Intelligent packaging production line automation control method and system

Abstract

The application relates to the technical field of intelligent packaging production line automatic control and multi-station collaborative regulation, and discloses an intelligent packaging production line automatic control method and system, the method comprises the following steps: collecting visual and force feedback data, fusing and preprocessing to obtain multi-station integration information, identifying an abnormal mode, determining product switching, and generating a parameter adjustment scheme; issuing a regulation instruction to correct an execution parameter and analyzing a collaborative deviation; collecting updated data to analyze a deviation value and iteratively optimizing the parameter; executing action coordination and monitoring packaging completeness to obtain an adaptive stable state of the production line. The method realizes multi-station working condition collaborative sensing, intelligent adaptation of product switching and dynamic correction of parameters, solves the problems of response lagging, parameter regulation rigidity and insufficient packaging quality stability of a traditional packaging production line, and improves the operation efficiency and finished product qualification rate of the intelligent packaging production line.

Description

Technical Field

[0001] This invention relates to the field of automated control and multi-station collaborative regulation technology for intelligent packaging production lines, and particularly to an automated control method and system for intelligent packaging production lines. Background Technology

[0002] Currently, with the continuous upgrading of the demand for flexible production and rapid product switching, the requirements for the real-time performance and adaptability of multi-station data integration, accurate anomaly identification, and coordinated action control of intelligent packaging production lines are also increasing.

[0003] In existing technologies, the control of intelligent packaging production lines mostly adopts traditional fixed parameter preset modes, relying on a single type of sensor to collect basic operating data. The supporting industrial control system can only realize simple command issuance and action execution functions, and cannot effectively fuse and process multi-source data from visual and force feedback, or perform synchronous calibration. It is difficult to accurately capture dynamic features such as product switching and workstation action deviations. At the same time, existing control methods lack in-depth identification of abnormal patterns and historical data matching, and cannot dynamically adjust parameters according to environmental changes and the degree of deviation. When faced with complex working conditions of rapid switching of multiple product specifications, problems such as uncoordinated actions, delayed response, and poor control adaptability are prone to occur, making it impossible to ensure the efficient and stable operation of packaging production.

[0004] Therefore, existing technologies suffer from poor adaptability and low motion coordination accuracy in intelligent packaging production lines due to insufficient multi-source data fusion capabilities and a lack of adaptive identification and dynamic collaborative control capabilities in the control methods. This makes it difficult to achieve efficient and stable control of the entire process under rapid product switching. Summary of the Invention

[0005] This invention provides an automated control method and system for intelligent packaging production lines to solve the problems in existing technologies, such as insufficient multi-source visual and force data fusion processing capabilities, lack of dynamic adaptability of industrial control systems, and lack of adaptive collaborative regulation and deviation iterative optimization capabilities under product switching, which lead to lagging product switching response, low motion coordination accuracy, and difficulty in achieving efficient and stable adaptive control in intelligent packaging production lines.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides an automated control method for an intelligent packaging production line, comprising:

[0007] Visual feedback data and force feedback data are collected, preprocessed and fused to obtain multi-station integrated information;

[0008] Based on the integrated information of the multi-workstation, an abnormal pattern is identified. If the abnormal pattern matches the preset historical switching data, the feature vector of the abnormal pattern is extracted, and the change classification label is determined based on the feature vector.

[0009] If the change classification label is product switching, then the equipment response time constraint of the production line is obtained, and key feature data is extracted from the feature vector. The key feature data is denoised to obtain product denoised data. Based on the product denoised data and the equipment response time constraint, the parameter optimization value of the equipment is predicted to obtain the parameter adjustment scheme.

[0010] According to the parameter adjustment scheme, control instructions are issued, and the gripping force of the robotic arm and the speed of the conveyor belt are corrected according to the control instructions to obtain the execution coordination deviation. Based on the execution coordination deviation, a hierarchical analysis is performed to generate update instructions.

[0011] The update command is executed to collect visual update data and force update data. The visual update data is analyzed for deviation according to priority, and the force update data is analyzed for deviation based on a dynamic threshold. The analysis results are then fused to obtain the data deviation value.

[0012] If the data deviation value exceeds the preset deviation threshold, then the preset deviation library and the data deviation value are used for iterative calculation to obtain a parameter adjustment scheme;

[0013] The actions are coordinated according to the parameter adjustment scheme, and the integrity of the packaging is monitored in real time to obtain the adaptive stable state of the production line.

[0014] Secondly, the present invention provides an automated control system for an intelligent packaging production line, comprising:

[0015] The data acquisition module is used to collect visual feedback data and force feedback data, preprocess and fuse the visual feedback data and force feedback data to obtain multi-station integrated information;

[0016] An anomaly identification module is used to identify anomaly patterns based on the multi-workstation integrated information. If the anomaly pattern matches preset historical switching data, the feature vector of the anomaly pattern is extracted, and a change classification label is determined based on the feature vector.

[0017] The parameter optimization module is used to obtain the equipment response time constraint of the production line if the change classification label is product switching, extract key feature data from the feature vector, denoise the key feature data to obtain product denoised data, predict the parameter optimization value of the equipment based on the product denoised data and the equipment response time constraint, and obtain the parameter adjustment scheme.

[0018] The instruction control module is used to issue control instructions according to the parameter adjustment scheme, correct the gripping force of the robotic arm and the speed of the conveyor belt according to the control instructions, obtain the execution coordination deviation, perform hierarchical analysis based on the execution coordination deviation, and generate update instructions.

[0019] The deviation analysis module is used to execute the update instruction to collect visual update data and force update data, analyze the deviation of the visual update data according to priority, analyze the deviation of the force update data based on dynamic threshold, and fuse the analysis results to obtain the data deviation value.

[0020] The iterative optimization module is used to perform iterative calculations by combining the preset deviation library with the data deviation value if the data deviation value exceeds a preset deviation threshold, so as to obtain a parameter adjustment scheme.

[0021] The stability determination module is used to adjust the coordination of the execution of the scheme according to the parameters, monitor the integrity of the packaging in real time, and obtain the adaptive stable state of the production line.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) This invention constructs multi-station integrated information by synchronously collecting visual and force feedback data and performing fusion preprocessing. This breaks through the limitations of traditional packaging production lines that rely solely on single data monitoring and lack sufficient judgment on the coordination of work station conditions. It achieves accurate integration of multi-dimensional operation data of the production line and effectively improves the comprehensiveness and authenticity of packaging condition monitoring.

[0024] (2) This invention identifies abnormal patterns and matches historical switching data to determine product switching characteristics, extracts feature vectors in a targeted manner and optimizes equipment parameters, overcoming the problems of slow product switching response and blind parameter adjustment in existing packaging production lines, realizing rapid identification of product switching conditions and accurate parameter planning, and greatly improving the efficiency of production line changeover.

