Intelligent glass bottle packaging line operation management system

Abstract

This invention relates to the field of intelligent manufacturing technology, specifically disclosing an intelligent glass bottle packaging line operation management system. The system synchronously collects operational data from multiple production lines, including production status and material physical properties. By comparing real-time data with historical trajectories to identify collaborative deviation patterns, iteratively generates local control commands. These commands and key feature data are encapsulated into standard encrypted units using homomorphic and standard encryption, and uploaded to a central node for homomorphic aggregation and decryption analysis without decryption, constructing and continuously updating a centralized experience knowledge set. When a production line processes new materials, optimized control commands under similar historical operating conditions are matched from this set based on the material's physical parameters. After virtual simulation and physical security verification, these commands are deployed to the main process. Based on feedback from the actual effects of command application, the historical association weights in the knowledge set are corrected using confidence weighting. This invention achieves adaptive and reliable control of production lines for variable materials and secure evolution of global knowledge.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, specifically to an intelligent glass bottle packaging line operation and management system. Background Technology

[0002] With the development of industrial automation and intelligence, glass bottle packaging production lines have widely adopted programmable logic controllers (PLCs), industrial robots, and sensor networks to achieve basic automation. To further improve efficiency, quality, and flexibility, technologies such as digital twins, the Internet of Things (IoT), and big data analytics have been introduced into operations management, forming monitoring and optimization systems at the individual production line level. These systems collect production line operation data by deploying sensors and perform simulations, predictions, and optimizations based on preset or self-learning models to achieve control over processes such as filling, capping, and labeling.

[0003] The existing technology has the following shortcomings: When glass bottle packaging production lines deal with rapid changes in glass bottles of different materials, the slight and unknown changes in the physical properties of the materials (such as the coefficient of friction and the modulus of elasticity) can cause the control system based on a fixed model to malfunction, leading to technical problems such as product breakage, equipment failure, or frequent shutdowns for debugging. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent glass bottle packaging line operation management system to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: The intelligent glass bottle packaging line operation management system includes: The operational data collaborative acquisition module simultaneously acquires real-time operational data from multiple glass bottle packaging lines. The operational data includes production parameters that characterize the operating status of the production line, as well as physical parameters that characterize the properties of the processed materials. The abnormal operating condition identification and local optimization module analyzes the operational data acquired by each glass bottle packaging line to determine whether there are any abnormal operating conditions that deviate from the preset standards. When an abnormal operating condition is identified, an independent optimization process is initiated based on the current operational data to generate a set of local control instructions adapted to the corresponding abnormal operating condition. The knowledge fragment generation and encryption module encapsulates and converts local control commands, along with the operational data that triggers the generation of local control commands, into standardized encrypted data units. The distributed knowledge aggregation and construction module gathers multiple encrypted data units from different glass bottle packaging lines, and performs fusion analysis through data processing to build a centrally stored set of experience knowledge. The similar working condition strategy matching module, when any glass bottle packaging line needs to process materials with new physical parameter characteristics, matches the control instructions corresponding to the historical records of similar parameters from the experience knowledge set based on the physical parameters of the materials. The strategy verification and deployment execution module simulates and verifies the obtained control commands on the corresponding glass bottle packaging line. After the verification is successful, the commands are put into actual application, and new operational data after the application is collected to evaluate the actual effect of the control commands. The effect evaluation and knowledge feedback module generates feedback information based on the actual effect of the evaluation and applies it to the experience knowledge set to correct and enhance the accuracy of the correlation between parameters and instructions within the experience knowledge set.

[0006] As a further aspect of the present invention: the generation of a set of local control instructions adapted to corresponding abnormal operating conditions specifically includes: Based on real-time acquired operational data, a reference data set is constructed that includes instantaneous data points and recent historical data points; By comparing real-time operational data with a set of reference data in multiple dimensions over time, a collaborative deviation pattern is identified that deviates continuously in the physical parameter dimension and is accompanied by synchronous disturbances in the production parameter dimension. Guided by the feature vector of the cooperative deviation mode, the parameter combination of the control command is fine-tuned in a targeted iterative manner within the preset safety parameter space; When the simulated output of operational data prediction results after iterative fine-tuning indicates that the collaborative deviation mode is suppressed, the current parameter combination is locked as a local control command adapted to the corresponding abnormal operating condition.

[0007] As a further aspect of the present invention: the process for identifying the cooperative deviation pattern is as follows: Extract the change trajectories of physical parameters and production parameters within the same time window from real-time operational data and reference data sets respectively; Calculate the morphological differences between the real-time change trajectory and the reference change trajectory across multiple consecutive time segments, and generate a set of difference data arranged in a time series. Correlation analysis was performed on the difference data to identify the time period in which the difference in physical parameters and the difference in production parameters increased synchronously and continued to exceed the preset threshold over time. The combination of multiple parameter change trajectories that are strongly correlated within the corresponding time period is identified as a cooperative deviation mode.

