Ethylene oxide sterilization control system and method based on ai dynamic parameter optimization

Abstract

The application provides an ethylene oxide sterilization control system and method based on AI dynamic parameter optimization, and belongs to the technical field of industrial process control. The system is composed of an AI optimization module and a multidimensional acquisition module, a man-machine interaction module and a collaborative regulation module in communication connection with the AI optimization module. The multidimensional acquisition module is arranged inside and outside the sterilization cabin to collect working condition data such as temperature, pressure and EO concentration in real time. The man-machine interaction module inputs the instrument category parameters. The collaborative regulation module is connected with the execution components in the cabin. During sterilization, the real-time working conditions and instrument category parameters are obtained first. Then the K-Means clustering is used to match the reference process parameter group of the corresponding category. Then the DNN neural network is used to fit the working condition and process mapping relationship. The optimized target set adapted to the current working condition is obtained through dynamic optimization. Finally, the fuzzy self-adaptive incremental algorithm is used to close-loop regulate the execution component operating parameters, so that the actual working condition approaches the optimization target, and then the stable and good sterilization effect is effectively ensured through dynamic parameter optimization.

Description

Technical Field

[0001] This invention belongs to the field of industrial process control technology, specifically relating to an ethylene oxide sterilization control system and method based on AI dynamic parameter optimization. Background Technology

[0002] Ethylene oxide, also known as ethylene oxide or oxane, is abbreviated as EO in English. It has broad-spectrum bactericidal properties and can effectively kill various microorganisms such as bacteria, viruses, and spores. Its low-temperature sterilization properties are suitable for medical products such as plastics, rubber, and precision instruments that are not resistant to high temperatures and pressures, and it is the mainstream sterilization process in the medical industry.

[0003] Currently, in the process of ethylene oxide sterilization, as described in the Chinese invention patent application number "202411532097.0", the sterilization process adopts fixed interval parameter control, and the entire process relies on preset fixed temperature, pressure, concentration and time parameters to run mechanically according to the established process.

[0004] However, during the sterilization process, the real-time operating data in the sterilization chamber fluctuates dynamically, and the sterilization process parameters differ depending on the type of instrument. Simply using fixed process parameters to mechanically execute the sterilization process and relying on rough fixed-value control can easily result in poor sterilization effect. Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to effectively ensure a stable and good sterilization effect through dynamic parameter optimization. In view of the shortcomings of the prior art, an ethylene oxide sterilization control system and method based on AI dynamic parameter optimization is provided.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present invention provides an ethylene oxide sterilization control system based on AI dynamic parameter optimization, comprising an AI optimization module and a multi-dimensional acquisition module, a human-machine interaction module, and a collaborative adjustment module, all communicatively connected to the AI ​​optimization module; the multi-dimensional acquisition module is configured to be distributed inside and outside the sterilization chamber; the collaborative adjustment module is configured inside the sterilization chamber and communicatively connected to the execution components inside the sterilization chamber; the AI ​​optimization module is used for:

[0008] After the instruments to be sterilized are moved into the sterilization chamber, the instrument category parameters entered by the human-computer interaction module are acquired, the execution component is driven to work through the collaborative adjustment module, and real-time working condition data collected by the multi-dimensional acquisition module at a fixed frequency is acquired.

[0009] Preset clustering categories and use the K-Means clustering algorithm to cluster and divide historical operating data in the preset historical database, and filter to obtain a set of benchmark process parameters that match the parameters of the instrument category;

[0010] Based on the pre-built three-layer DNN deep neural network, the benchmark process parameter set is optimized with the real-time operating data as a constraint to obtain the optimization target set;

[0011] The optimization target set is used as a fixed value, and the instantaneous operating condition data transmitted in real time by the collaborative adjustment module is used as a feedback value. The real-time deviation between the fixed value and the feedback value is used as input. The adjustment control quantity is obtained through the fuzzy adaptive incremental control algorithm and transmitted to the collaborative adjustment module to adjust the operating parameters of the actuator accordingly.

[0012] Compared to existing technologies, the beneficial effects of the AI-based dynamic parameter optimization ethylene oxide sterilization control system of the present invention include: The system comprises an AI optimization module and a multi-dimensional acquisition module, a human-machine interaction module, and a collaborative adjustment module, all communicatively connected to the AI ​​optimization module. The multi-dimensional acquisition module is configured and distributed inside and outside the sterilization chamber, enabling real-time acquisition of operating condition data such as temperature, pressure, vacuum, and EO concentration. This provides accurate global operating condition data for subsequent AI-driven optimization of the ethylene oxide sterilization process of medical devices. The human-machine interaction module allows input of device category parameters, providing a basis for matching and clustering the baseline process parameters. Simultaneously, the collaborative adjustment module is configured inside the sterilization chamber and communicatively connected to the execution components within it. This enables real-time acquisition of instantaneous operating condition data at the execution components and real-time control of their operation. This facilitates AI-driven decision-making based on instantaneous operating condition data to finely and dynamically adjust the operating parameters of the execution components, thereby optimizing the ethylene oxide sterilization process of medical devices. Building upon this foundation, the AI ​​optimization module, acting as the control and decision-making center, firstly drives the execution components to carry out the sterilization process after the instruments to be sterilized are moved into the sterilization chamber, through the collaborative adjustment module. Simultaneously, it acquires real-time operating condition data and instrument category parameters, providing a data foundation for subsequent dynamic parameter optimization of the sterilization process. Next, based on preset clustering categories, it uses the K-Means clustering algorithm to cluster historical operating condition data in a preset historical database, filtering out a set of baseline process parameters that match the instrument category parameters. This addresses the issue of significant differences in sterilization characteristics among different instrument categories by automatically matching the most suitable baseline process for the current instrument from historical operating condition data through process clustering, laying the foundation for further optimization of process adaptability. Then, since the baseline process parameter set is a static process parameter based on historical operating condition data, to adapt to the dynamically fluctuating real-time operating conditions of the current instrument sterilization process, a pre-built three-layer D... The nonlinear fitting capability of deep neural networks (NNs) establishes a mapping relationship between real-time operating data and a set of baseline process parameters. Using real-time operating data as a constraint reference, the baseline process parameters are dynamically adjusted to obtain an optimized target set suitable for the current operating conditions. This provides a basis for subsequent dynamic adjustment of the operating parameters of the actuators. Finally, the optimized target set is used as a setpoint, and the instantaneous operating data transmitted in real time by the collaborative adjustment module is used as a feedback value. The real-time deviation between the setpoint and the feedback value is used as input, and a fuzzy adaptive incremental control algorithm outputs an adjustment control quantity. This allows the collaborative adjustment module to adjust the operating parameters of the actuators accordingly. This forms a closed-loop dynamic adjustment that considers the existence of real-time deviation, outputs adjusted operating parameters, and determines whether real-time deviation exists. This allows the instantaneous operating data to gradually approach the optimized target set through adjustment, ensuring a stable sterilization environment while achieving dynamic parameter optimization and precise control of the ethylene oxide sterilization process, thus guaranteeing a stable and effective sterilization result.

[0013] Optionally, the real-time operating data includes chamber temperature, chamber pressure, chamber vacuum, chamber EO concentration, chamber flow rate, chamber loading density, and instrument material characteristic values; the multi-dimensional acquisition module includes a temperature sensor, a pressure sensor, a vacuum transmitter, an EO concentration infrared sensor, an airflow velocity sensor, an ambient temperature and humidity sensor, a loading density detection sensor array, and a material characteristic detection element; the instrument category parameters include the instrument category, instrument material attribute data, instrument packaging type data, and instrument loading operating condition data.

[0014] Optionally, the categories of instruments include endoscopes, implants, medical dressings, and precision surgical instruments. The AI ​​optimization module is specifically used in the process of acquiring the baseline process parameter set to:

[0015] Retrieve the historical operating condition data of the preset batch, and use the complete sterilization process data of a single batch as a sample. Construct the following multi-dimensional feature vector for each sample to form multi-dimensional sample data:

[0016] ,

[0017] Among them, the For the first The sample mentioned above, the For the first The average temperature of the sample, the For the first The working pressure of the sample described in the article, the For the first The vacuum degree of the sample mentioned above, the For the first The sample mentioned in the article equilibrium concentration, the For the first The material characteristic values ​​of the sample mentioned in the article;

[0018] The extreme values ​​of specific dimension features of the samples are obtained by traversing the multidimensional sample data. The multidimensional sample data is then normalized based on the following formula, mapped to the interval [0, 1], to form a normalized sample set:

[0019] ,

[0020] Among them, the For normalized eigenvalues, the For the first The sample described in the article is number one. The original eigenvalues, the The first in the multidimensional sample data The maximum value of the dimensional feature, the The first in the multidimensional sample data The minimum value of a feature;

[0021] Pre-defined clustering categories are established. Based on the number of clustering categories, a corresponding number of samples are randomly selected from the normalized sample set as initial cluster centers. For each normalized sample, the spatial Euclidean distance between it and each of the initial cluster centers is calculated using the following formula. Then, the samples are classified according to the minimum distance criterion to form multiple clusters:

[0022] ,

[0023] Among them, the Let Euclidean distance be the spatial distance. For the normalized sample, the For the first Cluster centers, the For the first The first cluster center Dimensional feature values;

[0024] The clusters are integrated and iteratively converged to form a historical cluster dataset. Based on the instrument category parameters, the corresponding cluster category is matched to select and obtain the matching benchmark process parameter group from the historical cluster dataset.

