Circuit board processing control method and system applied to industrial Internet of Things

Abstract

The invention provides a circuit board processing control method and system applied to an industrial internet of things, and the method comprises the steps: collecting the real-time equipment state data of a plurality of processing equipment in the processing process of a target circuit board, and calling a path generation model to generate an initial processing path based on the historical processing path data in a processing parameter set; inputting the path node data and a physical sensing data set into a parameter adjustment model for matching, and generating a path node adjustment instruction; performing parameter optimization on the processing nodes in the initial processing path based on the path node adjustment instruction through a parameter adjustment model, generating a target processing path, extracting a process constraint condition of each processing node in the target processing path, performing real-time control instruction distribution on the plurality of processing devices, and generating a target processing path; and performing joint iterative updating on the path generation model and the parameter adjustment model based on the feedback signal until the processing error rate of the target processing path is lower than a preset threshold value. The circuit board machining precision and the overall production efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more particularly, to a circuit board processing control method and system applied to the industrial Internet of Things. Background Art

[0002] The processing control of circuit boards in the industrial Internet of Things aims to achieve high-precision production through real-time monitoring and path planning. In existing processing control methods, a fixed processing path is usually generated based on preset historical path data, or static adjustment of local parameters is relied on sensor feedback to adapt to the device state. However, in existing methods, the heterogeneity of historical path data and real-time physical sensing data makes it difficult to fuse multi-source information, and it is impossible to dynamically identify path offsets caused by device anomalies or environmental interferences. Moreover, in the traditional model independent operation mode, the path generation and parameter adjustment links are separated, lacking a collaborative optimization mechanism, resulting in the accumulation of processing errors that are difficult to eliminate. At the same time, the parameter correction strategy driven by fixed rules is difficult to adapt to the dynamic process constraints under complex working conditions, and frequent device response delays and path conflicts seriously restrict the improvement of processing efficiency and product qualification rate. Summary of the Invention

[0003] The present invention provides a circuit board processing control method and system applied to the industrial Internet of Things.

[0004] In a first aspect, an embodiment of the present invention provides a circuit board processing control method applied to the industrial Internet of Things. The method includes: collecting real-time device state data of multiple processing devices during the processing of a target circuit board, where the device state data includes a set of processing parameters and a set of physical sensing data; based on the historical processing path data in the set of processing parameters, calling a path generation model to generate an initial processing path, and inputting the initial processing path and the set of physical sensing data into a parameter adjustment model for matching to generate a path node adjustment instruction; optimizing the parameters of the processing nodes in the initial processing path by the parameter adjustment model based on the path node adjustment instruction to generate a target processing path, and extracting the process constraint conditions of each processing node in the target processing path; allocating real-time control instructions to the multiple processing devices according to the process constraint conditions, and monitoring the feedback signals of the processing devices after executing the control instructions; jointly iteratively updating the path generation model and the parameter adjustment model based on the feedback signals until the processing error rate of the target processing path is lower than a preset threshold.

[0005] In a second aspect, an embodiment of the present invention provides a circuit board processing control system, including: a memory in which a computer program is stored; a processor configured to load the computer program to implement the circuit board processing control method applied to the industrial Internet of Things as described above.

[0006] The industrial Internet of Things circuit board processing control method provided by the present invention collects the device status data of the processing equipment in real time and extracts the processing parameter set and the physical sensing data set. Based on the historical processing path data, it calls the path generation model to generate the initial processing path. Combining the physical sensing data set, it matches and optimizes the processing nodes in the initial processing path through the parameter adjustment model, generates the target processing path that conforms to the real-time working conditions, and extracts the process constraint conditions. According to the process constraint conditions, it allocates real-time control instructions and monitors the device feedback signal, and jointly iteratively updates the path generation model and the parameter adjustment model until the processing error rate is lower than the preset threshold. In this way, the device status data can comprehensively reflect the dynamic changes in the processing environment. The path generation model constructs the initial path based on historical data to ensure the coherence of the processing logic. The parameter adjustment model realizes the matching by fusing real-time sensing data and path node information, effectively identifies and corrects the path deviation caused by device anomalies or environmental interference, and improves the adaptability of the path to complex working conditions. At the same time, the dynamic extraction of process constraint conditions and instruction allocation transform the optimized path parameters into executable device control strategies, ensuring the precise synchronization of processing actions and path planning. Through the model joint iteration mechanism driven by the feedback signal, the path generation model and the parameter adjustment model continuously optimize the parameters and path structure under the constraint of the error rate, which not only ensures the stability of the global processing path but also enhances the flexibility of local node adjustment, thereby significantly improving the circuit board processing accuracy and overall production efficiency while reducing equipment energy consumption and transmission conflicts. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0008] Figure 1 It is a flowchart of a circuit board processing control method applied to the industrial Internet of Things provided by an embodiment of the present invention.

[0009] Figure 2 It is a schematic diagram of the composition of a circuit board processing control system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0010] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all of them. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0011] Please refer to Figure 1 , Figure 1 which is a flowchart of a circuit board processing control method applied to the industrial Internet of Things provided by an embodiment of the present invention. The circuit board processing control method applied to the industrial Internet of Things can be executed by a circuit board processing control system. The circuit board processing control method applied to the industrial Internet of Things may include the following steps: A circuit board processing control method applied to the industrial Internet of Things provided by the present invention specifically includes the following steps: Step S100: Collect real-time device status data of multiple processing devices during the processing of the target circuit board. The device status data includes a set of processing parameters and a set of physical sensing data.

[0012] In this step, the target circuit board refers to the circuit board that is being processed, which is the specific object of processing control. The processing device is a machine used to perform various processing operations on the circuit board, such as a drilling machine, a chip mounter, a welding device, etc. The real-time device status data is information reflecting the operating condition of the processing device at the current moment, and it consists of two parts: a set of processing parameters and a set of physical sensing data. The set of processing parameters covers various parameters directly related to the processing process, including but not limited to text-based processing parameters, such as process descriptions, process steps, etc.; image-based processing parameters, such as image information of the processing area, etc. The set of physical sensing data is data reflecting the physical state of the device collected by sensors, such as temperature, pressure, vibration frequency, etc. In order to collect these data, various sensors can be reasonably arranged on the processing device to ensure that the status information of the device can be accurately obtained. For example, a temperature sensor is installed on the drilling machine to monitor the temperature change of the drill bit during operation in real time; a pressure sensor is installed to obtain the pressure value during drilling. The analog signals collected by these sensors are converted into digital signals and transmitted to the data acquisition system. The data acquisition system preliminarily processes and stores these data for subsequent analysis and use. This way of collecting real-time device status data can timely master the operating status of the processing device and provide accurate data support for subsequent path planning and parameter adjustment.

[0013] As an implementation manner, before step S100 of collecting real-time device status data of multiple processing devices during the processing of the target circuit board, the method provided by the embodiment of the present invention further includes the following series of steps: Step S101: Based on the process type and historical failure records of each processing device, configure the dynamic parameter adjustment rules of the data acquisition unit for the processing device, and generate a signal acquisition channel list matching the device model.

[0014] In the embodiments of the present application, the process type refers to the specific processing process executed by the processing equipment, and different process types have different requirements for data acquisition. For example, for the welding process, it may be necessary to focus on collecting data such as temperature and current; while for the chip mounter process, more attention is paid to parameters such as position accuracy and chip placement speed. The historical fault record is the relevant information about the faults that have occurred in the processing equipment in the past. By analyzing these records, the links and parameters where the equipment is prone to problems can be understood, so as to provide targeted guidance for data acquisition.

[0015] The data acquisition unit is a hardware device used to collect equipment status data, and it can process and transmit the data collected by the sensors. The dynamic parameter adjustment rule is formulated according to the process type and historical fault record, and is used to adjust the acquisition parameters of the data acquisition unit to ensure that the collected data is more accurate and effective. For example, if a certain processing equipment has often had the fault of overheating temperature in history, then the data acquisition unit can be set to collect temperature data more frequently and expand the temperature acquisition range.

[0016] The signal acquisition channel list is generated according to the equipment model. Each equipment model has its signal acquisition requirements, and the list contains information about each signal acquisition channel that matches the equipment, such as channel number, data type to be collected, etc. By configuring the dynamic parameter adjustment rule and generating the signal acquisition channel list for the processing equipment, the data acquisition can be made more accurate and efficient, laying a good foundation for subsequent data processing and analysis.

[0017] Step S102: Activate the data stream capture function of the corresponding sensor according to the signal acquisition channel list, and perform multi-level validity verification on the initially collected physical sensing data to identify data segments with signal jump anomalies or continuous zero values.

[0018] The data stream capture function refers to the ability of the sensor to collect and transmit data in real time. According to the signal acquisition channel list generated in step S101, activate the data stream capture function of the corresponding sensor to ensure that the required physical sensing data can be accurately obtained.