[0008] As a further aspect of the present invention: the conversion process of the encrypted data unit is as follows: From the operational data that triggers the generation of local control commands, extract key feature data to characterize the causal relationship between the collaborative deviation pattern and the command generation logic; By using encryption methods that support ciphertext computation, key feature data is encrypted and transformed to generate a first ciphertext data block that supports aggregation analysis without decryption. Convert local control commands and other operational data into a second encrypted data block; The first ciphertext data block, the second ciphertext data block, and the metadata header describing the data structure and encryption method are combined according to a preset encapsulation format to generate a standardized encrypted data unit.

[0009] As a further aspect of the present invention: the construction of a centrally stored set of experiential knowledge specifically includes: The aggregated encrypted data units are parsed to separate the first ciphertext data block generated by the encryption method that supports ciphertext computation; Without decryption, homomorphic aggregation operation is performed on the first ciphertext data block to generate aggregated ciphertext that represents the association between common deviation patterns and successful instructions; In the secure environment of the central processing system, the aggregated ciphertext is decrypted to obtain readable data on common relationships. Common relationship data is integrated with information from each data unit that describes the data structure and context, and stored in the empirical knowledge set in the form of structured mapping entries.

[0010] As a further aspect of the present invention: the process of generating the aggregated ciphertext is as follows: The parsed first ciphertext data blocks are format-aligned and blindly concatenated to form a unified ciphertext input sequence; By leveraging the homomorphic encryption algorithm's support for ciphertext addition and multiplication, the ciphertext components representing the same combination of physical and production parameters in the ciphertext input sequence are summed and averaged. Each ciphertext component, after being accumulated and averaged, is correlated with the corresponding local control command's encrypted identifier within the ciphertext space, and its weight is calculated accordingly. The calculated ciphertext result, which integrates statistical features and correlation weights from multiple data sources, is then repackaged into aggregated ciphertext.

[0011] As a further aspect of the present invention: the step of matching control commands with similar parameter historical records from the set of empirical knowledge specifically includes: The physical parameters of new materials are constructed into a multidimensional query vector, while the physical parameters of historical materials are extracted from the structured mapping entries of the experience knowledge set as a multidimensional reference vector. In the metric space composed of physical parameters, the composite similarity between the query vector and each reference vector is calculated. The composite similarity is composed of a weighted average of the proximity of parameter values ​​and the similarity of the association patterns between parameters. Based on the ranking results of composite similarity, historical mapping entries that exceed the preset similarity threshold are filtered out; The control commands or command combinations recorded in the filtered historical mapping entries are output as the matched control commands.

[0012] As a further aspect of the present invention: the obtained control commands are simulated and verified on the corresponding glass bottle packaging line, and after successful verification, they are put into practical application, specifically including: In a virtual operating environment built based on the current production line's precise parameters, the matched control commands and real-time collected production line status data are loaded to generate a predicted operating trajectory for the next production cycle. In the safety verification sub-units defined in the physical production line that do not affect the main process, the operation nodes in the control instructions are executed in sequence, and the actual feedback data during the execution process is collected synchronously. The actual feedback data is compared with the expected data of the corresponding nodes in the predicted running trajectory in multiple dimensions to calculate the degree of conformity between the overall running trend and key indicators. When the compliance rate exceeds the preset verification threshold, the verification is deemed successful. The control instructions are then fully deployed to the main process of the corresponding production line, and new operational data after the application is continuously collected.

[0013] As a further aspect of the present invention: the accuracy of the correlation between the internal parameters of the empirical knowledge set and the instructions is corrected and enhanced, specifically including: The new operational data after application is compared with the initial target expected data before the execution of the control instructions to generate an efficiency deviation vector characterizing the degree of improvement or degradation in each dimension. Based on the performance deviation vector, the historical association weights in the source mapping entries that triggered this instruction matching and application are calculated and reversed to generate weight adjustment amounts. Based on the reliability of the control commands verified in actual applications, a confidence level weight is assigned to the weight adjustment amount generated this time; Using the weight adjustment amount adjusted by confidence weight, the historical association weights of the corresponding source mapping entries in the empirical knowledge set are merged and updated, and the entry contribution scores of the empirical knowledge set are recalculated.

[0014] The beneficial effects of this invention are: (1) This invention identifies the “cooperative deviation pattern” between physical parameters and production parameters in real time, and uses this as a guide to quickly iterate and generate optimized control commands within a safe space. More importantly, it constructs a continuously evolving “set of experiential knowledge”. When dealing with new materials, it can immediately match successful strategies from similar historical working conditions as a high starting point, greatly shortening the self-learning and debugging cycle, and transforming “unknown problems” into “quasi-known problems”, thereby reducing downtime and product loss caused by production changeover debugging.