[0025] Optionally, the AI ​​optimization module is specifically used in the process of integrating the clusters to iteratively converge and form a historical cluster dataset for:

[0026] After completing the initial partitioning and classification of each normalized sample to form the cluster, the partitioning and classification process for each normalized sample within the cluster is repeated. At the end of each round of partitioning and classification, the mean of all normalized samples within the cluster is obtained based on the following formula, and the cluster center corresponding to the cluster is iteratively updated:

[0027] ,

[0028] Among them, the For the first The first iteration after the update The cluster centers, the For the first The aforementioned clusters;

[0029] Based on the cluster centers and corresponding normalized samples updated in each round, the offset distance of the current cluster center compared to the previous cluster center and the sum of errors in the current round are obtained according to the following formula. When the offset distance is less than a preset offset threshold, or the difference between the current error and the sum of errors in the previous round approaches zero, or the number of iterations is a preset maximum number of iterations, the iteration update converges and the cluster centers are output. Each cluster center is then labeled according to the category of the medical device to form the historical clustering dataset.

[0030] ,

[0031] ,

[0032] Among them, the For the first The cluster centers at the in Compared to the first round of updates The offset distance of the round update, the For the first The first iteration after the update The first of the cluster centers dimensional features, the No. The first iteration after the update The first of the cluster centers dimensional features, the The sum of the total squared errors of the clustering, the The number of cluster categories, the It is an adjacent 2-norm.

[0033] Optionally, the three-layer DNN deep neural network includes an input layer, a hidden layer, and an output layer connected in sequence. The AI ​​optimization module is specifically used to:

[0034] The baseline process parameter set and the real-time multidimensional operating condition data are transmitted to the input layer to form the following total network input feature vector:

[0035] ,

[0036] Among them, the The total input feature vector of the network, the The reference process parameter set is, and The To set a temperature for the product category benchmark, the As a category benchmark working pressure, the As the standard vacuum level for the product category, the As the category benchmark EO concentration, the - These are the baseline times for the four stages of the sterilization process: preheating, permeation, sterilization, and desorption. The real-time multidimensional operating condition data, and The The cabin temperature is [the temperature inside the cabin]. The pressure inside the cabin, the The vacuum level inside the cabin, the The EO concentration inside the chamber, the The loading density inside the cabin, the The material characteristic value of the instrument;

[0037] The total input feature vector of the network is transmitted to the hidden layer. The connection weights and biases of the hidden layer, obtained by training and convergence based on the historical working condition data, are retrieved. Linear weighted summation and nonlinear activation are performed sequentially based on the following formula to obtain the nonlinear output value of the neuron:

[0038] ,

[0039] ,

[0040] Among them, the For the first hidden layer The weighted sum of neurons, the For the input layer The node to the hidden layer The node connection weight, the The first of the total input feature vectors of the network Dimensional input feature value, the For the first hidden layer The bias of each neuron, the For the first The nonlinear output value of the neuron, the It is a natural constant;

[0041] The nonlinear output value of the neuron is transmitted to the output layer. The weights and biases of the output layer, obtained by training and convergence based on the historical working condition data, are retrieved. The optimization target parameters are output based on the following formula to form the optimization target set:

[0042] ,

[0043] Among them, the For the optimization target set, the For the output layer weights, the The nonlinear output value of the neuron, the The output layer bias, the To optimize the temperature inside the target cabin, the To optimize the pressure inside the target chamber, the To optimize the vacuum level inside the target chamber, the To optimize the EO concentration in the target chamber.

[0044] Optionally, the AI ​​optimization module is specifically used in the process of acquiring the adjustment control amount to:

[0045] Based on the difference between the optimization target set and the instantaneous operating condition data, the operating condition deviation value is obtained. Based on the time series of the instantaneous operating condition data collected in real time by the collaborative adjustment module, the change in operating condition deviation before and after the time series is obtained.

[0046] Based on the sterilization process characteristics and historical qualified batch data statistically integrated from the historical database, the initial gain coefficient of PID, the maximum allowable deviation threshold, and the maximum deviation change threshold are calibrated to form a fuzzy rule base. The operating condition deviation value and the operating condition deviation change are used as fuzzy inputs, and the PID gain coefficient is updated and output in real time through fuzzy inference.

[0047] The regulating control quantity is calculated using the following formula based on the PID gain coefficient, the operating condition deviation value, and the change in operating condition deviation:

[0048] ,

[0049] ,

[0050] Among them, the For the first The control increment at time, the The above The above These are the proportional gain coefficient, integral gain coefficient, and derivative gain coefficient in the PID gain coefficients, respectively. For the first The operating condition deviation value at time, the For the first The change in the operating condition deviation at time t, For the first The adjustment control quantity at time, the For the first The adjustment and control quantity at any given time.

[0051] Optionally, the ethylene oxide sterilization control system based on AI dynamic parameter optimization further includes a multi-factor analysis module. This module incorporates a multi-factor SAL prediction model. The device category parameters also include the initial bioburden value of the device. The AI ​​optimization module is communicatively connected to the multi-factor analysis module and is used for:

[0052] During the sterilization process in the sterilization chamber, the sterilization efficiency correction coefficient, calibrated by the sterilization process characteristics, is retrieved according to the type of the instrument. The initial bioburden value of the instrument, the real-time operating data, and the sterilization efficiency correction coefficient are input to the multi-factor analysis module to output a predicted value through the multi-factor SAL prediction model as follows:

[0053] ,

[0054] ,

[0055] Among them, the The cumulative sterilization effect of the sterilization process, the for Real-time EO concentration inside the chamber, the for Real-time temperature inside the cabin, the for Real-time pressure inside the cabin, the aforementioned The sterilization process is a continuous time variable. The total time for the sterilization process has been accumulated. The predicted value is... The initial bioburden value of the device, the This is the sterilization efficiency correction factor;

[0056] When the predicted value is greater than the qualified threshold, sterilization is determined to be substandard. The delay compensation ratio coefficient and concentration compensation ratio coefficient calibrated by the sterilization process characteristics are retrieved, and the sterilization compensation time and EO concentration increment are obtained based on the following formula, and transmitted to the AI ​​optimization module to update the target optimization set:

[0057] ,

[0058] ,

[0059] Among them, the For the sterilization compensation time, the The delay compensation ratio coefficient is A, and the qualified threshold is A. For the EO concentration increment, the This is the concentration compensation ratio coefficient.

[0060] Optionally, the multi-factor analysis module also includes a built-in EO residue inference model, and the AI ​​optimization module is further used for:

[0061] When the sterilization process is completed, if the predicted value is greater than the qualified threshold, the batch of the instrument corresponding to the current sterilization process is determined to be an unqualified batch, the current batch is locked and marked, and an alarm signal and a prompt to restart the sterilization process are output to the human-computer interaction module.

[0062] If the predicted value is less than the qualified threshold, the EO analysis correction coefficient, calibrated by combining the material attribute data and packaging type data of the device in the historical database, is retrieved according to the type of the device. The effective volume of the sterilization chamber, the EO analysis correction coefficient, and the real-time operating condition data are then transmitted to the multi-factor analysis module so that the EO residue estimation model can output the EO residue amount as follows:

[0063] ,

[0064] Among them, the The residual amount of EO, the For the effective volume, the The total duration of the sterilization process, the The cumulative amount of EO concentration coupled with pressure in the sterilization process, the The EO analytical correction coefficient, the This refers to the cumulative temperature increase during the sterilization process.

[0065] When the residual EO amount is greater than the preset residual threshold, the batch of the instrument corresponding to the current sterilization process is determined to be an unqualified batch, the current batch is locked and marked, and an alarm signal and a prompt to restart the sterilization process are output to the human-machine interaction module.

[0066] Optionally, the ethylene oxide sterilization control system based on AI dynamic parameter optimization further includes an encrypted traceability module. This module incorporates a hash encryption algorithm and a dual-link evidence storage unit, and is communicatively connected to both the human-machine interaction module and the AI ​​optimization module. The encrypted traceability module is used for:

[0067] The system receives data records from the AI ​​optimization module for each batch of instruments during the sterilization process in real time, and adds timestamps to form a sterilization dataset. The data record set includes real-time acquired data, optimization adjustment records, human-computer interaction records, and alarm lock records.

[0068] The sterilization dataset is encrypted using the hash encryption algorithm to form an encrypted dataset, which is then transmitted to the dual-link evidence storage unit. When a retrieval command is received from the human-computer interaction module, the timestamp and data record set corresponding to the sterilization dataset are read according to the retrieval command, and the encrypted dataset is verified using the encryption algorithm.

