Die cutting workshop production line intelligent management method and system

Abstract

The invention provides a die cutting workshop production line intelligent management method and system, and the method comprises the steps: collecting multi-dimensional state data to form an original matrix, carrying out the preprocessing and fusion of the original matrix, and generating a fused matrix; inputting the fused matrix into a pre-constructed depth perception model, and outputting a state coding vector and a channel attention vector; inputting the state coding vector as an environment state, constructing a deep reinforcement learning model, and outputting an action sequence and dynamic strategy distribution; combining the action sequence and the dynamic strategy distribution, constructing a dynamic process offset index, generating a process offset value, adjusting a current process parameter through a parameter correction network when a parameter correction value exceeds a preset threshold value, and triggering a fault response program when the real-time monitoring index is continuously higher than a dynamic threshold value for a certain period. And outputting a real-time early warning instruction sequence. According to the invention, automation and optimization of production scheduling, equipment fault prediction and quality control are realized.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent management of production lines, and in particular relates to an intelligent management method and system for a production line in a die-cutting workshop. Background Art

[0002] As manufacturing automation continues to improve, die-cutting workshops, as a crucial component of industrial production, face numerous challenges in terms of production efficiency, product quality, and equipment reliability. The core mission of a die-cutting workshop is to efficiently and precisely cut raw materials using die-cutting machines to produce products that meet quality requirements. However, existing die-cutting production lines still face several pressing challenges in actual operation, particularly in areas such as production scheduling, equipment failure prediction, quality control, and intelligent management.

[0003] Currently, production scheduling in die-cutting workshops primarily relies on manual operations or rule-based scheduling algorithms. These traditional scheduling methods are often unable to cope with the dynamic changes and complexity of production lines, resulting in inefficient production execution. Especially in complex production environments with multiple devices working together, multi-task scheduling, and complex production environments, traditional production scheduling methods struggle to effectively respond to changes in equipment status, fluctuations in materials, and urgent adjustments to production tasks. This leads to low production efficiency, prolonged equipment idle time, and frequent production bottlenecks. Furthermore, equipment failures and downtime remain a significant factor in the production process. Although some workshops have implemented sensor-based fault detection systems, these systems mostly rely on preset rules and thresholds, lacking sufficient flexibility and predictive power. They are unable to proactively identify potential equipment failures, resulting in frequent production line downtime and maintenance. Furthermore, equipment failures often impact subsequent production processes, delaying delivery times and impacting customer satisfaction.

[0004] In terms of quality control, although some workshops have introduced automated quality inspection equipment, such as visual inspection systems, most still rely on manual inspection or empirical judgment to ensure product quality. This approach is subject to uncertainty, especially when product quality fluctuates during the production process. Traditional quality control methods often fail to identify and adjust the production process in a timely manner, leading to the accumulation of quality issues and increased scrap rates. Current quality management flaws are mainly manifested in the following aspects: low inspection efficiency, delayed feedback, and the inability to adjust process parameters in real time to adapt to changes in production.

[0005] Therefore, existing technologies still face numerous bottlenecks in intelligent production line management, necessitating innovative solutions to improve production efficiency, equipment reliability, and quality control in die-cutting workshops. Traditional scheduling algorithms and quality management methods have limited effectiveness in dynamic production environments, failing to precisely schedule complex production tasks or comprehensively predict equipment status. Furthermore, they lack sufficient adaptive capabilities and real-time feedback mechanisms for quality control. Therefore, how to improve production efficiency, reduce downtime, and optimize quality control through intelligent means, particularly data-driven automated control methods, has become a pressing challenge for current die-cutting workshop management systems. Summary of the Invention

[0006] The purpose of this invention is to design an intelligent management method and system for the production line of a die-cutting workshop. Based on advanced intelligent perception and adaptive control technology, it monitors the production line status in real time and combines technologies such as deep learning and reinforcement learning to achieve automation and optimization of production scheduling, equipment failure prediction, and quality control.

[0007] In order to achieve the above-mentioned object, a first aspect of the present invention provides an intelligent management method for a die-cutting workshop production line, the method comprising:

[0008] Deploy multiple sensors at key workstations of the die-cutting production line to collect multi-dimensional state data to form an original matrix, pre-process and fuse the original matrix to generate a fused matrix;

[0009] Input the fused matrix into a pre-built depth perception model, extract local features using a dual-domain perception convolution kernel, process it with a channel attention module based on physical prior knowledge, and output a state encoding vector and a channel attention vector;

[0010] Taking the state encoding vector as the environment state input, building a deep reinforcement learning model, designing a reward function that includes a task-resource cross-fit index, and outputting an action sequence and a dynamic policy distribution;

[0011] Combining the action sequence and the dynamic strategy distribution, a dynamic process offset index is constructed to generate a process offset value. When the parameter correction value exceeds a preset threshold, the current process parameter is adjusted through a parameter correction network, and the adjusted new parameter value is output to form a parameter correction sequence.