[0015] (2) This invention uses a “encrypted data unit” structure to process key feature data using homomorphic encryption technology, enabling the central node to perform secure aggregation operations (such as averaging and calculating correlation weights) on multi-source encrypted data without decryption, thereby extracting common and valuable knowledge patterns. This approach not only ensures the privacy and security of operational data from each production base, but also integrates the experience of multiple parties to build a more comprehensive and reliable knowledge base. This allows abnormal operating conditions encountered by a single production line to be transformed into the collective wisdom of the entire manufacturing network, achieving a simultaneous improvement in overall production capacity and knowledge level under the premise of safety. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 As shown, this invention is an intelligent glass bottle packaging line operation management system, comprising: The operational data collaborative acquisition module simultaneously acquires real-time operational data from multiple glass bottle packaging lines. The operational data includes production parameters that characterize the operating status of the production line, as well as physical parameters that characterize the properties of the processed materials. The abnormal operating condition identification and local optimization module analyzes the operational data acquired by each glass bottle packaging line to determine whether there are any abnormal operating conditions that deviate from the preset standards. When an abnormal operating condition is identified, an independent optimization process is initiated based on the current operational data to generate a set of local control instructions adapted to the corresponding abnormal operating condition. The knowledge fragment generation and encryption module encapsulates and converts local control commands, along with the operational data that triggers the generation of local control commands, into standardized encrypted data units. The distributed knowledge aggregation and construction module gathers multiple encrypted data units from different glass bottle packaging lines, and performs fusion analysis through data processing to build a centrally stored set of experience knowledge. The similar working condition strategy matching module, when any glass bottle packaging line needs to process materials with new physical parameter characteristics, matches the control instructions corresponding to the historical records of similar parameters from the experience knowledge set based on the physical parameters of the materials. The strategy verification and deployment execution module simulates and verifies the obtained control commands on the corresponding glass bottle packaging line. After the verification is successful, the commands are put into actual application, and new operational data after the application is collected to evaluate the actual effect of the control commands. The effect evaluation and knowledge feedback module generates feedback information based on the actual effect of the evaluation and applies it to the experience knowledge set to correct and enhance the accuracy of the correlation between parameters and instructions within the experience knowledge set.

[0020] In the operational data collaborative acquisition module, sensing devices and measurement units are deployed at key physical locations on each glass bottle packaging production line to simultaneously capture two types of data streams. Production parameters characterizing the production line's operating status are obtained through real-time measurements of the conveyor belt speed by encoders, filling head pressure monitored by pressure transmitters, internal temperature of the sterilization tunnel measured by thermocouples, and bottle counting pulses triggered by photoelectric sensors.

[0021] Physical parameters characterizing the properties of the processed material are acquired using specialized measuring equipment deployed at the feeding end or online inspection station. For glass bottles, the surface friction coefficient is obtained through online sampling measurement using a customized inclined plane testing unit. This unit automatically tests a sample bottle on the production line every 60 seconds, recording the critical angle at which it begins to slide and converting it into a static friction coefficient value according to Coulomb's friction formula. The equivalent elastic modulus of the bottle body is estimated using a non-contact laser micro-displacement sensor in conjunction with a known micro-pneumatic loading device. The calculation is performed by measuring the deformation of a specific point on the bottle body before and after applying a small force, combined with the theoretical thickness of the bottle wall.

[0022] Data streams from different sources on the same production line need to be synchronized and aligned in time. Using a unified time server as a reference, each collected data point is timestamped with millisecond-level precision. For non-continuously measured physical parameters (such as the friction coefficient), their data points are inserted into the continuously collected production parameter data sequence according to the timestamp, forming a time-sorted, multi-dimensional hybrid data sequence, known as the real-time operation data stream of the production line.

[0023] Real-time operational data streams undergo preprocessing to ensure data quality. Production parameter sequences first pass through a real-time filter based on the absolute deviation of the median to remove transient spikes caused by electromagnetic interference. For physical parameters, their values ​​are calculated using a moving average and standard deviation based on the most recent 10 measurements. If the latest measurement exceeds the boundary of plus or minus three times the standard deviation of the moving average, the value is considered an outlier and marked, but is still retained in the data stream for subsequent analysis.

[0024] The real-time operational data stream from each production line is encapsulated into standard-format data packets. Each packet consists of a header and a payload. The header includes the production line number, start and end timestamps of data acquisition, and data dimension information; the payload is a pre-processed sequence of numerical parameters arranged in a preset order. These data packets are simultaneously and continuously transmitted to a designated central data processing node via a dedicated industrial communication network, providing consistent input for subsequent analysis and decision-making.