[0019] Multi-level validity verification is a process of verifying the initially collected physical sensing data multiple times to ensure the accuracy and reliability of the data. The verification methods can include data range check, data change rate check, etc. For example, for the data collected by the temperature sensor, check whether it is within a reasonable temperature range; for the data collected by the pressure sensor, check whether its change rate conforms to the normal physical law.

[0020] Signal jump anomaly refers to a large change in data within a short period of time, which may be caused by sensor failures, interference, etc. A data segment with continuous zero values refers to data that remains zero for a period of time, which may indicate sensor failure or connection problems. Through multi-level validity checks, these abnormal data segments can be detected in a timely manner, providing a basis for subsequent data repair and processing.

[0021] Step S103: Classify and store the data packets that pass the check into the temporary buffer according to the device identifier. At the same time, mark the data segments that fail the check as data to be repaired, and trigger a redundant data completion request for adjacent data acquisition units.

[0022] The device identifier is the unique identifier of each processing device, used to distinguish different devices. Classifying the data packets that pass the check according to the device identifier facilitates subsequent management and analysis of data for different devices. The temporary buffer is an area for temporarily storing data, which can play a buffering role during the data processing process to avoid data loss.

[0023] Data to be repaired refers to data segments that fail the multi-level validity check. These data may be abnormal and need to be repaired. The redundant data completion request is a request sent to adjacent data acquisition units, requesting them to provide relevant redundant data to complete the data to be repaired. For example, when the data collected by a certain sensor is abnormal, it can request the relevant data collected by adjacent sensors, and try to repair the abnormal data by comparing and analyzing these redundant data. In this way, the integrity and accuracy of the data can be improved.

[0024] Step S104: Monitor the data accumulation rate in the temporary buffer. When the rate exceeds the device processing capacity threshold, start the data shunt forwarding mechanism to distribute the overflow data packets to idle storage nodes for temporary storage.

[0025] The data accumulation rate refers to the speed at which the data volume in the temporary buffer increases. By monitoring the data accumulation rate in real time, the usage of the temporary buffer can be understood. The device processing capacity threshold refers to the upper limit of the data volume that the processing device can handle. If the data accumulation rate exceeds this threshold, it may lead to insufficient device processing capacity and affect the data processing efficiency.

[0026] The data shunt forwarding mechanism is a mechanism started when the data accumulation rate exceeds the threshold. It transfers the overflow data packets from the temporary buffer to idle storage nodes for temporary storage. An idle storage node refers to a storage device that is not fully utilized at present, such as a spare hard disk, cloud storage, etc. By starting the data shunt forwarding mechanism, it is possible to avoid excessive data accumulation in the temporary buffer and ensure the smoothness and stability of data processing.

[0027] Step S105: Create a mirror queue in the idle storage node that is consistent with the structure of the temporary buffer area, and configure the data life cycle policy and abnormal power-off recovery mechanism for the mirror queue.

[0028] The mirror queue is a queue created in the idle storage node with the same structure as the temporary buffer area. It is used to store the data packets diverted from the temporary buffer area. By creating the mirror queue, data backup and redundant storage can be achieved, improving data security.

[0029] The data life cycle policy refers to the policy for managing the storage time, deletion rules, etc. of the data in the mirror queue. For example, it can be set to automatically delete the data after it has been stored in the mirror queue for a preset time to save storage space. The abnormal power-off recovery mechanism is a mechanism that can ensure that the data in the mirror queue is not lost in the event of abnormal power-off and can resume normal data processing after the power is restored. For example, technical means such as regular backup and data recovery algorithms can be used to achieve abnormal power-off recovery. By configuring the data life cycle policy and abnormal power-off recovery mechanism, the data in the mirror queue can be effectively managed and protected.

[0030] Step S106: When the processing device enters the standby state, perform format standardization and reorganization on the device status data in the mirror queue, and transmit the reorganized data packets back to the temporary buffer area for subsequent processing.

[0031] When the processing device enters the standby state, it means that the device has temporarily stopped the processing operation. At this time, the device status data in the mirror queue can be processed. Format standardization and reorganization refer to organizing and converting the data in the mirror queue according to a unified format to facilitate subsequent analysis and use.

[0032] Specifically, Step S106 performs format standardization and reorganization on the device status data in the mirror queue, including the following steps: Step S1061: Perform steady-state baseline fitting on the temperature sampling values in the physical sensing data, identify transient noise pulses deviating from the baseline amplitude, and use context-dependent interpolation to replace the transient noise pulses.

[0033] Steady-state baseline fitting is to analyze the temperature sampling values in the physical sensing data to find a stable baseline value. For example, methods such as the least squares method can be used to fit the temperature sampling values over a period of time to obtain an approximate steady-state baseline. Transient noise pulses refer to the pulse signals that suddenly appear in the temperature sampling values and deviate significantly from the baseline amplitude. These pulse signals may be caused by external interference, sensor errors, etc.

[0034] Context-dependent interpolation estimates and replaces transient noise pulses based on the data information before and after the transient noise pulses, using the correlation of adjacent data. For example, if an abnormal pulse appears in the temperature sampling value at a certain moment, and the temperature values before and after this moment change relatively smoothly, a reasonable temperature value can be obtained by linearly interpolating the temperature values before and after, and this pulse can be replaced with this value. In this way, the noise interference in the physical sensing data can be removed, and the data quality can be improved.

[0035] Step S1062: Perform keyword integrity verification on the text-type parameters in the set of processing parameters, locate the text paragraphs missing the process description fields, and retrieve similar contexts from the historical processing logs for semantic completion.

[0036] Keyword integrity verification checks the text-type parameters in the set of processing parameters to ensure that the keywords in them are complete. For example, for a process description text, check whether it contains necessary keywords such as processing procedures and processing materials. The text paragraphs missing the process description fields refer to the parts of the text-type parameters that lack key process description information.

[0037] The historical processing log is a log file that records relevant information in the past processing process, which contains a large amount of process descriptions and context information. By retrieving similar contexts from the historical processing log, information related to the text paragraphs missing the process description fields can be found and semantic completion can be performed. For example, if a specific operation step of a certain processing procedure is missing in a text paragraph, the operation steps of a similar procedure can be found from the historical processing log for supplementation and improvement. In this way, the integrity and accuracy of the text-type parameters in the set of processing parameters can be improved.

[0038] Step S1063: Perform illumination intensity equalization analysis on the image-type parameters in the set of processing parameters, segment the local image areas with excessive light and dark differences, and perform texture smoothing reconstruction based on the adjacent pixel gradients.

[0039] Illumination intensity equalization analysis processes the image-type parameters in the set of processing parameters to make the illumination intensity of the image more uniform. Since the image may be affected by different illumination conditions during the acquisition process, resulting in areas with large light and dark differences in the image, which will affect the analysis and recognition of the image. Through illumination intensity equalization analysis, the brightness and contrast of the image can be adjusted to make the overall image clearer. The local image areas with excessive light and dark differences refer to the partial areas in the image where the light and dark differences exceed the preset threshold. For these areas, a segmentation algorithm can be used to separate them from the image for separate processing. The adjacent pixel gradient refers to the brightness change rate between adjacent pixels in the image. By analyzing the adjacent pixel gradient, the texture information of the image can be understood.

[0040] Texture smoothing reconstruction processes the segmented local image regions based on the gradients of adjacent pixels to make the texture of the image smoother. For example, methods such as Gaussian filtering can be used to smooth the local image regions, removing noise and jagged edges in the image. Through illumination intensity equalization analysis, local image region segmentation, and texture smoothing reconstruction, the quality of image-based parameters can be improved, facilitating subsequent image analysis and recognition.

[0041] Step S1064: Align the processed temperature sampling values, text-based parameters, and image-based parameters along the device time axis, and perform device state backtracking interpolation for the timestamp breakpoints found during the alignment process.

[0042] The device time axis is a time axis established according to the operating time sequence of the device. Aligning the processed temperature sampling values, text-based parameters, and image-based parameters along the device time axis can make different types of data consistent in time, facilitating subsequent comprehensive analysis. A timestamp breakpoint refers to the situation where, during the data alignment process, the timestamps of some data are discontinuous and there are interruptions. Device state backtracking interpolation estimates the data values at the timestamp breakpoints through the analysis of the device's historical state data and inserts them into the corresponding positions. For example, if the temperature sampling value at a certain moment is missing, the temperature value at that moment can be estimated based on the temperature values at the previous and subsequent moments and the operating state of the device, and inserted into the time axis. Through device state backtracking interpolation, the continuity and integrity of the data can be ensured.

[0043] Step S1065: Split the complete data stream after interpolation into independent data blocks according to the processing operation stages, and add an operation stage identifier and a data quality scoring label to each data block.

