Medical dialysis dry powder production whole process intelligent supervision method

Abstract

The invention discloses a medicine dialysis dry powder production whole process intelligent supervision method, and belongs to the technical field of medicine production intelligent monitoring. The method comprises the following steps: collecting a raw material purity parameter, a mixing process microscopic parameter and a finished product performance index parameter in real time through a sensor array module; the data processing module performs preprocessing and feature extraction on the collected parameters; the deep learning model module carries out training based on the preprocessed parameters to generate a dynamic ratio adjustment scheme; the control module controls the raw material conveying amount of the production equipment according to the adjustment scheme; and the closed loop feedback module feeds back the adjusted finished product performance index parameters to the deep learning model module to form a dynamic optimization closed loop. The method further comprises an abnormity early warning step and a model iteration step, the whole-process intelligent supervision of the medicine dialysis dry powder production process is realized, the proportion can be automatically adjusted according to the purity fluctuation of the raw materials, the stable product quality is ensured, the production efficiency is improved, and the manual intervention requirement is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical production supervision, and particularly relates to a medical dialysis dry powder production whole-process intelligent supervision method. BACKGROUND

[0002] Medical dialysis dry powder is a key consumable in hemodialysis treatment, and its production quality is directly related to the treatment effect and safety of patients.

[0003] At present, medical dialysis dry powder production supervision mainly adopts the traditional manual operation combined with offline detection mode, and this mode has obvious deficiencies in production efficiency and product quality stability.

[0004] Firstly, in the traditional medical dialysis dry powder production supervision process, the raw material purity detection mainly depends on manual operation and offline detection, which needs to be carried out by the way of regular sampling and testing. This way is not only inefficient, but also easy to cause unstable product quality due to human operation errors. Especially in the case of large batch changes of raw materials, it is difficult to adjust the production parameters in time, which affects the quality of the final product.

[0005] Secondly, the traditional supervision method is difficult to realize real-time capture and collaborative analysis of multi-dimensional and high dynamic parameters. In the production process of medical dialysis dry powder, parameters such as temperature, stirring rate and reaction time have important influence on product quality, but the existing technology is difficult to realize real-time monitoring and comprehensive analysis of these parameters, which leads to lack of data support for production process optimization and inability to dynamically adjust according to the characteristics of raw materials and environmental changes.

[0006] In addition, although there are some intelligent supervision systems in the existing technology, most of them lack special design for the characteristics of medical dialysis dry powder production, and it is difficult to meet the special production requirements of this field. Especially in the case of large fluctuation of raw material purity, there is a lack of effective dynamic proportioning adjustment mechanism, which cannot guarantee the stability of product quality.

[0007] Therefore, it is urgent to develop a whole-process intelligent supervision method for the characteristics of medical dialysis dry powder production, to realize real-time monitoring, data analysis and dynamic adjustment of the production process, and to improve the stability of product quality and production efficiency.

[0008] Therefore, we need to design a medical dialysis dry powder production whole-process intelligent supervision method to solve these problems. SUMMARY

[0009] In order to solve the technical problems that the traditional medical dialysis dry powder production supervision relies on manual operation and offline detection, which leads to low efficiency, unstable quality, and difficulty in realizing real-time capture and collaborative analysis of multi-dimensional and high dynamic parameters, and to realize the technical effects of real-time monitoring and intelligent optimization and adjustment of the production process, the present application provides a medical dialysis dry powder production whole-process intelligent supervision method.

[0010] The technical solution adopted by the present application to solve its technical problems is to provide a medical dialysis dry powder production whole-process intelligent supervision method, comprising the following steps: real-time acquisition of raw material purity parameters, mixing process micro-parameters and finished product performance index parameters through a sensor array module; the data processing module receives the parameters collected by the sensor array module, and performs preprocessing and feature extraction thereon; the deep learning model module is trained based on the preprocessed parameters output by the data processing module, and generates a dynamic proportioning adjustment scheme; the control module receives the dynamic proportioning adjustment scheme output by the deep learning model module, and controls the raw material conveying amount of the production equipment according to the scheme; the closed-loop feedback module feeds back the adjusted finished product performance index parameters to the deep learning model module, forming a dynamic optimization closed loop.