[0025] In the abnormal operating condition identification and local optimization module, a reference data set for comparison is constructed. Using the current moment as a baseline, real-time operational data collected within the last 30 seconds is selected consecutively to form the instantaneous data point set. Simultaneously, the 10 most recent continuous production segments marked as "normal production" under the same product specifications are retrieved from the historical database. Each segment is also 30 seconds long. The operational data parameter values ​​at the same time points in these segments are arithmetically averaged to generate an averaged historical data trajectory, which serves as the recent historical data point set. The instantaneous data point set and the averaged historical data trajectory together constitute the reference data set.

[0026] Multi-dimensional time-series comparisons are performed to identify collaborative deviation patterns. Data within the most recent 5-second time window is extracted from the average historical trajectory of the current real-time operational data stream and the reference data set. For each monitored parameter, such as filling pressure and friction coefficient, a series of values ​​within 5 seconds are connected chronologically to form the parameter's change trajectory. The morphological difference between the real-time change trajectory and the reference change trajectory is calculated. Specifically, the sum of the absolute values ​​of the differences between the two at each corresponding time point is calculated, and then divided by the total number of time points to obtain a difference value representing the overall degree of deviation. This calculation is performed separately for the friction coefficient (physical parameter dimension) and the filling pressure (production parameter dimension), resulting in a friction coefficient difference sequence and a filling pressure difference sequence over time.

[0027] Analyze the difference data to locate time periods of coordination deviation. Perform correlation analysis on the two difference sequences mentioned above, calculating their product at each sampling time. Set a judgment threshold of 0.05, which is determined by taking the average value plus three standard deviations of the difference product values ​​in the past 100 normal production cycles. If the difference product value exceeds this judgment threshold for two consecutive seconds, and the values ​​of both difference sequences are consistently higher than their respective historical baselines, then this two-second period is judged as a potential abnormal period. If such periods occur consecutively and the total duration exceeds 5 seconds, then a complete coordination deviation pattern is confirmed. The characteristics of this pattern are described by the original change trajectories of all parameters within this period.

[0028] Iterative fine-tuning of control commands is performed based on the characteristics of cooperative deviation patterns. The characteristics of the currently identified cooperative deviation patterns are converted into a set of direction vectors to guide the search. A safety parameter space is preset, defined by the upper and lower limits of key control command parameters such as robot gripping force and vacuum suction pressure. Starting from the currently used combination of control command parameters, a new parameter combination is generated according to the correction direction indicated by the aforementioned direction vectors, with each adjustment not exceeding 5% of the original value. This new parameter combination is input into a simulation program based on the precise dynamic relationship of the current production line to predict the changes in production and physical parameters over the next 10 seconds.

[0029] Lock the optimized local control command. The simulation program outputs the parameter change trajectory obtained from the deduction, and the morphological difference between this trajectory and the normal reference trajectory is recalculated. If, after a maximum of 10 iterations of fine-tuning, the parameter differences involved in the cooperative deviation mode in the deduction results corresponding to the new parameter combination all decrease to below 50% of the judgment threshold, then the deviation mode is determined to be effectively suppressed in the simulation. At this time, the parameter combination generated in this iteration is locked as the local control command adapted to the current abnormal operating condition and is prepared to be issued to the production line execution unit. If the suppression condition is still not met after reaching the maximum number of iterations, the parameter combination with the largest decrease in difference is selected as the output command.

[0030] In the knowledge fragment generation and encryption module, key feature data is extracted from the operational data that triggers the generation of local control commands. This data aims to establish a causal logical relationship between the cooperative deviation pattern and the ultimately generated control commands. Specifically, the extracted content includes: a list of the names of the specific physical and production parameters involved in the identified cooperative deviation pattern; the total duration of the pattern; the peak and trough values ​​of the parameter variation trajectories within the pattern; and the main component values ​​of the guiding vector used to indicate the correction direction when generating control commands. These data collectively constitute a structured feature set.

[0031] Using the Paillier homomorphic encryption algorithm, which supports ciphertext computation, the aforementioned key feature data is encrypted and transformed to generate the first ciphertext data block. The specific process is as follows: A Paillier public key, pre-agreed upon by the central processing node and all local nodes, is obtained. Each numerical data in the feature set is converted into a fixed-length decimal number string. Using the public key, each number string is independently encrypted to obtain a corresponding ciphertext. All these ciphertexts are arranged and concatenated in a predefined order, with an index table indicating the meaning and length of the original ciphertext data appended to the header, ultimately forming the first ciphertext data block. The key feature of this data block is that it allows homomorphic addition and multiplication operations to be performed directly on the ciphertext without decryption.

[0032] Local control commands and other non-critical operational data are converted into a second ciphertext data block. Control commands are presented as parameter lists, while other operational data includes contextual information such as device serial numbers and complete timestamps. An Advanced Encryption Standard (AES) algorithm is used, employing a locally unique, secure session key periodically distributed by the central processing node to encrypt the entire data set. The encrypted binary data stream constitutes the second ciphertext data block.

