Intelligent adjusting type medical cotton ball forming method and system

Abstract

The invention relates to the technical field of intelligent forming, in particular to an intelligent adjusting type medical cotton ball forming method and system.The intelligent adjusting type medical cotton ball forming method comprises the following steps that pressure data are collected through a distributed pressure sensor array, pressure feature vectors are generated through Gaussian filtering denoising, a pressure change trend is analyzed based on the feature vectors, and a gradient adjusting instruction is output; and controlling the valve to adjust air pressure to obtain compensation parameters, driving the servo motor to adjust a mold gap, collecting offset data to generate a feedback signal, and inputting a neural network to evaluate a forming result in combination with feedback and adjustment instructions. According to the method, a distributed sensor array collects data, multi-dimensional denoising improves feature extraction, a dynamic function analyzes a pressure trend, an offset adjustment parameter is identified, closed-loop compensation is realized through cooperative control, a displacement sensor feeds back gap change, parameter modeling optimizes precision consistency, a convolutional network extracts time sequence features, and a self-adaptive system is constructed through fusion analysis. And intelligent prediction and optimization of the forming quality are realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent molding technology, and in particular to an intelligent adjustable medical cotton ball molding method and system. Background Art

[0002] The field of intelligent molding technology includes the integrated application of medical material processing equipment and automated control systems. Its core lies in achieving precise control of material morphology through real-time feedback of sensor data and dynamic adjustment of actuators. In the production of medical dressings, it mainly involves three key technologies: cotton fiber directional arrangement control, laminated structure molding precision adjustment, and porous structure density gradient control. The technological development in this field shows an evolutionary path from mechanical mold shaping to intelligent parameter adaptive adjustment. At this stage, the technical system covers three basic components: pressure gradient control module, temperature and humidity coordinated control module, and visual detection feedback module.

[0003] Among them, the intelligent adjustable medical cotton ball forming method refers to optimizing and controlling the forming process of medical cotton balls through intelligent adjustment means. This topic addresses the shortcomings of existing cotton ball forming methods and proposes a method for accurately controlling the size, density and shape of cotton balls through intelligent adjustment equipment. Specifically, this method uses intelligent sensors and control systems to monitor the forming state of cotton balls in real time, and automatically adjusts key parameters such as forming pressure and temperature to ensure that the quality of each cotton ball meets the requirements. This method also combines advanced automated forming equipment to achieve standardized production of cotton balls through precisely controlled robotic arms and forming molds.

[0004] Existing technologies rely on manual experience to set fixed molding parameters and lack a real-time monitoring and feedback mechanism for dynamic changes in pressure distribution. This results in the inability to promptly correct quality deviations caused by fluctuations in material properties or environmental interference during the molding process. Traditional pressure sensors use a single-point detection mode, which makes it difficult to fully reflect the spatial distribution characteristics of the pressure field inside the mold cavity, easily causing local pressure overload or uneven molding density. Mechanical adjustment systems often use open-loop control strategies, and there is a lack of coordination between valve opening and mold gap adjustment. The coupling of air pressure fluctuations and mechanical deformation can easily cause molding dimensional deviations. The quality assessment process relies on offline spot checks and manual visual inspections, which cannot capture transient abnormal signals during the molding process, resulting in an increased risk of defective products flowing into subsequent links. For example, if the micron-level offset of the mold gap is not detected in time during the molding of medical cotton balls, the cotton balls will not be able to meet the fluffiness standards, affecting the medical hemostatic effect. Existing technologies, due to the lack of a high-precision closed-loop feedback system, are difficult to avoid such problems. Summary of the Invention

[0005] In order to solve the problem that the existing technology relies on manual experience to set fixed molding parameters, lacks a real-time monitoring and feedback mechanism for dynamic changes in pressure distribution, resulting in the inability to correct quality deviations caused by fluctuations in material properties or environmental interference during the molding process in a timely manner, traditional pressure sensors use a single-point detection mode, which is difficult to fully reflect the spatial distribution characteristics of the pressure field inside the mold cavity, and can easily cause local pressure overload or uneven molding density. Mechanical adjustment systems mostly use open-loop control strategies, and the valve opening and mold gap adjustment lack synergy. When air pressure fluctuations and mechanical deformation are coupled, it is easy to cause molding size deviations. The quality assessment link relies on offline sampling and manual visual inspection, and cannot capture transient abnormal signals in the molding process, resulting in an increased risk of defective products flowing into subsequent links. For example, if the micron-level offset of the mold gap is not detected in time during the molding of medical cotton balls, the fluffiness of the cotton balls will not meet the standards, affecting the medical hemostasis effect. Due to the technical problem of the lack of a high-precision closed-loop feedback system in the existing technology, the embodiment of the present invention provides an intelligent adjustable medical cotton ball molding method and system. The technical solution is as follows:

[0006] In one aspect, a method for forming an intelligently adjustable medical cotton ball is provided, the method comprising:

[0007] S1: Regional pressure data is collected through a distributed pressure sensor array. The collected raw pressure data is input into a Gaussian filtering algorithm that dynamically adjusts kernel parameters based on the noise level, and multi-dimensional spatial denoising is performed to generate a regional pressure feature vector.

[0008] S2: Based on the regional pressure characteristic vector, a dynamic gradient determination function is called to perform directional analysis on the pressure change trend in the region, and when the trend change satisfies the continuous directional offset condition, a dynamic gradient adjustment instruction is output;

[0009] S3: According to the dynamic gradient adjustment instruction, the valve opening of the compressed gas channel in the target area is controlled to adjust the air pressure in the mold cavity, and the current valve opening parameter is modified by adjusting the amplitude to obtain the molding pressure compensation parameter;

[0010] S4: Based on the molding pressure compensation parameter, the servo motor is driven to adjust the mold gap. The adjusted gap data is collected in real time by the displacement sensor. The offset is calculated with the reference gap value. The offset direction and change rate are combined to obtain a molding state feedback signal.

[0011] S5: Call the molding state feedback signal and dynamic gradient adjustment instruction, input them into the trained convolutional neural network model, perform time series feature extraction and analysis, and output the molding evaluation result.

[0012] As a further solution of the present invention, the dynamic gradient determination function selects a time window based on the noise level, and the continuous direction shift condition is that the gradient direction continues to change within at least three consecutive time steps, and the shift angle exceeds a set threshold. The set threshold is optimized using the gradient descent method on historical data samples;

[0013] The offset direction and the rate of change are subjected to Z-score normalization to unify the dimensions;

[0014] The convolutional neural network model includes two convolution extraction layers, an LSTM time series processing layer and two fully connected layers, and is trained based on batch raw molding data samples of the experimental device;

[0015] The regional pressure characteristic vector includes the pressure distribution matrix, the contact area coordinates, and the effective pressure value. The dynamic gradient adjustment instruction specifically includes the pressure gradient direction angle, the displacement accumulation, and the critical rate threshold. The molding pressure compensation parameters include the linear displacement of the valve opening, the cavity pressure gradient coefficient, and the pulse width modulation parameters. The molding state feedback signal specifically refers to the gap displacement deviation, the real-time displacement rate, and the deformation convergence rate. The molding result evaluation includes the porosity distribution diagram, the density standard deviation, and the stress concentration factor distribution.