[0025] (3) This invention, through deviation analysis and iterative optimization adjustment after execution of instructions, combined with packaging integrity monitoring to form an adaptive control closed loop, overcomes the shortcomings of traditional packaging production lines that lack dynamic correction capabilities and insufficient packaging quality stability, and realizes full-process adaptive and stable control of intelligent packaging production lines, effectively ensuring the qualified rate of finished packaging products and the continuous and stable operation of the production line. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of an automated control method for an intelligent packaging production line provided in the first embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of the structure of an automated control system for an intelligent packaging production line provided in the second embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Reference Figure 1 The first embodiment of the present invention provides an automated control method for an intelligent packaging production line, comprising the following steps:

[0030] S1, collect visual feedback data and force feedback data, preprocess and fuse the visual feedback data and the force feedback data to obtain multi-station integrated information;

[0031] S2, Identify abnormal patterns based on the multi-station integrated information. If the abnormal pattern matches the preset historical switching data, extract the feature vector of the abnormal pattern and determine the change classification label based on the feature vector.

[0032] S3, if the change classification label is product switching, then obtain the equipment response time constraint of the production line, extract key feature data from the feature vector, denoise the key feature data to obtain product denoised data, and predict the parameter optimization value of the equipment based on the product denoised data and the equipment response time constraint to obtain the parameter adjustment scheme.

[0033] S4, issue control instructions according to the parameter adjustment scheme, correct the gripping force of the robotic arm and the speed of the conveyor belt according to the control instructions, obtain the execution coordination deviation, perform hierarchical analysis according to the execution coordination deviation, and generate update instructions;

[0034] S5, execute the update instruction to collect visual update data and force update data, analyze the deviation of the visual update data according to priority, analyze the deviation of the force update data based on dynamic threshold, and fuse the analysis results to obtain the data deviation value;

[0035] S6. If the data deviation value exceeds the preset deviation threshold, then the preset deviation library and the data deviation value are used for iterative calculation to obtain a parameter adjustment scheme.

[0036] S7, adjust the execution of actions according to the parameters, monitor the integrity of the packaging in real time, and obtain the adaptive stable state of the production line.

[0037] In step S1, visual feedback data and force feedback data are collected, preprocessed, and fused to obtain multi-station integrated information, including:

[0038] S11, collect visual feedback data and force feedback data, and perform outlier removal and normalization processing on the visual feedback data and force feedback data to obtain cleaned data;

[0039] S12, The cleaned data is aligned according to a unified timestamp to construct a data fusion matrix;

[0040] S13, perform feature fusion on the data fusion matrix to obtain multi-dimensional fusion working condition features, and integrate the working condition data of each workstation according to the multi-dimensional fusion working condition features to obtain multi-workstation integrated information.

[0041] The cleaned data includes parameter data from two different acquisition frequencies: visual and force sensor data. Using the sampling time of the force sensor as the baseline time axis, the visual data is resampled to each force sensor sampling time using linear interpolation, resulting in time-aligned parameter pairs. These aligned multi-dimensional parameters are then arranged by workstation number, sensor type, and parameter type to construct a data fusion matrix. The linear interpolation formula is: the interpolation result equals the visual value at the previous time step plus the difference between the visual value at the next time step and the previous time step multiplied by the difference between the interpolation time and the previous time step, then divided by the difference between the next time step and the previous time step.

[0042] For example, if the force sampling time is 15:20:00.333, and there is no visual data at this time, the normalized values ​​of the workpiece size at the two visual sampling points before and after are taken as 0.10 (corresponding to 15:20:00.325) and 0.12 (corresponding to 15:20:00.825), respectively. The normalized value of the workpiece size at 15:20:00.333 is calculated as 0.10032 by linear interpolation. This value is then used to form an alignment parameter pair with the force data at this time. After processing all force sampling times in sequence, a data fusion matrix is ​​obtained.

[0043] Preliminary feature correlation analysis was performed on the historical dataset. The historical dataset contained 100 packaging samples. Each sample recorded six feature values ​​for each workstation during the production process: workpiece size, positional offset, sealing alignment, gripping force fluctuation, peak contact force, and torque change rate, along with the corresponding packaging integrity detection result (a value between 0 and 1, with 1 indicating complete compliance). The Pearson correlation coefficient between each feature and packaging integrity was calculated, and the Pearson correlation coefficients between all pairs of features were also calculated. Feature pairs with an absolute correlation coefficient greater than 0.9 were defined as redundant pairs. This threshold was set based on industrial experience; when the correlation coefficient exceeds 0.9, the information overlap between the two features is usually higher than 80%, and removing one of them can effectively reduce feature dimensionality without significant information loss. Features with low correlation to packaging integrity were removed. Based on the historical analysis results, the feature set retained for online operation was determined to be the normalized values ​​of workpiece size, positional offset, gripping force fluctuation, peak contact force, and torque change rate. The normalized value of the sealing alignment was rejected because it was too highly correlated with the workpiece size and had a low correlation with the integrity.

[0044] During online operation, the aforementioned retained features are extracted from the data of each workstation in the data fusion matrix. All feature values ​​have been mapped to the [0,1] interval through min-max normalization, forming the multi-dimensional fusion working condition features of that workstation.

[0045] For example, the normalized values ​​of the workpiece size, position offset, gripping force fluctuation, peak contact force, and torque change rate at a certain workstation are 0.02, 0.15, 0.23, 0.28, and 0.05, respectively, which constitute the multi-dimensional integrated working condition characteristics of that workstation.

[0046] The multi-dimensional integrated working condition features of all workstations are aggregated to form multi-workstation integrated information. Each workstation has the same feature dimensions, which are stored in a dictionary structure, with the workstation number as the key and the feature vector of that workstation as the value.

[0047] For example, the multi-dimensional fusion working condition features of workstation A are [0.02, 0.15, 0.23, 0.28, 0.05]; the features of workstation B are [0.01, 0.10, 0.18, 0.25, 0.04], where each component corresponds to the normalized value of workpiece size, the normalized value of position offset, the normalized value of gripping force fluctuation, the normalized value of contact force peak value, and the normalized value of torque change rate, respectively. After summarizing, the integrated information of multiple workstations is obtained.

[0048] In step S2, an abnormal pattern is identified based on the multi-station integrated information. If the abnormal pattern matches preset historical switching data, the feature vector of the abnormal pattern is extracted, and a change classification label is determined based on the feature vector, including:

[0049] S21, compare the integrated information of the multi-station with the preset normal fluctuation range in a time sequence, filter out continuous data segments that exceed the preset normal fluctuation range, extract the amplitude, frequency and trend characteristics of the continuous data segments, and construct an abnormal mode.

[0050] S22, calculate the similarity between the abnormal pattern and the preset historical switching data to obtain a comparison analysis value;

[0051] S23, if the comparison analysis value reaches the preset similarity threshold, then the feature vector of the abnormal pattern is extracted, and the change classification label is determined according to the feature vector; if the comparison analysis value does not meet the preset similarity threshold, then the control parameters remain unchanged.