[0012] A real-time monitoring indicator is constructed based on the process offset value and parameter correction sequence, and a dynamic threshold is set. When the real-time monitoring indicator is continuously higher than the dynamic threshold for a certain period, a fault response program is triggered and a real-time warning instruction sequence is output.

[0013] Furthermore, the preprocessing and fusing of the original matrix to generate a fused matrix specifically includes:

[0014] Obtaining the original matrix;

[0015] The original matrix is ​​subjected to time alignment of sensor data of different frequencies by linear interpolation, dimension differences are eliminated by normalization, and then the original matrix is ​​subjected to Frobenius norm and Regular sparse recovery calculation is performed to generate the fused matrix.

[0016] Furthermore, the dual-domain perception convolution kernel includes a short perception domain and a long perception domain, which respectively capture abnormal disturbances and trend changes.

[0017] Furthermore, the fused matrix is ​​input into a pre-built depth perception model, local features are extracted using a dual-domain perception convolution kernel, and a channel attention module based on physical prior knowledge is used for processing to output a state encoding vector and a channel attention vector, specifically including:

[0018] Obtaining the fused matrix;

[0019] Using the short receptive field and the long receptive field to perform convolution operations on the fused matrix and perform channel splicing to generate local features;

[0020] The local features are input into the channel attention model, and the state encoding vector and the channel attention vector are output; wherein the loss function of the channel attention model is Expressed as:

[0021] ;

[0022] in, The current training process The channel attention coefficient of each sensor, is the prior probability of the physical importance of the sensor given in the process drawing or experience library, is the Kullback-Leibler divergence, which measures the deviation between the attention distribution learned by the model and the physical assumption. is the weight adjustment coefficient of each channel, which is used to control its penalty intensity.

[0023] Furthermore, the deep reinforcement learning model consists of two fully connected layers, each equipped with a ReLU activation function and a normalization layer, whose output is a probability distribution of a set of discrete actions, each action corresponding to a set of scheduling decisions;

[0024] Among them, the scheduling strategy of the deep reinforcement learning model is completed through deep reinforcement learning training, and the PPO algorithm is used to update the strategy.

[0025] Furthermore, the reward function is expressed as:

[0026] ;

[0027] in, It is Instant rewards for every step; is the scheduling action generated by the policy network at the current time step; Indicates the resource utilization score, which measures the device utilization and average waiting time under the current action; The adaptability score measures the degree of match between the selected task and the device capabilities and sensitive channels in the current state; and is an adjustable weight parameter.

[0028] Furthermore, the combining of the action sequence and the dynamic strategy distribution to construct a dynamic process offset indicator and generate a process offset value specifically includes:

[0029] Using the multiple sensors to perform multi-dimensional data collection to generate fusion features for the current task;

[0030] Obtain the optimal features corresponding to the current task in the historical stable output cycle;

[0031] The degree of deviation of the current process state from its historical optimal working condition is calculated based on the optimal feature and the fusion feature as a process offset value.

[0032] Furthermore, the real-time monitoring indicator is used to reflect the system's recent performance based on the parameter correction amount. The state stability in a cycle; the dynamic threshold is obtained based on the mean and standard deviation statistics of the real-time monitoring indicators during the stable operation period of the system.

[0033] Furthermore, the real-time warning instructions include: sending load reduction / shutdown instructions to the target equipment, pushing warning notifications to on-site personnel, automatically deploying spare equipment to take over subsequent tasks, and packaging and archiving fault-related fragments.

[0034] In a second aspect of the present invention, there is provided an intelligent management system for a die-cutting workshop production line, the system comprising:

[0035] The data acquisition and fusion module is used to deploy multiple sensors at key workstations of the die-cutting workshop production line to collect multi-dimensional status data to form an original matrix, pre-process and fuse the original matrix to generate a fused matrix;

[0036] A depth perception modeling module is used to input the fused matrix into a pre-built depth perception model, extract local features using a dual-domain perception convolution kernel, process the local features using a channel attention module based on physical prior knowledge, and output a state encoding vector and a channel attention vector.

[0037] a scheduling optimization module, configured to take the state encoding vector as an environment state input, construct a deep reinforcement learning model, design a reward function including a task-resource cross-fit index, and output an action sequence and a dynamic policy distribution;

[0038] a quality control module, configured to combine the action sequence and the dynamic strategy distribution to construct a dynamic process deviation index, generate a process deviation value, and adjust the current process parameters through a parameter correction network when the parameter correction value exceeds a preset threshold, output the adjusted new parameter value, and form a parameter correction sequence;

[0039] The fault warning module is used to construct a real-time monitoring indicator based on the process offset value and parameter correction sequence, and set a dynamic threshold. When the real-time monitoring indicator is continuously higher than the dynamic threshold for a certain period, the fault response program is triggered and a real-time warning instruction sequence is output.

[0040] The beneficial technical effects of the present invention are at least as follows:

[0041] First, multi-dimensional sensor data fusion and a deep perception network are used to comprehensively perceive and monitor the production status of the die-cutting workshop in real time. By integrating data from multiple sensors, including temperature, humidity, pressure, vibration, and vision, the system accurately captures the health of equipment during production, changes in the production environment, and product quality, addressing the lack of comprehensive real-time monitoring and dynamic data analysis in traditional technologies.