[0016] As a further solution of the present invention, the specific steps of S1 include:

[0017] S101: Collect regional pressure data through a distributed pressure sensor array, collect pressure at multiple points in the region according to a preset fixed period, record the pressure values ​​and timestamps of the sensor nodes, and aggregate the sampled values ​​to form a unified data set to obtain an original pressure data set;

[0018] S102: Based on the original pressure data set, calling the Gaussian filter algorithm to perform spatial domain denoising, identifying the Gaussian weight distribution of each data point and its neighboring points, calculating the pressure response value and adjusting the node pressure value to obtain denoised pressure distribution data;

[0019] The kernel size and standard deviation of the Gaussian filter algorithm are configured according to the sensor noise level and the target resolution;

[0020] S103: Based on the denoised pressure distribution data, extract the spatial distribution characteristics of the nodes, analyze the high-dimensional feature space under the positional relationship between the nodes, calculate the coordinate values ​​of the nodes in the feature space and unify them into a standard vector form, and integrate the node output regional pressure feature vectors.

[0021] As a further solution of the present invention, the specific steps of S2 include:

[0022] S201: Based on the regional pressure feature vector, extract the pressure value at the time node, construct a pressure change matrix, call the dynamic gradient judgment function to calculate the directional offset of adjacent units, and generate a directional change index sequence;

[0023] S202: extracting continuous direction change segments based on the direction change indicator sequence, screening segments that meet angle change and length thresholds, calculating segment direction switching frequency and angle offset, and obtaining a direction offset trend parameter set;

[0024] S203: Calling the paragraph change parameter in the direction shift trend parameter set, performing alignment comparison with the corresponding time period direction code in the regional pressure feature vector, recording the paragraph start and end time and regional number that meet the continuous direction shift condition, and obtaining a dynamic gradient adjustment instruction.

[0025] As a further solution of the present invention, the specific steps of S3 include:

[0026] S301: Based on the dynamic gradient adjustment instruction, the current opening parameter of the compressed gas channel in the target area is called, and the opening parameter is adjusted within a set range according to the adjustment amplitude to generate a valve adjustment opening value;

[0027] S302: adjusting the valve opening value to control the real-time opening and closing state of the compressed gas channel valve in the mold cavity, collecting the gas channel flow rate data after opening and closing, and calculating the interval difference between the gas channel flow rate data and the initial value to generate a flow rate change value;

[0028] S303: Based on the flow rate change value, normalization processing is performed in combination with the mold cavity volume data, and the unit volume flow rate change rate and the amplification amplitude corresponding to the set pressure adjustment coefficient are calculated to obtain the molding pressure compensation parameter.

[0029] As a further solution of the present invention, the flow rate change value is calculated using the formula:

[0030]

[0031] Among them, ΔQ represents the change in flow rate, the unit is L / min, Q t1 Represents the flow rate data value at the current time t1, in L / min, Q t0 represents the initial flow rate data value at time t0, in L / min, V represents the mold cavity volume data, in L, λ j represents the real-time flow rate deviation of the compressed gas channel in the jth cycle, in L / min, γ j represents the instantaneous opening and closing amplitude of the valve in the jth cycle, in percentage, k represents the total number of sampling cycles, and η represents the gas compression response sensitivity coefficient, which is a dimensionless parameter.

[0032] As a further solution of the present invention, the specific steps of S4 include:

[0033] S401: Based on the molding pressure compensation parameter, drive the servo motor to adjust the mold gap, extract the difference between the control displacement and the target instruction, perform a synchronous correction operation and record the response rate and feedback delay to generate a mold gap adjustment value;

[0034] S402: calling the mold gap adjustment value, collecting mold gap change data through a displacement sensor, calculating the offset between the mold gap and the reference gap, and performing time sequence organization based on the data collection time points to obtain the mold gap offset change;

[0035] S403: Extract the gap change direction and change rate based on the mold gap offset change, build a trend offset map based on the time series data, perform map key trend identification and state determination operations, and generate a molding state feedback signal.

[0036] As a further solution of the present invention, the specific steps of S5 include:

[0037] S501: Based on the forming state feedback signal and the dynamic gradient adjustment instruction, they are synchronously input into the trained convolutional neural network model, the feedback signal is spatially expanded through the convolution layer, the grayscale difference of adjacent pixels is extracted, mean filtering is performed on a regional basis, the trend of regional pixel difference values ​​is calculated, and the signal gradient change rate is generated;

[0038] S502: Calling the signal gradient change rate, reconstructing the gradient change sequence in the time axis direction according to the time sequence recognition module in the convolutional neural network model, superimposing the node weight ratio in the dynamic gradient adjustment instruction, calculating the node lateral offset amplitude, and generating the node gradient offset;

[0039] S503: Call the node gradient offset, compare the node offset result with the gradient tolerance limit value according to the evaluation function parameter of the model output end, determine whether the offset amplitude deviates, extract the segment deviation ratio to perform overall trend fitting, and obtain the forming evaluation result.

[0040] As a further solution of the present invention, the gradient change rate of the forming signal is calculated using the formula:

[0041]

[0042] Among them, ΔG represents the rate of change of signal gradient, P i Represents the grayscale value of the i-th pixel, X i Represents the horizontal coordinate value of the i-th pixel, Y i Represents the vertical coordinate value of the i-th pixel, w iRepresents the weight factor of the i-th pixel, the adaptive weight factor w i The feature parameters learned by the model are dynamically set and regulated based on the importance of different pixels. n represents the number of pixels in the image area, and i is the pixel index.