[0052] It should be noted that the preset normal fluctuation range is calculated based on the mean and standard deviation of historical normal operation data, and the mean plus or minus twice the standard deviation is taken as the upper and lower limits. For visual position deviation, the set range is [-0.5mm, 0.5mm]; for force load fluctuation, the set range is [-5N, 5N]. The real-time data in the multi-station integrated information is traversed sequentially, and the data at each moment is compared with the range one by one. Three consecutive sampling points exceeding the range are identified as continuous data segments (this threshold is set based on the run test principle, and the probability of three consecutive points exceeding the normal range is less than 0.3%, which can effectively filter out single-point random noise). The amplitude (difference between the maximum and minimum values), fluctuation frequency (the number of times the data crosses the mean of the segment per unit second, with the sensor sampling frequency as the time base), and change trend (the slope of the least squares linear fitting) of the segment are extracted and combined to form an abnormal pattern.

[0053] For example, by comparing the integrated information of multiple workstations on the production line in a time sequence, it was found that the grasping force data of four consecutive sampling points exceeded the preset normal fluctuation range. The amplitude of this segment was 2N, the frequency was 1.2Hz, and the trend slope was 0.5, thus completing the construction of the abnormal pattern.

[0054] The preset historical switchover data contains at least 50 sets of abnormal pattern samples from past product switchovers. Each set of samples records three feature values: amplitude, frequency, and trend, along with their corresponding product switchover labels. Min-max normalization is performed on the amplitude, frequency, and trend features, with the minimum and maximum values ​​for normalization taken from the minimum and maximum values ​​of the feature in the historical data. The minimum value for amplitude is set to 0N, and the maximum value to 5N; the minimum value for frequency is set to 0Hz, and the maximum value to 2Hz; and the minimum value for trend is set to 0, and the maximum value to 1. Each feature is mapped to the [0,1] interval to eliminate the influence of dimensions. The cosine similarity is calculated between the normalized feature vector of the current abnormal pattern and the normalized feature vector of the historical samples. The similarity is equal to the dot product of the current feature vector and the historical feature vector divided by the product of their magnitudes, yielding a comparative analysis value (range 0 to 1). This value reflects the degree of matching between the current abnormality and the product switchover condition.

[0055] For example, the amplitude of the current abnormal pattern is 2N, the frequency is 1.2Hz, and the trend slope is 0.5. After normalization according to the above range, the vector is [0.4, 0.6, 0.5]. The original features of a certain historical switching sample [1.8N, 1.1Hz, 0.4] are normalized to [0.36, 0.55, 0.4], and the cosine similarity calculation result is 0.99.

[0056] The preset threshold for the comparative analysis value is 0.85. This threshold is set by rounding up the lowest similarity (0.83) of the product switching samples in historical data to ensure no cases are missed. If the comparative analysis value is greater than 0.85, it is determined to meet the characteristics of product switching. All three features (amplitude, frequency, and trend) in the abnormal pattern are extracted to form a feature vector, and the change classification label is determined to be "product switching". If the comparative analysis value is less than or equal to 0.85, it is determined to be a non-switching anomaly, and the original control parameters such as the robotic arm gripping force and conveyor belt speed are kept unchanged.

[0057] For example, if a calculation yields a comparative analysis value of 0.92, which is higher than 0.85, it confirms that the operating condition meets the product switching characteristics. The feature vector [2N, 1.2Hz, 0.5] is extracted and labeled with a product switching classification tag. If the value is 0.75, the original control parameters are maintained for stable operation.

[0058] In step S3, if the change classification label is product switching, then the equipment response time constraint of the production line is obtained, and key feature data is extracted from the feature vector. The key feature data is then denoised to obtain product denoised data. Based on the product denoised data and the equipment response time constraint, the optimized parameter values ​​of the equipment are predicted to obtain a parameter adjustment scheme, including:

[0059] S31, if the change classification label is product switching, then principal component analysis is performed on the feature vector to obtain key feature data;

[0060] S32, Denoise the key feature data by filtering out environmental noise and equipment vibration noise to obtain product noise-denoising data;

[0061] S33, obtain the response time constraint parameters of the production line execution equipment, predict the equipment parameter optimization value based on the product denoising data and the response time constraint parameters, and obtain the parameter optimization constraint;

[0062] S34, Input the product denoising data into the preset parameter prediction model, and solve it in combination with the parameter optimization constraints to obtain the parameter adjustment scheme.

[0063] It should be noted that when the change classification label indicates a product switch, principal component analysis is performed on the extracted feature vectors. The original feature vectors contain six dimensions: workpiece size, positional offset, sealing alignment, gripping force fluctuation, peak contact force, and torque change rate. Each feature has been normalized to the [0,1] interval using min-max normalization to eliminate dimensional differences. The correlation coefficient matrix is ​​calculated for the normalized feature vectors, and the eigenvalues ​​and eigenvectors are solved. The eigenvalues ​​are sorted from largest to smallest, and the top k principal components with a cumulative variance contribution rate of 95% are selected as retained features. In this example, the cumulative variance contribution rate of the first three principal components is 96%, so the original six-dimensional features are reduced to three dimensions to obtain the key feature data.

[0064] For example, principal component analysis is performed on the 6-dimensional normalized feature vector corresponding to product switching. The first 3 principal components are retained, and the output consists of three core control parameters: workpiece size equivalence factor, gripping force comprehensive coefficient, and speed response factor, forming key feature data.

[0065] Median filtering was applied to each time series in the key feature data for denoising. The window size was set to 5 sampling points (approximately 1.67 seconds for the stress sensor). This window effectively filtered out isolated impulse noise while preserving signal edges. After denoising, discrete outliers exceeding ± three standard deviations of the sequence mean were removed, retaining valid data reflecting the actual production conditions to obtain the denoised product data.

[0066] For example, applying a mean filter to the grasping force comprehensive coefficient sequence in the key feature data with a window size of 5 removes isolated outliers caused by sudden changes in ambient light and fluctuation peaks caused by equipment vibration. After sorting and filtering, noise-reduced product data without interference is obtained.

[0067] The response time constraints of the production line equipment are determined by the equipment's rated parameters: the robotic arm's gripping action response time is 50 milliseconds, the conveyor belt's speed adjustment response time is 100 milliseconds, and the servo drive's action response time is 20 milliseconds. Dividing the target gripping force and target speed from the product's denoised data by the response time yields the maximum rate of change constraint for parameter adjustment. Considering the new product packaging process requirements (e.g., gripping force not exceeding 100N, conveyor belt speed not exceeding 0.5m / s), a reasonable parameter adjustment range is calculated: the robotic arm's gripping force adjustment range is no more than 5N per adjustment, and the conveyor belt speed adjustment range is no more than 0.02m / s per adjustment. This is used to construct parameter optimization constraints, limiting the parameter adjustment range to within the equipment's response capability.