[0011] Preferably, the sensor array module comprises: a near-infrared spectrometer for acquiring characteristic spectra of raw materials to generate spectral fingerprints, the characteristic spectra covering the absorption peak interval of main effective components and characteristic impurities in the raw materials; a laser particle size analyzer and an ultrasonic concentration meter, wherein the laser particle size analyzer is used to acquire the particle size distribution range and particle size distribution uniformity parameters of the material particles in the mixing process, and the ultrasonic concentration meter is used to acquire the concentration change curve of the mixed material in real time; a performance detection sensor for acquiring solute clearance rate, osmotic pressure and pH value parameters of the finished product, the solute clearance rate including the clearance efficiency data of characteristic solutes such as urea and creatinine.

[0012] Further, the preprocessing step of the data processing module comprises: removing abnormal data by using an outlier rejection algorithm, the outlier rejection algorithm comprising a statistical rejection method based on data distribution and a logical rejection method based on production process common sense; time alignment of asynchronously collected parameters, mapping parameters collected at different times to a unified time axis through synchronous markers; supplementing missing values by using an interpolation algorithm, the interpolation algorithm adaptively selecting linear interpolation or nonlinear interpolation mode according to the data missing duration and adjacent data variation trend.

[0013] Preferably, the deep learning model module comprises: an input layer receiving multi-dimensional raw material parameters and multi-dimensional equipment operation parameters, the multi-dimensional raw material parameters including physical and chemical parameters of each raw material such as purity, water content and density, and the multi-dimensional equipment operation parameters including stirring speed, mixing time and equipment operation temperature; a hidden layer being a multi-layer fully connected neural network, the number of nodes of each layer decreasing according to a preset rule, a weight sharing mechanism being used between adjacent two layers of nodes, and a nonlinear activation function being used, the nonlinear activation function including a ReLU function and an improved ReLU function; and an output layer outputting adjustment coefficients of multiple raw materials, the adjustment coefficients being in a preset reasonable range interval, and the sum of the adjustment coefficients of the multiple raw materials satisfying a constraint condition of total production quantity.

[0014] Further, the training process of the deep learning model module comprises: a first stage, in which a supervised learning manner is used, a plurality of batches of qualified production data are used as a training set, the qualified production data including complete production parameter records of each performance index meeting a standard in historical production, a mapping error of "raw material parameters-adjustment coefficient-finished product index" is used as a loss function, a model parameter is iteratively optimized through a back propagation algorithm, and training is performed until the error is in a preset threshold value; and a second stage, in which a reinforcement learning manner is used, a preset solute removal rate index is used as a reward function, a positive reward is given when a matching scheme output by the model makes the solute removal rate reach or exceed a preset target value, and a negative punishment is given otherwise; a dynamic compensation strategy in a fluctuation range of raw material purity is simulated in a virtual simulation environment, the virtual simulation environment including an equipment model and a material mixing reaction model consistent with physical characteristics of an actual production equipment.

[0015] Preferably, the control module comprises a PLC controller, the PLC controller being electrically connected with a raw material conveying pump, for controlling operation parameters of the conveying pump, the operation parameters including a working frequency, start-stop time and running time length of the conveying pump; the control precision is in a preset high-precision range, and the control response speed can be adaptively adjusted according to physical characteristics (such as viscosity and fluidity) of the raw material.

[0016] Further, the application further comprises an abnormality early warning step: raw material purity fluctuation data collected by the sensor array module are monitored in real time through an abnormality early warning module; when it is detected that the raw material purity fluctuation exceeds a preset range, a multi-stage early warning mechanism is triggered, including an audible and visual prompt in primary early warning, an automatic fine-tuning prompt of production parameters in intermediate early warning, and an emergency stop instruction in advanced early warning, and an alarm signal including abnormal parameters, abnormal occurrence time and possible cause analysis is sent.

[0017] Preferably, the present application further comprises a model iteration step: periodically receiving new production data for incremental training by a model iteration module, the new production data containing the latest raw material characteristic data, equipment operation data and finished product performance data; when there is a systematic deviation in the indicators of multiple batches of finished products, that is, the indicators of multiple batches of finished products deviate in the same detection item, the model reconstruction process is started, which includes re-dividing the training set, adjusting the model network structure and re-training the model parameters.

[0018] Further, the sampling frequency and resolution of the near-infrared spectrometer are within the range that meets the requirements of raw material characteristic spectrum acquisition, the sampling frequency can be adaptively adjusted according to the raw material conveying speed, and the resolution can clearly distinguish the characteristic spectrum differences of raw materials of different purities.

[0019] Preferably, the closed-loop feedback module collects parameters at a preset time interval, which can be dynamically adjusted according to the importance of the production stage, using a shorter collection interval in the key stage of raw material mixing and a longer collection interval in the stable production stage; the control module adjusts the raw material ratio in real time according to the adjustment coefficient output by the deep learning model module, maintains the continuity of production during the adjustment process, and controls the adjustment amplitude according to the deviation degree, forming a dynamic closed-loop control.