[0043] On the other hand, an intelligent adjustable medical cotton ball forming system is provided, wherein the intelligent adjustable medical cotton ball forming system is used to perform the above-mentioned intelligent adjustable medical cotton ball forming method, and the system comprises:

[0044] The pressure acquisition module is used to obtain real-time pressure data of the target area of ​​the mold cavity through a distributed pressure sensor array, and input the raw pressure data into the Gaussian filter algorithm for spatial dimension denoising, generate regional pressure feature vectors, and pass them to the pressure trend determination module;

[0045] a pressure trend determination module, configured to receive the regional pressure characteristic vector, and perform directional judgment on the pressure change trend in the region by calling a dynamic gradient judgment function, to determine whether the characteristic vector satisfies a continuous directional offset condition, and output a dynamic gradient adjustment instruction when the condition is satisfied, and transmit the instruction to the pressure regulation module;

[0046] A pressure regulating module is configured to receive the dynamic gradient regulating instruction, perform amplitude regulation on the compressed gas control valve, calculate a molding pressure compensation parameter, and transmit the calculated parameter to the gap adjusting module;

[0047] a gap adjustment module, configured to receive the molding pressure compensation parameters and drive a servo motor to adjust the mold gap based on the parameters, collect the adjusted gap data through a displacement sensor, perform an offset operation with the reference gap value, generate a molding state feedback signal, and transmit it to the molding evaluation module;

[0048] The molding evaluation module is used to receive the molding state feedback signal and the dynamic gradient adjustment instruction, and input them into the convolutional neural network model for time series feature extraction and analysis, and output the molding evaluation result.

[0049] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0050] A distributed pressure sensor array collects data at a fixed interval, combined with multidimensional spatial denoising, to improve the accuracy of pressure feature extraction and eliminate environmental noise interference on the molding process. A dynamic gradient decision function directionally analyzes pressure variation trends, accurately identifying continuous offset conditions and enabling real-time dynamic adjustment of molding parameters, avoiding the hysteresis error caused by traditional static threshold control. Coordinated control of the compressed gas channel valve opening and servo motors creates a closed-loop pressure compensation mechanism, ensuring synchronization between mold cavity pressure and gap adjustment, enhancing molding process stability. Displacement sensors collect gap data in real time and calculate offsets against reference values. This information is then combined with the rate of change to generate feedback signals, building a multidimensional parameter correlation model to optimize molding accuracy and consistency. A convolutional neural network deeply extracts temporal features, integrating dynamic adjustment commands with feedback signals for analysis. This system constructs an adaptive evaluation system, overcoming the limitations of traditional experience-based process optimization and enabling intelligent prediction and closed-loop optimization of molding quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0052] Figure 2 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0054] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0055] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0056] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0057] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0058] See also Figure 1 The embodiment of the present invention provides an intelligent adjustable medical cotton ball forming method, the processing flow of the method may include the following steps:

[0059] S1: Regional pressure data is collected through a distributed pressure sensor array. The collected raw pressure data is input into a Gaussian filtering algorithm that dynamically adjusts kernel parameters based on the noise level, and multi-dimensional spatial denoising is performed to generate a regional pressure feature vector.

[0060] S2: Based on the regional pressure feature vector, the dynamic gradient judgment function is called to perform directional analysis on the pressure change trend in the region. When the trend change meets the continuous direction offset condition, the dynamic gradient adjustment instruction is output;

[0061] S3: According to the dynamic gradient adjustment instruction, the valve opening of the compressed gas channel in the target area is controlled to adjust the air pressure in the mold cavity. The current valve opening parameter is modified by adjusting the amplitude to obtain the molding pressure compensation parameter;

[0062] S4: Based on the molding pressure compensation parameters, the servo motor is driven to adjust the mold gap. The adjusted gap data is collected in real time through the displacement sensor, and the offset is calculated with the reference gap value. The molding state feedback signal is obtained by combining the offset direction and change rate.

[0063] S5: Call the molding state feedback signal and dynamic gradient adjustment instruction, input them into the trained convolutional neural network model, perform time series feature extraction and analysis, and output the molding evaluation results;

[0064] The regional pressure characteristic vector includes the pressure distribution matrix, contact area coordinates, and effective pressure value. The dynamic gradient adjustment instructions are specifically the pressure gradient direction angle, displacement accumulation, and critical rate threshold. The molding pressure compensation parameters include the linear displacement of the valve opening, the cavity pressure gradient coefficient, and the pulse width modulation parameters. The molding state feedback signal specifically refers to the gap displacement deviation, real-time displacement rate, and deformation convergence rate. The molding result evaluation includes the porosity distribution map, density standard deviation, and stress concentration factor distribution.

[0065] Specifically, the steps of S1 are:

[0066] S101: Collect regional pressure data through a distributed pressure sensor array, collect pressure at multiple points in the region according to a preset fixed period, record the pressure values ​​and timestamps of the sensor nodes, and aggregate the sampled values ​​to form a unified data set to obtain an original pressure data set;

[0067] First, the sensor array layout must be rational. In medical cotton ball molding equipment, the sensor array can be evenly distributed across the inner surface of the mold, ensuring comprehensive monitoring of pressure changes across the entire pressing area. For example, if the spacing between sensor nodes is set to 1 cm, each sensor will be evenly distributed, ensuring timely acquisition of pressure changes in every area. The acquisition cycle setting needs to be optimized based on the equipment's operating speed and the dynamic characteristics of the molding process. Setting the acquisition cycle to 100 ms means collecting data 10 times per second. Assuming that in a real-world application, sensor node A records a pressure value of 15.2 kPa with a timestamp of 1050 ms, each sensor will record its pressure value and the corresponding timestamp. For example, if the pressure value of node A1 at 1050 ms is 15.2 kPa, the corresponding data item is (A1, 1050, 15.2). The collected data from all sensor nodes is transmitted to the central processing unit via serial communication and aggregated into a three-tuple data set consisting of the node number, timestamp, and pressure value. This data set is updated once per acquisition cycle.

[0068] S102: Based on the original pressure data set, a Gaussian filter algorithm is called to perform spatial domain denoising, identify the Gaussian weight distribution of each data point and its neighboring points, calculate the pressure response value and adjust the node pressure value to obtain denoised pressure distribution data;

[0069] The kernel size and standard deviation of the Gaussian filter algorithm are configured according to the sensor noise level and target resolution;

[0070] In practice, we first need to determine the kernel size and standard deviation σ of the Gaussian filter. We choose a kernel size of 3×3 and set σ to 1.0 to accommodate the spatial resolution of the sensor array. For each sensor node, we extract data from its 3×3 neighborhood and calculate the corresponding Gaussian weight matrix. Assume the weight of the central node is 0.1336, and the weights of neighboring nodes decrease with distance from the central node. For example, the pressure values ​​for the neighborhood around central node B2 are shown in Table 1:

[0071] Table 1: Gaussian filter neighborhood pressure values

[0072]

[0073]

[0074] As shown in Table 1, although the differences between the pressure values ​​are not large, through sophisticated weighted processing, abnormal noise or local mutations can be buffered, thereby improving the stability and representativeness of the overall pressure data.

[0075] According to the calculation rules of Gaussian filtering, the pressure value of the central node B2 is obtained by multiplying the neighborhood data with the corresponding Gaussian weight and then summing them up, which can be expressed as follows:

[0076]

[0077] Among them, P B2 is the pressure value of the central node, w ij is the Gaussian weight of the node in row i and column j, P ij is the pressure value of the node in the i-th row and j-th column.