[0068] For example, based on the rated maximum force change rate of the robotic arm servo driver of 80 N / s and the control cycle of 0.1 s, the upper limit of the single adjustment step is 80 N / s × 0.1 s = 8 N; at the same time, due to process safety restrictions, the gripping force should be less than or equal to 100 N, and it is determined that each adjustment should not exceed 4 N.

[0069] The preset parameter prediction model employs a three-layer fully connected neural network. The number of nodes in the input layer equals the dimension of the denoised product data (3 core parameters in this example), the number of nodes in the hidden layer is 10, the activation function is ReLU, and the number of nodes in the output layer is 3 (corresponding to the robotic arm's gripping force, conveyor belt speed, and servo drive acceleration). Model training uses 500 sets of historical product switching data, with 80% used for training and 20% for validation. The loss function is mean squared error, the optimizer is Adam, the learning rate is 0.001, the training epochs are 200, and the early stop epochs are 10. The denoised product data is input into the trained model, which outputs initial parameter values. These values ​​are then constrained by parameter optimization constraints: if the output value exceeds the constraint range, the constraint boundary value is applied. The final output is a robotic arm gripping force, conveyor belt speed, and servo drive parameters that match the new product specifications, forming a parameter adjustment scheme that can be directly sent to the control system for execution.

[0070] For example, by inputting the workpiece size equivalence factor of 0.6, the gripping force comprehensive coefficient of 0.7, and the speed response factor of 0.5 from the product noise reduction data into the neural network model, the model outputs a gripping force of 72N, a conveyor belt speed of 0.35m / s, and a servo acceleration of 0.2m / s². After constraint checks (gripping force not exceeding 80N and speed not exceeding 0.4m / s), the model meets the requirements and forms the final parameter adjustment scheme.

[0071] In step S4, a control command is issued according to the parameter adjustment scheme. The robotic arm's gripping force and conveyor belt speed are corrected according to the control command to obtain the execution coordination deviation. A hierarchical analysis is performed based on the execution coordination deviation to generate an update command, including:

[0072] S41, parse the parameter adjustment scheme, generate executable control instructions, correct the gripping force of the robotic arm and the running speed of the conveyor belt according to the control instructions, and collect the running parameters of the robotic arm and the conveyor belt.

[0073] S42, compare the operating parameters with the optimized values ​​of the preset parameter adjustment scheme, calculate the deviation, and obtain the execution coordination deviation;

[0074] S43, the execution coordination deviation is analyzed by workstation and parameter type to obtain error feature vector, and the adjustment strategy is determined based on the error feature vector to generate update instructions.

[0075] It should be noted that the parameter adjustment scheme undergoes standardized analysis. The target values ​​for the robotic arm's gripping force, the conveyor belt's running speed, and the execution timing (each instruction is spaced 20 milliseconds apart, this interval is set according to the control system's communication cycle) are converted into message formats conforming to the industrial Ethernet communication protocol and sent to the control system. After receiving the control instructions, the control system drives the robotic arm and conveyor belt actuators to adjust their output parameters in real time. Simultaneously, it collects the actual gripping force of the robotic arm and the actual running speed of the conveyor belt through force sensors (sampling frequency 100Hz) and speed sensors (sampling frequency 50Hz) mounted on the actuators.

[0076] For example, the gripping force of 80N and the conveyor belt speed of 0.35m / s in the parameter adjustment scheme are parsed into control messages and sent out. After the system executes the message, the actual gripping force of 78N and the actual speed of 0.33m / s are collected, thus completing the execution of the instruction and the acquisition of the measured data.

[0077] The actual values ​​collected by the sensors are compared with the target values ​​in the parameter adjustment scheme item by item. The deviation is defined as the actual value minus the target value, resulting in a signed deviation value: the deviation of the robotic arm's gripping force is -2N (negative indicates that it is lower than the target), and the deviation of the conveyor belt speed is -0.02m / s. The deviation values ​​of each station and each actuator are organized into a deviation vector according to the station number and parameter type, and recorded as the execution coordination deviation. Its structure is in dictionary form: the key is the station number, and the value is [gripping force deviation (N), speed deviation (m / s)].

[0078] For example, the actual gripping force of the robotic arm at workstation 1 is 78N, the target is 80N, and the deviation is -2N; the actual speed of the conveyor belt is 0.33m / s, the target is 0.35m / s, and the deviation is -0.02m / s. Then the execution coordination deviation of workstation 1 is [-2, -0.02].

[0079] The execution coordination deviation is broken down into layers based on workstation and parameter type. For each deviation value, its amplitude (absolute value of the deviation) and direction of change (sign of the deviation) are calculated. Simultaneously, based on the deviation values ​​at the three most recent sampling times, the slope is fitted using the least squares method to obtain the deviation change rate (unit: N / s or m / s²). The amplitude, sign, and change rate are combined to form the error feature vector of the deviation, in the format [amplitude, sign, change rate].

[0080] For example, if the gripping force deviation of station 1 is -2N, the amplitude is 2, and the sign is negative, and the deviations are -1 N, -1.2 N, and -1.4 N for three consecutive times with a time interval of 0.5 s, then the rate of change is approximately -0.4N / s, which meets the threshold of 0.2 N / s. Therefore, the error feature vector is [2, -1, -0.5], where +1 represents positive and -1 represents negative.

[0081] The adjustment strategy is determined based on the error feature vector. Thresholds are set according to equipment accuracy and process requirements: amplitude threshold 1N (adjustment is required if this value is exceeded), rate of change threshold 0.2N / s (exceeding this value is considered dynamic deviation), duration period 2 periods (to avoid false triggering from single fluctuations), and timing correction step size 10 milliseconds (based on the minimum response time of the servo drive). Specific rules are as follows:

[0082] If the deviation amplitude is greater than 1N and the absolute value of the rate of change is greater than 0.2N / s, then parameter fine-tuning is performed. The adjustment amount is equal to 0.5 times the deviation amplitude, and the adjustment direction is opposite to the deviation sign (i.e., increase the parameter value for negative deviations and decrease the parameter value for positive deviations).

[0083] If the deviation amplitude is greater than 1N and the absolute value of the rate of change is less than or equal to 0.2N / s, and this state lasts for 2 sampling periods, then timing correction is performed. The correction direction is determined by the sign of the deviation: a negative deviation indicates execution lag, advancing the robotic arm's movement by 10 milliseconds; a positive deviation indicates execution lead, delaying the robotic arm's movement by 10 milliseconds.

[0084] The adjustment amount is converted into an incremental instruction (the format is the same as the original control instruction, but the field meaning is the adjustment increment), and an update instruction is generated and sent to the corresponding execution agency.

[0085] For example, if the gripping force deviation of station 1 is 2N, negative in sign, and has a change rate of -0.5N / s, it meets the fine-tuning conditions. The adjustment amount is 2×0.5=1N, and the direction is increasing. The update instruction "gripping force increased by 1N" is generated. If the deviation amplitude is 1.5N, negative in sign, and has a change rate of 0.1N / s for two consecutive cycles, the timing correction instruction "robotic arm movement advanced by 10ms" is generated.