[0033] The two ciphertext data blocks mentioned above are combined and encapsulated with the metadata header. The metadata header is in plaintext and includes: the length of the first ciphertext data block, the length of the second ciphertext data block, the Paillier public key version identifier used, the Advanced Encryption Standard (AES) algorithm identifier used, the generation timestamp of the encrypted data unit, and its sequence number. The data is then concatenated in binary format according to a fixed order: "metadata header length value, metadata header content, first ciphertext data block, second ciphertext data block," to generate the final standardized encrypted data unit. This unit serves as a complete knowledge carrier, suitable for subsequent transmission and aggregation processing.

[0034] In the distributed knowledge aggregation and construction module, multiple encrypted data units are parsed and separated. After receiving encrypted data units from different production lines, the central processing node parses them according to the fixed format recorded in the metadata header of each unit. From each unit, based on the length information of the first ciphertext data block, the first ciphertext data block generated by the Paiye homomorphic encryption algorithm is precisely separated. Simultaneously, descriptive information such as the production line number, timestamp, and public key version identifier corresponding to the unit is extracted from the metadata header for subsequent use.

[0035] Without decryption, homomorphic aggregation is performed on the separated first ciphertext data blocks. Specifically, based on the index table attached to each first ciphertext data block, ciphertext components representing the same physical or production parameter (such as "friction coefficient_mean") in all data blocks are located and aligned. Since the Paiye algorithm allows direct mathematical operations on the ciphertext, the central processing node uses the corresponding public key to perform ciphertext addition on all ciphertext components representing the same parameter from different data blocks, obtaining the sum ciphertext for that parameter. Then, this sum ciphertext is divided by the total number of ciphertext components involved in the calculation (this "division by a constant" operation is achieved in the Paiye algorithm by performing a specific number of exponentiations on the ciphertext), thus obtaining the average ciphertext for that parameter. This accumulation and averaging operation is performed on all relevant parameters.

[0036] The process involves calculating correlation weights and encapsulating aggregated ciphertext. In ciphertext mode, a correlation weight value is calculated, quantifying the correlation strength between a specific parameter deviation pattern and its corresponding control command. Specifically, each parameter ciphertext, after averaging, is paired with an encrypted control command identifier from the same original data unit. Within the ciphertext space, the product of each pair of "parameter mean ciphertext" and "command identifier ciphertext" is calculated; the value of this product ciphertext is used as the ciphertext representation of the correlation weight. All calculated parameter mean ciphertexts and their corresponding correlation weight product ciphertexts are recombine in a predetermined order, and new aggregation metadata (such as the number of units involved in the calculation and the aggregation time) is appended, ultimately encapsulating them into a new, larger ciphertext data packet, i.e., the aggregated ciphertext.

[0037] Within the secure decryption environment of the central processing node, the aggregated ciphertext is decrypted using a pre-stored Paiye private key paired with the public key. The decryption operation restores the statistical averages of all parameters (e.g., average friction coefficient deviation) and the correlation weights between each average and the control command identifier in a single operation. These restored statistical and correlational data, derived from data from multiple production lines, constitute readable common correlation data.

[0038] The common correlation data is integrated with the contextual description information of each original encrypted data unit. Using the decrypted common data as the core, a structured mapping entry is created. This entry contains the following fields: a feature summary of the described common deviation pattern (described by the average deviation value of key parameters), a list of identifiers of several control instructions most strongly correlated with this pattern, the correlation weight of each instruction, the generation time of this entry, and the number and source number of the original data units on which this entry was based. This mapping entry is added to a searchable database table, which serves as a continuously updated set of experiential knowledge.

[0039] In the similar operating condition strategy matching module, query vectors and reference vectors are constructed. When a packaging line needs to process new materials, the physical parameters of the material are obtained, including the surface static friction coefficient, equivalent elastic modulus, and bottle weight. These parameter values ​​are arranged in a fixed order to form a numerical list, called the multidimensional query vector. Simultaneously, from each structured mapping entry in the empirical knowledge set, the average values ​​of the key physical parameters corresponding to the common deviation patterns described by that entry are extracted and arranged in exactly the same order to form multiple multidimensional reference vectors.

[0040] Calculate the composite similarity between the query vector and each reference vector. This similarity consists of a weighted sum of two parts. The first part is the parameter value proximity score, calculated by first calculating the absolute value of the difference between the query vector and the reference vector in each corresponding dimension, then summing the absolute values ​​of the differences across all dimensions and taking the reciprocal. The second part is the parameter association pattern similarity score, calculated by treating the query vector and reference vector as separate entities, calculating their cosine similarity (i.e., summing the products of corresponding components of the two vectors, then dividing by the product of the square roots of the sum of the squares of all components of each vector). The final composite similarity score is the sum of the parameter value proximity score (70% weight) and the association pattern similarity score (30% weight).