[0042] Secondly, by introducing a reinforcement learning algorithm to dynamically optimize production scheduling, the system can autonomously adjust the priority of production tasks and the allocation of production line resources based on real-time status data from the production process, ensuring efficient execution of production tasks. This approach overcomes the problems of traditional scheduling algorithms, such as their inability to respond to changes in the production environment in real time and their difficulty in coping with complex scheduling requirements, thereby improving the overall operational efficiency of the production line.

[0043] Finally, this invention incorporates an adaptive control mechanism to achieve real-time optimization of die-cutting process parameters. The system automatically adjusts process parameters (such as tool pressure and cutting speed) based on real-time data feedback during the production process, thereby achieving precise quality control. This innovative approach overcomes the lag and difficulty in timely adjustment associated with traditional quality control methods, significantly improving product quality consistency and production efficiency.

[0044] In summary, this invention, through the combination of deep perception networks and reinforcement learning, provides an intelligent solution that dynamically optimizes production scheduling, accurately predicts equipment failures, and adaptively controls quality during production, ensuring efficient and stable operation of die-cutting workshops in complex and changing production environments. This solution not only improves production efficiency and reduces downtime, but also optimizes quality control, demonstrating its high level of intelligence and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0046] Figure 1 This is a flow chart of an intelligent management method for a die-cutting workshop production line according to the present invention. DETAILED DESCRIPTION

[0047] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0048] In one or more embodiments, Figure 1 As shown, a method for intelligent management of a die-cutting workshop production line is disclosed, and the method includes the following steps S1 to S5:

[0049] S1. Deploy multiple sensors at key workstations of the die-cutting workshop production line to collect multi-dimensional state data to form an original matrix, pre-process and fuse the original matrix to generate a fused matrix.

[0050] Specifically, this step deploys temperature, pressure, vibration, and vision sensors at multiple key workstations in the die-cutting workshop production line to collect multi-dimensional status data during the production process in real time. Considering that the die-cutting equipment will be affected by physical pressure, thermal deformation, and mechanical vibration during operation, single-dimensional data is not enough to fully describe its status. Therefore, we need to build a data structure that integrates multi-dimensional information to lay a data foundation for subsequent state perception modeling and scheduling optimization. The original sensor data is transmitted to the central data system in real time via the serial port, Modbus, or image acquisition card. At a certain point in time For example, from The data collected by the sensors constitute a feature vector , the observation data at all time points form the following matrix:

[0051] ;

[0052] in, Indicates in The first time point collected The data value of each sensor; is the number of sampling moments, is the total number of sensor types (such as pressure, temperature, image features, vibration RMS value, etc.).

[0053] The data dimensions, frequencies, and noise characteristics of the different dimensions in the original matrix vary significantly, necessitating unified preprocessing and fusion. First, we perform time alignment. For example, if a visual sensor captures an image every 5 seconds, while a vibration sensor collects data every second, we use linear interpolation to upscale the low-frequency data to a second-level resolution, ensuring that all sensor data corresponds to the same timeline. Next, we perform normalization on the data of different dimensions to unify their numerical scales and eliminate model bias caused by dimensional differences.

[0054] Next, to enhance the discriminability of the data in subsequent models, we introduce a structural sparse regularization term in the fusion stage to emphasize key signal channels and suppress redundant interference signals. The preprocessed data matrix is ​​as follows:

[0055] ;

[0056] in, is the new matrix after fusion; It is A vector of columns corresponding to each sensor; is the Frobenius norm, which is used to measure the data reconstruction error; the second term is Regularization, used to sparse channels, is a coefficient that balances data retention and channel selection. This regularization term is designed for the unique scenario of a die-cutting workshop: when die-cutting equipment fluctuates, only a few key sensors (such as vibration and pressure) typically react significantly to the change, while other sensors may simply be noise sources. Through structured sparse fusion, we enable the model to automatically compress redundant inputs while preserving the primary signal, improving subsequent modeling efficiency.

[0057] Among them, the fused matrix The output of this step has the following characteristics: 1) unified time series, 2) data standardization, 3) prominent key features, and 4) easy to input into subsequent deep models.

[0058] For example, in a typical production cycle, the vibration signal collected every second might show a spike at the moment the die-cutting knife contacts the material; while the output of the pressure sensor might indicate whether the die-cutting head is positioned consistently. By aligning these physically relevant high-frequency signals and introducing a sparse fusion structure, the discriminative capabilities of the state modeling stage can be significantly improved.

[0059] S2. Input the fused matrix into a pre-built depth perception model, use the dual-domain perception convolution kernel to extract local features, process it in combination with the channel attention module of physical prior knowledge, and output the state encoding vector and channel attention vector.