[0086] In step S5, the update instruction is executed to collect visual update data and force update data. The visual update data is analyzed for deviation according to priority, and the force update data is analyzed for deviation based on a dynamic threshold. The analysis results are fused to obtain the data deviation value, including:

[0087] S51, execute the update instruction and collect visual update data and force update data;

[0088] S52, the visual update data and the force update data are parsed and processed to obtain the visual deviation set and the force deviation set;

[0089] S53, extract visual deviation features from the visual deviation set, and construct a dynamic threshold based on the force deviation set;

[0090] S54, prioritize the visual deviation features to obtain ranked visual deviation features; compare the ranked visual deviation features with a preset visual threshold, compare the force deviation set with the dynamic threshold, calculate the deviation amount respectively and integrate them with weight to obtain the data deviation value.

[0091] It should be noted that the control system's actuators respond to update commands by simultaneously activating industrial vision sensors and six-dimensional force sensors, ensuring coordinated action between the two types of sensors and the actuators. The vision sensors primarily collect visual update data such as workpiece position deviation and packaging seal alignment, while the force sensors primarily collect force update data such as fluctuations in the robotic arm's gripping force and changes in the conveyor belt's force. The acquisition frequency is twice per second for the vision sensors and three times per second for the force sensors. This frequency setting accurately captures real-time changes in operating conditions after command execution while avoiding data redundancy. The raw update data collected by the sensors is stored in a four-dimensional structure: timestamp, sensor number, parameter type, and parameter value. The timestamp is accurate to milliseconds, ensuring data synchronization.

[0092] For example, after executing the update command, the vision and force sensors are activated. The vision update data collected during a certain period is a workpiece position deviation of 0.8mm and a sealing alignment deviation of 0.5mm. The force update data is a gripping force fluctuation of 2.3N and a conveyor belt force fluctuation of 1.8N. After storing the data in a four-dimensional structure, the data collection for that period is completed.

[0093] The collected visual and force update data are analyzed and processed separately. First, isolated outliers are removed using a three-standard-deviation criterion: the mean and standard deviation are calculated for each parameter sequence, and points exceeding the mean ± three standard deviations are removed. This threshold is set based on the principle of normal distribution (the probability of exceeding the standard deviation is less than 0.3%). Then, the two types of data are compared item by item with the target parameters corresponding to the update instructions, and the difference between the actual value and the target value is recorded as the deviation. The deviations of all visual parameters are summarized to form a visual deviation set, and the deviations of all force parameters are summarized to form a force deviation set.

[0094] For example, if the target values ​​for visual parameters are 0mm for workpiece position deviation and 0mm for sealing alignment, and the actual values ​​are 0.8mm and 0.5mm, then the set of visual deviations is [0.8, 0.5]. If the target values ​​for force parameters are 0N for gripping force fluctuation and 0N for conveyor belt force fluctuation, and the actual values ​​are 2.3N and 1.8N, then the set of force deviations is [2.3, 1.8].

[0095] Core parameters with significant impact on packaging quality are extracted from the visual deviation set as visual deviation features. The extraction criteria are: workpiece position deviation directly affects packaging sealing performance, and seal alignment deviation affects appearance; both are retained. When constructing a dynamic threshold based on the force deviation set, a sliding window mean method is applied to each force parameter. The lower limit of the dynamic threshold is equal to 0.5 times the historical mean deviation of that parameter, and the upper limit is equal to 2 times the historical mean deviation of that parameter. The historical mean is taken from the most recent 50 sampling periods. This threshold can adaptively adjust with changes in operating conditions, capturing reasonable fluctuations while identifying anomalies.

[0096] For example, workpiece position deviation of 0.8mm and sealing alignment deviation of 0.5mm are extracted as visual deviation features from the visual deviation set. The historical average of gripping force fluctuation is 1.5N, so its dynamic threshold has a lower limit of 0.75N and an upper limit of 3.0N; the historical average of conveyor belt force fluctuation is 1.2N, so its dynamic threshold has a lower limit of 0.6N and an upper limit of 2.4N. The current gripping force fluctuation of 2.3N is within the threshold, and the conveyor belt force fluctuation of 1.8N is also within the threshold.

[0097] The extracted visual deviation features were prioritized based on their contribution to the finished product pass rate. This contribution was determined through regression analysis of 500 historical samples: the contribution of workpiece position deviation was 0.65, and the contribution of sealing alignment deviation was 0.35, thus workpiece position deviation had a higher priority. Each prioritized visual deviation feature was compared to a preset visual threshold, set according to packaging process standards: 0.5mm for workpiece position deviation and 0.3mm for sealing alignment deviation. The difference between each visual deviation feature and its corresponding preset threshold was calculated (0 if the deviation is less than the threshold), and then weighted and summed according to priority contribution to obtain the visual deviation amount. Simultaneously, each deviation value in the force perception deviation set was compared to its corresponding dynamic threshold. If the deviation value was within the threshold range, the deviation amount for that parameter was 0; otherwise, the deviation amount equaled the portion exceeding the threshold (i.e., deviation value minus the upper limit, or lower limit minus the deviation value). The deviation amounts of all force perception parameters were summed to obtain the force perception deviation amount. Finally, the impact of visual deviation and force deviation on the stable operation of the production line is weighted and integrated. The weights are based on historical failure statistics: visual deviation accounts for 70% of packaging defects, and force deviation accounts for 30%. Therefore, the weight of visual deviation is set at 70%, and the weight of force deviation is set at 30%.

[0098] For example, the workpiece position deviation is 0.8mm, the preset threshold is 0.5mm, and the difference is 0.3mm; the sealing alignment deviation is 0.5mm, the preset threshold is 0.3mm, and the difference is 0.2mm. Visual deviation = 0.3 × 0.65 + 0.2 × 0.35 = 0.195 + 0.07 = 0.265. The gripping force fluctuation of 2.3N is within the threshold [0.75, 3.0], with an out-of-tolerance value of 0; the conveyor belt force fluctuation of 1.8N is within the threshold [0.6, 2.4], with an out-of-tolerance value of 0; force perception deviation = 0 + 0 = 0. Data deviation value = 0.265 × 0.7 + 0 × 0.3 = 0.1855.

[0099] In step S6, if the data deviation value exceeds the preset deviation threshold, the preset deviation library and the data deviation value are used for iterative calculation to obtain a parameter adjustment scheme.

[0100] When the data deviation value obtained in step S5 exceeds the preset deviation threshold, the iterative calculation process begins. The preset deviation threshold is set based on the distribution of data deviation values ​​in historical normal production batches, taking the 95th percentile of the data deviation values ​​of 200 sets of qualified product batches as the threshold, which is 0.25. If the current data deviation value is greater than 0.25, optimization calculation is triggered; if it is less than or equal to 0.25, it is determined that the current parameter does not need to be adjusted, and the process proceeds directly to step S7.