[0078] For example, the weight values ​​around node B2 are:

[0079] w 11 =0.1336, w 12 =0.1065, w 13 =0.0755, w 21 =0.1065, w 22 =

[0080] 0.1592, w 23 =0.1065, w 31 =0.0755, w 32 =0.1065, w 33 =0.1336;

[0081] After substituting the pressure value, the denoised pressure value of node B2 is calculated to be 15.2 kPa. This value replaces the original pressure value, completing the denoising process.

[0082] S103: Based on the denoised pressure distribution data, extract the spatial distribution characteristics of the nodes, analyze the high-dimensional feature space under the positional relationship between the nodes, calculate the coordinate values ​​of the nodes in the feature space and unify them into a standard vector form, and integrate the nodes to output the regional pressure feature vector;

[0083] The spatial positions of the sensor nodes are mapped into a two-dimensional coordinate system. For example, node A1 is set to (0, 0), node A2 to (1, 0), and so on, forming the spatial coordinates of the nodes. For each node, a vector containing its coordinates and the denoised pressure value is constructed. For example, the vector for node A1 can be represented as (0, 0, 14.8). Next, the vectors of all nodes are combined into a matrix, and principal component analysis (PCA) is used for dimensionality reduction to extract the main eigenvalues. Assuming that the PCA analysis finds that the first two principal components can explain 95% of the data variance, these two principal components are selected as the basis for the eigenvalues. The original three-dimensional vector is projected onto these two principal components to obtain a standardized eigenvalue. For example, the eigenvalue of node A1 is (0.85, -0.12). Finally, the eigenvalues ​​of all nodes are combined into a regional pressure eigenvalue for subsequent analysis and processing.

[0084] Specifically, the steps of S2 are:

[0085] S201: Based on the regional pressure feature vector, extract the pressure value at the time node, construct a pressure change matrix, call the dynamic gradient judgment function to calculate the directional offset of adjacent units, and generate a directional change indicator sequence;

[0086] First, by acquiring pressure feature vector data within the region, the pressure values ​​at the corresponding time nodes are extracted, based on the time nodes. Pressure values ​​are obtained through a real-time monitoring and data acquisition system. Pressure data is acquired through regular measurements of pressure points installed by sensors in different regions. Assume that at a certain moment, the pressure values ​​within the region are P1 = 10 Pa, P2 = 15 Pa, and P3 = 18 Pa, respectively. This forms a pressure matrix. For example, for data at consecutive time nodes t1, t2, t3, …, the pressure values ​​corresponding to each node are recorded and formed into a vector. The pressure matrix is ​​then {[P1(t1), P2(t1), P3(t1)], [P1(t2), P2(t2), P3(t2)], …}, representing the pressure distribution at multiple points within the region at different time nodes.

[0087] Next, the dynamic gradient in the area is calculated, and the pressure change is calculated using the gradient judgment function. Assuming that the pressure change in adjacent areas is from P1 (t1) to P2 (t2), its directional offset can be determined by comparing the pressure changes of adjacent units. For example, if the pressure change is greater than a preset threshold, it is considered that a large offset has occurred in that direction. The dynamic gradient judgment function calculates the gradient value based on the change between specific pressure values ​​and generates a sequence of directional change indicators. The key to this process is to dynamically adjust the pressure gradient of the area to ensure accurate changes in the pressure value. Assuming that the gradient change rate ΔP = P2-P1 = 15Pa-10Pa = 5Pa at this time, the directional change of the offset will be calculated using this difference.

[0088] S202: Extracting continuous direction-changing segments based on the direction-changing indicator sequence, screening segments that meet the angle change and length thresholds, calculating the segment direction switching frequency and angle offset, and obtaining a direction offset trend parameter set;

[0089] By extracting a sequence of directional change indicators, we identify segments of continuous directional change. The key to this process is defining what constitutes a "continuous directional change." Generally speaking, a segment is considered a valid directional change if the angle change exceeds a preset threshold (e.g., an angle change greater than 15°) and lasts for a certain length of time (e.g., more than 10 seconds).

[0090] Next, we filter out paragraphs that meet the angle change and length threshold requirements. For example, if a paragraph changes direction from θ1 = 10° to θ2 = 30° within the time interval [t1, t2], and its angle change is 20°, then the angle change of the paragraph meets the preset threshold and can be selected.

[0091] Then, the direction switching frequency and angle offset of each paragraph are calculated. Assuming that there are multiple small direction switching segments within a paragraph, and the angle offset of each switching is Δθ1 = 5° and Δθ2 = 3°, the formula for calculating the direction switching frequency is:

[0092]

[0093] Where Δt is the time interval between each switch, and T is the total duration of the segment. Assuming the total duration of the segment is 30 seconds and the number of switches is 3, the switching frequency is:

[0094]

[0095] This result can reflect the frequency of direction changes within the segment. The direction offset is calculated by accumulating the angle offset of each switch to obtain the total angle change.

[0096] S203: calling the segment change parameter in the direction shift trend parameter set, performing a positional comparison with the corresponding time period direction code in the regional pressure feature vector, recording the start and end times of the segments and the regional numbers that meet the continuous direction shift conditions, and obtaining a dynamic gradient adjustment instruction;

[0097] By calling the segment change parameters in the directional offset trend parameter set and combining them with the directional code in the regional pressure feature vector, a positional comparison is performed. The key to this process is to accurately align the directional offset parameters with the pressure changes in the actual time period, ensuring that the pressure changes in different time periods accurately match the directional changes. Assuming that the directional code for a certain area is from time t1 to t3, the direction code within this period is [θ1, θ2, θ3], and the corresponding pressure data is [P1, P2, P3]. By comparing the directional changes with the pressure data, it is possible to determine whether the segment meets the continuous directional offset condition.

[0098] If the conditions are met, the start and end times of the segment, as well as the region number, are recorded to generate dynamic gradient adjustment instructions. This allows precise control of pressure changes within the region and generates corresponding adjustment instructions. For example, if a region experiences continuous directional shifts from time periods t2 to t4, and the pressure changes meet the requirements, adjustment instructions will be generated based on this information to optimize the region's pressure state.