[0069] Secondly, the present invention also provides a method for controlling ethylene oxide sterilization based on AI dynamic parameter optimization, comprising:

[0070] S1. After the instruments to be sterilized are moved into the sterilization chamber, the real-time operating data collected at a fixed frequency by the multi-dimensional acquisition module of the ethylene oxide sterilization control system based on AI dynamic parameter optimization is obtained, and the instrument category parameters entered by the human-computer interaction module of the ethylene oxide sterilization control system based on AI dynamic parameter optimization are obtained.

[0071] S2. Preset clustering categories and use the K-Means clustering algorithm to cluster and divide the historical operating condition data in the preset historical database, and filter to obtain the benchmark process parameter group that matches the instrument category parameters;

[0072] S3. Based on the pre-built three-layer DNN deep neural network, optimize the benchmark process parameter set with the real-time operating data as constraints to obtain the optimization target set;

[0073] S4. The optimization target set is used as a fixed value, and the instantaneous operating condition data transmitted in real time by the collaborative adjustment module of the ethylene oxide sterilization control system based on AI dynamic parameter optimization is used as a feedback value. The real-time deviation between the fixed value and the feedback value is used as input, and the adjustment control quantity is obtained through the fuzzy adaptive incremental control algorithm and transmitted to the collaborative adjustment module to adjust the operating parameters of the execution component accordingly.

[0074] Compared to existing technologies, the beneficial effects of the AI-based dynamic parameter optimization-based ethylene oxide sterilization control method of the present invention are the same as those of the AI-based dynamic parameter optimization-based ethylene oxide sterilization control system described above, and will not be repeated here. Attached Figure Description

[0075] The present invention will now be described in further detail with reference to the accompanying drawings.

[0076] Figure 1 : A schematic diagram of the structure of the ethylene oxide sterilization control system based on AI dynamic parameter optimization in this embodiment of the invention;

[0077] Figure 2 : A schematic diagram of the process of the ethylene oxide sterilization control method based on AI dynamic parameter optimization in this embodiment of the invention;

[0078] Among them, 1-AI optimization module, 2-multi-dimensional data acquisition module, 3-human-computer interaction module, 4-coordinated adjustment module, 5-multi-factor analysis module, and 6-encrypted traceability module. Detailed Implementation

[0079] To better understand the present invention, the following embodiments further illustrate the content of the invention, but the scope of protection of the present invention is not limited to the following embodiments. Numerous specific details are set forth in the following description to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the present invention can be practiced without one or more of these details.

[0080] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0081] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0082] In a first aspect, an embodiment of the present invention provides an ethylene oxide sterilization control system based on AI dynamic parameter optimization, comprising an AI optimization module 1 and a multi-dimensional acquisition module 2, a human-machine interaction module 3, and a collaborative adjustment module 4, all communicatively connected to the AI ​​optimization module 1; the multi-dimensional acquisition module 2 is configured to be distributed inside and outside the sterilization chamber; the collaborative adjustment module 4 is configured inside the sterilization chamber and communicatively connected to the execution components inside the sterilization chamber; the AI ​​optimization module 1 is used to: after the instruments to be sterilized are moved into the sterilization chamber, acquire the instrument category parameters entered by the human-machine interaction module 3, drive the execution components to work through the collaborative adjustment module 4, and then acquire the multi-dimensional acquisition module 2 at a fixed frequency The system collects real-time operating condition data; it pre-sets clustering categories and uses the K-Means clustering algorithm to cluster historical operating condition data in a pre-set historical database, and filters out a set of benchmark process parameters that match the parameters of the instrument category; based on a pre-built three-layer DNN deep neural network, it optimizes the set of benchmark process parameters with real-time operating condition data as constraints to obtain an optimization target set; the optimization target set is used as a setpoint, and the instantaneous operating condition data transmitted in real time by the collaborative adjustment module 4 is used as a feedback value. The real-time deviation between the setpoint and the feedback value is used as input, and the adjustment control quantity is obtained through a fuzzy adaptive incremental control algorithm and transmitted to the collaborative adjustment module 4 to adjust the operating parameters of the corresponding actuator.

[0083] Specifically, the AI ​​optimization module 1 serves as the core control center, consisting of a hardware unit comprised of an industrial computer, a high-speed data acquisition card, and an industrial Ethernet switch, and a software unit comprised of a data preprocessing unit, a K-Means clustering algorithm unit, a three-layer DNN deep neural network unit, and a fuzzy adaptive incremental PID control unit; the multi-dimensional acquisition module 2 serves as the data acquisition center, consisting of sensor hardware comprised of multiple different types of sensors distributed inside and outside the sterilization chamber, and receiving and transmitting hardware comprised of a sensor signal conditioning unit and a data acquisition and transmission unit; the human-machine interaction module 3 serves as the interaction interface, consisting of interactive hardware comprised of a display, touch screen, or keyboard, and software comprised of display, query, and input interfaces; the collaborative adjustment module 4 serves as the execution control center of the sterilization chamber, consisting of a hardware unit comprised of a PLC controller, actuator drivers, and a signal acquisition unit.

[0084] In this embodiment, as Figure 1As shown, an ethylene oxide sterilization control system based on AI dynamic parameter optimization is constructed, consisting of an AI optimization module 1, a multi-dimensional acquisition module 2 (all of which are communicatively connected to the AI ​​optimization module 1), a human-machine interaction module 3, and a collaborative adjustment module 4. The multi-dimensional acquisition module 2 is configured to be distributed inside and outside the sterilization chamber, enabling real-time acquisition of operating condition data such as temperature, pressure, vacuum, and EO concentration. This provides accurate global operating condition data for subsequent AI-driven optimization of the ethylene oxide sterilization process of medical devices. The human-machine interaction module 3 can input device category parameters, providing a basis for matching and clustering the baseline process parameters. Simultaneously, the collaborative adjustment module 4 is configured inside the sterilization chamber and communicatively connected to the actuators within it. This allows for real-time acquisition of instantaneous operating condition data at the actuators and real-time control of their operation. This facilitates AI-driven decision-making based on instantaneous operating condition data to finely and dynamically adjust the operating parameters of the actuators, thereby optimizing the ethylene oxide sterilization process of medical devices. Building upon this foundation, AI optimization module 1, acting as the control and decision-making center, firstly drives the execution components to carry out the sterilization process after the instruments to be sterilized are moved into the sterilization chamber via the collaborative adjustment module 4, while simultaneously acquiring real-time operating condition data and instrument category parameters. This provides a data foundation for the dynamic optimization of parameters in the subsequent sterilization process. Next, based on preset clustering categories, the K-Means clustering algorithm is used to cluster and divide historical operating condition data in the preset historical database, filtering out a set of benchmark process parameters that match the instrument category parameters. This addresses the issue of significant differences in sterilization characteristics among different instrument categories by automatically matching the most suitable benchmark process for the current instrument from historical operating condition data through process clustering, laying the foundation for further optimization of process adaptability. Then, since the benchmark process parameter set is a static process parameter based on historical operating condition data, in order to adapt to the dynamic fluctuations in real-time operating conditions of the current instrument sterilization, a pre-built three-layer D... The nonlinear fitting capability of deep neural networks (NNs) establishes a mapping relationship between real-time operating data and a set of baseline process parameters. Using real-time operating data as a constraint reference, the baseline process parameters are dynamically adjusted to obtain an optimized target set adapted to the current operating conditions. This provides a basis for subsequent dynamic adjustment of the operating parameters of the actuators. Finally, the optimized target set is used as a setpoint, and the instantaneous operating data transmitted in real time by the collaborative adjustment module 4 is used as a feedback value. The real-time deviation between the setpoint and the feedback value is used as input, and a fuzzy adaptive incremental control algorithm outputs an adjustment control quantity. This allows the collaborative adjustment module 4 to adjust the operating parameters of the actuators accordingly. This forms a closed-loop dynamic adjustment that considers the existence of real-time deviation, outputs adjusted operating parameters, and determines whether real-time deviation exists. This allows the instantaneous operating data to gradually approach the optimized target set through adjustment, ensuring a stable sterilization environment while achieving dynamic parameter optimization and precise control of the ethylene oxide sterilization process, thus guaranteeing a stable and effective sterilization result.

[0085] It should be noted that when the real-time operating data is stabilized through closed-loop dynamic adjustment using the fuzzy adaptive incremental control algorithm, the real-time operating data is compared with the optimized target set in real time. During this process: when slight fluctuations occur in the real-time operating data due to reasons such as equipment start-up and shutdown, gas convection, slight sensor drift, and normal heat exchange, such as slight temperature fluctuations, small pressure fluctuations, instantaneous slight concentration fluctuations, and normal airflow disturbances, the optimized target set remains unchanged. The operating parameters are adjusted through the fuzzy adaptive incremental control algorithm to resolve the slight fluctuations in the real-time operating data by adjusting the instantaneous operating data, avoiding the need to regenerate a new optimized target set, which would increase computing power consumption and ensure response speed. When the real-time operating data deviates from the optimized target set by more than the allowable deviation, or when there is a stage switch in the sterilization process, a sudden change in the environment of the sterilization chamber, a significant change in the characteristics of the materials inside the chamber, or a continuous steady-state shift in the air pressure data, i.e., when the real-time operating data fluctuates significantly, a new optimized target set needs to be obtained based on a three-layer DNN deep neural network. Then, the fuzzy adaptive incremental control algorithm is used to perform closed-loop dynamic adjustment based on the new optimized target set.