[0101] The preset deviation library is a pre-built database. Specifically, it is constructed by manually injecting known deviations, such as robotic arm gripping force deviations of 1N, 2N, 3N, 4N, and 5N, and conveyor belt speed deviations of 0.01m / s, 0.02m / s, 0.03m / s, 0.04m / s, and 0.05m / s. The corresponding data deviation values ​​are calculated by executing steps S1 to S5, and the data deviation values ​​after each adjustment are recorded. Adjustment schemes that reduce the data deviation value to below 0.25 are then added to the database. Deviation combinations that fail to meet the target on the first attempt are not recorded. After calibration experiments, 100 valid records were obtained. Each record contains three fields: the data deviation value before iteration, the control parameter adjustment scheme used after iteration, and the adjusted data deviation value. The control parameter adjustment scheme only includes the adjustment amount of the robotic arm gripping force (in N) and the adjustment amount of the conveyor belt speed (in m / s), because these two parameters have the greatest impact on the data deviation value obtained in step S5. Other parameters remain unchanged from the output value of step S3. The sign of the adjustment amount indicates an increase or decrease.

[0102] The specific method for iterative calculation is as follows: The current data deviation value is recorded as the current value. The two historical records closest to the current value are searched in the preset deviation library. These two records must have data deviation values ​​before iteration that are less than and greater than the current value, respectively, and their corresponding adjusted data deviation values ​​must both be less than 0.25. If two such records are found, denoted as Record A and Record B, linear interpolation is used to calculate the required adjustment scheme. The linear interpolation calculation method is as follows: the first component of the adjustment scheme equals the first component of Record A plus the difference between the first component of Record B and the first component of Record A, multiplied by the difference between the current value and the data deviation value of Record A, and then divided by the difference between the data deviation values ​​of Record B and Record A; the second component is calculated similarly. If only one record with a data deviation value less than the current value before iteration and meeting the standard after adjustment can be found in the library, the adjustment scheme for that record is directly adopted. If no record meeting the conditions is found, the default adjustment scheme is adopted: the robotic arm gripping force adjustment is +5N, and the conveyor belt speed adjustment is -0.02m / s. The default adjustment scheme is set based on the equipment's rated adjustment capacity. The maximum safe adjustment amount for a single robotic arm is 5N, and the maximum safe adjustment amount for a single conveyor belt is 0.02m / s. The direction of adjustment is towards improving the integrity of the packaging.

[0103] For example, the current data deviation value is 0.38. Two records are found in the preset deviation library: the data deviation value of record A before iteration is 0.30, and the adjustment plan is to increase the gripping force of the robotic arm by 2N and decrease the conveyor belt speed by 0.008m / s; the data deviation value of record B before iteration is 0.45, and the adjustment plan is to increase the gripping force of the robotic arm by 4N and decrease the conveyor belt speed by 0.015m / s. The calculated adjustment amount of the robotic arm's gripping force is 2 plus the difference of 4 minus 2, multiplied by the difference of 0.38 minus 0.30, and then divided by the difference of 0.45 minus 0.30, which is 2 plus 2 multiplied by 0.08 divided by 0.15, approximately equal to 2 plus 1.07 equals 3.07N, rounded to 3N; the adjustment amount of the conveyor belt speed is -0.008, plus the product of the difference of -0.015 minus -0.008 multiplied by the difference of 0.38 minus 0.30, and then divided by the difference of 0.45 minus 0.30, which is -0.008 plus -0.007 multiplied by 0.08 divided by 0.15, approximately equal to -0.008 minus 0.0037 equals -0.0117m / s, rounded to three significant figures as -0.012m / s. The final parameter adjustment scheme is as follows: increase the gripping force of the robotic arm by 3N and decrease the conveyor belt speed by 0.012m / s. This scheme is then executed in step S7.

[0104] If only one record is found in the preset deviation database where the data deviation value before iteration is 0.30 and meets the standard after adjustment, and the adjustment scheme is to increase the robotic arm's gripping force by 2N and decrease the conveyor belt speed by 0.008m / s, and there are no records with values ​​greater than the current value, then this scheme will be directly adopted. If no record is found, then the default adjustment scheme will be adopted, which is to increase the robotic arm's gripping force by 5N and decrease the conveyor belt speed by 0.02m / s.

[0105] The above calculation process is executed only once, directly outputting the parameter adjustment scheme without the need for looping. If the packaging integrity still does not meet the standard after step S7, the data deviation value will be recalculated in step S5 of the next main loop, and the process will enter step S6 again, thus forming a closed-loop iteration.

[0106] In step S7, the action coordination is performed according to the parameter adjustment scheme, and the integrity of the packaging is monitored in real time to obtain the adaptive stable state of the production line, including:

[0107] S71, generate a servo drive sequence based on the device execution parameters according to the parameter adjustment scheme, analyze the dynamic response delay of the actuator through the servo drive sequence, and generate an action coordination command based on the dynamic response delay;

[0108] S72, drive the packaging equipment to operate according to the action coordination command, collect packaging sealing tension and edge bonding gap data, and perform feature fusion of the packaging sealing tension and edge bonding gap data to obtain material deformation characteristics;

[0109] S73, calculate the tension fluctuation amplitude and gap closing rate based on the material deformation characteristics to obtain the packaging quality characterization characteristics; construct an integrity evaluation matrix based on the packaging quality characterization characteristics to obtain the packaging integrity value;

[0110] S74. Compare the packaging integrity value with the preset state judgment threshold to obtain the adaptive stable state of the production line.

[0111] It should be noted that the parameter adjustment scheme clearly defines core control parameters such as the robotic arm's grasping angle, running speed, and action timing. When generating the servo drive sequence, these parameters are converted into pulse signals recognizable by the actuators. The sequence includes the motion angle (in degrees), running speed (in degrees / second), and action interval (in milliseconds). When parsing the servo drive sequence, the dynamic response delay of each actuator is obtained through offline calibration: the robotic arm is excited with a step signal, and the time difference from the issuance of the command to the start of the actual action is measured. The average of 10 measurements is taken as the robotic arm's response delay (e.g., 60ms); similarly, the conveyor belt's response delay is calibrated (e.g., 40ms). Based on the delay difference (the robotic arm is 20ms slower than the conveyor belt), the robotic arm command is issued 20ms in advance to generate the motion coordination command.

[0112] For example, in the parameter adjustment scheme, the robotic arm grasping angle is 15 degrees and the conveyor belt running speed is 0.35m / s. After calibration, the robotic arm response delay is 60ms and the conveyor belt response delay is 40ms. The robotic arm command is issued 20ms in advance to achieve synchronous start-up.