[0044] The complete data stream after interpolation is a data stream that is continuous and complete in time after being processed for timestamp breakpoints. Splitting it according to the processing operation stages is to facilitate the separate management and analysis of data in different operation stages. The operation stage identifier is a label used to identify the processing operation stage to which each data block belongs, such as the drilling operation, the chip mounting operation, etc.

[0045] The data quality scoring label is a label added after scoring each data block according to indicators such as data accuracy, integrity, and consistency. For example, for data blocks with high data accuracy and good integrity, a higher score can be given; for data blocks with errors or missing data, a lower score can be given. By adding an operation stage identifier and a data quality scoring label, the data can be better classified and evaluated.

[0046] Step S1066: Classify and store the data blocks into the training dataset or the test dataset according to the data quality scoring label, and configure a data augmentation strategy and a sample weight balancing coefficient for the training dataset.

[0047] The training dataset is a collection of data used to train a machine learning model, and the test dataset is a collection of data used to test the performance of the trained model. Classifying data blocks into the training dataset or the test dataset according to the data quality scoring labels can ensure the data quality of the training dataset and the test dataset.

[0048] The data augmentation strategy is a strategy for expanding and transforming the data in the training dataset to increase the diversity and richness of the data. For example, operations such as rotation, flipping, and scaling can be performed on image data, and operations such as synonym replacement and sentence restructuring can be performed on text data. The sample weight balancing coefficient is a coefficient set to balance the weights of different samples in the training dataset. In some cases, there may be a large difference in the number of samples of different classes in the training dataset. By setting the sample weight balancing coefficient, the model can pay more attention to the classes with fewer samples and improve the generalization ability of the model. By configuring the data augmentation strategy and the sample weight balancing coefficient, the training effect and performance of the machine learning model can be improved.

[0049] Step S200: Based on the historical processing path data in the set of processing parameters, call the path generation model to generate an initial processing path, and input the initial processing path and the physical sensing data set into the parameter adjustment model for matching to generate a path node adjustment instruction.

[0050] The historical processing path data refers to the processing path information adopted in the past during the processing of printed circuit boards. This data includes the positions, sequences of each processing node, and the connection relationships between nodes, etc. The path generation model is a machine learning model that can learn the generation rules of the processing path based on the input historical processing path data and generate an initial processing path. For example, the path generation model can adopt a neural network model such as a recurrent neural network (RNN) or a long short-term memory network (LSTM). By training on a large amount of historical processing path data, it masters the path generation pattern.

[0051] The initial processing path is a preliminary processing path generated by the path generation model based on the historical processing path data, which provides a basic plan for subsequent processing operations. The parameter adjustment model is another machine learning model used to optimize and adjust the initial processing path. Inputting the initial processing path and the physical sensing data set into the parameter adjustment model for matching is to correct the initial processing path according to the actual device state and environmental conditions. The path node adjustment instruction is an instruction generated by the parameter adjustment model according to the matching result, used to adjust the node parameters in the initial processing path to improve the rationality and effectiveness of the processing path.

[0052] Specifically, in step S200, based on the historical processing path data in the set of processing parameters, a path generation model is called to generate an initial processing path, including the following steps: Step S210: Perform semantic segmentation on the text-based processing parameters, extract multiple process description segments, and perform edge contour detection on the image-based processing parameters to generate a corresponding processing area positioning map.

[0053] Semantic segmentation analyzes the text-based processing parameters, segments them according to semantic information, and extracts multiple process description segments with independent meanings. For example, for a text describing the circuit board processing technology, through semantic segmentation, it can be split into multiple segments such as drilling process description, chip placement process description, etc.

[0054] Edge contour detection processes the image-based processing parameters, detects the edge contours of objects in the image. For the image-based parameters of circuit board processing, through edge contour detection, the boundaries of the processing area can be determined, and a corresponding processing area positioning map can be generated. The processing area positioning map can intuitively display the area that needs to be processed, providing important reference information for subsequent path planning.

[0055] Step S220: Align the spatial coordinates of the process description segments with the processing area positioning map, construct a process feature sequence, and input the process feature sequence into the time series encoding layer of the path generation model.

[0056] Spatial coordinate alignment matches the position information in the process description segments with the coordinate information in the processing area positioning map to make them correspond spatially. For example, for a certain processing position mentioned in the process description segment, through spatial coordinate alignment, the corresponding coordinate position can be found in the processing area positioning map.

[0057] The process feature sequence integrates the process description segments and the processing area positioning map information after spatial coordinate alignment to form a sequence containing process features. This sequence can reflect information such as the order and position of the processing processes, providing rich input features for the path generation model.

[0058] The time series encoding layer is an important part of the path generation model. It is used to encode the process feature sequence and convert it into a vector representation that the model can process. The time series encoding layer can use a recurrent neural network (RNN) or its variant, such as GRU (Gated Recurrent Unit), to perform time series modeling on the process feature sequence and capture the dependencies between processes.

[0059] Specifically, in step S220, aligning the spatial coordinates of the process description segments with the processing area positioning map to construct a process feature sequence includes the following steps: Step S221: Extract the text paragraphs containing location identifiers from the process description fragments, generate a set of location keywords, and identify the coordinate region boundaries corresponding to the location identifiers in the processing area positioning map.

[0060] The location identifier is a keyword used in the process description fragment to represent the processing location, such as the upper left corner, lower right corner, etc. Extracting the text paragraphs containing location identifiers from the process description fragment and generating a set of location keywords can clarify the key information of the processing location.

[0061] In the processing area positioning map, through image processing algorithms such as template matching and feature extraction, identify the coordinate region boundaries corresponding to the location identifiers. For example, if the location identifier is the upper left corner, find the coordinate region in the upper left corner of the processing area positioning map and determine its boundary.

[0062] Step S222: According to the geometric center point coordinates of the coordinate region boundaries, assign region anchor point coordinates to each location keyword to generate a keyword list with coordinate tags.

[0063] The geometric center point coordinates refer to the center point coordinates of the region enclosed by the coordinate region boundaries. According to the geometric center point coordinates of the coordinate region boundaries, assign a region anchor point coordinate to each location keyword to associate the location keyword with specific coordinate information.

[0064] The keyword list with coordinate tags is a list formed by arranging the location keywords with coordinate tags in order, which provides accurate location information for subsequent path planning.

[0065] Step S223: Perform a topological traversal on the contour segmentation lines in the processing area positioning map to determine the processing order priority of each contour region, and sort the keyword list with coordinate tags based on the processing order priority.

[0066] The contour segmentation line is a line used to divide different processing regions in the processing area positioning map. Performing a topological traversal on the contour segmentation line is to traverse the contour segmentation line to determine the processing order priority of each contour region. For example, the priority can be determined according to factors such as the connection relationship of the contour regions and the sequence of processing processes.

[0067] Sort the keyword list with coordinate tags based on the processing order priority, so that the location information in the keyword list is arranged according to the processing order, providing an ordered input for subsequent path generation.

[0068] Step S224: Match the region anchor point coordinates in the sorted keyword list with the actual processing path start coordinates in the processing area positioning map, and filter out the candidate coordinate pairs with a distance deviation less than the tolerance threshold.

[0069] The starting coordinate of the actual machining path refers to the coordinate position where the machining operation begins. The distance between the regional anchor coordinates in the sorted keyword list and the starting coordinate of the actual machining path is matched, and the distance between the two is calculated.

[0070] The tolerance threshold is a pre-set distance threshold. If the distance deviation between the regional anchor coordinates and the starting coordinate of the actual machining path is less than the tolerance threshold, then this coordinate pair is used as a candidate coordinate pair. Screening out candidate coordinate pairs can reduce unnecessary calculations and interference, and improve the efficiency and accuracy of path generation.

[0071] Step S225: According to the process description content of the keywords in the candidate coordinate pairs, perform semantic annotation on the contour area of the machining area positioning map to generate an annotated process space distribution map.

[0072] The process description content is the description information about the machining process contained in the keywords in the candidate coordinate pairs, such as drilling, chip mounting, etc. According to the process description content, perform semantic annotation on the contour area of the machining area positioning map, and associate different machining processes with the corresponding contour areas.

[0073] The annotated process space distribution map is the machining area positioning map after semantic annotation. It intuitively shows the process types and positional relationships of each machining area, and provides clear spatial information for path generation.

[0074] Step S226: According to the distribution density of the annotated areas in the process space distribution map, temporally bind the text content in the process description segment to the corresponding annotated areas to form a process feature sequence containing spatial coordinate dependency relationships.

[0075] The distribution density refers to the degree of spatial distribution density of the annotated areas in the process space distribution map. According to the distribution density, temporally bind the text content in the process description segment to the corresponding annotated areas, that is, determine the temporal sequence and association relationship between the text content and the annotated areas.

[0076] The process feature sequence containing spatial coordinate dependency relationships is a sequence formed after temporal binding. It not only contains the text description and spatial coordinate information of the process, but also reflects the temporal and spatial dependency relationships between processes, providing more comprehensive input features for the path generation model.