[0060] Specifically, this step aims to generate the fusion matrix based on the output of step 1. , building a deep perception model that accurately and finely reflects the state of the die-cutting production line. This model must not only be able to identify subtle fluctuations in equipment operation in real time, but also possess a certain level of physical constraint awareness to distinguish the impact of operational variation from environmental disturbances. Our goal is not to build a general neural network, but to design a perception framework that is highly structurally aware, has a clear physical interpretation, and possesses self-constraint capabilities, incorporating the state response characteristics of the die-cutting process.

[0061] Input is the standardized fusion data matrix generated in step 1 ,in is the number of sampling moments, is the number of effective sensor channels. Each column in the matrix is ​​the normalized time series of a certain type of sensor (such as pressure, vibration RMS, image texture entropy, etc.), and each row represents the multi-channel observation at a certain moment.

[0062] Considering the two typical characteristics of the operating state of die-cutting equipment: local fluctuations with high frequency and fineness, and global trends with low frequency and stability, we use a set of dual-domain perception convolution kernels to extract local features. These convolution kernels are divided into short perception domain ( ) and long receptive field ( ) to capture abnormal disturbances and trend changes respectively. Perform parallel multi-receptive domain convolution:

[0063] ;

[0064] in, Represents local features, Indicates channel splicing operation, and These are convolution operations for the short and long perception kernels, respectively. This design can simultaneously extract high-frequency oscillations in the second range, such as those caused by die-cutting head imbalance, and pressure attenuation trends in the hour range, such as those caused by die-cutting mold aging.

[0065] After batch normalization and activation function processing, the above features are fed into a channel attention module that incorporates physical prior knowledge. This module incorporates a penalty term during the learning process to limit the physical rationality of the attention weights based on the importance of different signals in the physical model of the die-cutting equipment. The regularization term is constructed as follows:

[0066] ;

[0067] in, The current training process The channel attention coefficient of each sensor, is the prior probability of the physical importance of the sensor given in the process drawing or experience library, is the Kullback-Leibler divergence, which measures the deviation between the attention distribution learned by the model and the physical assumption. is the weight adjustment coefficient of each channel, which is used to control its penalty intensity.

[0068] It is understandable that the introduction of this regularization term solves the problem that traditional attention mechanisms are prone to "misfocusing" in multi-sensor tasks. Take the die-cutting pressure channel as an example. It may be weakly correlated with the output quality in most states, but it is extremely decisive in the microsecond moment when the die enters and exits. If it relies entirely on training data, the model may ignore this channel because it is rare, and add After that, the network will be forced to learn the importance of the channel during the critical period, thereby more realistically simulating the experience judgment of human craftsmen.

[0069] The final output of the model is the state encoding vector , its dimension Much smaller than . It is the state embedding of die-cutting production at the current moment in the low-dimensional feature space, with the following characteristics: (1) reflecting the global overview of equipment operation; (2) being able to calculate the difference with the historical state; (3) being easy to input into the subsequent scheduling and optimization modules.

[0070] Also output is the channel attention vector ,This vector can be used by system operation and maintenance personnel to explain the model decision logic, such as whether the fluctuation of high-weight channels is related to the current decline in die-cutting accuracy.

[0071] For example, in actual applications, we observed that a batch of materials had obvious static electricity accumulation, which led to an abnormal increase in adhesion residue after die cutting. Traditional methods are difficult to distinguish between mechanical intervention abnormalities and material changes, but this model can The changing trend related to visual signals in the dimension, combined with the phenomenon of increased image channel weight in channel attention, successfully prompted the system to switch the static removal device and make surface treatment adjustments, reflecting the model's high sensitivity and explanatory power for complex production conditions.

[0072] S3. Using the state encoding vector as the environment state input, constructing a deep reinforcement learning model, designing a reward function including a task-resource cross-fit index, and outputting an action sequence and a dynamic policy distribution.

[0073] Specifically, this step is based on the state encoding vector output in step 2 and channel attention vector , a reinforcement learning strategy model for multi-process scheduling and resource allocation in die-cutting workshops is constructed. This model is designed to deal with the typical characteristic problems of "frequent changes in equipment health status", "simultaneous scheduling of multiple tasks", and "strong resource competition coupling" in die-cutting workshops. Traditional fixed scheduling strategies or heuristic-based job sorting algorithms can often only maintain local optimality under set conditions, and it is difficult to cope with sudden interference situations in the production line, such as high equipment vibration or temperature overload. Therefore, this scheme proposes a state-driven, self-evolving reinforcement learning system as a bridge between the perception module and the control module to achieve real-time adaptive optimization of key links such as die-cutting task sequence, equipment configuration, and personnel scheduling.

[0074] The model is obtained in step 2. (indicates the comprehensive operating status of the current equipment and environment) as the environmental status input, and uses This indicates the current importance of each sensor channel, thereby inferring which process variables are most sensitive in the current task. For example, when the channel weight of a pressure sensor is significantly higher than that of other channels at a certain moment, the system should prioritize completing precision die-cutting tasks that are highly correlated with pressure control. Scheduling actions are generated by a policy network, which consists of two fully connected layers, each equipped with a ReLU activation function and a normalization layer. Its output is a probability distribution of a set of discrete actions, each corresponding to a set of scheduling decisions, such as "assign task A to device 3" or "pause device 2 for maintenance."