[0086] For example, a batch of precision surgical instruments undergoes a sterilization process. The K-Means clustering algorithm yields a baseline set of process parameters, including a target temperature of 55°C, a target pressure of −20 kPa, and a target EO concentration of 8 mg / m³. However, during sterilization, the ambient temperature in the sterilization room is low, the instruments are densely packed, and the chamber dissipates heat quickly. The collected real-time operating data shows a real-time temperature of 51°C and a real-time pressure of −17 kPa. When this real-time operating data is fed into a DNN (Digital Neural Network), because the on-site conditions are not met, forced heating would result in overshoot, waste, and uneven sterilization. The DNN does not rigidly set the temperature to 55°C, while setting it to 51°C would lead to insufficient sterilization intensity. Therefore, the DNN automatically compromises and optimizes, outputting an optimized target set adapted to the current situation, including a target temperature of 53°C and a target pressure of −17 kPa. The target pressure is 9 kPa and the target EO concentration is 7.6 mg / m³. Next, using the optimized target set as a constant, a fuzzy adaptive incremental control algorithm continuously outputs adjustment control quantities to bring the instantaneous operating data closer to the optimized target set. Specifically, during the sterilization process, the instantaneous operating data fluctuates wildly; the instantaneous temperature fluctuates between 52.2℃ and 53.8℃, and the instantaneous pressure fluctuates between -18 kPa and -20.5 kPa. At this time, the optimized target set remains constant, and the fuzzy PID continuously compares the optimized target set with the feedback instantaneous operating data. By adjusting the control quantities, it regulates the operating parameters of the actuators, such as adjusting the heating, valves, and vacuum pump, bringing the fluctuating instantaneous operating data to a small steady state around 53℃. This achieves dynamic parameter adjustment to match the current sterilization conditions, ensuring the sterilization effect.

[0087] Optionally, real-time operating data includes chamber temperature, chamber pressure, chamber vacuum, chamber EO concentration, chamber flow rate, chamber loading density, and instrument material characteristics; the multi-dimensional acquisition module 2 includes a temperature sensor, a pressure sensor, a vacuum transmitter, an EO concentration infrared sensor, an airflow velocity sensor, an ambient temperature and humidity sensor, a loading density detection sensor array, and a material characteristic detection element; instrument category parameters include the instrument's category, instrument material properties, instrument packaging type, and instrument loading operating conditions.

[0088] Specifically, temperature sensors are configured to be distributed in multiple layers along the height direction within the sterilization chamber, and are located inside the air inlet and outlet channels of the sterilization chamber; pressure sensors are configured at the standard pressure measurement interface on the side wall of the sterilization chamber and at the branch node of the main air inlet pipe; vacuum transmitters are configured at the main vacuum pumping pipeline of the sterilization chamber and at the pressure stabilizing pipeline at the front end of the vacuum pump outlet; and EO concentration infrared sensors are configured at the gas retention area in the middle of the sterilization chamber and the circulating air inside the chamber. The airflow velocity sensor is configured at the internal circulation air duct and the gas diffusion guide port of the sterilization chamber, respectively; the ambient temperature and humidity sensor is configured at the side of the air inlet of the machine room where the sterilization chamber is located and the external air circulation area of ​​the sterilization chamber, respectively; the loading density detection sensor array is configured to be distributed below the layered tray of the internal shelf of the sterilization chamber and inside the instrument loading inlet; the material characteristic detection element is configured to be distributed on both sides of the instrument loading inlet and the feeding end of the shelf of the sterilization chamber.

[0089] In this optional embodiment, the multi-dimensional acquisition module 2 can collect real-time data on the chamber temperature, chamber pressure, chamber vacuum, chamber EO concentration, chamber flow rate, chamber loading density, and instrument material characteristics. The multi-dimensional acquisition module 2 comprises: a temperature sensor with multiple layers and channels on the inner side to fully capture vertical temperature differences and airflow disturbances; a chamber pressure sensor to simultaneously monitor steady-state pressure and dynamic inlet pressure changes; and a chamber vacuum transmitter to monitor dynamic changes during the vacuum extraction process and equipment status. Together, these two sensors enable monitoring across the entire pressure range from atmospheric pressure to vacuum; a chamber EO concentration infrared sensor to simultaneously monitor localized high-concentration areas and overall uniformity; and a chamber airflow velocity sensor to monitor gas circulation efficiency and... Diffusion uniformity, when combined with a temperature sensor, allows for analysis of the impact of gas flow on concentration distribution. Furthermore, the combination of both sensors, along with the temperature sensor, enables analysis of the impact of airflow on concentration and temperature distribution. An environmental temperature and humidity sensor monitors external environmental interference with the sterilization process and, when used with an internal temperature sensor, distinguishes whether temperature changes within the chamber are caused by internal process factors or external environmental factors. A loading density detection sensor array monitors the overall distribution and local density of the loaded instruments, and, when used with an internal temperature sensor and an internal EO concentration infrared sensor, analyzes the impact of loading density on temperature and concentration distribution. A material characteristic detection element performs batch testing of the materials of all instruments to be sterilized, and, when combined with the instrument category parameters entered in the human-machine interaction module 3, cross-validates the material properties of the instruments. This multi-dimensional data acquisition module 2 addresses the issues of incomplete sterilization monitoring and inaccurate data, providing reliable, multi-dimensional data for AI decision-making and ensuring the comprehensiveness, accuracy, and authenticity of real-time operational data. Based on this, the components of the medical device category parameters include: the device category (defining the device's classification), the material attribute data (describing the device's material characteristics), and the packaging type data (describing the device's packaging form). These three can serve as partitioning constraints in the clustering process. The device loading condition data (describing the device's loading method and distribution) can be combined with the material attribute data and packaging type data as constraints for DNN optimization. This setup provides complete constraints for clustering and DNN optimization, improving the process adaptability for dynamic parameter optimization during sterilization and effectively enhancing the sterilization process.

[0090] Optionally, the categories of instruments include endoscopes, implants, medical dressings, and precision surgical instruments. In the process of obtaining the baseline process parameter set, AI optimization module 1 specifically retrieves historical operating condition data for a preset batch, using the complete sterilization process data of a single batch as samples. For each sample, a multi-dimensional feature vector is constructed to form multi-dimensional sample data:

[0091] ,

[0092] in, For the first Sample, For the first The average temperature of the sample, For the first The workload of each sample For the first Vacuum degree of the sample, For the first bar sample equilibrium concentration For the first The material feature values ​​of each sample; the extreme values ​​of specific dimension features of the samples are obtained by traversing the multidimensional sample data and statistically analyzing them. The multidimensional sample data is then normalized based on the following formula, mapped to the interval [0, 1], to form a normalized sample set:

[0093] ,

[0094] in, For normalized eigenvalues, For the first The first sample dimensional original eigenvalues, For the first multidimensional sample data The maximum value of the dimensional feature. For the first multidimensional sample data The minimum value of a feature;

[0095] Predetermine cluster categories, and randomly select a corresponding number of samples from the normalized sample set as initial cluster centers based on the number of cluster categories. For each normalized sample, calculate the spatial Euclidean distance between it and each initial cluster center using the following formula, and classify it according to the minimum distance criterion to form multiple clusters:

[0096] ,

[0097] in, For spatial Euclidean distance, For normalized samples, For the first Cluster centers, For the first The first cluster center Dimensional feature values; integrate clusters and iteratively converge to form a historical cluster dataset, and match the cluster category to which the instrument category parameter belongs, so as to filter and obtain the matching benchmark process parameter group from the historical cluster dataset.