[0113] Driven by motion coordination commands, the packaging equipment operates simultaneously, activating the data acquisition module to collect real-time packaging sealing tension data (in N) and edge bonding gap data (in mm). The acquisition frequency is 3 times per second from the force sensor (consistent with the update data acquisition frequency after command execution). A data window consisting of 10 continuously acquired sampling points (approximately 3.33 seconds) is used to apply median filtering (window size 3, which effectively filters out single-point impulse noise without losing trend information) to both the tension and gap sequences, eliminating isolated outliers. The filtered results in valid tension and gap sequences, serving as material deformation characteristics.

[0114] For example, the collected tension sequence is [25,26,27,100,26,25,27,26,25,26]N. After median filtering (window 3) and removing 100N, we get [25,26,27,26,25,27,26,25,26]N. The gap sequence [0.3,0.2,0.4,0.3,0.2,0.3,0.4,0.3,0.2,0.3]mm remains unchanged after filtering.

[0115] Based on the material deformation characteristics (tension sequence and gap sequence), the tension fluctuation amplitude is calculated, which is the difference between the maximum and minimum values ​​in the sequence (unit: N); the gap closing rate is calculated, which is the number of times the gap value changes per unit time (an absolute difference between adjacent sampling points greater than 0.05 mm is counted as one change, and the total number of changes is divided by the window time length, unit: times / second). A numerical calculation model for packaging integrity is constructed: Integrity = 100 × [1 - (tension fluctuation amplitude / 10N × 0.6 + gap closing rate / 5 times / second × 0.4)], where 10N is the maximum allowable tension fluctuation of the equipment (set according to 10% of the rated load of the robotic arm), and 5 times / second is the maximum allowable closing rate of the process (set according to the heat sealing time of the packaging material). The weights 0.6 and 0.4 are determined based on the contribution ratio of the two types of factors to the sealing failure in historical failure statistics. The integrity result is truncated between 0 and 100, with higher values ​​indicating more intact packaging.

[0116] For example, if the tension fluctuation amplitude is 5N and the gap closing rate is 0.1 times / second, then the integrity is equal to 100×[1-(5 / 10×0.6+0.1 / 5×0.4)]=100×(1-(0.3+0.008))=69.2 points.

[0117] The preset threshold for judging packaging integrity is 85 points. This threshold is set based on the lower quartile (80 points) of the integrity distribution of 200 historical qualified products, rounded up and with a safety margin of 5 points, to ensure that no less than 95% of qualified products meet the standard. If the calculated packaging integrity value is higher than 85 points, the production line is considered to be in an adaptive stable state; if the value is lower than or equal to 85 points, the gripping force of the robotic arm is increased (by 5N each time) or the conveyor belt speed is decreased (by 0.05m / s each time) according to the degree of deviation, and steps S4 to S7 are repeated until the integrity standard is met. At the same time, the current working condition data is recorded for subsequent parameter optimization.

[0118] For example, if the preset integrity threshold is 85 points and the current integrity score is 69.2 points, which is lower than the threshold, the gripping force of the robotic arm will be increased from 80N to 85N, and the parameter adjustment and detection will be re-executed; if the integrity score reaches 90 points after adjustment, it will be considered stable.

[0119] In summary, this invention discloses an automated control method for an intelligent packaging production line, including multi-station visual and force data acquisition and fusion, abnormal pattern recognition and product switching feature determination, key feature extraction and equipment parameter optimization, control command execution and collaborative deviation analysis, updated data parsing and deviation iterative optimization, servo drive coordination and packaging quality detection, integrity assessment and adaptive stable state determination, etc., achieving multi-dimensional working condition collaborative perception, intelligent product switching adaptation, dynamic correction of execution parameters, precise control of packaging quality, and adaptive stable operation and efficient production assurance of the entire production line.

[0120] Reference Figure 2 The second embodiment of the present invention provides an automated control system for an intelligent packaging production line, comprising:

[0121] The data acquisition module is used to collect visual feedback data and force feedback data, preprocess and fuse the visual feedback data and force feedback data to obtain multi-station integrated information;

[0122] An anomaly identification module is used to identify anomaly patterns based on the multi-workstation integrated information. If the anomaly pattern matches preset historical switching data, the feature vector of the anomaly pattern is extracted, and a change classification label is determined based on the feature vector.

[0123] The parameter optimization module is used to obtain the equipment response time constraint of the production line if the change classification label is product switching, extract key feature data from the feature vector, denoise the key feature data to obtain product denoised data, predict the parameter optimization value of the equipment based on the product denoised data and the equipment response time constraint, and obtain the parameter adjustment scheme.

[0124] The instruction control module is used to issue control instructions according to the parameter adjustment scheme, correct the gripping force of the robotic arm and the speed of the conveyor belt according to the control instructions, obtain the execution coordination deviation, perform hierarchical analysis based on the execution coordination deviation, and generate update instructions.

[0125] The deviation analysis module is used to execute the update instruction to collect visual update data and force update data, analyze the deviation of the visual update data according to priority, analyze the deviation of the force update data based on dynamic threshold, and fuse the analysis results to obtain the data deviation value.

[0126] The iterative optimization module is used to perform iterative calculations by combining the preset deviation library with the data deviation value if the data deviation value exceeds a preset deviation threshold, so as to obtain a parameter adjustment scheme.

[0127] The stability determination module is used to adjust the coordination of the execution of the scheme according to the parameters, monitor the integrity of the packaging in real time, and obtain the adaptive stable state of the production line.

[0128] It should be noted that the intelligent packaging production line automation control system provided in this embodiment of the invention is used to execute all process steps of the intelligent packaging production line automation control method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0129] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0130] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An automated control method for an intelligent packaging production line, characterized in that, include: Visual feedback data and force feedback data are collected, preprocessed and fused to obtain multi-station integrated information; Based on the integrated information of the multi-workstation, an abnormal pattern is identified. If the abnormal pattern matches the preset historical switching data, the feature vector of the abnormal pattern is extracted, and the change classification label is determined based on the feature vector. If the change classification label is product switching, then the equipment response time constraint of the production line is obtained, and key feature data is extracted from the feature vector. The key feature data is denoised to obtain product denoised data. Based on the product denoised data and the equipment response time constraint, the parameter optimization value of the equipment is predicted to obtain the parameter adjustment scheme. According to the parameter adjustment scheme, control instructions are issued, the gripping force of the robotic arm and the speed of the conveyor belt are corrected according to the control instructions, the operating parameters of the robotic arm and the conveyor belt are collected, and the deviation value between the operating parameters and the target value in the parameter adjustment scheme is calculated. For each deviation value, calculate the magnitude and direction of change, and fit the deviation change rate based on the deviation values ​​at the most recent multiple sampling times, combining them into an error feature vector; Based on the combination of deviation magnitude and rate of change in the error feature vector, a fine-tuning strategy or a timing correction strategy is determined, and an update instruction is generated. The update instruction is executed to collect visual update data and force update data. The visual update data is analyzed for deviation according to priority, and the force update data is analyzed for deviation based on dynamic threshold. The analysis results are fused to obtain the data deviation value. If the data deviation value exceeds the preset deviation threshold, the deviation data exceeding the deviation threshold are filtered to construct a deviation feature vector. Two historical records in the preset deviation library are found whose data deviation values ​​before the iteration are less than and greater than the current data deviation value, respectively. Linear interpolation is used to calculate the current adjustment amount of the robotic arm gripping force and the adjustment amount of the conveyor belt speed to obtain the parameter adjustment scheme. The actions are coordinated according to the parameter adjustment scheme, and the integrity of the packaging is monitored in real time to obtain the adaptive stable state of the production line.