[0077] Step S230: Through the temporal encoding layer, perform topological sorting on the process dependency relationships in the process feature sequence to generate a process node connection graph.

[0078] Topological sorting is an algorithm for sorting a directed acyclic graph (DAG), which can determine the order of nodes in the graph such that for any directed edge (u, v) in the graph, node u appears before node v in the sorting result. In this step, topological sorting of the process dependency relationships in the process feature sequence is performed through the timing encoding layer to determine the execution order of each process node.

[0079] The process node connection graph is a directed graph, where nodes represent process nodes and edges represent the dependency relationships between processes. Generating the process node connection graph through topological sorting can clearly show the order and dependency relationships between processes, providing a basis for subsequent path generation.

[0080] Specifically, step S230 performs topological sorting on the process dependency relationships in the process feature sequence through the timing encoding layer to generate a process node connection graph, including the following steps: Step S231: Extract the node identifier and corresponding processing action description of each process node from the process feature sequence to generate a set of process node features containing the input-output relationships between nodes.

[0081] The node identifier is the unique identifier of each process node, used to distinguish different process nodes. The processing action description is a detailed description of the processing actions performed by the process node, such as drilling, welding, etc. Extracting the node identifier and corresponding processing action description of each process node from the process feature sequence can clarify the specific information of each process node.

[0082] The set of process node features is a set formed by integrating the node identifiers, processing action descriptions, and input-output relationships between nodes of each process node. This set contains important feature information of the process nodes, providing a data basis for subsequent topological sorting.

[0083] Step S232: Traverse the set of process node features, analyze the predecessor process nodes and successor process nodes of each process node, establish a process dependency relationship chain, and mark the process transmission direction between nodes in the process dependency relationship chain.

[0084] The predecessor process node refers to the process node that must be completed before a certain process node in the process execution order; the successor process node refers to the process node that needs to be executed after a certain process node in the process execution order. By traversing the set of process node features, analyzing the predecessor process nodes and successor process nodes of each process node, a process dependency relationship chain is established.

[0085] The process transmission direction refers to the transmission direction of materials, information, etc. between processes. Marking the process transmission direction between nodes in the process dependency relationship chain can more accurately describe the relationship between processes, providing a basis for subsequent directed graph reconstruction.

[0086] Step S233: Reconstruct the process dependency chain into a directed graph according to the process transmission direction, and determine the initial in-degree value and the triggerable execution status of each process node.

[0087] The directed graph reconstruction is the process of converting the process dependency chain into a directed graph, and adding directions to the edges of the directed graph according to the process transmission direction. The initial in-degree value refers to the number of incoming edges of each node in the directed graph, which represents the number of predecessor process nodes of the node.

[0088] The triggerable execution status refers to the status of whether the process node meets the execution conditions. According to the initial in-degree value and the dependency relationship between processes, the triggerable execution status of each process node can be determined. For example, if the initial in-degree value of a process node is 0, it means that the node has no predecessor process nodes and can be triggered for execution immediately.

[0089] Step S234: Screen the candidate starting nodes with an in-degree of zero from the process node feature set based on the initial in-degree value, and add the candidate starting nodes to the queue of nodes to be processed.

[0090] Candidate starting nodes refer to the process nodes with an initial in-degree value of zero, and these nodes are the starting points of the process execution. Screen the candidate starting nodes from the process node feature set based on the initial in-degree value and add them to the queue of nodes to be processed to prepare for the subsequent topological sorting operation.

[0091] Step S235: Take out the current processing node from the queue of nodes to be processed in turn, traverse the successor process nodes of the current processing node, update the dynamic in-degree values of the successor process nodes, and add the successor process nodes with the dynamic in-degree value reduced to zero to the queue of nodes to be processed.

[0092] The current processing node refers to the current process node being processed taken out from the queue of nodes to be processed. Traversing the successor process nodes of the current processing node is to update the dynamic in-degree values of these successor process nodes. The dynamic in-degree value refers to the fact that during the topological sorting process, as the process nodes are executed, the in-degree values of the successor process nodes will change.

[0093] When the dynamic in-degree value of a certain successor process node is reduced to zero, it means that all the predecessor process nodes of the node have been executed, and the node can be triggered for execution. Add it to the queue of nodes to be processed and continue the topological sorting operation.

[0094] Step S236: Record the connection order and the process transmission direction between the current processing node and the successor process nodes, and generate a set of directed edges with weight identifiers.

[0095] During the topological sorting process, record the connection order and process transmission direction between the currently processed node and the successor process nodes, and add a weight identifier to each directed edge. The weight identifier can represent information such as the transmission time and resource consumption between processes, and it can more comprehensively describe the relationship between processes.

[0096] The set of directed edges with weight identifiers is a set formed by integrating all the recorded directed edges, which provides detailed edge information for the subsequent generation of the process node connection graph.

[0097] Step S237: Generate a process node connection graph containing a topological hierarchical relationship according to the arrangement order of the nodes in the set of directed edges and the set of process node characteristics, where the topological hierarchical relationship is used to indicate the parallel execution interval and serial dependency depth of the process nodes.

[0098] Generate a process node connection graph according to the arrangement order of the nodes in the set of directed edges with weight identifiers and the set of process node characteristics. The topological hierarchical relationship refers to the hierarchical structure of the process nodes in the graph, which can indicate the parallel execution interval and serial dependency depth of the process nodes.

[0099] The parallel execution interval refers to the time period during which multiple process nodes can be executed simultaneously, and the serial dependency depth refers to the depth of the serial dependency relationship between process nodes. Through the topological hierarchical relationship, the execution order of processes can be better planned to improve processing efficiency.

[0100] Step S240: Generate the arrangement order of the nodes and the transmission time between nodes of the initial processing path according to the process weights of the nodes in the process node connection graph.

[0101] The process weight is the weight value assigned to each node in the process node connection graph, which reflects factors such as the importance, complexity, and time consumption of the node in the processing process. For example, for some key processing process nodes, a higher process weight can be assigned.

[0102] Generate the arrangement order of the nodes and the transmission time between nodes of the initial processing path according to the process weights of the nodes in the process node connection graph. The arrangement order of the nodes determines the sequence of the processing operations, and the transmission time between nodes represents the time required to go from one node to another.

[0103] Specifically, step S240 generates the arrangement order of the nodes and the transmission time between nodes of the initial processing path according to the process weights of the nodes in the process node connection graph, including the following steps: Step S241: Obtain the processing time weight, equipment switching weight, and path continuity weight included in the process weight, and generate a weight type association table based on the weight type.

[0104] The processing time-consuming weight refers to the weight value in the process weight that is related to the time required for processing operations, which reflects the processing time-consuming situation of each process node. The equipment switching weight is the weight value related to the equipment switching process. Equipment switching may cause consumption of time and resources, so the equipment switching weight needs to be considered. The path continuity weight is the weight value used to measure path continuity. A continuous processing path can improve processing efficiency.

[0105] Generate a weight type association table based on the weight types. This table records the corresponding relationships between different weight types and process nodes, providing convenience for subsequent calculations.

[0106] Step S242: Traverse all the nodes in the process node connection graph, normalize the processing time-consuming weight of each node according to the weight type association table, and generate a comprehensive priority score for the nodes.

[0107] Normalization processing is to convert the processing time-consuming weights in different ranges into values within a unified range for comparison and calculation. By traversing all the nodes in the process node connection graph and normalizing the processing time-consuming weight of each node according to the weight type association table, the dimensional differences between the processing time-consuming weights of different nodes are eliminated.

[0108] The comprehensive priority score of a node is a comprehensive score generated for each node after comprehensively considering factors such as the processing time-consuming weight, equipment switching weight, and path continuity weight. This score is used to determine the priority of the node in the processing path.

[0109] Step S243: Extract the set of predecessor nodes of the current node from the process node connection graph, and determine the trigger execution condition of the current node according to the comprehensive priority scores of the set of predecessor nodes.

[0110] The set of predecessor nodes refers to the set composed of all the predecessor process nodes of the current node in the process node connection graph. According to the comprehensive priority scores of the set of predecessor nodes, the trigger execution condition of the current node can be determined. For example, if the comprehensive priority score of a certain predecessor node is relatively high and the node has not been executed completely, the current node cannot be triggered to execute.

[0111] Step S244: Screen candidate nodes that meet the path continuity weight requirements based on the trigger execution condition, and sort the candidate nodes in descending order according to the comprehensive priority scores of the nodes to generate a candidate node execution queue.

[0112] Based on the trigger execution condition, screen candidate nodes that meet the path continuity weight requirements from the process node connection graph. The path continuity weight requirements can include factors such as the distance and connection relationship between nodes.

[0113] Arrange the candidate nodes in descending order according to the comprehensive priority score of the nodes, with the nodes having higher scores ranked in the front, and generate a candidate node execution queue. This queue determines the execution order of the candidate nodes.