[0098] In this optional embodiment, the categories of instruments include endoscopes, implants, medical dressings, and precision surgical instruments. Clustering categories can be set based on the categories of instruments; for example, the number of clustering categories can be set to 4. Based on this, during the process of obtaining the baseline process parameter set in AI optimization module 1, firstly, since the effect of EO sterilization is determined by multiple factors such as temperature, pressure, vacuum degree, EO concentration, loading density, and instrument material, the process parameters of a single sterilization batch are a multi-dimensional whole. Therefore, after retrieving historical sterilization data from the historical database, a complete single-batch process is used as a sample to construct the following... The multidimensional feature vectors transform unstructured batch data into structured multidimensional sample data, laying the foundation for subsequent mathematical processing and cluster analysis. Then, the multidimensional sample data is traversed, and the extreme values ​​of each dimension's features are statistically analyzed using the formula... The data is mapped to the interval [0, 1] to form a normalized sample set. This eliminates the dimensional and numerical differences between different feature dimensions through linear transformation, ensuring that subsequent Euclidean distance calculations fairly reflect the similarity between samples and guaranteeing the accuracy of clustering. Then, the number of cluster categories is set according to the number of medical device categories. A corresponding number of samples are randomly selected from the normalized sample set as initial cluster centers, thus determining the target number of groups and initial center points for clustering. This allows the algorithm to automatically group historical samples according to medical device categories, providing a foundation for subsequent iterative clustering. Subsequently, for each normalized sample, the following formula is used: The distances to each initial cluster center are calculated, and the samples are categorized according to the minimum distance criterion to form multiple clusters. This ensures that the similarity of process parameters among samples within the same cluster is the highest, while the differences between samples in different clusters are the greatest. Thus, all historical samples are divided into multiple clusters corresponding to the number of medical device categories, with each cluster representing a set of process parameters for a type of medical device. Finally, the clusters are integrated and iteratively converged to form a historical cluster dataset. Based on the medical device category parameters, the corresponding cluster category is matched, and the matching baseline process parameter set is obtained. This iterative convergence method allows multiple clusters to form historical cluster datasets for multiple medical device categories. The cluster center is the baseline process parameter for that type of medical device. The multidimensional feature vector of the cluster center is used as the baseline process parameter set for that type of medical device. This matches the current medical device category with the optimal baseline process parameter set obtained from historical data clustering, providing a reliable initial parameter foundation for subsequent DNN optimization and closed-loop control.

[0099] Optionally, the AI ​​optimization module 1, in the process of integrating clusters and iteratively converging to form a historical cluster dataset, is specifically used for: after completing the initial partitioning and classification of each normalized sample to form a cluster, repeating the partitioning and classification process of each normalized sample within the cluster; and at the end of each round of partitioning and classification, obtaining the mean of all normalized samples within the cluster based on the following formula, and iteratively updating the cluster centers corresponding to the clusters:

[0100] ,

[0101] in, For the first The first iteration after the update Cluster centers, For the first There are several clusters. Based on the updated cluster centers and corresponding normalized samples in each round, the offset distance between the current cluster center and the previous cluster center and the sum of errors in this round are obtained according to the following formula. When the offset distance is less than a preset offset threshold, or the difference between the current error and the sum of errors in the previous round approaches zero, or the number of iterations is the preset maximum number of iterations, the iteration update is completed and the cluster centers are output. Each cluster center is labeled according to the category of the medical device to form a historical cluster dataset.

[0102] ,

[0103] ,

[0104] in, For the first The cluster centers at the in Compared to the first round of updates The offset distance of the round update, For the first The first iteration after the update The first cluster center Dimensional features, No. The first iteration after the update The first cluster center Dimensional features, This is the sum of the total squared errors of the clustering. The number of cluster categories, It is an adjacent 2-norm.

[0105] In this optional embodiment, during the process of AI optimization module 1 integrating clusters and iteratively converging to form a historical cluster dataset, firstly, after completing the initial classification to form clusters, the partitioning and classification process of each normalized sample is repeated. At the end of each round of partitioning and classification, based on the formula... The mean of all normalized samples within a cluster is obtained, and the cluster centers are iteratively updated. This causes the cluster centers to move towards the centroid of the samples within the cluster, gradually correcting the initial center deviation and converging the cluster centers to the true distribution center of the samples within the cluster, thus improving the representativeness of the baseline process parameters. Then, based on the updated cluster centers and corresponding normalized samples in each round, the parameters are further refined using formula... and The algorithm calculates the offset distance of the cluster centers and the sum of the errors in this iteration. When one of the three preset iteration conditions is met, the iteration converges and outputs the cluster centers. This setting allows for monitoring changes in the cluster centers (offset distance) and the error function (error and difference) to determine whether the algorithm has found the optimal cluster centers or reached the maximum number of iterations. It also allows for reasonable termination of the iteration while ensuring the stability of the clustering results, balancing algorithm accuracy and computational efficiency. Finally, each cluster center is labeled according to the medical device category, forming a historical clustering dataset. This binds each converged cluster center to its corresponding medical device category, establishing a mapping relationship between category and process parameters. This standardized historical clustering dataset provides a reliable foundation for subsequent process optimization.

[0106] Optionally, the three-layer DNN deep neural network includes an input layer, a hidden layer, and an output layer connected sequentially. Specifically, the AI ​​optimization module 1, in acquiring the optimization target set, is used to: transmit the baseline process parameter set and real-time multi-dimensional operating condition data to the input layer, forming the following total network input feature vector:

[0107] ,

[0108] in, This is the total input feature vector of the network. As the baseline process parameter set, and , Set the temperature for the category benchmark. Working pressure to set category benchmarks, As the benchmark vacuum level for the product category, As the category benchmark EO concentration, - These are the baseline times for the four stages of the sterilization process: preheating, permeation, sterilization, and desorption. For real-time multi-dimensional operating condition data, and , The cabin temperature, For cabin pressure, The vacuum level inside the cabin, The concentration of EO inside the chamber. For the internal loading density, The material characteristics of the equipment are used; the total input feature vector of the transmission network is sent to the hidden layer, and the connection weights and biases of the hidden layer are obtained by training and converging based on historical working condition data. Linear weighted summation and nonlinear activation are then performed sequentially based on the following formula to obtain the nonlinear output value of the neuron:

[0109] ,

[0110] ,

[0111] in, For the hidden layer's first A weighted sum of neurons, For the input layer The node is connected to the hidden layer. Each node connection weight The first feature vector is the total input feature vector of the network. 3D input feature values, For the hidden layer Neuron bias For the first The nonlinear output value of a neuron. The natural constant is used; the nonlinear output value of the transmission neuron is sent to the output layer, and the weights and biases of the output layer obtained by training and convergence based on historical working condition data are retrieved. The target parameters are then output based on the following formula to form the optimization target set:

[0112] ,

[0113] in, To optimize the target set, For output layer weights, This represents the nonlinear output value of the neuron. This is the output layer bias. To optimize the temperature inside the target cabin, To optimize the pressure inside the target cabin, To optimize the vacuum level inside the target chamber, To optimize the EO concentration in the target chamber.

[0114] In this optional embodiment, a three-layer DNN deep neural network is constructed by sequentially connecting an input layer, a hidden layer, and an output layer. During the process of the AI ​​optimization module 1 acquiring the optimization target set, firstly, the baseline process parameter set and real-time operating condition data are transmitted to the input layer, constructing a network as shown in the formula... The network's total input feature vector is set up in such a way that the baseline process parameters are used as static optimal parameters, combined with dynamic real-time operating data as input data for the input layer. This facilitates the DNN network's subsequent learning of how the baseline process parameters dynamically adjust with real-time operating conditions, and then outputs optimized parameters adapted to the current operating conditions through the coupling relationship between the two. Next, the input feature vector is transmitted to the hidden layer, and the trained hidden layer weights and biases are retrieved, and then processed using the formula... Linear weighted summation can achieve a combined transformation of input features, mapping the input feature vector to a high-dimensional feature space, and then using the formula... Nonlinear activation introduces nonlinear characteristics into the DNN network, allowing the acquisition of nonlinear output values ​​from neurons. This enables the DNN network to fit the complex nonlinear relationship between process parameters and operating conditions, providing feature support for subsequent output optimization target sets. Finally, the nonlinear output values ​​of the hidden layer neurons are transmitted to the output layer, and the trained output layer weights and biases are retrieved using the formula... The output optimizes the target parameters, thereby mapping the features extracted from the hidden layer back to the physical space of the actual process parameters through the linear transformation of the output layer, obtaining the optimized target data adapted to the current operating conditions, forming an optimized target set, and providing accurate target values ​​for subsequent fuzzy adaptive closed-loop control.

[0115] Optionally, the AI ​​optimization module 1 is specifically used in the process of obtaining the regulation control quantity to: obtain the operating condition deviation value based on the difference between the optimization target set and the instantaneous operating condition data; obtain the change in operating condition deviation based on the time series of instantaneous operating condition data collected in real time by the collaborative adjustment module 4; calibrate the initial gain coefficient, maximum allowable deviation threshold, and maximum deviation change threshold of the PID based on the sterilization process characteristics and historical qualified batch data statistically integrated from the historical database, form a fuzzy rule base, use the operating condition deviation value and the change in operating condition deviation as fuzzy input, update and output the PID gain coefficient in real time through fuzzy inference; and calculate the regulation control quantity according to the PID gain coefficient, the operating condition deviation value, and the change in operating condition deviation using the following formula:

[0116] ,

[0117] ,

[0118] in, For the first Control increment at any time, , , These are the proportional gain coefficient, integral gain coefficient, and derivative gain coefficient in the PID gain coefficients, respectively. For the first Operating condition deviation value at time, For the first The change in operating condition deviation at any given time. For the first The amount of adjustment and control at any given time. For the first The amount of adjustment and control at any given time.

[0119] Specifically, the actuators connected to the coordinated adjustment module 4 include the EO flow regulating valve, the EO inlet valve, the pressure stabilizing buffer tank inlet gas ratio valve, the heating component, the cooling heat exchange circuit regulating valve, the vacuum pump group control valve, the vacuum pressure relief regulating valve, and the chamber pressure balance valve. The operating parameters include the regulating valve opening, the inlet valve on / off, the inlet flow ratio, the heating component power, the cooling valve opening, the vacuum pump operating level, and the balance valve opening. The instantaneous operating data include instantaneous temperature, instantaneous pressure, instantaneous vacuum degree, and instantaneous EO concentration.