[0041] Based on the calculation results, select historical mapping entries that meet the criteria. Sort the composite similarity calculation results corresponding to all reference vectors from highest to lowest. The preset similarity threshold is set to 0.75, which is determined based on the statistical analysis of one hundred successful matching cases in history, taking the lower quarter quantile of their similarity values. From the sorted results, select all historical mapping entries with a composite similarity value greater than or equal to 0.75. If the selection results are empty, relax the threshold to the first-ranked similarity value, but not lower than 0.6.

[0042] Output the matched control commands. Read the identifiers of the top three control commands with the highest association weights recorded in all filtered historical mapping entries. Output the composite similarity values ​​of these command identifiers and their source entries as the result of this matching. This result provides a clear, knowledge-based validated set of candidate control commands for subsequent simulation verification.

[0043] In the strategy verification and deployment execution module, a virtual operating environment is constructed and a predicted operating trajectory is generated. This environment is built based on the precise physical parameters of the current production line, obtained through equipment files and sensor calibration data. Specifically, these parameters include the length and mass distribution of each joint of the robotic arm, the conveyor belt transmission ratio, and the maximum acceleration and deceleration of each motion axis. Matched control commands, such as the target clamping force and motion speed curves, along with real-time collected production line status data, such as the current bottle position and robotic arm posture, are input into a deduction program based on production line dynamics and kinematics. This program calculates the expected state of each actuator and the expected motion path of the bottle within the next ten seconds in 1%-second time steps, and connects the expected values ​​of all key parameters in chronological order to form the predicted operating trajectory.

[0044] Commands are executed and feedback data is collected on the physical production line safety verification subunit. This subunit is an independent, small-scale, reproducible testing device located bypassing the main process. Its mechanical structure is identical to the key workstations in the main process, but its operating speed and range are limited. Control commands are decomposed into a series of ordered operation nodes, such as "move to position A" and "apply force F". These nodes are executed strictly in sequence on this subunit. Simultaneously, its high-frequency force sensor and encoder collect the actual clamping force values, actual position coordinates, and timestamps during execution at a sampling rate of no less than 500 Hz, forming an actual feedback data sequence.

[0045] The actual data is compared with the predicted data, and the degree of conformity is calculated. The actual feedback data sequence is compared with the expected data at the same time points in the predicted operating trajectory in multiple dimensions. For each parameter being compared, such as clamping force, the absolute value of the difference between the actual and predicted values ​​is calculated at each comparison time point. Then, the average of these differences over all time points is calculated as the deviation of that parameter. The degree of conformity between the overall operating trend and key indicators is obtained by calculating the weighted average of the deviations of all compared parameters and then subtracting it, where the weights are pre-set based on the degree of influence of each parameter on production quality.

[0046] Deployment and data collection are executed based on the compliance assessment results. The preset verification threshold is 90%. If the calculated compliance rate is greater than or equal to this threshold, the verification is considered successful. Subsequently, the control instruction is fully issued to the actual execution agency of the corresponding main process on the production line, overriding the original instruction. From the time the new instruction takes effect, continuous collection of main process operational data resumes. The data collected at this time is the new operational data used to evaluate the effectiveness of the control instruction in actual application. If the compliance rate does not reach the threshold, the instruction is marked as a verification failure, triggering a re-matching or manual intervention process.

[0047] In the effectiveness evaluation and knowledge feedback module, an effectiveness deviation vector is generated to characterize the actual effect of control instructions. New operational data collected after applying control instructions is compared with the initial target data set before executing the instructions. For each monitored key indicator, such as the percentage increase in product qualification rate, the rate of change in production capacity per unit time, or the stability of a specific production parameter, the difference between the actual achieved value and the expected target value is calculated, and then divided by the absolute value of the expected target value to obtain the percentage improvement or decline of that indicator. All percentage values ​​of the evaluated indicators are arranged in a fixed order to form a numerical list, which is the effectiveness deviation vector. If the actual value of an indicator is better than expected, the corresponding percentage is positive; otherwise, it is negative.

[0048] Based on the aforementioned performance deviation vector, the weight adjustment for the relevant source mapping entries is calculated. From the empirical knowledge set, the historical mapping entry from which the control command of this application originates is located, and the current historical association weight of the control command in that entry is obtained. For that entry, the list of key physical parameters it records is read. Based on the percentage value of the indicators most relevant to these parameters in the performance deviation vector (e.g., if the entry involves the friction coefficient, a production stability indicator related to friction is selected), the historical association weight of that entry is reversed. Specifically, a temporary adjustment is generated, equal to the current historical association weight multiplied by the negative of the selected performance deviation percentage value, and then multiplied by 0.5. This adjustment is the weight adjustment generated for this application.

[0049] A confidence weight is assigned to the calculated weight adjustment. This confidence weight is determined based on the overall compliance rate calculated during the verification phase prior to deployment of the control instruction. This compliance rate is a value between 0 and 1. The confidence weight is calculated as the square of the compliance rate. This means that the more accurate the instruction performs in simulation verification, the greater the impact of its subsequent practical effects on the correction of the knowledge base.