[0114] Step S245: Sequentially select the head node from the candidate node execution queue and add it to the initial processing path sequence, and calculate the equipment switching time consumption between the head node and the previous node based on the equipment switching weight.

[0115] The head node is the node ranked at the very front in the candidate node execution queue. Sequentially select the head node from the candidate node execution queue and add it to the initial processing path sequence to gradually construct the initial processing path.

[0116] Calculate the equipment switching time consumption between the head node and the previous node based on the equipment switching weight. The equipment switching time consumption includes the time required for processes such as equipment startup, adjustment, and calibration.

[0117] Step S246: According to the set of successor nodes of the head node in the process node connection graph, update the node arrangement order in the candidate node execution queue, and record the transmission time consumption increment from the current node to the successor nodes.

[0118] The set of successor nodes refers to the set composed of all successor process nodes of the head node in the process node connection graph. According to the set of successor nodes, update the node arrangement order in the candidate node execution queue to ensure that the nodes in the queue are arranged according to the latest priority.

[0119] Record the transmission time consumption increment from the current node to the successor nodes. The transmission time consumption increment represents the additional time required from the current node to the successor nodes.

[0120] Step S247: Repeat the steps of node selection, switching time consumption calculation, and queue update until the initial processing path sequence contains all the nodes in the process node connection graph, and synchronously output the cumulative value of the transmission time consumption between nodes.

[0121] Repeat Step S245 and Step S246, continuously select nodes and add them to the initial processing path sequence, calculate the equipment switching time consumption, and update the candidate node execution queue until the initial processing path sequence contains all the nodes in the process node connection graph.

[0122] Synchronously output the cumulative value of the transmission time consumption between nodes. This cumulative value represents the total transmission time consumption of the entire initial processing path.

[0123] In Step S200, input the initial processing path and the physical sensing data set into the parameter adjustment model for matching to generate a path node adjustment instruction, which specifically includes the following steps: Step S250: Obtain the processing node coordinate sequence in the initial processing path and the preset process parameters for each node, and extract the real-time device position data and environmental monitoring data corresponding to the processing node coordinates from the physical sensing data set.

[0124] The processing node coordinate sequence is the coordinate information of each processing node in the initial processing path, which determines the position of the processing node. The preset process parameters are the process parameters preset for each processing node, such as processing speed, processing temperature, etc.

[0125] Extract the real-time device position data and environmental monitoring data corresponding to the processing node coordinates from the physical sensing data set. The real-time device position data represents the actual position of the processing device at the current moment, and the environmental monitoring data includes data on environmental factors such as temperature, pressure, and humidity. By extracting these data, the actual operating state and environmental conditions of the processing device can be understood.

[0126] Step S260: Align the processing node coordinate sequence and the real-time device position data according to the time stamp to generate a spatio-temporal alignment mapping table, which contains the actual temperature value, pressure value, and vibration frequency value of each processing node within the corresponding time window.

[0127] The time stamp is the time mark of data acquisition. Aligning the processing node coordinate sequence and the real-time device position data according to the time stamp is to make them correspond in time. Through the alignment operation, a spatio-temporal alignment mapping table is generated.

[0128] The spatio-temporal alignment mapping table contains the actual temperature value, pressure value, and vibration frequency value of each processing node within the corresponding time window. These actual values reflect the actual operating state and environmental conditions of the device during the processing, providing a basis for subsequent parameter adjustment.

[0129] Step S270: Compare the actual temperature value in the spatio-temporal alignment mapping table with the preset temperature safety threshold, mark the processing nodes exceeding the safety threshold as high-temperature risk nodes, and compare the actual pressure value with the preset pressure fluctuation range, mark the processing nodes exceeding the fluctuation range as pressure abnormal nodes.

[0130] The preset temperature safety threshold is the preset temperature safety range. If the actual temperature value exceeds this threshold, it indicates that there is a high-temperature risk at the processing node. Compare the actual temperature value in the spatio-temporal alignment mapping table with the preset temperature safety threshold, and mark the processing nodes exceeding the safety threshold as high-temperature risk nodes.

[0131] The preset pressure fluctuation range is a pre-set pressure fluctuation range. If the actual pressure value exceeds this range, it indicates that there is a pressure anomaly at the processing node. Compare the actual pressure value with the preset pressure fluctuation range, and mark the processing nodes that exceed the fluctuation range as pressure anomaly nodes. By marking the high-temperature risk nodes and pressure anomaly nodes, potential problems in the processing process can be discovered in a timely manner.

[0132] Step S280: According to the distribution density of the high-temperature risk nodes and pressure anomaly nodes, identify the transmission conflict segments formed by consecutive anomaly nodes, and extract the maximum offset distance between the real-time position data of the device and the coordinates of the processing nodes in the transmission conflict segments.

[0133] The distribution density refers to the degree of concentration of high-temperature risk nodes and pressure anomaly nodes in the processing path. According to the distribution density of the high-temperature risk nodes and pressure anomaly nodes, identify the transmission conflict segments formed by consecutive anomaly nodes. The transmission conflict segments represent areas where there may be conflicts or problems during the processing.

[0134] Extract the maximum offset distance between the real-time position data of the device and the coordinates of the processing nodes in the transmission conflict segments. The maximum offset distance reflects the deviation degree between the actual position of the device and the preset position of the processing nodes, providing important reference information for subsequent path adjustment.

[0135] Step S290: Based on the time-domain change trend of the maximum offset distance and the vibration frequency value, generate a speed reduction adjustment instruction for the high-temperature risk nodes, a device switching instruction for the pressure anomaly nodes, and a path re-planning instruction for the transmission conflict segments, forming a set of path node adjustment instructions that includes instruction types, target node identifiers, and parameter adjustment ranges.

[0136] The time-domain change trend refers to the change of the vibration frequency value over time. Based on the time-domain change trend of the maximum offset distance and the vibration frequency value, generate corresponding adjustment instructions.

[0137] The speed reduction adjustment instruction is an instruction generated for the high-temperature risk nodes, which reduces the heat generation by reducing the processing speed and reduces the high-temperature risk. The device switching instruction is an instruction generated for the pressure anomaly nodes, which solves the pressure anomaly problem by switching devices. The path re-planning instruction is an instruction generated for the transmission conflict segments, which avoids conflicts by re-planning the path.

[0138] The set of path node adjustment instructions is a set formed by integrating the speed reduction adjustment instruction, the device switching instruction, and the path re-planning instruction. This set includes information such as instruction types, target node identifiers, and parameter adjustment ranges, providing clear guidance for subsequent parameter optimization.

[0139] In the embodiments of the present invention, the parameter adjustment model is obtained through multi-stage training. The process of multi-stage training includes the following steps: Step S10: Obtain the first-stage training data set, which includes multiple sets of device status sample data and their corresponding standard machining path labels.

[0140] The first-stage training data set is a data set used for the first-stage training of the parameter adjustment model. The device status sample data is the device status data collected from the actual machining process, including the machining parameter set and the physical sensing data set, etc. The standard machining path label is the correct machining path information corresponding to the device status sample data, and it is used as the training target to guide the model learning.

[0141] Step S20: Input the device status sample data into the initial parameter adjustment model to generate a predicted machining path, and calculate the path coincidence loss value between the predicted machining path and the standard machining path label.

[0142] The initial parameter adjustment model is the initial state of the parameter adjustment model, and it has not been fully trained. Inputting the device status sample data into the initial parameter adjustment model, the model will generate a predicted machining path according to the input data.

[0143] The path coincidence loss value is an index used to measure the difference degree between the predicted machining path and the standard machining path label. By calculating the path coincidence loss value, the prediction performance of the model can be evaluated, providing a basis for subsequent model optimization.

[0144] Specifically, calculating the path coincidence loss value between the predicted machining path and the standard machining path label in step S20 includes the following steps: Step S21: Extract the position coordinate sequence of each machining node in the predicted machining path, and perform point-by-point matching of the position coordinate sequence with the reference coordinate sequence in the standard machining path label.

[0145] The position coordinate sequence is the coordinate information of each machining node in the predicted machining path, and the reference coordinate sequence is the coordinate information in the standard machining path label. Performing point-by-point matching of the position coordinate sequence with the reference coordinate sequence is to compare the differences between the two.

[0146] Step S22: Calculate the Euclidean distance for the unmatched coordinate points to generate the first local loss component, and calculate the machining time difference for the matched coordinate points to generate the second local loss component.

[0147] The Euclidean distance is a commonly used method for calculating the distance between two points. The Euclidean distance is calculated for the unmatched coordinate points to obtain the first local loss component. The first local loss component reflects the difference in position between the predicted machining path and the standard machining path label.