[0120] In this optional embodiment, the coordinated adjustment module 4 can adjust the corresponding operating parameters through a communication connection with the execution component, and can also directly read the controller feedback signal of the execution component, i.e., the operating parameters, to determine whether the control command is effectively executed, thereby realizing real-time adjustment of sterilization parameters and providing action feedback for the control algorithm. The instantaneous operating condition data obtained by the coordinated adjustment module 4 can be collected through the execution component or a nearby installed sensor, which facilitates ensuring that the instantaneous operating condition data always tracks the optimization target parameters during dynamic adjustment. On this basis, in the process of the AI ​​optimization module 1 obtaining the adjustment control quantity, firstly, since the core of the PID control algorithm is based on deviation adjustment, knowing only the deviation value cannot reflect the trend of deviation change, such as whether the deviation is increasing or decreasing. Therefore, while obtaining the operating condition deviation value based on the difference between the optimization target set and the instantaneous operating condition data, the change in operating condition deviation is obtained based on the real-time collected instantaneous operating condition data time sequence, thereby providing a basis for the fuzzy PID control algorithm. The control algorithm provides two-dimensional input to achieve comprehensive perception of operating condition deviations. Then, due to the strong nonlinearity and time-varying nature of the operating conditions within the sterilization chamber (e.g., rapid concentration changes during EO intake and temperature lag during heating), fixed PID parameters cannot adapt to the characteristics of different operating conditions. Fuzzy control is needed to adjust parameters in real time. Therefore, based on historical database statistics of sterilization process characteristics and qualified batch data, the initial PID gain coefficient, maximum permissible deviation threshold, and maximum deviation change threshold are calibrated to form a fuzzy rule base. Then, using the operating condition deviation value and deviation change as fuzzy input, fuzzy inference updates and outputs the PID gain coefficient in real time, thereby dynamically adjusting the PID parameters according to the current deviation state, achieving adaptive parameter tuning. This ensures the PID control algorithm maintains optimal regulation performance in different operating conditions, quickly eliminating deviations and avoiding overshoot. Finally, based on the PID gain coefficient, operating condition deviation value, and deviation change, the algorithm uses the formula... Calculate the control increment using the formula The control quantity at the current moment is calculated by adding the control quantity from the previous moment and transmitted to the working condition coordination adjustment module. This drives the actuators to adjust the operating parameters, thereby ensuring the output of stable and accurate control commands, enabling the actual working conditions inside the cabin to quickly and stably track and optimize the target parameters.

[0121] Optionally, the ethylene oxide sterilization control system based on AI dynamic parameter optimization also includes a multi-factor analysis module 5. The multi-factor analysis module 5 has a built-in multi-factor SAL prediction model. The device category parameters also include the initial bioburden value of the device. The AI ​​optimization module 1 is communicatively connected to the multi-factor analysis module 5 and is used to: during the sterilization process in the sterilization chamber, retrieve the sterilization efficiency correction coefficient calibrated by the sterilization process characteristics according to the device category, and input the initial bioburden value of the device, real-time operating data, and sterilization efficiency correction coefficient to the multi-factor analysis module 5, so as to output a predicted value through the following multi-factor SAL prediction model:

[0122] ,

[0123] ,

[0124] in, This represents the cumulative sterilization effect of the sterilization process. for Real-time EO concentration inside the chamber. for Real-time temperature inside the cabin for Real-time pressure inside the cabin. For the continuous time variable of the sterilization process, The sterilization process has accumulated time. As a predicted value, This represents the initial bioburden value of the device. This is the sterilization efficiency correction coefficient; when the predicted value is greater than the qualified threshold, sterilization is deemed unqualified. The delay compensation ratio coefficient and concentration compensation ratio coefficient, calibrated by the sterilization process characteristics, are retrieved. Based on the following formula, the sterilization compensation time and EO concentration increment are obtained and transmitted to AI optimization module 1 to update the target optimization set:

[0125] ,

[0126] ,

[0127] in, To compensate for the sterilization time, Here, A is the delay compensation ratio coefficient, and A is the acceptable threshold. For the increment of EO concentration, This is the concentration compensation ratio coefficient.

[0128] Specifically, the sterilization efficiency correction factor is retrieved from the historical database based on the type of medical device, reflecting the impact of different materials and packaging on the sterilization effect; the initial bioburden value of the medical device is the bacterial load of the device before sterilization, which is obtained by external equipment detection.

[0129] In this optional embodiment, such as Figure 1 As shown, the ethylene oxide sterilization control system based on AI dynamic parameter optimization also includes a multi-factor analysis module 5. This module 5 incorporates a multi-factor SAL prediction model, which is derived from the equation... Japanese style The system consists of several components, including the initial bioburden value of the medical device category. The AI ​​optimization module 1 communicates with the multi-factor analysis module 5, enabling real-time quality prediction of the sterilization process. In this process, firstly, during the sterilization process, the sterilization efficiency correction coefficient is retrieved based on the medical device category. The initial bioburden value, real-time operating data, and sterilization efficiency correction coefficient are input to the multi-factor analysis module 5 to construct a multi-factor input of bioburden, real-time operating data, and category correction coefficient. This provides complete sterilization process information for the SAL prediction model, ensuring the accuracy of the SAL prediction results. Then, when the predicted SAL value is greater than the acceptable threshold, sterilization is deemed substandard. The delay compensation ratio coefficient and concentration compensation ratio coefficient are retrieved, and based on the formula... and The sterilization compensation time and EO concentration increment are calculated separately and transmitted to AI optimization module 1 to update the target optimization set. This setting allows the required compensation time and concentration increment to be calculated based on the deviation between the predicted value and the qualified threshold through a logarithmic relationship. The larger the deviation, the larger the compensation amount, thus achieving on-demand compensation and issuing compensation instructions. Considering that the compensation time and concentration increment are subject to drastic fluctuations in real-time operating data, the target optimization set is updated through AI optimization module 1. Based on the target optimization set and PID control algorithm, the actuator is driven to extend the sterilization time or increase the EO concentration to supplement the sterilization effect, achieving stable dynamic compensation and improving the sterilization qualification rate and production efficiency.

[0130] Optionally, the multi-factor analysis module 5 also includes a built-in EO residue prediction model, and the AI ​​optimization module 1 is further used for: when the sterilization process is completed, if the predicted value is greater than the qualified threshold, determining that the batch of instruments corresponding to the current sterilization process is an unqualified batch, locking and marking the current batch, and outputting an alarm signal and a prompt to restart the sterilization process to the human-machine interaction module; if the predicted value is less than the qualified threshold, retrieving the EO analysis correction coefficient based on the instrument category and the instrument material attribute data and instrument packaging type data from the historical database, and transmitting the effective volume of the sterilization chamber, the EO analysis correction coefficient, and real-time operating condition data to the multi-factor analysis module 5, so that the following EO residue prediction model can output the EO residue amount:

[0131] ,

[0132] in, This refers to the residual amount of EO. For effective volume, This is the total duration of the sterilization process. This represents the cumulative EO concentration coupled with pressure during the sterilization process. For EO analysis correction coefficients, The temperature accumulation during the sterilization process is used to determine the batch of instruments corresponding to the current sterilization process as an unqualified batch when the residual EO amount exceeds the preset residual threshold. The current batch is locked and marked, and an alarm signal and a prompt to restart the sterilization process are output to the human-machine interaction module 3.

[0133] Specifically, the cumulative amount of EO concentration coupled with pressure during the sterilization process reflects the total amount of EO adsorbed by the device. That is, the higher the concentration and the greater the pressure, the greater the adsorption. As for the cumulative amount of temperature during the sterilization process, the temperature promotes the desorption of EO during the analysis process. The higher the temperature, the greater the desorption.

[0134] In this optional embodiment, in addition to the built-in multi-factor SAL prediction model, the multi-factor judgment module 5 also has a built-in EO residual inference model, which is derived from the formula... The AI ​​optimization module 1, through the multi-factor analysis module 5, performs EO residue estimation. First, after the sterilization process is completed, if the SAL (Self-Altered Acid) prediction value is greater than the acceptable threshold, the batch is deemed unqualified, the current batch is locked, and an alarm signal and a prompt to restart the sterilization process are output to the human-machine interaction module 3 for operator processing. This ensures that only batches that have passed sterilization can proceed to subsequent stages. Clear marking and prompts also prevent batch confusion and quality incidents. Next, if the SAL prediction value is less than the acceptable threshold, material attribute data and packaging type data are retrieved according to the medical device category. The system collects EO analysis correction coefficients from the sterilization chamber's effective volume, EO analysis correction coefficients, and real-time multi-dimensional operating conditions data, and transmits these data to the multi-factor analysis module 5. This provides input for the EO residue prediction model, enabling the construction of multi-factor inputs for sterilization process conditions, instrument materials, and packaging types. The problem of predicting EO residue is transformed into a quantitative calculation problem based on multi-dimensional data. When the EO residue exceeds the threshold, the batch is also deemed unqualified, the current batch is locked, and an alarm signal and a prompt to restart the sterilization process are output to the human-machine interaction module 3, further ensuring sterilization quality.