2. The automated control method for an intelligent packaging production line according to claim 1, characterized in that, The collected visual feedback data and force feedback data are preprocessed and fused to obtain multi-station integrated information, including: Visual feedback data and force feedback data are collected, and outlier removal and normalization are performed on the visual feedback data and force feedback data to obtain cleaned data. The cleaned data is aligned sequentially according to a unified timestamp to construct a data fusion matrix; The data fusion matrix is ​​subjected to feature fusion to obtain multi-dimensional fused working condition features. The working condition data of each workstation are integrated based on the multi-dimensional fused working condition features to obtain multi-workstation integrated information.

3. The automated control method for an intelligent packaging production line according to claim 1, characterized in that, The step of identifying abnormal patterns based on the multi-workstation integrated information, and if the abnormal pattern matches preset historical switching data, then extracting the feature vector of the abnormal pattern, and determining a change classification label based on the feature vector, includes: The multi-station integrated information is compared with the preset normal fluctuation range in a time sequence. Continuous data segments that exceed the preset normal fluctuation range are filtered out, and the amplitude, frequency and trend characteristics of the continuous data segments are extracted to construct an abnormal pattern. The similarity between the abnormal pattern and the preset historical switching data is calculated to obtain a comparative analysis value. If the comparison analysis value reaches the preset similarity threshold, the feature vector of the abnormal pattern is extracted, and the change classification label is determined based on the feature vector. If the comparison analysis value does not meet the preset similarity threshold, the control parameters remain unchanged.

4. The automated control method for an intelligent packaging production line according to claim 1, characterized in that, If the change classification label is product switching, then the equipment response time constraint of the production line is obtained, and key feature data is extracted from the feature vector. The key feature data is then denoised to obtain product denoised data. Based on the product denoised data and the equipment response time constraint, the optimized parameter values ​​of the equipment are predicted to obtain a parameter adjustment scheme, including: If the change classification label is product switching, then principal component analysis is performed on the feature vector to obtain key feature data; The key feature data is denoised to filter out environmental noise and equipment vibration noise, resulting in denoised product data. Obtain the response time constraint parameters of the production line execution equipment, predict the optimized values ​​of the equipment parameters based on the product denoising data and the response time constraint parameters, and obtain the parameter optimization constraints; The product denoising data is input into a preset parameter prediction model, and the parameter optimization constraints are combined to solve the problem, thereby obtaining a parameter adjustment scheme.

5. The automated control method for an intelligent packaging production line according to claim 1, characterized in that, The process of executing the update instruction involves collecting visual update data and force update data, analyzing the deviation of the visual update data according to priority, analyzing the deviation of the force update data based on a dynamic threshold, and fusing the analysis results to obtain a data deviation value, including: Execute the update command to collect visual update data and force update data; The visual update data and the force update data are parsed and processed to obtain the visual deviation set and the force deviation set; Visual deviation features are extracted from the set of visual deviations, and a dynamic threshold is constructed based on the set of force deviations. The visual deviation features are prioritized to obtain ranked visual deviation features; the ranked visual deviation features are compared with a preset visual threshold, and the force perception deviation set is compared with the dynamic threshold. The deviation amount is calculated and weighted and integrated to obtain the data deviation value.

6. The automated control method for an intelligent packaging production line according to claim 1, characterized in that, The process of coordinating actions based on the parameter adjustment scheme, real-time monitoring of packaging integrity, and obtaining the adaptive stable state of the production line includes: The device execution parameters are generated according to the parameter adjustment scheme to form a servo drive sequence. The dynamic response delay of the actuator is analyzed through the servo drive sequence, and an action coordination command is generated based on the dynamic response delay. The packaging equipment is driven to operate according to the action coordination command, and the packaging sealing tension and edge bonding gap data are collected. The packaging sealing tension and edge bonding gap data are fused to obtain the material deformation characteristics. Based on the material deformation characteristics, the tension fluctuation amplitude and gap closing rate are calculated to obtain the packaging quality characterization characteristics; based on the packaging quality characterization characteristics, an integrity evaluation matrix is ​​constructed to obtain the packaging integrity value; By comparing the packaging integrity value with the preset state determination threshold, the adaptive stable state of the production line is obtained.

7. An automated control system for an intelligent packaging production line, characterized in that, include: The data acquisition module is used to collect visual feedback data and force feedback data, preprocess and fuse the visual feedback data and force feedback data to obtain multi-station integrated information; An anomaly identification module is used to identify anomaly patterns based on the multi-workstation integrated information. If the anomaly pattern matches preset historical switching data, the feature vector of the anomaly pattern is extracted, and a change classification label is determined based on the feature vector. The parameter optimization module is used to obtain the equipment response time constraint of the production line if the change classification label is product switching, extract key feature data from the feature vector, denoise the key feature data to obtain product denoised data, predict the parameter optimization value of the equipment based on the product denoised data and the equipment response time constraint, and obtain the parameter adjustment scheme. The instruction control module is used to issue control instructions according to the parameter adjustment scheme, correct the gripping force of the robotic arm and the speed of the conveyor belt according to the control instructions, collect the operating parameters of the robotic arm and the conveyor belt, and calculate the deviation value between the operating parameters and the target value in the parameter adjustment scheme. For each deviation value, calculate the magnitude and direction of change, and fit the deviation change rate based on the deviation values ​​at the most recent multiple sampling times, combining them into an error feature vector; Based on the combination of deviation magnitude and rate of change in the error feature vector, a fine-tuning strategy or a timing correction strategy is determined, and an update instruction is generated. The deviation analysis module is used to execute the update instruction to collect visual update data and force update data, analyze the deviation of the visual update data according to priority, analyze the deviation of the force update data based on dynamic threshold, and fuse the analysis results to obtain the data deviation value. The iterative optimization module is used to filter out the deviation data that exceeds the preset deviation threshold to construct a deviation feature vector if the data deviation value exceeds the preset deviation threshold. It also searches for two historical records in the preset deviation library where the data deviation value before the iteration is less than and greater than the current data deviation value, respectively. Linear interpolation is used to calculate the current adjustment amount of the robotic arm gripping force and the adjustment amount of the conveyor belt speed to obtain the parameter adjustment scheme. The stability determination module is used to adjust the coordination of the execution of the scheme according to the parameters, monitor the integrity of the packaging in real time, and obtain the adaptive stable state of the production line.

Images