[0075] Furthermore, to strengthen the model's understanding of the associated costs between different resources, we designed a task-resource cross-fitting indicator in the reward function to measure whether the current action fully considers the coupling between the equipment status and the task process characteristics. The reward function is as follows:

[0076] ;

[0077] in It is Instant rewards for every step; is the scheduling action generated by the policy network at the current time step; Indicates the resource utilization score, which measures the device utilization and average waiting time under the current action; The adaptability score measures the degree of match between the selected task and the device capabilities and sensitive channels in the current state; and is an adjustable weight parameter.

[0078] Furthermore, taking the actual scenario as an example, assume that the die-cutting tasks include three types: Class A (requiring high pressure and precision), Class B (long cycle and large mold), and Class C (high requirements for image recognition). The die-cutting machine 1 shows a slight vibration increase trend, channel attention This indicates a significant increase in the vibration channel weight, leading the system to conclude that the equipment is prone to structural anomalies. The reinforcement learning model automatically reduces the probability of Equipment 1 accepting Class A tasks, prioritizing them for Class B tasks, which are less sensitive to vibration. It also assigns Class A tasks to the more stable Equipment 2 or 3, and prompts the system to schedule a brief maintenance outage for Equipment 1.

[0079] Furthermore, the scheduling strategy is trained through deep reinforcement learning and updated using the Proximal Policy Optimization (PPO) algorithm. To improve sample utilization and training stability, we introduced the following mechanisms: First, an experience replay pool stores recent decision trajectories and reward feedback to prevent overfitting of recent states during training; second, a dynamic sampling mechanism increases the sampling frequency of samples with high process sensitivity (such as equipment anomalies or mission critical points), focusing training more on key scenarios; and third, a long-term average reward baseline dynamically adjusts model preferences to prevent short-term high-efficiency behaviors from misleading the overall strategy.

[0080] The two key outputs of the model are:

[0081] Action Sequence : Represents the scheduling behavior under continuous time steps, which is used to directly drive the execution of the workshop, including specific operations such as task sequence, equipment selection, and mold change scheduling;

[0082] Dynamic Strategy Distribution : It can be used to evaluate the current model's preference for scheduling behaviors under different states and assist manual intervention in strategy correction.

[0083] The core innovations of this step are reflected in three aspects: first, it systematically connects die-cutting process perception and task scheduling, avoiding the problem of "state recognition-scheduling control" fault; second, by introducing sensor channel attention to participate in the construction of scheduling preferences, the model has the ability to explain the process dimension; finally, through the dual reward structure and strategy stabilization mechanism, it realizes the long-term optimal exploration of the dynamic task-equipment mapping relationship, adapting to the highly changing environment and resource coupling relationship in die-cutting production.

[0084] S4. Combining the action sequence and dynamic strategy distribution, constructing a dynamic process offset index, generating a process offset value, and when the parameter correction value exceeds a preset threshold, adjusting the current process parameter through the parameter correction network, outputting the adjusted new parameter value, and forming a parameter correction sequence.

[0085] Specifically, this step is based on the previous reinforcement learning scheduling module and combines its output scheduling action sequence Stateful Strategy Distribution , enabling online quality feedback and adaptive adjustment of process parameters for die-cutting production scenarios. Its goal is to establish a control mechanism that can dynamically respond to operating condition fluctuations and proactively correct key process parameters during the production process, thereby addressing typical pain points in traditional die-cutting production, such as the difficulty in detecting quality fluctuations in real time, rigid and static parameter settings, and delayed correction of process anomalies.

[0086] First, the two outputs of step 3 are fully utilized in this step: Represents the specific scheduling decisions made by the system at each moment, involving execution behaviors such as task allocation, equipment selection, and mold activation; This represents the decision confidence of the current policy, assessing whether the current scheduling decision is in a sensitive state of "high uncertainty" and potentially risky. Based on these two outputs and combined with real-time quality monitoring data, we identify quality fluctuations and drive the generation and deployment of parameter adjustment actions.

[0087] In die-cutting production, defective products are often caused by certain key parameters not being adapted to the current working conditions in a timely manner, such as: the knife pressure not being fine-tuned according to the material thickness, the feeding beat not being adjusted according to the vibration state of the equipment, the scraping angle and pressure not matching, resulting in missed cutting or sticking, etc. Therefore, we construct a "dynamic process deviation index" in real time , quantifies the degree of deviation of the current process state from its historical optimal working condition. The construction method is as follows:

[0088] ;

[0089] in, It is The fusion features extracted from the quality monitoring module at all times (such as image clarity, edge contour, pressure consistency index, etc.) It is the optimal feature reference corresponding to the task in the historical stable output cycle. is the sliding window length. This indicator is updated once a second and is used to drive the decision of whether to trigger parameter correction.