[0099] Specifically, the steps of S3 are:

[0100] S301: Based on the dynamic gradient adjustment instruction, the current opening parameter of the compressed gas channel in the target area is called, and the opening parameter is adjusted within a set range according to the adjustment amplitude to generate a valve adjustment opening value;

[0101] Based on the dynamic gradient adjustment instruction, first identify the specific position of the compressed gas channel in the target area referenced in the adjustment instruction, locate its control valve and read the current opening parameter. Assume that the opening is 30% when the mold number is M01, and the adjustment amplitude is set to 10%. Confirm whether the current opening is within the allowable range from the set adjustment range of 20%-70%. If it is judged that 30% is within the lower range of the set range, the opening parameter is adjusted upward. During the adjustment process, the current opening value A is set to 0. d The numerical superposition is performed with the adjustment amplitude ΔA, and the calculation formula is:

[0102] A x =A d +ΔA;

[0103] Among them A d =30%, ΔA=10%, then we can calculate:

[0104] A d=30%+10%=40%;

[0105] If this value falls within the set range of 20%-70%, it is determined that the adjustable condition is met, and the adjustment value is recorded and applied as the current new valve opening. For mold number M02, its current opening is 45%, the adjustment amplitude is -5%, and the set range is 30%-60%, then:

[0106] A x =45%-5%=40%,

[0107] The set interval constraint is also satisfied. If there is an instance, such as M03, the current opening is 60% and the adjustment amplitude is 15%, then the direct addition is:

[0108] A x =60%+15%=75%,

[0109] Although this value is greater than the initial value, it has not exceeded the maximum threshold of 80% and is still a valid adjustment value. If it exceeds the maximum value, it must be limited to the upper limit of the interval. During the execution process, the current opening data is obtained by reading the valve current status register, and arithmetic operations are performed through the adjustment amplitude field in the instruction to limit the result to the set interval. If it exceeds, it is trimmed, that is: if the new opening is greater than the maximum value, the maximum value is taken; if it is less than the minimum value, the minimum value is taken; the final output of this adjustment operation is the adjusted opening value of the valve, which is used to drive the actuator to perform the next mechanical response operation.

[0110] Table 2: Valve adjustment data table

[0111]

[0112]

[0113] As shown in Table 2, by comparing the given adjustment amplitude with the set range, the target opening value corresponding to each execution of the dynamic gradient adjustment instruction can be clearly obtained, ensuring that it is within the control range and providing basic parameters for subsequent valve opening and closing control and flow analysis.

[0114] S302: Controlling the real-time opening and closing state of the compressed gas channel valve in the mold cavity according to the valve opening value adjustment, collecting the gas channel flow rate data after opening and closing, and calculating the interval difference between the gas channel flow rate data and the initial value to generate a flow rate change value;

[0115] After receiving and executing the new adjusted opening value of the valve, the real-time control mechanism of the compressed gas channel in the mold cavity is immediately started. The control mechanism first reads the control ID corresponding to the channel in the control bus, and synchronously transmits the valve drive signal to the physical execution end through the control logic unit. The driver sets its rotation angle according to the input opening value, thereby physically changing the effective cross-sectional area of ​​the compressed gas channel. Assuming that the current mold is M01 and the adjusted opening value is 40%, the valve core is controlled at the corresponding opening position by the actuator, and the feedback elements such as the Hall sensor are used to confirm in real time whether the current physical position of the valve core reaches the target set opening. If the target is not reached, the drive is repeated and monitored until it is achieved; after completing the physical position adjustment of the valve, the gas flow rate data of the compressed gas channel is collected in real time. The sensor is arranged in the middle of the channel, and the actual flow volume of the gas per minute is obtained by differential pressure measurement and thermal flow detection. Assuming that the initial flow rate Q t0 =100L / min, the flow rate Q measured at the current moment t1 =110L / min, mold cavity volume V = 2.0L, set the total number of sampling cycles k = 3, the compressed gas flow rate deviation within the cycle λ j are 0.3, 0.2, and 0.1 respectively, and the valve opening and closing amplitude γ j The values ​​are 10%, 8%, and 12% respectively, and the compression response sensitivity coefficient η = 1.5. The flow rate change value is calculated using the formula:

[0116]

[0117] Among them, ΔQ represents the change in flow rate, the unit is L / min, Q t1 Represents the flow rate data value at the current time t1, in L / min, Q t0 represents the initial flow rate data value at time t0, in L / min, V represents the mold cavity volume data, in L, λ j represents the real-time flow rate deviation of the compressed gas channel in the jth cycle, in L / min, γ j represents the instantaneous opening and closing amplitude of the valve in the jth cycle, in percentage, k represents the total number of sampling cycles, and η represents the gas compression response sensitivity coefficient, which is a dimensionless parameter.

[0118] First calculate the deviation terms and the average value:

[0119]

[0120] Then calculate the main difference term:

[0121]

[0122] ΔQ=|7.07+2.90|=9.97L / min;

[0123] This value represents the change in flow rate per unit condition in the compressed gas channel caused by valve opening adjustment. This change serves as a key input for subsequent pressure regulation and compensation. By controlling the valve opening and sensing the resulting flow rate change in real time, combined with data from the valve's dynamic behavior over multiple cycles, it is possible to accurately reproduce the instantaneous and average flow rate changes caused by the opening adjustment, thus providing quantitative input for the mold gas pressure system.

[0124] S303: Based on the flow rate change value, normalization is performed in combination with the mold cavity volume data to calculate the unit volume flow rate change rate and the amplification value corresponding to the set pressure adjustment coefficient to obtain the molding pressure compensation parameter;

[0125] The calculated flow change value ΔQ = 9.97 L / min is normalized with the volume parameter of the mold cavity to quantify the degree of flow change per unit volume. Here, the mold volume is V = 2.0 L, and the unit volume flow change rate is: ΔQ / V = 9.97 / 2.0 = 4.985 L / min / L. This result indicates that the gas flow rate per liter of mold space changes by about 4.985 liters per minute. Subsequently, the set pressure adjustment coefficient K = 1.2 is introduced, which represents the pressure amplification response ratio of the gas in the mold to the flow change. The molding pressure compensation value P b The calculation is as follows: b =4.985×1.2=5.982L / min / L. This value is the additional pressure reference value applied during the mold molding process according to the current valve adjustment. It is used to adjust the output set value of the pressure control system to compensate for the mold cavity pressure fluctuation caused by flow rate changes. By comparing this value with the established target pressure change range (such as 4.5~6.5L / min / L), it can be seen that the result is within the target range, indicating that the pressure change caused by the current opening adjustment can be effectively accepted and can maintain a stable pressure state of the mold cavity within the expected control range, thereby providing a reliable guarantee for subsequent molding operations.