[0135] Optionally, the ethylene oxide sterilization control system based on AI dynamic parameter optimization also includes an encrypted traceability module 6. The encrypted traceability module 6 has a built-in hash encryption algorithm and a dual-link evidence storage unit, and is communicatively connected to the human-machine interaction module 3 and the AI ​​optimization module 1, respectively. The encrypted traceability module 6 is used to: receive the data record set of each batch of instruments during the sterilization process transmitted by the AI ​​optimization module 1 in real time, add a timestamp to form a sterilization dataset, wherein the data record set includes real-time acquired data, optimization adjustment records, human-machine interaction records, and alarm lock records; encrypt the sterilization dataset using a hash encryption algorithm to form an encrypted dataset and transmit it to the dual-link evidence storage unit; when a retrieval command is received from the human-machine interaction module 3, the timestamp and data record set of the corresponding sterilization dataset are read according to the retrieval command, and the encrypted dataset is verified by the encryption algorithm.

[0136] In this optional embodiment, such as Figure 1 As shown, an encrypted traceability module 6 is also provided. This module incorporates a hash encryption algorithm and a dual-link evidence storage unit, and is communicatively connected to the human-machine interaction module 3 and the AI ​​optimization module 1, respectively. This configuration allows the encrypted traceability module 6 to receive data records from the AI ​​optimization module 1 in real time during the sterilization process of each batch of instruments, and then timestamp them to form a sterilization dataset. The data record set includes real-time collected data, optimization adjustment records, human-machine interaction records, and alarm lock records, ensuring the integrity and timeliness of the sterilization dataset. Based on this, the encrypted traceability module 6 can... The hash encryption algorithm encrypts the sterilization dataset, forming an encrypted dataset, which is then transmitted to the dual-link evidence storage unit. This ensures the authenticity, integrity, and tamper-proof nature of the sterilization dataset through a unique hash value, while the dual-link evidence storage enhances the reliability and security of data storage. When a retrieval command is received from the human-machine interface module 3, the timestamp and data record set of the corresponding sterilization dataset are read according to the retrieval command, and the encrypted dataset is verified through the encryption algorithm. This achieves secure retrieval and verification of sterilization data, ensuring the authenticity and integrity of the data and providing a reliable basis for product quality traceability and compliance inspection.

[0137] Secondly, one embodiment of the present invention provides an ethylene oxide sterilization control method based on AI dynamic parameter optimization, comprising: S1, after the instruments to be sterilized are moved into the sterilization chamber, acquiring real-time operating condition data collected at a fixed frequency by the multi-dimensional acquisition module 2 of the ethylene oxide sterilization control system based on AI dynamic parameter optimization, and acquiring instrument category parameters entered by the human-computer interaction module 3 of the ethylene oxide sterilization control system based on AI dynamic parameter optimization; S2, pre-setting clustering categories and using the K-Means clustering algorithm to cluster and divide the historical operating condition data in the pre-set historical database. S3. Based on the pre-built three-layer DNN deep neural network, the benchmark process parameter set is optimized with real-time operating data as a constraint to obtain the optimization target set; S4. The optimization target set is used as a setpoint, and the instantaneous operating data transmitted in real time by the collaborative adjustment module 4 of the ethylene oxide sterilization control system based on AI dynamic parameter optimization is used as a feedback value. The real-time deviation between the setpoint and the feedback value is used as input, and the adjustment control quantity is obtained through the fuzzy adaptive incremental control algorithm and transmitted to the collaborative adjustment module 4 to adjust the operating parameters of the corresponding actuator.

[0138] like Figure 2 As shown, the technical effect of the ethylene oxide sterilization control method based on AI dynamic parameter optimization in this embodiment is similar to that of the ethylene oxide sterilization control system based on AI dynamic parameter optimization described above, and will not be repeated here.

[0139] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An ethylene oxide sterilization control system based on AI dynamic parameter optimization, characterized in that, It includes an AI optimization module (1) and a multi-dimensional acquisition module (2), a human-computer interaction module (3), and a collaborative adjustment module (4), all of which are communicatively connected to the AI ​​optimization module (1); the multi-dimensional acquisition module (2) is configured to be distributed inside and outside the sterilization chamber; the collaborative adjustment module (4) is configured inside the sterilization chamber and is communicatively connected to the execution components inside the sterilization chamber; the AI ​​optimization module (1) is used for: After the instruments to be sterilized are moved into the sterilization chamber, the instrument category parameters entered by the human-machine interaction module (3) are obtained, the execution component is driven to work through the collaborative adjustment module (4), and the real-time working condition data collected by the multi-dimensional acquisition module (2) at a fixed frequency is obtained. Preset clustering categories and use the K-Means clustering algorithm to cluster and divide historical operating data in the preset historical database, and filter to obtain a set of benchmark process parameters that match the parameters of the instrument category; Based on the pre-built three-layer DNN deep neural network, the benchmark process parameter set is optimized with the real-time operating data as a constraint to obtain the optimization target set; The optimization target set is used as a fixed value, and the instantaneous operating condition data transmitted in real time by the collaborative adjustment module (4) is used as a feedback value. The real-time deviation between the fixed value and the feedback value is used as input. The adjustment control quantity is obtained through the fuzzy adaptive incremental control algorithm and transmitted to the collaborative adjustment module (4) to adjust the operating parameters of the execution component accordingly.

2. The ethylene oxide sterilization control system based on AI dynamic parameter optimization as described in claim 1, characterized in that, The real-time operating data includes cabin temperature, cabin pressure, cabin vacuum, cabin EO concentration, cabin flow rate, cabin loading density, and instrument material characteristic values; the multi-dimensional acquisition module (2) includes a temperature sensor, a pressure sensor, a vacuum transmitter, an EO concentration infrared sensor, an airflow velocity sensor, an ambient temperature and humidity sensor, a loading density detection sensor array, and a material characteristic detection element; the instrument category parameters include the instrument category, instrument material attribute data, instrument packaging type data, and instrument loading operating data.

3. The ethylene oxide sterilization control system based on AI dynamic parameter optimization as described in claim 2, characterized in that, The categories of instruments include endoscopes, implants, medical dressings, and precision surgical instruments. The AI ​​optimization module (1) is specifically used in the process of acquiring the baseline process parameter set for: Retrieve the historical operating condition data of the preset batch, and use the complete sterilization process data of a single batch as a sample. Construct the following multi-dimensional feature vector for each sample to form multi-dimensional sample data: ， Among them, the For the first The sample mentioned above, the For the first The average temperature of the sample, the For the first The working pressure of the sample described in the article, the For the first The vacuum degree of the sample mentioned above, the For the first The sample mentioned in the article equilibrium concentration, the For the first The material characteristic values ​​of the sample mentioned in the article; The extreme values ​​of specific dimension features of the samples are obtained by traversing the multidimensional sample data. The multidimensional sample data is then normalized based on the following formula, mapped to the interval [0, 1], to form a normalized sample set: ， Among them, the For normalized eigenvalues, the For the first The sample described in the article is number one. The original eigenvalues, the The first in the multidimensional sample data The maximum value of the dimensional feature, the The first in the multidimensional sample data The minimum value of a feature; Pre-defined clustering categories are established. Based on the number of clustering categories, a corresponding number of samples are randomly selected from the normalized sample set as initial cluster centers. For each normalized sample, the spatial Euclidean distance between it and each of the initial cluster centers is calculated using the following formula. Then, the samples are classified according to the minimum distance criterion to form multiple clusters: ， Among them, the Let Euclidean distance be the spatial distance. For the normalized sample, the For the first Cluster centers, the For the first The first cluster center Dimensional feature values; The clusters are integrated and iteratively converged to form a historical cluster dataset. Based on the instrument category parameters, the corresponding cluster category is matched to select and obtain the matching benchmark process parameter group from the historical cluster dataset.

4. The ethylene oxide sterilization control system based on AI dynamic parameter optimization as described in claim 3, characterized in that, The AI ​​optimization module (1) is specifically used in the process of integrating the clusters to iteratively converge and form the historical cluster dataset for: After completing the initial partitioning and classification of each normalized sample to form the cluster, the partitioning and classification process for each normalized sample within the cluster is repeated. At the end of each round of partitioning and classification, the mean of all normalized samples within the cluster is obtained based on the following formula, and the cluster center corresponding to the cluster is iteratively updated: ， Among them, the For the first The first iteration after the update The cluster centers, the For the first The aforementioned clusters; Based on the cluster centers and corresponding normalized samples updated in each round, the offset distance of the current cluster center compared to the previous cluster center and the sum of errors in the current round are obtained according to the following formula. When the offset distance is less than a preset offset threshold, or the difference between the current error and the sum of errors in the previous round approaches zero, or the number of iterations is a preset maximum number of iterations, the iteration update converges and the cluster centers are output. Each cluster center is then labeled according to the category of the medical device to form the historical clustering dataset. ， ， Among them, the For the first The cluster centers at the in Compared to the first round of updates The offset distance of the round update, the For the first The first iteration after the update The first of the cluster centers dimensional features, the No. The first iteration after the update The first of the cluster centers dimensional features, the The sum of the total squared errors of the clustering, the The number of cluster categories, the It is an adjacent 2-norm.

