include detection time constraints, maximum equipment throughput constraints, and reagent and consumable cost constraints, including: Detection time limit constraints: in: To be allocated to equipment The total number of tasks For equipment Execute the The single detection time for each task For equipment superior Number of tests for each task For the task The final deadline, The current system time; Maximum throughput constraint of the device: in: For equipment Maximum theoretical throughput For the total testing time window, For the task In the equipment Total time spent on; Reagent and consumable cost constraints: in: For the task In the equipment The nominal reagent consumption is listed above. The reagent consumption rate, For equipment The maximum capacity of the reagent storage compartment.

8. The artificial intelligence-based inspection and testing management system according to claim 1, characterized in that, When the availability index of any device is detected to be lower than the first threshold, a dynamic adjustment strategy is triggered, including: The affected tasks are split into sub-task sets and redistributed to adjacent available devices based on the remaining device capacity. For tasks that cannot be split, activate the priority preemption mechanism to pause low-priority tasks and release resources.

Images