[0050] The empirical knowledge set is updated using the adjusted weights, and the contribution of each entry is recalculated. The weight adjustment obtained in the previous step is multiplied by its confidence weight to obtain a final adjusted weight. This final adjusted weight is added to the historical association weight of the instruction in the source mapping entry to obtain the updated weight. Subsequently, the contribution score of each entry is recalculated based on the latest weights of all mapping entries, the number of times it has been successfully applied in retrieval, and the recentity of its generation time. The contribution score is calculated by multiplying the entry's weight by 1.0, adding its number of successful applications multiplied by 0.3, and adding a decay coefficient based on time recentity (e.g., the base contribution score decreases by 0.05 for every 30 days). These three values ​​are then added together to obtain the total score. The empirical knowledge set is then sorted according to the new contribution scores to optimize subsequent retrieval efficiency.

[0051] The working principle of this invention is as follows: Real-time operational data, including production status and material physical properties, is synchronously collected from multiple production lines and then time-aligned and preprocessed for quality. By comparing real-time data with historical normal trajectories in a multi-dimensional time series, abnormal operating conditions where physical parameters and production parameters deviate in tandem are identified. Based on this, local control commands are generated through iterative fine-tuning within a safety parameter space. The commands and associated abnormal feature data are encapsulated into aggregable encrypted data units using a combination of homomorphic and standard encryption. Homomorphic operations and aggregation of multi-source encrypted units are performed at the central node without decryption. After decryption, a new data structure is constructed. It updates the centralized set of experience knowledge; then, when the production line processes new materials, it matches optimized control instructions from similar historical working conditions based on their physical parameters from the knowledge set, and first conducts simulation verification in a virtual environment and physical safety verification unit based on precise dynamics. After passing the verification, it is deployed to the main process; feedback is generated based on the actual effect data after the instruction is applied, and the corresponding historical association weights in the knowledge set are corrected and updated with confidence weighting, thus forming a complete self-learning operation management system from distributed perception and privacy protection collaborative knowledge construction to precise matching verification and final closed-loop feedback optimization.

[0052] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent glass bottle packaging line operation management system, characterized in that, include: The operational data collaborative acquisition module simultaneously acquires real-time operational data from multiple glass bottle packaging lines. The operational data includes production parameters that characterize the operating status of the production line, as well as physical parameters that characterize the properties of the processed materials. The abnormal operating condition identification and local optimization module analyzes the operational data acquired by each glass bottle packaging line to determine whether there are any abnormal operating conditions that deviate from the preset standards. When an abnormal operating condition is identified, an independent optimization process is initiated based on the current operational data to generate a set of local control instructions adapted to the corresponding abnormal operating condition. The knowledge fragment generation and encryption module encapsulates and converts local control commands, along with the operational data that triggers the generation of local control commands, into standardized encrypted data units. The distributed knowledge aggregation and construction module brings together multiple encrypted data units from different glass bottle packaging lines, and performs fusion analysis through data processing to build a centrally stored set of experience knowledge. The similar working condition strategy matching module, when any glass bottle packaging line needs to process materials with new physical parameter characteristics, matches the control instructions corresponding to the historical records of similar parameters from the experience knowledge set based on the physical parameters of the materials. The strategy verification and deployment execution module simulates and verifies the obtained control commands on the corresponding glass bottle packaging line. After successful verification, the commands are put into actual application, and new operational data is collected after application to evaluate the actual effect of the control commands. The effect evaluation and knowledge feedback module generates feedback information based on the actual evaluation results and applies it to the experience knowledge set to correct and enhance the accuracy of the correlation between parameters and instructions within the experience knowledge set.

2. The intelligent glass bottle packaging line operation management system according to claim 1, characterized in that, The generation of a set of local control instructions adapted to corresponding abnormal operating conditions specifically includes: Based on real-time acquired operational data, a reference data set is constructed that includes instantaneous data points and recent historical data points; By comparing real-time operational data with a set of reference data in multiple dimensions over time, a collaborative deviation pattern is identified that deviates continuously in the physical parameter dimension and is accompanied by synchronous disturbances in the production parameter dimension. Guided by the feature vector of the cooperative deviation mode, the parameter combination of the control command is fine-tuned in a targeted iterative manner within the preset safety parameter space; When the simulated output of operational data prediction results after iterative fine-tuning indicates that the collaborative deviation mode is suppressed, the current parameter combination is locked as a local control command adapted to the corresponding abnormal operating condition.