[0148] The difference in machining time consumption is calculated for the matched coordinate points, that is, the difference in machining time consumption between the matched coordinate points in the predicted machining path and the corresponding coordinate points in the standard machining path label is calculated to generate the second local loss component. The second local loss component reflects the difference in machining time consumption between the predicted machining path and the standard machining path label.

[0149] Step S23: The first local loss component and the second local loss component are weighted and summed, and a dynamic adjustment coefficient is generated according to the weighted sum result.

[0150] Weighted summation is to add the first local loss component and the second local loss component according to the preset weights to obtain a comprehensive loss value. A dynamic adjustment coefficient is generated according to the weighted sum result, and the dynamic adjustment coefficient is used to adjust the loss weights of the unmatched coordinate points and the matched coordinate points.

[0151] Step S24: Based on the dynamic adjustment coefficient, the loss weights of the unmatched coordinate points and the matched coordinate points are redistributed to obtain the path coincidence degree loss value.

[0152] Based on the dynamic adjustment coefficient, the loss weights of the unmatched coordinate points and the matched coordinate points are redistributed to make the model pay more attention to the parts with larger differences. After redistributing the loss weights, the path coincidence degree loss value is obtained.

[0153] Specifically, step S24 redistributes the loss weights of the unmatched coordinate points and the matched coordinate points based on the dynamic adjustment coefficient to obtain the path coincidence degree loss value, including the following steps: Step S2401: Extract the error amplification factor of the unmatched coordinate points and the error suppression factor of the matched coordinate points from the dynamic adjustment coefficient, and associate the first local loss component of the unmatched coordinate points and the second local loss component of the matched coordinate points.

[0154] The error amplification factor is a coefficient used to amplify the loss weight of the unmatched coordinate points, and the error suppression factor is a coefficient used to suppress the loss weight of the matched coordinate points. The error amplification factor of the unmatched coordinate points and the error suppression factor of the matched coordinate points are extracted from the dynamic adjustment coefficient, and they are respectively associated with the first local loss component of the unmatched coordinate points and the second local loss component of the matched coordinate points.

[0155] Step S2402: Weight-amplify the first local loss component according to the error amplification factor to generate an adjusted unmatched loss component, and at the same time weight-decay the second local loss component according to the error suppression factor to generate an adjusted matched loss component.

[0156] Weight-amplify the first local loss component according to the error amplification factor, so that the loss of unmatched coordinate points occupies a greater proportion in the path coincidence degree loss value. At the same time, weight-decay the second local loss component according to the error suppression factor to reduce the influence of the loss of matched coordinate points on the path coincidence degree loss value.

[0157] Step S2403: Traverse each coordinate point in the adjusted unmatched loss component, calculate the direction offset angle between it and the nearest neighbor coordinate point in the reference coordinate sequence, and perform direction consistency compensation on the adjusted unmatched loss component based on the direction offset angle.

[0158] The direction offset angle refers to the angular deviation between the coordinate point in the adjusted unmatched loss component and the nearest neighbor coordinate point in the reference coordinate sequence. Traverse each coordinate point in the adjusted unmatched loss component, calculate its direction offset angle, and perform direction consistency compensation on the adjusted unmatched loss component based on the direction offset angle. Direction consistency compensation can make the model pay more attention to the direction deviation of coordinate points and improve the evaluation accuracy of path coincidence degree.

[0159] Step S2404: Superimpose the unmatched loss component after direction consistency compensation and the adjusted matched loss component to generate a total path deviation loss value, and generate a normalized loss ratio according to the ratio of the total path deviation loss value to the preset path error upper limit.

[0160] Superimpose the unmatched loss component after direction consistency compensation and the adjusted matched loss component to obtain the total path deviation loss value. The total path deviation loss value reflects the comprehensive difference between the predicted processing path and the standard processing path label.

[0161] Generate a normalized loss ratio according to the ratio of the total path deviation loss value to the preset path error upper limit. The normalized loss ratio maps the total path deviation loss value to a unified range, which is convenient for subsequent processing and comparison.

[0162] Step S2405: Perform piecewise linear mapping on the total path deviation loss value based on the normalized loss ratio to generate a path coincidence degree loss value that meets the loss value range constraint, and use the path coincidence degree loss value as the objective function input for backpropagation optimization.

[0163] Piecewise linear mapping is a method that converts the normalized loss ratio into a path coincidence loss value that satisfies the loss value range constraint. Through piecewise linear mapping, the path coincidence loss value is ensured to be within a reasonable range.

[0164] Taking the path coincidence loss value as the input of the objective function for backpropagation optimization, the backpropagation algorithm will update the parameters of the initial parameter adjustment model according to the path coincidence loss value to improve the prediction performance of the model.

[0165] Step S30: Obtain the second-stage training dataset, which contains multiple groups of abnormal process sample data and their corresponding path correction labels.

[0166] The second-stage training dataset is a data set used for the second-stage training of the parameter adjustment model. The abnormal process sample data is equipment status data containing abnormal process conditions, such as high-temperature risk, abnormal pressure, etc. The path correction label is the correct path correction information for the abnormal process sample data, which is used to guide the model to learn how to handle abnormal situations.

[0167] Step S40: Input the abnormal process sample data into the initial parameter adjustment model trained in the first stage, generate a corrected predicted processing path, and calculate the error compensation loss value between the corrected predicted processing path and the path correction label.

[0168] Inputting the abnormal process sample data into the initial parameter adjustment model trained in the first stage, the model will generate a corrected predicted processing path according to the input data. The error compensation loss value is an index used to measure the degree of difference between the corrected predicted processing path and the path correction label. By calculating the error compensation loss value, the performance of the model in handling abnormal situations can be evaluated.

[0169] Step S50: Fuse the path coincidence loss value and the error compensation loss value according to a preset ratio to obtain a comprehensive training loss value, and perform backpropagation optimization on the initial parameter adjustment model according to the comprehensive training loss value until the comprehensive training loss value converges.

[0170] The preset ratio is the fusion ratio of the path coincidence loss value and the error compensation loss value set in advance. Fusing the path coincidence loss value and the error compensation loss value according to the preset ratio to obtain a comprehensive training loss value. The comprehensive training loss value comprehensively considers the prediction performance of the model in normal and abnormal situations. Performing backpropagation optimization on the initial parameter adjustment model according to the comprehensive training loss value, the backpropagation algorithm will continuously adjust the parameters of the model to gradually reduce the comprehensive training loss value until it converges. When the comprehensive training loss value converges, it indicates that the model has achieved a good training effect.

[0171] Step S300: Optimize the parameters of the machining nodes in the initial machining path based on the path node adjustment instruction through the parameter adjustment model, generate the target machining path, and extract the process constraint conditions of each machining node in the target machining path.

[0172] Parameter optimization refers to adjusting and optimizing the parameters of the machining nodes in the initial machining path according to the path node adjustment instruction to improve the rationality and effectiveness of the machining path. The target machining path is the final machining path obtained after parameter optimization. It takes into account the actual equipment status and environmental conditions and can better meet the machining requirements.

[0173] The process constraint condition is the condition that each machining node needs to meet during the machining process, such as temperature range, pressure range, machining accuracy, etc. Extracting the process constraint conditions of each machining node in the target machining path provides a basis for subsequent equipment control.

[0174] Specifically, step S300 optimizes the parameters of the machining nodes in the initial machining path based on the path node adjustment instruction through the parameter adjustment model to generate the target machining path, including the following steps: Step S310: Analyze the speed reduction adjustment instruction in the path node adjustment instruction set, obtain the target node identifier of the high-temperature risk node and the allowed maximum speed reduction ratio, and calculate the actual speed reduction coefficient according to the real-time temperature value in the physical sensing data set.

[0175] The speed reduction adjustment instruction is an instruction in the path node adjustment instruction set used to reduce the machining speed of the high-temperature risk node. Analyze the speed reduction adjustment instruction to obtain the target node identifier of the high-temperature risk node and the allowed maximum speed reduction ratio. The actual speed reduction coefficient is a coefficient calculated according to the real-time temperature value in the physical sensing data set, which represents the proportion of the speed that actually needs to be reduced. By comparing the real-time temperature value with the preset temperature threshold, calculate the actual speed reduction coefficient to ensure that the speed reduction operation can effectively reduce the high-temperature risk.

[0176] Step S320: Linearly adjust the machining speed parameter of the high-temperature risk node based on the actual speed reduction coefficient to generate the node speed parameter after speed reduction, and synchronously update the transmission time compensation value of the adjacent nodes in the initial machining path.

[0177] Linear adjustment is to linearly transform the machining speed parameter of the high-temperature risk node according to the actual speed reduction coefficient to obtain the node speed parameter after speed reduction. The node speed parameter after speed reduction can reduce the machining speed of the high-temperature risk node and reduce the generation of heat. Synchronously update the transmission time compensation value of the adjacent nodes in the initial machining path. Since the machining speed of the high-temperature risk node is reduced, it may affect the transmission time of the adjacent nodes. Therefore, it is necessary to update the transmission time compensation value to ensure the overall rationality of the machining path.