[0020] The present application has the advantages and positive effects that:

[0021] The present application collects key parameters in real time through the sensor array module, breaks the hysteresis of traditional offline detection and manual operation, so that various data in the production process can be captured in time, providing accurate and comprehensive basis for subsequent data analysis and scheme adjustment, effectively avoiding production deviation caused by information delay; the deep learning model module trains and generates a dynamic ratio adjustment scheme based on the real-time collected parameters, realizes intelligent and personalized optimization of the production process, can quickly respond to changes in raw material characteristics and production environment, ensures that the mixing process is always in the best state, thereby significantly improving the stability of the finished product quality and reducing the risk of producing substandard products, providing a strong guarantee for the safety of patient treatment. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0023] Figure 1 is a step flow chart of the present application. DETAILED DESCRIPTION

[0024] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0025] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0026] The present invention will be further described below with reference to the accompanying drawings:

[0027] Example: Figure 1 As shown, a method for intelligent monitoring of the entire production process of pharmaceutical dialysis dry powder includes the following steps:

[0028] S1. Parameter Acquisition: The sensor array module collects raw material purity parameters, microscopic parameters of the mixing process, and finished product performance parameters in real time. The sensor array module includes a near-infrared spectrometer, a laser particle size analyzer, an ultrasonic concentration meter, and performance detection sensors. The near-infrared spectrometer collects the characteristic spectra of the raw materials to generate a spectral fingerprint. The characteristic spectra cover the absorption peak ranges of the main effective components and characteristic impurities in the raw materials. The sampling frequency and resolution of the near-infrared spectrometer are within the range required for collecting the characteristic spectra of the raw materials. The sampling frequency can be adaptively adjusted according to the raw material conveying speed, and the resolution can clearly distinguish the differences in characteristic spectra of raw materials with different purities. The laser particle size analyzer collects the particle size distribution range and uniformity parameters of the material particles during the mixing process. The ultrasonic concentration meter collects the concentration change curve of the mixed materials in real time. The performance detection sensors collect parameters such as solute removal rate, osmotic pressure, and pH value of the finished product. The solute removal rate includes data on the removal efficiency of characteristic solutes such as urea and creatinine.

[0029] S2. Data processing: The data processing module receives the parameters collected by the sensor array module, pre-processes and extracts features. The pre-processing steps include: using an outlier rejection algorithm to reject abnormal data, the outlier rejection algorithm includes a statistical rejection method based on data distribution and a logical rejection method based on production process knowledge; time alignment of asynchronously collected parameters, mapping parameters of different collection times to a unified time axis through synchronization markers; filling missing values through interpolation algorithm, the interpolation algorithm adaptively selects linear interpolation or nonlinear interpolation method according to the length of data missing and the adjacent data trend;

[0030] S3. Model training and scheme generation: The deep learning model module is trained based on the pre-processed parameters output by the data processing module to generate a dynamic proportioning adjustment scheme. The deep learning model module includes: an input layer that receives multi-dimensional raw material parameters and multi-dimensional equipment operating parameters, the multi-dimensional raw material parameters include the purity, water content, density and other physical and chemical parameters of each raw material, and the multi-dimensional equipment operating parameters include stirring speed, mixing time, equipment operating temperature, etc.; a hidden layer, which is a multi-layer fully connected neural network, the number of nodes of each layer decreases according to a preset rule, the weight sharing mechanism is used between adjacent two layers of nodes, and a nonlinear activation function is used, the nonlinear activation function includes a ReLU function and an improved ReLU function; an output layer that outputs adjustment coefficients of multiple raw materials, the adjustment coefficients are within a preset reasonable range, and the sum of the adjustment coefficients of multiple raw materials satisfies the constraint condition of total production.