[0090] Further, when Exceeding the set threshold And the corresponding strategy distribution If there is obvious skewness (for example, the distribution entropy exceeds the set confidence limit), the system determines that the current process may be in a risk state of "uncertain scheduling strategy + significant quality fluctuations", and the fine-tuning mechanism is triggered. We construct a parameter correction network, which is a two-layer perceptron with inputs including , Scheduling Actions and strategy distribution characteristics (such as entropy, maximum probability value), output a process parameter correction vector The correction method is as follows:

[0091] ;

[0092] in is the current process parameters (such as die cutting pressure, feed speed, cutter head frequency, etc.), is the tuning direction and magnitude of the model prediction, The parameter correction value can be tailored to physical constraints (such as maximum pressure limit) to ensure device safety.

[0093] It is understandable that in the specific implementation process, for example, a task A is assigned to device 2. After running continuously for 10 seconds, the quality monitoring module finds repeated burrs on the cutting edge contour through image recognition, and the quality deviation index It reaches 0.85, far exceeding the normal value of 0.35. At the same time, the scheduling strategy distribution The high entropy distribution in this time period indicates that the system has insufficient confidence in the scheduling choice. After combining these two signals, the model determines that there is a potential mismatch risk, starts the correction network, and outputs , indicating that the die cutting pressure should be reduced by 0.2 units, the feed rate should be increased by 0.05 units, and the scraping frequency should remain unchanged. It is immediately written into the PLC control unit and takes effect. The quality monitoring module reports that the cutting burrs have been reduced after 5 seconds. It dropped to 0.42, indicating that the adjustment was successful.

[0094] The two outputs of this step are:

[0095] Parameter correction sequence , used to replace the original parameters to drive process execution, and is a responsive control of quality fluctuations after strategic behavior.

[0096] Dynamic offset curve It is used to monitor in real time whether the current production task deviates from the historical normal working conditions, and can also be used as a scheduling stability evaluation indicator.

[0097] It is important to note that the innovation of this step is reflected in the following three aspects:

[0098] First, it introduces strategy distribution characteristics as judgment trigger conditions, enabling the system to have the ability to "preemptively intervene in uncertain states" rather than just react based on results; second, by comparing the sliding window dynamic offset indicator with the historical best feature reference point, the quality evaluation is more targeted and real-time; finally, the parameter correction mechanism is not based on manually set rules, but on regression prediction by fusing multiple input states to construct a neural network, which has adaptive and sustainable evolution capabilities.

[0099] S5. Construct a real-time monitoring indicator based on the process offset value and parameter correction sequence, and set a dynamic threshold. When the real-time monitoring indicator is continuously higher than the dynamic threshold for a certain period, a fault response program is triggered to output a real-time warning instruction sequence.

[0100] Specifically, this step is the parameter correction sequence output in the previous stage and Based on this, we established an online fault warning mechanism and closed-loop optimization feedback control process for the die-cutting production system. Its core goal is to provide practical monitoring of the system status during operation, promptly identify potential fault trends, and drive actual production intervention operations, thereby forming a final control closed loop based on "status assessment + risk triggering + control execution."

[0101] Among them, in terms of input, first, the parameter correction sequence It reflects the process adjustment behavior that the system has made in the past production cycle to maintain quality. Frequent and large adjustments are usually early signals of abnormal working conditions. Secondly, the quality deviation trend The fluctuation range of product quality within a certain time window is quantified. The trend of continuous deviation from the historical stable state indicates that there are anomalies in the current process that are difficult to correct through conventional adjustments.

[0102] In order to integrate these two inputs into a decision-oriented fault warning signal, we build a real-time monitoring indicator , comprehensive reflection system in the recent State stability in cycles:

[0103] ;

[0104] in, Indicates the The parameter adjustment amount at the moment, is the mass shift indicator, The weight balance coefficient for parameter adjustment and mass offset term (e.g. ), is the time window length (such as 30 seconds). and All have been normalized in the previous steps (including maximum and minimum scaling and zero mean standardization), so when constructing The dimensionality is unified, and no additional unit alignment operation is required, thus avoiding the deviation caused by cross-physical quantity comparison.

[0105] Furthermore, in order to ensure that the early warning mechanism can cover significant anomalies and avoid frequent false alarms, we Setting dynamic thresholds , its initial value can be obtained by The mean and standard deviation Statistics are obtained and set as:

[0106] ;

[0107] in Usually takes the value to , can be adjusted according to the stability of different process lines. In subsequent operation, It can be updated on a rolling basis based on weekly / monthly operating data to adapt it to the long-term degradation patterns of different task groups and different equipment, ensuring that the early warning mechanism remains sensitive and robust in different scenarios.

[0108] Among them, when Continuously above the threshold Reaching the warning cycle number If the system fails to respond to a fault (e.g., three consecutive beat cycles), the system triggers the actual fault response procedure. The response mechanism no longer remains at the modeling or inference level, but directly enters the action control execution phase, initiating the following operations:

[0109] Send load shedding / shutdown instructions to target equipment;

[0110] Push early warning notifications to on-site personnel, including task number, equipment ID, current , the main source of offset (such as frequent parameter adjustment of a certain channel);

[0111] Automatically allocate spare equipment to take over subsequent tasks, and the scheduling layer refers to historical task allocation strategies to prioritize low-risk batches;

[0112] Package and archive the fault-related fragments for subsequent expert review or maintenance records.