[0178] Step S330: Analyze the device switching instructions in the path node adjustment instruction set, determine the spare device number and switching response delay of the pressure anomaly node, and calculate the coordinate calibration offset after device switching based on the real-time device position data.

[0179] The device switching instruction is another type of instruction in the path node adjustment instruction set, which is used to switch the processing device of the pressure anomaly node. Analyze the device switching instruction to determine the spare device number and switching response delay of the pressure anomaly node.

[0180] The coordinate calibration offset is the offset for coordinate calibration required after device switching calculated based on the real-time device position data. Since the position of the spare device may be different from that of the original device, it is necessary to calculate the coordinate calibration offset to ensure the accuracy of the processing operation.

[0181] Step S340: Perform position compensation on the processing coordinates of the pressure anomaly node based on the coordinate calibration offset, generate the calibrated node coordinate parameters, and accumulate the switching response delay to the transmission time compensation value.

[0182] Perform position compensation on the processing coordinates of the pressure anomaly node based on the coordinate calibration offset to obtain the calibrated node coordinate parameters. The calibrated node coordinate parameters can ensure that the spare device performs processing operations at the correct position.

[0183] Accumulate the switching response delay to the transmission time compensation value. Since device switching takes a certain amount of time, it is necessary to consider the switching response delay in the transmission time and update the transmission time compensation value.

[0184] Step S350: Analyze the path replanning instruction in the path node adjustment instruction set, identify the start node and end node of the transmission conflict segment, and insert the candidate coordinate set of the transfer node between the start node and the end node.

[0185] The path replanning instruction is a type of instruction in the path node adjustment instruction set, which is used to replan the processing path of the transmission conflict segment. Analyze the path replanning instruction to identify the start node and end node of the transmission conflict segment. The candidate coordinate set of the transfer node is the coordinate set of the transfer nodes that may be inserted between the start node and the end node. By inserting transfer nodes, transmission conflicts can be avoided and the processing path can be optimized.

[0186] Step S360: According to the device load rate and idle state time series in the physical sensing data set, screen the transfer node coordinates that meet the load constraints from the candidate coordinate set to generate the replanned path topology structure.

[0187] The equipment load rate refers to the load condition of the processing equipment, and the idle state time series refers to the idle time series of the processing equipment. According to the equipment load rate and the idle state time series in the physical sensing data set, the transfer node coordinates that meet the load constraints are screened from the candidate coordinate set. The re-planned path topology structure is the new path topology structure formed by the transfer node coordinates determined after screening, which can avoid transmission conflicts and improve processing efficiency.

[0188] Step S370: Integrate the decelerated node speed parameters, the calibrated node coordinate parameters, and the re-planned path topology structure to generate the complete node sequence of the target processing path and the transmission time between nodes, and verify whether the node parameters of the target processing path meet the dynamic process constraint conditions.

[0189] Integrate the decelerated node speed parameters, the calibrated node coordinate parameters, and the re-planned path topology structure to obtain the complete node sequence of the target processing path and the transmission time between nodes. The complete node sequence determines the order of processing operations, and the transmission time between nodes represents the transmission time between nodes. Verify whether the node parameters of the target processing path meet the dynamic process constraint conditions, and the dynamic process constraint conditions are conditions that change in real time according to the actual processing situation. If the node parameters do not meet the dynamic process constraint conditions, it is necessary to further adjust the target processing path to ensure the smooth progress of processing operations.

[0190] Step S400: Allocate real-time control instructions to multiple processing devices according to the process constraint conditions, and monitor the feedback signals after the processing devices execute the control instructions.

[0191] Real-time control instruction allocation is to allocate corresponding control instructions to multiple processing devices according to the process constraint conditions of each processing node in the target processing path. These control instructions include temperature control, pressure control, power control, etc., to ensure that the processing devices can perform processing operations as required.

[0192] The feedback signal is the signal returned after the processing device executes the control instruction, which reflects the actual operating state of the device. Monitoring the feedback signal can timely understand the execution situation of the processing device, discover potential problems, and take corresponding measures for adjustment.

[0193] Specifically, step S400 allocates real-time control instructions to multiple processing devices according to the process constraint conditions, including the following steps: Step S410: Analyze the temperature threshold range, pressure threshold range, and processing accuracy level in the process constraint conditions, and generate a temperature control instruction set for the heating device based on the temperature threshold range.

[0194] The temperature threshold range is the allowable range of temperature during the processing specified in the process constraints, the pressure threshold range is the allowable range of pressure, and the machining accuracy grade is the requirement for machining accuracy. Analyzing these parameters in the process constraints provides a basis for subsequent instruction generation.

[0195] Generate a temperature control instruction set for the heating equipment based on the temperature threshold range. The temperature control instruction set contains temperature control instructions that the heating equipment needs to execute, such as heating, heat preservation, cooling and other instructions. Through the temperature control instruction set, the temperature during the processing can be accurately controlled.

[0196] Step S420: Segmentally calibrate the pressure parameters of the stamping equipment according to the pressure threshold range to generate a pressure gradient adjustment instruction.

[0197] Segmentally calibrate the pressure parameters of the stamping equipment according to the pressure threshold range, divide the pressure range into multiple intervals, and calibrate the pressure parameters of each interval. The pressure gradient adjustment instruction is an instruction generated according to the calibration result, which is used to adjust the pressure gradient of the stamping equipment to ensure that the pressure during the stamping process meets the process requirements.

[0198] Step S430: Modulate the output power of the laser engraving equipment based on the machining accuracy grade to generate a power adaptation instruction sequence.

[0199] The output power is an important parameter of the laser engraving equipment, which directly affects the engraving accuracy and effect. Modulate the output power of the laser engraving equipment based on the machining accuracy grade, and adjust the output power according to different machining accuracy requirements.

[0200] The power adaptation instruction sequence is an instruction sequence generated according to the modulation result, which is used to control the output power of the laser engraving equipment to make it match the machining accuracy grade.

[0201] Step S440: Verify the instruction synchronization of the temperature control instruction set, the pressure gradient adjustment instruction, and the power adaptation instruction sequence according to the processing time sequence, and delete the instruction segments with time conflicts.

[0202] The processing time sequence is the time sequence of processing operations. Verify the instruction synchronization of the temperature control instruction set, the pressure gradient adjustment instruction, and the power adaptation instruction sequence according to the processing time sequence to check whether these instructions are coordinated with each other in time.

[0203] Delete the instruction segments with time conflicts. Time conflicts may cause the processing equipment to execute in a chaotic manner and affect the processing quality. By deleting the instruction segments with time conflicts, ensure the synchronization and coordination of the instructions.

[0204] Step S450: Distribute the verified instruction set to the corresponding processing equipment, and record the instruction reception timestamp and execution status code of each processing equipment.

[0205] Distribute the verified instruction set to the corresponding processing equipment to ensure that each processing equipment can receive the correct control instructions. Record the instruction reception timestamp and execution status code of each processing equipment. The instruction reception timestamp is used to record the time when the instruction arrives at the equipment, and the execution status code is used to reflect the situation of the equipment executing the instruction, such as successful execution, execution failure, etc.

[0206] Step S460: Real-time collect the current fluctuation data and vibration frequency data of the processing equipment when executing the instruction set.

[0207] The current fluctuation data and vibration frequency data are important indicators reflecting the operating state of the processing equipment. Real-time collect the current fluctuation data and vibration frequency data of the processing equipment when executing the instruction set. By analyzing these data, the operating condition of the equipment can be understood, and potential faults and problems can be discovered.

[0208] Step S470: Perform frequency-domain transformation on the current fluctuation data, extract the energy distribution characteristics within a preset frequency band, and judge whether the equipment is in an overloaded state according to the energy distribution characteristics.

[0209] Frequency-domain transformation is the process of converting the current fluctuation data from the time domain to the frequency domain. Common frequency-domain transformation methods include Fourier transform, etc. Through frequency-domain transformation, the current fluctuation data can be decomposed into components of different frequencies, and the energy distribution characteristics within a preset frequency band can be extracted. The energy distribution characteristics reflect the energy distribution of the current at different frequencies. According to the energy distribution characteristics, it can be judged whether the equipment is in an overloaded state. If the energy within the preset frequency band is too high, it may indicate that the equipment is in an overloaded state.

[0210] Step S480: Perform time-domain segmentation on the vibration frequency data, obtain the peak amplitudes within multiple time windows, and calculate the equipment stability index based on the peak amplitudes.

[0211] Time-domain segmentation is to segment the vibration frequency data according to time to obtain multiple time windows. Obtain the peak amplitude within each time window. The peak amplitude reflects the vibration intensity of the equipment within that time window.