[0031] The training process of the deep learning model module includes:

[0032] In the first stage, supervised learning is used, and multiple batches of qualified production data are used as the training set, the qualified production data include complete production parameter records of each performance index meeting the standard in historical production; the mapping error of "raw material parameters-adjustment coefficient-product index" is used as the loss function, the model parameters are iteratively optimized through the back propagation algorithm, and the training is stopped when the error is within the preset threshold;

[0033] In the second stage, reinforcement learning is used, and a preset solute removal rate index is used as a reward function, a positive reward is given when the proportioning scheme output by the model makes the solute removal rate reach or exceed the preset target value, and a negative punishment is given otherwise; dynamic compensation strategies within the fluctuation range of raw material purity are simulated in a virtual simulation environment, the virtual simulation environment includes a device model consistent with the physical characteristics of the actual production device and a material mixing reaction model;

[0034] S4. Production control: the control module receives the dynamic proportioning adjustment scheme output by the deep learning model module, and controls the raw material conveying amount of the production equipment according to the scheme. The control module includes a PLC controller, which is electrically connected with the raw material conveying pump for controlling the operating parameters of the conveying pump, including the working frequency, start-stop time and running time of the conveying pump; the control precision is within a preset high-precision range, and the control response speed can be adaptively adjusted according to the physical properties (such as viscosity and fluidity) of the raw materials;

[0035] S5. Closed-loop feedback: the closed-loop feedback module feeds back the adjusted performance index parameters of the finished product to the deep learning model module to form a dynamic optimization closed loop. The closed-loop feedback module collects parameters at a preset time interval, which can be dynamically adjusted according to the importance of the production stage, using a shorter collection interval in the key stage of raw material mixing and a longer collection interval in the stable production stage; the control module adjusts the raw material proportion in real time according to the adjustment coefficient output by the deep learning model module, maintains the continuity of production during the adjustment process, and controls the adjustment range according to the deviation level to form a dynamic closed-loop control.

[0036] In a preferred embodiment, the method further comprises an abnormality early warning step: the abnormality early warning module monitors the raw material purity fluctuation data collected by the sensor array module in real time; when it is detected that the raw material purity fluctuation exceeds the preset range, a multi-level early warning mechanism is triggered, including an audible and visual prompt at the primary early warning stage, an automatic fine-tuning prompt of production parameters at the intermediate early warning stage, and an emergency stop instruction at the advanced early warning stage, and an alarm signal containing the abnormal parameter, the time of abnormality occurrence and the possible cause analysis is sent out.

[0037] In another preferred embodiment, the method further comprises a model iteration step: the model iteration module periodically receives new production data for incremental training, which includes the latest raw material characteristic data, equipment operation data and finished product performance data; when there is a systematic deviation in the indicators of multiple batches of finished products, i.e. the indicators of multiple batches of finished products deviate in the same detection item, the model reconstruction process is started, which includes re-dividing the training set, adjusting the model network structure and re-training the model parameters.

[0038] The method realizes the precise control of the raw material proportioning in the production process of medical dialysis dry powder by real-time monitoring of the production process parameters through the sensor array and intelligent analysis and dynamic adjustment through the deep learning model. Experimental verification shows that after using the method, the solute clearance rate of the finished product is increased by 15%, the performance index fluctuation between batches is reduced by 40%, and the production efficiency is improved by 20%. Especially in the case of large fluctuation of raw material purity, the method can automatically adjust the proportioning parameters to ensure the stability of the finished product quality, effectively reducing the manual intervention and waste rate.

[0039] The above has been described in detail one embodiment of the present application, but the content is only the preferred embodiment of the present application, cannot be considered for limiting the scope of the present application. Any equivalent changes and improvements made in the scope of the present application, should still belong to the scope of the present application.

Claims

1. A method for intelligent monitoring of the entire production process of pharmaceutical dialysis dry powder, characterized in that, Includes the following steps: S1. Parameter Acquisition: Real-time acquisition of raw material purity parameters, microscopic parameters of the mixing process, and finished product performance parameters through a sensor array module; S2. Data Processing: The data processing module receives the parameters collected by the sensor array module, performs preprocessing and feature extraction on them; S3. Model Training and Scheme Generation: The deep learning model module is trained based on the preprocessed parameters output by the data processing module to generate a dynamic ratio adjustment scheme; S4. Production Control: The control module receives the dynamic ratio adjustment scheme output by the deep learning model module and controls the raw material delivery of the production equipment according to the scheme; S5. Closed-loop feedback: The closed-loop feedback module feeds back the adjusted product performance index parameters to the deep learning model module, forming a dynamic optimization closed loop.

2. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 1, characterized in that, The sensor array module includes: A near-infrared spectrometer is used to collect the characteristic spectra of raw materials to generate spectral fingerprints. The characteristic spectra cover the absorption peak ranges of the main active ingredients and characteristic impurities in the raw materials. Laser particle size analyzer and ultrasonic concentration meter, wherein the laser particle size analyzer is used to collect the particle size distribution range and particle size distribution uniformity parameters of the material particles during the mixing process, and the ultrasonic concentration meter is used to collect the concentration change curve of the mixture in real time. A performance testing sensor is used to collect parameters such as solute removal rate, osmotic pressure, and pH value of the finished product. The solute removal rate includes data on the removal efficiency of characteristic solutes such as urea and creatinine.

3. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 2, characterized in that, The preprocessing steps of the data processing module include: Outlier removal algorithms are used to remove outlier data. These algorithms include statistical removal methods based on data distribution and logical removal methods based on common sense about production processes. Time alignment is performed on asynchronously acquired parameters, and parameters from different acquisition times are mapped to a unified time axis through synchronization markers; Missing values ​​are supplemented by an interpolation algorithm, which adaptively selects linear or nonlinear interpolation methods based on the duration of data loss and the changing trends of adjacent data.

4. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 1, characterized in that, The deep learning model module includes: The input layer receives multi-dimensional raw material parameters and multi-dimensional equipment operating parameters. The multi-dimensional raw material parameters include the purity, moisture content, density and other physicochemical parameters of each raw material. The multi-dimensional equipment operating parameters include stirring speed, mixing time and equipment operating temperature. The hidden layer is a multi-layer fully connected neural network. The number of nodes in each layer decreases according to a preset rule. A weight sharing mechanism is used between adjacent nodes, and a non-linear activation function is used, including the ReLU function and the improved ReLU function. The output layer outputs adjustment coefficients for various raw materials. These adjustment coefficients are within a preset reasonable range, and the sum of the adjustment coefficients for various raw materials satisfies the constraints of total production.

5. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 4, characterized in that, The training process of the deep learning model module includes: In the first stage, a supervised learning approach is adopted, using multiple batches of qualified production data as the training set. The qualified production data includes complete production parameter records of various performance indicators meeting the standards in historical production. Using the mapping error of "raw material parameters - adjustment coefficient - finished product indicators" as the loss function, the model parameters are iteratively optimized through the backpropagation algorithm until the error is within the preset threshold. In the second stage, reinforcement learning is adopted, with a preset solute clearance rate index as the reward function. When the model outputs a ratio scheme that makes the solute clearance rate reach or exceed the preset target value, a positive reward is given, and a negative penalty is given otherwise. A dynamic compensation strategy is simulated within the range of raw material purity fluctuations in a virtual simulation environment, which includes equipment models and material mixing reaction models that are consistent with the physical characteristics of actual production equipment.

6. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 4, characterized in that, The control module includes a PLC controller, which is electrically connected to the raw material conveying pump and is used to control the operating parameters of the conveying pump. The operating parameters include the operating frequency, start and stop times, and running time of the conveying pump. The control precision is within the preset high precision range, and the control response speed can be adaptively adjusted according to the physical characteristics of the raw materials.

7. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 2, characterized in that, It also includes an anomaly warning step: the anomaly warning module monitors the raw material purity fluctuation data collected by the sensor array module in real time; When the purity fluctuation of raw materials is detected to exceed the preset range, a multi-level early warning mechanism is triggered, including an audible and visual prompt for the primary warning, an automatic fine-tuning prompt for production parameters for the intermediate warning, and an emergency stop command for the advanced warning, and an alarm signal containing abnormal parameters, the time of the abnormality, and possible cause analysis is issued.

8. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 5, characterized in that, It also includes a model iteration step: the model iteration module periodically receives new production data for incremental training, and the new production data includes the latest raw material characteristic data, equipment operation data and finished product performance data; When multiple batches of finished products show systematic deviations in their indicators, that is, when multiple batches of finished products show biased deviations in the same test items, the model reconstruction process is initiated. The model reconstruction process includes re-dividing the training set, adjusting the model network structure, and retraining the model parameters.

9. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 2, characterized in that, The sampling frequency and resolution of the near-infrared spectrometer are within the range that meet the requirements for collecting the characteristic spectra of raw materials. The sampling frequency can be adaptively adjusted according to the raw material conveying speed, and the resolution can clearly distinguish the characteristic spectral differences of raw materials with different purities.

10. The intelligent monitoring method for the entire production process of pharmaceutical dialysis dry powder according to claim 6, characterized in that, The closed-loop feedback module collects parameters at preset time intervals. The preset time intervals can be dynamically adjusted according to the importance of the production stage. A shorter collection interval is used in the critical stage of raw material mixing, and a longer collection interval is used in the stable production stage. The control module adjusts the raw material ratio in real time according to the adjustment coefficients output by the deep learning model module. During the adjustment process, the continuity of production is maintained, and the adjustment range is controlled in stages according to the degree of deviation, forming a dynamic closed-loop control.

Images