[0126] Specifically, the steps of S4 are:

[0127] S401: Based on the molding pressure compensation parameter, the servo motor is driven to adjust the mold gap, the difference between the control displacement and the target instruction is extracted, a synchronous correction operation is performed, and the response rate and feedback delay are recorded to generate a mold gap adjustment value;

[0128] Based on the forming pressure compensation parameter P b =5.982L / min / L, the mold gap adjustment mechanism is driven by the servo motor control module, and the target displacement instruction D of the servo motor is first read. m=0.5mm, and at the same time, the encoder is used to collect the actual displacement value D of the current mold gap in real time s =0.45mm, calculate the displacement difference ΔD=|0.5-0.45|=0.05mm. If the difference exceeds the preset threshold of 0.03mm, the correction operation is triggered. The correction amount is ΔD×H=0.05×1.2=0.06mm (where the correction coefficient H=1.2 is set according to the response characteristics of the servo motor). Record the correction instruction issuance time t1=10:00:00.000 and the actuator feedback completion time t2=10:00:00.150, calculate the response delay Δt=150ms, and the response rate is v=ΔD / Δt=0.06mm / 0.15s=0.4mm / s. Finally, the mold gap adjustment value D is generated. t =0.45+0.06=0.51mm, this value is used as the control input for the next stage.

[0129] Table 3: Mold gap adjustment parameters

[0130] Parameter name Numerical unit Target displacement 0.50 mm Actual displacement 0.45 mm Correction factor 1.2 none Response delay 150 ms

[0131] As shown in Table 3, by calculating the displacement difference and applying the correction coefficient, the dynamic adjustment process of the servo motor can be accurately quantified, and a data basis can be provided for subsequent gap offset analysis.

[0132] S402: Calling the mold gap adjustment value, collecting mold gap change data through the displacement sensor, calculating the offset between the mold gap and the reference gap, and performing time sequence organization based on the data collection time points to obtain the mold gap offset change;

[0133] Call mold gap adjustment value D t =0.51mm, the gap data of 5 time points are continuously collected by the laser displacement sensor at a sampling frequency of 1000Hz:

[0134] [0.50, 0.52, 0.49, 0.53, 0.51]mm;

[0135] Reference gap D c = 0.50mm, calculate the time point offset Δd i =|d i -0.50|, we get [0.00, 0.02, 0.01, 0.03, 0.01] mm;

[0136] Organize the data in the time series t = [0, 0.1, 0.2, 0.3, 0.4]s, generate the time offset sequence {(0, 0.00), (0.1, 0.02), (0.2, 0.01), (0.3, 0.03), (0.4, 0.01)}, and calculate the average offset rate ∑Δd by linear interpolation. i / ∑ti =0.07 / 1.0=0.07mm / s. The result shows that the mold gap shows a fluctuation convergence trend after adjustment.

[0137] S403: Extract the gap change direction and change rate based on the mold gap offset change, build a trend offset map based on the time series data, perform key trend identification and state determination operations on the map, and generate a molding state feedback signal;

[0138] Based on the timing offset sequence data, the direction of the gap change is extracted as alternating positive and negative (the symbol sequence is [0, +1, -1, +1, -1]), and the change rate of adjacent time points is calculated:

[0139] r=[NaN,+0.2,-0.15,+0.4,-0.2]mm / s, take the average rate of change of absolute value after eliminating invalid values When constructing the trend deviation map, the stability judgment threshold is set to 0.25 mm / s (based on the 90th percentile of the original data statistics). The current gap state is determined to be "stable transition stage", and a forming state feedback signal S=1 is generated (code 1 means that production can continue). If The benefit of this formula is that it accurately characterizes the dynamic behavior of the mold by quantifying the coupling relationship between the direction and rate of gap change, thus avoiding the limitations of a single parameter criterion.

[0140] Specifically, the steps of S5 are:

[0141] S501: Based on the forming state feedback signal and the dynamic gradient adjustment instruction, they are synchronously input into the trained convolutional neural network model, the feedback signal is spatially expanded through the convolution layer, the grayscale difference of adjacent pixels is extracted, mean filtering is performed on a regional basis, the trend of regional pixel difference values ​​is calculated, and the signal gradient change rate is generated;

[0142] Based on the feedback signal state S = 1 and the node weight ratio W = [0.3, 0.5, 0.2] set in the dynamic gradient adjustment instruction, the current feedback flux data is spatially expanded, and three-point monitoring flux data Q = [1.2, 1.5, 1.3] L / s are selected to represent the measurement values ​​of consecutive nodes. The corresponding coordinates are set as (X1, Y1) = (1, 1), (X2, Y2) = (2, 1), and (X3, Y3) = (3, 1) in sequence. In the convolutional layer of the convolutional neural network, the local gradient change rate between adjacent data points is calculated using the formula:

[0143]

[0144] Among them, ΔG represents the rate of change of signal gradient, P iRepresents the grayscale value of the i-th pixel, X i Represents the horizontal coordinate value of the i-th pixel, Y i Represents the vertical coordinate value of the i-th pixel, w i Represents the weight factor of the i-th pixel, the adaptive weight factor w i The feature parameters learned by the model are dynamically set and regulated based on the importance of different pixels. n represents the number of pixels in the image area, and i is the pixel index.

[0145] In the specific implementation, the first step is to calculate the flux change between point 2 and point 1: P2 = 1.5, P1 = 1.2, and the lateral distance between the two points is The local change rate is |(1.5-1.2) / 1.0|=0.3. Combined with the weight w2=0.5, the weighted value of this segment is 0.3×0.5=0.15. The second step is to calculate the change rate of the third point and the second point, P3=1.3, P2=1.5, the lateral distance is still 1.0, the flux change rate is |(1.3-1.5) / 1.0|=0.2, the corresponding weight is w3=0.2, the weighted value is 0.2×0.2=0.04, and the final signal gradient change rate is the sum of the two:

[0146] ΔG=0.15+0.04=0.19L / s 2 ;

[0147] At the same time, in order to ensure the rationality of the values, the sliding mean processing is performed with the standard filter window, the window length is set to 2, and the mean filtering results of the difference sequence [0.3, 0.2] are calculated as follows:

[0148]

[0149] The second term was padded with zeros due to the lack of subsequent data. Combined with trend analysis, the slope was calculated to be (0.1-0.25) / 1 = -0.15. The absolute value was taken as the overall trend assessment indicator, 0.15. This value is close to the result of 0.19 derived from the formula, verifying the accuracy of the spatial gradient reflection. The following table provides the core parameters and results of the signal processing process:

[0150] Table 4: Signal gradient change rate calculation table

[0151] Parameter name Numerical unit Flux monitoring point data 1.2，1.5，1.3 L / s Pixel coordinates (1，1)，(2，1)，(3，1) grid Difference 0.3，0.2 L / s Distance calculation 1.0，1.0 grid Weight value 0.5，0.2 none Weighted gradient term 0.15，0.04 <![CDATA[L / s 2 ]]> Filtering results 0.25，0.1 L / s Gradient change rate 0.19 <![CDATA[L / s 2 ]]>

[0152] As shown in Table 4, the gradient change rate ΔG = 0.19 L / s 2 The flux difference is divided by the distance and then weighted summed up, and the result is used in the subsequent temporal gradient reconstruction operation.