3. The intelligent glass bottle packaging line operation management system according to claim 2, characterized in that, The process for identifying the cooperative deviation pattern is as follows: Extract the change trajectories of physical parameters and production parameters within the same time window from real-time operational data and reference data sets respectively; Calculate the morphological differences between the real-time change trajectory and the reference change trajectory across multiple consecutive time segments, and generate a set of difference data arranged in a time series. Correlation analysis was performed on the difference data to identify the time period in which the difference in physical parameters and the difference in production parameters increased synchronously and continued to exceed the preset threshold over time. The combination of multiple parameter change trajectories that are strongly correlated within the corresponding time period is identified as a cooperative deviation mode.

4. The intelligent glass bottle packaging line operation management system according to claim 1, characterized in that, The conversion process of the encrypted data unit is as follows: From the operational data that triggers the generation of local control commands, extract key feature data to characterize the causal relationship between the collaborative deviation pattern and the command generation logic; By using encryption methods that support ciphertext computation, key feature data is encrypted and transformed to generate a first ciphertext data block that supports aggregation analysis without decryption. Convert local control commands and other operational data into a second encrypted data block; The first ciphertext data block, the second ciphertext data block, and the metadata header describing the data structure and encryption method are combined according to a preset encapsulation format to generate a standardized encrypted data unit.

5. The intelligent glass bottle packaging line operation management system according to claim 1, characterized in that, The set of experiential knowledge for building centralized storage specifically includes: The aggregated encrypted data units are parsed to separate the first ciphertext data block generated by the encryption method that supports ciphertext computation; Without decryption, homomorphic aggregation operation is performed on the first ciphertext data block to generate aggregated ciphertext that represents the association between common deviation patterns and successful instructions; In the secure environment of the central processing system, the aggregated ciphertext is decrypted to obtain readable data on common relationships. Common relationship data is integrated with information from each data unit that describes the data structure and context, and stored in the empirical knowledge set in the form of structured mapping entries.

6. The intelligent glass bottle packaging line operation management system according to claim 5, characterized in that, The process of generating the aggregated ciphertext is as follows: The parsed first ciphertext data blocks are format-aligned and blindly concatenated to form a unified ciphertext input sequence; By leveraging the homomorphic encryption algorithm's support for ciphertext addition and multiplication, the ciphertext components representing the same combination of physical and production parameters in the ciphertext input sequence are summed and averaged. Each ciphertext component, after being accumulated and averaged, is correlated with the corresponding local control command's encrypted identifier within the ciphertext space, and its weight is calculated accordingly. The calculated ciphertext result, which integrates statistical features and correlation weights from multiple data sources, is then repackaged into aggregated ciphertext.

7. The intelligent glass bottle packaging line operation management system according to claim 1, characterized in that, The step of matching control commands with similar parameter historical records from the set of empirical knowledge specifically includes: The physical parameters of new materials are constructed into a multidimensional query vector, while the physical parameters of historical materials are extracted from the structured mapping entries of the experience knowledge set as a multidimensional reference vector. In the metric space composed of physical parameters, the composite similarity between the query vector and each reference vector is calculated. The composite similarity is composed of a weighted average of the proximity of parameter values ​​and the similarity of the association patterns between parameters. Based on the ranking results of composite similarity, historical mapping entries that exceed the preset similarity threshold are filtered out; The control commands or command combinations recorded in the filtered historical mapping entries are output as the matched control commands.

8. The intelligent glass bottle packaging line operation management system according to claim 1, characterized in that, The process of simulating and verifying the obtained control commands on the corresponding glass bottle packaging line, and then putting them into practical application after successful verification, specifically includes: In a virtual operating environment built based on the current production line's precise parameters, the matched control commands and real-time collected production line status data are loaded to generate a predicted operating trajectory for the next production cycle. In the safety verification sub-units defined in the physical production line that do not affect the main process, the operation nodes in the control instructions are executed in sequence, and the actual feedback data during the execution process is collected synchronously. The actual feedback data is compared with the expected data of the corresponding nodes in the predicted running trajectory in multiple dimensions to calculate the degree of conformity between the overall running trend and key indicators. When the compliance rate exceeds the preset verification threshold, the verification is deemed successful. The control instructions are then fully deployed to the main process of the corresponding production line, and new operational data after the application is continuously collected.

9. The intelligent glass bottle packaging line operation management system according to claim 1, wherein, The correction and enhancement of the accuracy of the correlation between the internal parameters of the experiential knowledge set and the instructions specifically includes: The new operational data after application is compared with the initial target expected data before the execution of the control instructions to generate an efficiency deviation vector characterizing the degree of improvement or degradation in each dimension. Based on the performance deviation vector, the historical association weights in the source mapping entries that triggered this instruction matching and application are calculated and reversed to generate weight adjustment amounts. Based on the reliability of the control commands verified in actual applications, a confidence level weight is assigned to the weight adjustment amount generated this time; Using the weight adjustment amount adjusted by confidence weight, the historical association weights of the corresponding source mapping entries in the empirical knowledge set are merged and updated, and the entry contribution scores of the empirical knowledge set are recalculated.

Images