[0113] For example, in a multi-batch cutting task, a die-cutting machine starts to frequently generate pressure correction behavior after completing the fifth batch of tasks. The mean gradually increases, and the quality deviation index It shows a linear growth trend. The system calculates the cumulative , far above the set threshold This problem persisted for two production cycles, triggering an early warning after entering the third cycle. The control system immediately issued a "Pause Equipment 4 Task" command, and the on-site dashboard displayed "Abnormal Pressure + Quality Drift." Operations and maintenance personnel confirmed that the mold pressure head had loosened. In this case, the early warning mechanism not only provided early detection but also improved accuracy through a combined parameter-quality assessment, avoiding unnecessary line stops or false alarms.

[0114] Final output, real-time warning instruction sequence : Used for direct control layer, including control commands, notification content, emergency dispatch requests, etc.; fault data fragment record set : The system automatically archives the key parameter change trajectories and quality deviation trends before and after the warning is triggered for subsequent analysis or training enhancement.

[0115] In one or more embodiments, an intelligent management system for a die-cutting workshop production line is disclosed, the system comprising:

[0116] The data acquisition and fusion module is used to deploy multiple sensors at key workstations of the die-cutting workshop production line to collect multi-dimensional status data to form an original matrix, pre-process and fuse the original matrix to generate a fused matrix;

[0117] A depth perception modeling module is used to input the fused matrix into a pre-built depth perception model, extract local features using a dual-domain perception convolution kernel, process the local features using a channel attention module based on physical prior knowledge, and output a state encoding vector and a channel attention vector.

[0118] a scheduling optimization module, configured to take the state encoding vector as an environment state input, construct a deep reinforcement learning model, design a reward function including a task-resource cross-fit index, and output an action sequence and a dynamic policy distribution;

[0119] a quality control module, configured to combine the action sequence and the dynamic strategy distribution to construct a dynamic process deviation index, generate a process deviation value, and adjust the current process parameters through a parameter correction network when the parameter correction value exceeds a preset threshold, output the adjusted new parameter value, and form a parameter correction sequence;

[0120] The fault warning module is used to construct a real-time monitoring indicator based on the process offset value and parameter correction sequence, and set a dynamic threshold. When the real-time monitoring indicator is continuously higher than the dynamic threshold for a certain period, the fault response program is triggered and a real-time warning instruction sequence is output.

[0121] It is worth noting that the specific workflow of the intelligent management system for a die-cutting workshop production line provided by an embodiment of the present invention is the same as the workflow of the intelligent management method for a die-cutting workshop production line described in the above embodiment, and will not be repeated here.

[0122] An embodiment of the present invention further provides an intelligent management device for a die-cutting workshop production line, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the above-mentioned embodiment of the intelligent management method for a die-cutting workshop production line are implemented, for example Figure 1 or, the processor implements the functions of the modules in the above-mentioned system embodiments when executing the computer program.

[0123] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the intelligent management device for a die-cutting workshop production line.

[0124] The intelligent management device for a die-cutting workshop production line can be a computing device such as a desktop computer, laptop, PDA, or cloud server. The intelligent management device for a die-cutting workshop production line can include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the intelligent management device for a die-cutting workshop production line can also include input / output devices, network access devices, buses, and the like.

[0125] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor serves as the control center of the intelligent management device for the die-cutting workshop production line, and utilizes various interfaces and lines to connect various parts of the intelligent management device for the die-cutting workshop production line.

[0126] The memory can be used to store the computer programs and / or modules. The processor implements the various functions of the intelligent management device for a die-cutting workshop production line by running or executing the computer programs and / or modules stored in the memory and accessing the data stored in the memory. The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system, at least one application required for a function, and the data storage area may store data generated based on the operation of the air conditioning controller. Furthermore, the memory may include high-speed random access memory (RAM) and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a Smart Memory Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0127] If the module integrated with the intelligent management equipment for a die-cutting workshop production line is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium.

[0128] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0129] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A die-cutting workshop production line intelligent management method, characterized in that: The method comprises: Deploy multiple sensors at key workstations of the die-cutting production line to collect multi-dimensional state data to form an original matrix, pre-process and fuse the original matrix to generate a fused matrix; Input the fused matrix into a pre-built depth perception model, extract local features using a dual-domain perception convolution kernel, process it with a channel attention module based on physical prior knowledge, and output a state encoding vector and a channel attention vector; Taking the state encoding vector as the environment state input, building a deep reinforcement learning model, designing a reward function that includes a task-resource cross-fit index, and outputting an action sequence and a dynamic policy distribution; Combining the action sequence and the dynamic strategy distribution, a dynamic process offset index is constructed to generate a process offset value. When the parameter correction value exceeds a preset threshold, the current process parameter is adjusted through a parameter correction network, and the adjusted new parameter value is output to form a parameter correction sequence. A real-time monitoring indicator is constructed based on the process offset value and parameter correction sequence, and a dynamic threshold is set. When the real-time monitoring indicator is continuously higher than the dynamic threshold for a certain period, a fault response program is triggered and a real-time warning instruction sequence is output.