[0153] S502: Calling the signal gradient change rate, reconstructing the gradient change sequence in the time axis direction according to the time sequence recognition module in the convolutional neural network model, superimposing the node weight ratio in the dynamic gradient adjustment instruction, calculating the node lateral offset amplitude, and generating the node gradient offset;

[0154] The signal gradient change rate ΔG obtained by the previous calculation is called = 0.19L / s 2 , introduce time series information into the time series recognition module, sort the corresponding gradient monitoring value sequence S = [0.10, 0.15, 0.12] according to the timestamp sequence t = [0, 0.5, 1.0]s, superimpose the node weight ratio W = [0.3, 0.5, 0.2] set in the dynamic gradient adjustment instruction, and perform weighted calculation:

[0155] G=0.3×0.10+0.5×0.15+0.2×0.12=0.03+0.075+0.024=0.129;

[0156] The weighted results are used for interpolation reconstruction of the time series curve, and the linear interpolation method is used to fit the gradient changes between discrete time nodes, and then the local offset characteristics are analyzed. Taking the nodes between t = 0.5s and t = 1.0s as an example, the corresponding gradient values ​​are 0.15 and 0.12, and the lateral offset amplitude is the gradient difference multiplied by the scale factor, that is,

[0157] Δx=|0.15-0.12|×10=0.03×10=0.3mm,

[0158] Thus, the node gradient offset is 0.3 mm, which reflects the degree of local disturbance of the forming trajectory within the specified time interval.

[0159] S503: Calling the node gradient offset, comparing the node offset result with the gradient tolerance limit value according to the evaluation function parameters at the model output end, determining whether the offset amplitude deviates, extracting the segment deviation ratio, performing overall trend fitting, and obtaining the forming evaluation result;

[0160] Read the gradient tolerance limit value T set in the evaluation module r=0.5mm, compared with the currently calculated offset of 0.3mm, since 0.3<0.5, it is determined that the current node has not deviated beyond the limit and is marked as "not deviated" state. In order to further evaluate the stability of the overall forming process, the deviation ratio data of three consecutive time periods [0.1, 0.2, 0.25] are extracted to calculate their changing trends. Through the linear regression method, its trend slope k=(0.2-0.1)+(0.25-0.2) / 2=0.075 is estimated. This value is lower than the set threshold of 0.1 (based on the 85th percentile value obtained from historical data statistics), indicating that the current forming state has trend stability. Therefore, the generated forming evaluation result is E=qualified. This process effectively realizes the joint judgment of the dynamic characteristics and accuracy of the forming through the time series analysis of the one-dimensional sequence, while ensuring the response speed and improving the accuracy of the evaluation.

[0161] like Figure 2 As shown, an intelligent adjustable medical cotton ball forming system includes:

[0162] The pressure acquisition module is used to obtain real-time pressure data of the target area of ​​the mold cavity through a distributed pressure sensor array, and input the raw pressure data into the Gaussian filter algorithm for spatial dimension denoising, generate regional pressure feature vectors, and pass them to the pressure trend determination module;

[0163] The pressure trend determination module is used to receive the regional pressure feature vector and perform directional judgment on the pressure change trend in the region by calling the dynamic gradient judgment function to determine whether the feature vector meets the continuous direction offset condition. When the condition is met, it outputs a dynamic gradient adjustment instruction and transmits it to the pressure regulation module;

[0164] The pressure regulation module is used to receive the dynamic gradient adjustment instruction, perform amplitude adjustment on the compressed gas control valve, calculate the molding pressure compensation parameter, and transmit it to the gap adjustment module;

[0165] The gap adjustment module is used to receive the molding pressure compensation parameters and drive the servo motor to adjust the mold gap based on the parameters. The adjusted gap data is collected by the displacement sensor and offset with the reference gap value to generate a molding status feedback signal, which is transmitted to the molding evaluation module.

[0166] The molding evaluation module is used to receive the molding state feedback signal and dynamic gradient adjustment instruction, and input them into the convolutional neural network model for time series feature extraction and analysis, and output the molding evaluation results.

[0167] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. An intelligent adjustable medical cotton ball forming method, characterized in that: The following steps are involved: S1: Regional pressure data is collected through a distributed pressure sensor array. The collected raw pressure data is input into a Gaussian filtering algorithm that dynamically adjusts kernel parameters based on the noise level, and multi-dimensional spatial denoising is performed to generate a regional pressure feature vector. S2: Based on the regional pressure characteristic vector, a dynamic gradient determination function is called to perform directional analysis on the pressure change trend in the region, and when the trend change satisfies the continuous directional offset condition, a dynamic gradient adjustment instruction is output; S3: According to the dynamic gradient adjustment instruction, the valve opening of the compressed gas channel in the target area is controlled to adjust the air pressure in the mold cavity, and the current valve opening parameter is modified by adjusting the amplitude to obtain the molding pressure compensation parameter; S4: Based on the molding pressure compensation parameter, the servo motor is driven to adjust the mold gap. The adjusted gap data is collected in real time by the displacement sensor. The offset is calculated with the reference gap value. The offset direction and change rate are combined to obtain a molding state feedback signal. S5: Call the molding state feedback signal and dynamic gradient adjustment instruction, input them into the trained convolutional neural network model, perform time series feature extraction and analysis, and output the molding evaluation result.

2. The intelligent adjustable medical cotton ball forming method according to claim 1, characterized in that: The dynamic gradient judgment function selects a time window based on the noise level. The continuous direction shift condition is that the gradient direction continues to change within at least three consecutive time steps, and the shift angle exceeds a set threshold. The set threshold is optimized using the gradient descent method on historical data samples. The offset direction and the rate of change are subjected to Z-score normalization to unify the dimensions; The convolutional neural network model includes two convolution extraction layers, an LSTM time series processing layer and two fully connected layers, and is trained based on batch raw molding data samples of the experimental device; The regional pressure characteristic vector includes the pressure distribution matrix, the contact area coordinates, and the effective pressure value. The dynamic gradient adjustment instruction specifically includes the pressure gradient direction angle, the displacement accumulation, and the critical rate threshold. The molding pressure compensation parameters include the linear displacement of the valve opening, the cavity pressure gradient coefficient, and the pulse width modulation parameters. The molding state feedback signal specifically refers to the gap displacement deviation, the real-time displacement rate, and the deformation convergence rate. The molding result evaluation includes the porosity distribution diagram, the density standard deviation, and the stress concentration factor distribution.