5. The ethylene oxide sterilization control system based on AI dynamic parameter optimization as described in claim 2, characterized in that, The three-layer DNN deep neural network includes an input layer, a hidden layer, and an output layer connected in sequence. The AI ​​optimization module (1) is specifically used in the process of obtaining the optimization target set to: The baseline process parameter set and the real-time multidimensional operating condition data are transmitted to the input layer to form the following total network input feature vector: ， Among them, the The total input feature vector of the network, the The reference process parameter set is, and The To set a temperature for the product category benchmark, the As a category benchmark working pressure, the As the standard vacuum level for the product category, the As the category benchmark EO concentration, the - These are the baseline times for the four stages of the sterilization process: preheating, permeation, sterilization, and desorption. The real-time multidimensional operating condition data, and The The cabin temperature is [the temperature inside the cabin]. The pressure inside the cabin, the The vacuum level inside the cabin, the The EO concentration inside the chamber, the The loading density inside the cabin, the The material characteristic value of the instrument; The total input feature vector of the network is transmitted to the hidden layer. The connection weights and biases of the hidden layer, obtained by training and convergence based on the historical working condition data, are retrieved. Linear weighted summation and nonlinear activation are performed sequentially based on the following formula to obtain the nonlinear output value of the neuron: ， ， Among them, the The first of the hidden layers The weighted sum of neurons, the For the input layer The node to the hidden layer The node connection weight, the The first of the total input feature vectors of the network Dimensional input feature value, the The first of the hidden layers The bias of each neuron, the For the first The nonlinear output value of the neuron, the It is a natural constant; The nonlinear output value of the neuron is transmitted to the output layer. The weights and biases of the output layer, obtained by training and convergence based on the historical working condition data, are retrieved. The optimization target parameters are output based on the following formula to form the optimization target set: ， Among them, the For the optimization target set, the For the output layer weights, the The nonlinear output value of the neuron, the For the output layer bias, the To optimize the temperature inside the target cabin, the To optimize the pressure inside the target cabin, the To optimize the vacuum level inside the target chamber, the To optimize the EO concentration in the target chamber.

6. The ethylene oxide sterilization control system based on AI dynamic parameter optimization as described in claim 2, characterized in that, The AI ​​optimization module (1) is specifically used in the process of acquiring the adjustment control quantity to: Based on the difference between the optimization target set and the instantaneous operating condition data, the operating condition deviation value is obtained. Based on the time sequence of the instantaneous operating condition data collected in real time by the collaborative adjustment module (4), the change in operating condition deviation based on the time sequence before and after the time sequence is obtained. Based on the sterilization process characteristics and historical qualified batch data statistically integrated from the historical database, the initial gain coefficient of PID, the maximum allowable deviation threshold, and the maximum deviation change threshold are calibrated to form a fuzzy rule base. The operating condition deviation value and the operating condition deviation change are used as fuzzy inputs, and the PID gain coefficient is updated and output in real time through fuzzy inference. The regulating control quantity is calculated using the following formula based on the PID gain coefficient, the operating condition deviation value, and the change in operating condition deviation: ， ， Among them, the For the first The control increment at time, the The above The above These are the proportional gain coefficient, integral gain coefficient, and derivative gain coefficient in the PID gain coefficients, respectively. For the first The operating condition deviation value at time, the For the first The change in the operating condition deviation at time t, For the first The adjustment control quantity at time, the For the first The adjustment and control quantity at any given time.

7. The ethylene oxide sterilization control system based on AI dynamic parameter optimization as described in claim 2, characterized in that, It also includes a multi-factor analysis module (5), which has a built-in multi-factor SAL prediction model. The device category parameters also include the initial bioburden value of the device. The AI ​​optimization module (1) is communicatively connected to the multi-factor analysis module (5) and is used for: During the sterilization process in the sterilization chamber, the sterilization efficiency correction coefficient calibrated by the sterilization process characteristics is retrieved according to the type of the instrument. The initial bioburden value of the instrument, the real-time operating data, and the sterilization efficiency correction coefficient are input to the multi-factor analysis module (5) to output the predicted value through the multi-factor SAL prediction model as follows: ， ， Among them, the The cumulative sterilization effect of the sterilization process, the for Real-time EO concentration inside the chamber, the for Real-time temperature inside the cabin, the for Real-time pressure inside the cabin, the aforementioned The sterilization process is a continuous time variable. The total time for the sterilization process has been accumulated. The predicted value is... The initial bioburden value of the device, the This is the sterilization efficiency correction factor; When the predicted value is greater than the qualified threshold, sterilization is determined to be substandard. The delay compensation ratio coefficient and concentration compensation ratio coefficient calibrated by the sterilization process characteristics are retrieved, and the sterilization compensation time and EO concentration increment are obtained based on the following formula, and transmitted to the AI ​​optimization module (1) to update the target optimization set: ， ， Among them, the For the sterilization compensation time, the The delay compensation ratio coefficient is A, and the qualified threshold is A. For the EO concentration increment, the This is the concentration compensation ratio coefficient.

8. The ethylene oxide sterilization control system based on AI dynamic parameter optimization as described in claim 7, characterized in that, The multi-factor analysis module (5) also has a built-in EO residual inference model, and the AI ​​optimization module (1) is also used for: When the sterilization process is completed, if the predicted value is greater than the qualified threshold, the batch of the instrument corresponding to the current sterilization process is determined to be an unqualified batch, the current batch is locked and marked, and an alarm signal and a prompt to restart the sterilization process are output to the human-computer interaction module (3). If the predicted value is less than the qualified threshold, the EO analysis correction coefficient, which is calibrated by combining the material attribute data of the device and the packaging type data of the device in the historical database, is retrieved according to the category of the device. The effective volume of the sterilization chamber, the EO analysis correction coefficient, and the real-time operating condition data are transmitted to the multi-factor judgment module (5) so that the EO residue estimation model can output the EO residue amount as follows: ， Among them, the The residual amount of EO, the For the effective volume, the The total duration of the sterilization process, the The cumulative amount of EO concentration coupled with pressure in the sterilization process, the The EO analytical correction coefficient, the This refers to the cumulative temperature increase during the sterilization process. When the residual amount of EO is greater than the preset residual threshold, the batch of the instrument corresponding to the current sterilization process is determined to be an unqualified batch, the current batch is locked and marked, and an alarm signal and a prompt to restart the sterilization process are output to the human-machine interaction module (3).

9. The ethylene oxide sterilization control system based on AI dynamic parameter optimization as described in claim 1, characterized in that, It also includes an encrypted traceability module (6), which has a built-in hash encryption algorithm and a dual-link evidence storage unit, and is communicatively connected to the human-computer interaction module (3) and the AI ​​optimization module (1), respectively. The encrypted traceability module (6) is used for: The AI ​​optimization module (1) transmits data records of each batch of instruments during the sterilization process in real time, and timestamps them to form a sterilization dataset. The data record set includes real-time data acquisition, optimization adjustment records, human-computer interaction records and alarm lock records. The sterilization dataset is encrypted using the hash encryption algorithm to form an encrypted dataset and transmitted to the dual-link evidence storage unit. When the retrieval instruction is received from the human-computer interaction module (3), the timestamp and data record set corresponding to the sterilization dataset are read according to the retrieval instruction, and the encrypted dataset is verified by the encryption algorithm.

10. A method for controlling ethylene oxide sterilization based on AI dynamic parameter optimization, characterized in that, include: S1. After the instruments to be sterilized are moved into the sterilization working chamber, the real-time working condition data collected at a fixed frequency by the multi-dimensional acquisition module (2) of the ethylene oxide sterilization control system based on AI dynamic parameter optimization is obtained, and the instrument category parameters entered by the human-computer interaction module (3) of the ethylene oxide sterilization control system based on AI dynamic parameter optimization are obtained. S2. Preset clustering categories and use the K-Means clustering algorithm to cluster and divide the historical operating condition data in the preset historical database, and filter to obtain the benchmark process parameter group that matches the instrument category parameters; S3. Based on the pre-built three-layer DNN deep neural network, optimize the benchmark process parameter set with the real-time operating data as constraints to obtain the optimization target set; S4. The optimization target set is used as a fixed value, and the instantaneous operating condition data transmitted in real time by the collaborative adjustment module (4) of the ethylene oxide sterilization control system based on AI dynamic parameter optimization is used as a feedback value. The real-time deviation between the fixed value and the feedback value is used as input, and the adjustment control quantity is obtained through the fuzzy adaptive incremental control algorithm and transmitted to the collaborative adjustment module (4) to adjust the operating parameters of the execution component accordingly.

Images