2. The intelligent management method of a die-cutting workshop production line according to claim 1, characterized in that: Preprocessing and fusing the original matrix to generate a fused matrix specifically includes: Obtaining the original matrix; The original matrix is ​​subjected to time alignment of sensor data of different frequencies by linear interpolation, dimension differences are eliminated by normalization, and then the original matrix is ​​subjected to Frobenius norm and Regular sparse recovery calculation is performed to generate the fused matrix.

3. The intelligent management method of a die-cutting workshop production line according to claim 1, characterized in that: The dual-domain perception convolution kernel includes a short perception domain and a long perception domain, which respectively capture abnormal disturbances and trend changes.

4. The intelligent management method of a die-cutting workshop production line according to claim 3, characterized in that: The fused matrix is ​​input into a pre-built depth perception model, local features are extracted using a dual-domain perception convolution kernel, and the channel attention module based on physical prior knowledge is used for processing to output a state encoding vector and a channel attention vector, specifically including: Obtaining the fused matrix; Using the short receptive field and the long receptive field to perform convolution operations on the fused matrix and perform channel splicing to generate local features; The local features are input into the channel attention model, and the state encoding vector and the channel attention vector are output; wherein the loss function of the channel attention model is Expressed as: ； in, The current training process The channel attention coefficient of each sensor, is the prior probability of the physical importance of the sensor given in the process drawing or experience library, is the Kullback-Leibler divergence, which measures the deviation between the attention distribution learned by the model and the physical assumption. is the weight adjustment coefficient of each channel, which is used to control its penalty intensity.

5. The intelligent management method of a die-cutting workshop production line according to claim 1, characterized in that: The deep reinforcement learning model consists of two fully connected layers, each equipped with a ReLU activation function and a normalization layer. Its output is a probability distribution of a set of discrete actions, where each action corresponds to a set of scheduling decisions. Among them, the scheduling strategy of the deep reinforcement learning model is completed through deep reinforcement learning training, and the PPO algorithm is used to update the strategy.

6. The intelligent management method of a die-cutting workshop production line according to claim 1, characterized in that: The reward function is expressed as: ； in, It is Instant rewards for every step; is the scheduling action generated by the policy network at the current time step; Indicates the resource utilization score, which measures the device utilization and average waiting time under the current action; The adaptability score measures the degree of match between the selected task and the device capabilities and sensitive channels in the current state; and is an adjustable weight parameter.

7. The intelligent management method of a die-cutting workshop production line according to claim 1, characterized in that: The step of combining the action sequence and the dynamic strategy distribution to construct a dynamic process offset indicator and generate a process offset value specifically includes: Using the multiple sensors to perform multi-dimensional data collection to generate fusion features for the current task; Obtain the optimal features corresponding to the current task in the historical stable output cycle; The degree of deviation of the current process state from its historical optimal working condition is calculated based on the optimal feature and the fusion feature as a process offset value.

8. The intelligent management method of a die-cutting workshop production line according to claim 1, characterized in that: The real-time monitoring indicator is used to reflect the system's recent performance based on the parameter correction amount. The state stability in a cycle; the dynamic threshold is obtained based on the mean and standard deviation statistics of the real-time monitoring indicators during the stable operation period of the system.

9. The intelligent management method of a die-cutting workshop production line according to claim 1, characterized in that: The real-time warning instructions include: sending load reduction / shutdown instructions to target equipment, pushing warning notifications to on-site personnel, automatically deploying spare equipment to take over subsequent tasks, and packaging and archiving fault-related fragments.

10. An intelligent management system for a die-cutting workshop production line, characterized in that: The system comprises: The data acquisition and fusion module is used to deploy multiple sensors at key workstations of the die-cutting workshop production line to collect multi-dimensional status data to form an original matrix, pre-process and fuse the original matrix to generate a fused matrix; A depth perception modeling module is used to input the fused matrix into a pre-built depth perception model, extract local features using a dual-domain perception convolution kernel, process the local features using a channel attention module based on physical prior knowledge, and output a state encoding vector and a channel attention vector. a scheduling optimization module, configured to take the state encoding vector as an environment state input, construct a deep reinforcement learning model, design a reward function including a task-resource cross-fit index, and output an action sequence and a dynamic policy distribution; a quality control module, configured to combine the action sequence and the dynamic strategy distribution to construct a dynamic process deviation index, generate a process deviation value, and adjust the current process parameters through a parameter correction network when the parameter correction value exceeds a preset threshold, output the adjusted new parameter value, and form a parameter correction sequence; The fault warning module is used to construct a real-time monitoring indicator based on the process offset value and parameter correction sequence, and set a dynamic threshold. When the real-time monitoring indicator is continuously higher than the dynamic threshold for a certain period, the fault response program is triggered and a real-time warning instruction sequence is output.

Images