3. The intelligent adjustable medical cotton ball forming method according to claim 1, characterized in that: The specific steps of S1 include: S101: Collect regional pressure data through a distributed pressure sensor array, collect pressure at multiple points in the region according to a preset fixed period, record the pressure values ​​and timestamps of the sensor nodes, and aggregate the sampled values ​​to form a unified data set to obtain an original pressure data set; S102: Based on the original pressure data set, calling the Gaussian filter algorithm to perform spatial domain denoising, identifying the Gaussian weight distribution of each data point and its neighboring points, calculating the pressure response value and adjusting the node pressure value to obtain denoised pressure distribution data; The kernel size and standard deviation of the Gaussian filter algorithm are configured according to the sensor noise level and the target resolution; S103: Based on the denoised pressure distribution data, extract the spatial distribution characteristics of the nodes, analyze the high-dimensional feature space under the positional relationship between the nodes, calculate the coordinate values ​​of the nodes in the feature space and unify them into a standard vector form, and integrate the node output regional pressure feature vectors.

4. The intelligent adjustable medical cotton ball forming method according to claim 3, characterized in that: The specific steps of S2 include: S201: Based on the regional pressure feature vector, extract the pressure value at the time node, construct a pressure change matrix, call the dynamic gradient judgment function to calculate the directional offset of adjacent units, and generate a directional change index sequence; S202: extracting continuous direction change segments based on the direction change indicator sequence, screening segments that meet angle change and length thresholds, calculating segment direction switching frequency and angle offset, and obtaining a direction offset trend parameter set; S203: Calling the paragraph change parameter in the direction shift trend parameter set, performing alignment comparison with the corresponding time period direction code in the regional pressure feature vector, recording the paragraph start and end time and regional number that meet the continuous direction shift condition, and obtaining a dynamic gradient adjustment instruction.

5. The intelligent adjustable medical cotton ball forming method according to claim 4, characterized in that: The specific steps of S3 include: S301: Based on the dynamic gradient adjustment instruction, the current opening parameter of the compressed gas channel in the target area is called, and the opening parameter is adjusted within a set range according to the adjustment amplitude to generate a valve adjustment opening value; S302: adjusting the valve opening value to control the real-time opening and closing state of the compressed gas channel valve in the mold cavity, collecting the gas channel flow rate data after opening and closing, and calculating the interval difference between the gas channel flow rate data and the initial value to generate a flow rate change value; S303: Based on the flow rate change value, normalization processing is performed in combination with the mold cavity volume data, and the unit volume flow rate change rate and the amplification amplitude corresponding to the set pressure adjustment coefficient are calculated to obtain the molding pressure compensation parameter.

6. The intelligent adjustable medical cotton ball forming method according to claim 5, characterized in that: The flow rate change value is calculated using the formula: Among them, ΔQ represents the change in flow rate, the unit is L / min, Q t1 Represents the flow rate data value at the current time t1, in L / min, Q t0 represents the initial flow rate data value at time t0, in L / min, V represents the mold cavity volume data, in L, λ j represents the real-time flow rate deviation of the compressed gas channel in the jth cycle, in L / min, γ j represents the instantaneous opening and closing amplitude of the valve in the jth cycle, in percentage, k represents the total number of sampling cycles, and η represents the gas compression response sensitivity coefficient, which is a dimensionless parameter.

7. The intelligent adjustable medical cotton ball forming method according to claim 5, characterized in that: The specific steps of S4 include: S401: Based on the molding pressure compensation parameter, drive the servo motor to adjust the mold gap, extract the difference between the control displacement and the target instruction, perform a synchronous correction operation and record the response rate and feedback delay to generate a mold gap adjustment value; S402: calling the mold gap adjustment value, collecting mold gap change data through a displacement sensor, calculating the offset between the mold gap and the reference gap, and performing time sequence organization based on the data collection time points to obtain the mold gap offset change; S403: Extract the gap change direction and change rate based on the mold gap offset change, build a trend offset map based on the time series data, perform map key trend identification and state determination operations, and generate a molding state feedback signal.

8. The intelligent adjustable medical cotton ball forming method according to claim 7, characterized in that: The specific steps of S5 include: S501: Based on the forming state feedback signal and the dynamic gradient adjustment instruction, they are synchronously input into the trained convolutional neural network model, the feedback signal is spatially expanded through the convolution layer, the grayscale difference of adjacent pixels is extracted, mean filtering is performed on a regional basis, the trend of regional pixel difference values ​​is calculated, and the signal gradient change rate is generated; S502: Calling the signal gradient change rate, reconstructing the gradient change sequence in the time axis direction according to the time sequence recognition module in the convolutional neural network model, superimposing the node weight ratio in the dynamic gradient adjustment instruction, calculating the node lateral offset amplitude, and generating the node gradient offset; S503: Call the node gradient offset, compare the node offset result with the gradient tolerance limit value according to the evaluation function parameter of the model output end, determine whether the offset amplitude deviates, extract the segment deviation ratio to perform overall trend fitting, and obtain the forming evaluation result.

9. The intelligent adjustable medical cotton ball forming method according to claim 8, characterized in that: The gradient change rate of the forming signal is calculated using the formula: Among them, ΔG represents the rate of change of signal gradient, P i Represents the grayscale value of the i-th pixel, X i Represents the horizontal coordinate value of the i-th pixel, Y i Represents the vertical coordinate value of the i-th pixel, w i Represents the weight factor of the i-th pixel, the adaptive weight factor w i The feature parameters learned from the model are dynamically set and regulated based on the criticality of differentiated pixels. n represents the number of pixels in the image area, and i is the pixel index.

10. An intelligent adjustable medical cotton ball forming system, characterized in that: The system is used to implement the intelligent adjustable medical cotton ball forming method according to any one of claims 1 to 9, and the system includes: The pressure acquisition module is used to obtain real-time pressure data of the target area of ​​the mold cavity through a distributed pressure sensor array, and input the raw pressure data into the Gaussian filter algorithm for spatial dimension denoising, generate regional pressure feature vectors, and pass them to the pressure trend determination module; a pressure trend determination module, configured to receive the regional pressure characteristic vector, and perform directional judgment on the pressure change trend in the region by calling a dynamic gradient judgment function, to determine whether the characteristic vector satisfies a continuous directional offset condition, and output a dynamic gradient adjustment instruction when the condition is satisfied, and transmit the instruction to the pressure regulation module; A pressure regulating module is configured to receive the dynamic gradient regulating instruction, perform amplitude regulation on the compressed gas control valve, calculate a molding pressure compensation parameter, and transmit the calculated parameter to the gap adjusting module; a gap adjustment module, configured to receive the molding pressure compensation parameters and drive a servo motor to adjust the mold gap based on the parameters, collect the adjusted gap data through a displacement sensor, perform an offset operation with the reference gap value, generate a molding state feedback signal, and transmit it to the molding evaluation module; The molding evaluation module is used to receive the molding state feedback signal and the dynamic gradient adjustment instruction, and input them into the convolutional neural network model for time series feature extraction and analysis, and output the molding evaluation result.

Images