[0212] The equipment stability index is an index calculated based on the peak amplitude, and it is used to evaluate the stability of the equipment. The higher the equipment stability index, the more stable the operation of the equipment.

[0213] Step S490: When detecting an overloaded state or the equipment stability index is lower than the safety threshold, send a path interruption signal to the parameter adjustment model and trigger the path generation model to regenerate an alternative processing path.

[0214] The safety threshold is the lower limit value of the pre-set device stability index. When an overload state is detected or the device stability index is lower than the safety threshold, it indicates that there may be a fault or problem with the device, and measures need to be taken promptly.

[0215] Send a path interruption signal to the parameter adjustment model to interrupt the current processing path. At the same time, trigger the path generation model to regenerate an alternative processing path to ensure that the processing operation can continue and avoid processing interruption due to device problems.

[0216] Step S500: Jointly iteratively update the path generation model and the parameter adjustment model based on the feedback signal until the processing error rate of the target processing path is lower than the preset threshold.

[0217] Jointly iteratively updating the path generation model and the parameter adjustment model based on the feedback signal is a key step to improve the processing control accuracy and stability of the circuit board. The feedback signal contains the actual situation of the processing process. By analyzing and utilizing it, the model can be continuously optimized to make the target processing path more in line with the actual processing requirements and reduce the processing error rate.

[0218] Specifically, step S500 jointly iteratively updates the path generation model and the parameter adjustment model based on the feedback signal, including the following steps: Step S510: Extract the actual execution duration data of the processing path and the device energy consumption data from the feedback signal, and calculate the deviation ratio of the actual execution duration to the predicted duration.

[0219] The feedback signal is a series of data returned by the processing device after executing the control instruction, which includes the actual execution duration data of the processing path and the device energy consumption data. The actual execution duration data refers to the time actually spent by the processing device to complete the entire processing process according to the target processing path, which reflects the time consumption of the actual processing process. The device energy consumption data reflects the energy consumption of the processing device during operation, which is closely related to the working efficiency of the device and the rationality of the processing technology.

[0220] Calculate the deviation ratio of the actual execution duration to the predicted duration. The predicted duration is the processing time estimated by the path generation model when generating the target processing path. The calculation formula for the deviation ratio is: deviation ratio = (actual execution duration - predicted duration) / predicted duration × 100%. This deviation ratio can intuitively reflect the difference between the actual processing time and the expected time, providing an important basis for subsequent adjustment of the path generation model. For example, if the deviation ratio is positive and large, it means that the actual execution time is longer than the predicted time, and there may be problems such as an unreasonable processing path and low device operation efficiency; if the deviation ratio is negative and large, it may be that the predicted duration is set too long and the accuracy of the model needs to be improved.

[0221] Step S520: Compensate and correct the node transfer time-consuming parameter in the path generation model according to the deviation ratio to generate the first model update gradient.

[0222] The node transfer time-consuming parameter is a parameter in the path generation model used to describe the time required from one processing node to another processing node, and it is an important basis for generating the processing path and predicting the processing duration. Compensate and correct the node transfer time-consuming parameter in the path generation model according to the deviation ratio calculated in step S510.

[0223] If the deviation ratio is positive, it means that the actual transfer time is longer than the model's prediction. At this time, it is necessary to appropriately increase the value of the node transfer time-consuming parameter; on the contrary, if the deviation ratio is negative, it is necessary to appropriately decrease the value of the node transfer time-consuming parameter. In this way, the model can more accurately predict the processing time.

[0224] The first model update gradient is calculated based on the compensated and corrected node transfer time-consuming parameter, and it represents the change amount of the path generation model in the parameter update direction. In machine learning, the gradient is a vector that points in the direction of the fastest growth of the function. By updating the model parameters along the opposite direction of the gradient, the loss function of the model can be gradually reduced, thereby improving the performance of the model. For example, when using the stochastic gradient descent algorithm for model update, the parameters of the path generation model will be adjusted according to the first model update gradient, so that the predicted duration of the processing path generated by the model is closer to the actual execution duration.

[0225] Step S530: Extract the actual processing accuracy data of the processing node from the feedback signal and compare it with the expected processing accuracy in the target processing path to generate an accuracy error distribution map.

[0226] The actual processing accuracy data refers to the processing accuracy indicators actually achieved by the processing equipment at each processing node, such as dimensional accuracy, shape accuracy, etc. These data can be collected through high-precision measuring equipment, such as coordinate measuring machines, laser scanners, etc. The expected processing accuracy is the accuracy requirement set for each processing node in the target processing path, and it is determined according to the design requirements and process standards of the circuit board.

[0227] Compare the actual processing accuracy data with the expected processing accuracy, and calculate the accuracy error of each processing node. The calculation method of the accuracy error can be determined according to specific accuracy indicators. For example, for dimensional accuracy, the difference between the actual size and the expected size can be calculated. By statistically analyzing the accuracy errors of all processing nodes, an accuracy error distribution map is generated. The accuracy error distribution map visually shows the accuracy error conditions of each processing node in a graphical way, which can help quickly locate the nodes with large accuracy errors and provide a direction for the subsequent optimization of the parameter adjustment model.

[0228] Step S540: Iteratively optimize the rule for generating process constraint conditions in the parameter adjustment model according to the accuracy error distribution map to generate a second model update gradient.

[0229] The rule for generating process constraint conditions is the rule used in the parameter adjustment model to generate the process constraint conditions for each processing node. It determines the operating parameters and operation requirements of the processing equipment at each processing node. According to the accuracy error distribution map generated in step S530, the rule for generating process constraint conditions is iteratively optimized.

[0230] If the accuracy error of a certain processing node is large, it indicates that the process constraint conditions of this node may be unreasonable, and the rule for generating the process constraint conditions of this node needs to be adjusted. For example, if the actual hole diameter accuracy error at a certain drilling node is large, it may be necessary to adjust the rules for generating parameters such as drilling speed and drill bit rotation speed in the drilling process constraint conditions. By continuously optimizing the rule for generating process constraint conditions according to the accuracy error distribution map, the parameter adjustment model can generate more reasonable process constraint conditions and improve the processing accuracy.

[0231] The second model update gradient is calculated based on the iteratively optimized rule for generating process constraint conditions. It represents the change amount of the parameter adjustment model in the parameter update direction. Similar to the first model update gradient, the second model update gradient is used to guide the parameter update of the parameter adjustment model, so that the model can better adapt to the actual processing situation and reduce the processing accuracy error.

[0232] Step S550: Adopt an alternating training strategy to fuse the first model update gradient and the second model update gradient, and synchronously update the network parameters of the path generation model and the parameter adjustment model.

[0233] The alternating training strategy is a training method that alternates the update processes of the path generation model and the parameter adjustment model. In each iteration, first, the network parameters of the path generation model are updated according to the update gradient of the first model, so that the path generation model can more accurately estimate the processing time; then, the network parameters of the parameter adjustment model are updated according to the update gradient of the second model, so that the parameter adjustment model can generate more reasonable process constraints. Through this alternating training method, the two models cooperate with each other to jointly improve the accuracy and stability of printed circuit board processing control.

[0234] Network parameters are learnable parameters in the path generation model and the parameter adjustment model, such as weights and biases in a neural network. During the alternating training process, these network parameters are adjusted according to the update gradient of the first model and the update gradient of the second model. Specific update methods can use common optimization algorithms, such as Stochastic Gradient Descent (SGD), Adagrad, Adam, etc. Through continuous iterative updates, the performance of the path generation model and the parameter adjustment model is continuously improved, and finally the processing error rate of the target processing path is lower than the preset threshold.

[0235] The preset threshold is a pre-set processing error rate standard, which is an important indicator for measuring the processing quality of printed circuit boards. When the processing error rate of the target processing path is lower than the preset threshold, it indicates that the processing quality of the printed circuit board meets the expected requirements. At this time, it can be considered that the joint iterative update process of the path generation model and the parameter adjustment model is completed, and the processing control method has achieved good results.

[0236] In practical applications, this joint iterative update process may need to be carried out multiple times, and each iteration will further optimize the model according to the feedback signal. At the same time, in order to ensure the generalization ability and stability of the model, technical means such as cross-validation and regularization can also be used. For example, when using cross-validation, the dataset can be divided into a training set, a validation set, and a test set. During the training process, the performance of the model is evaluated through the validation set to avoid overfitting of the model; when using the regularization technique, a regularization term can be added to the loss function to limit the value range of the model parameters and improve the generalization ability of the model.

[0237] In addition, with the use of processing equipment and the change of environmental conditions, the characteristics of the processing process may also change. Therefore, it is necessary to regularly collect and analyze the feedback signal, and continuously perform joint iterative updates on the path generation model and the parameter adjustment model to adapt to the changing processing requirements. For example, when the processing equipment shows wear, aging, etc., its operating performance may change. At this time, by updating the model in a timely manner, the processing accuracy and efficiency can be ensured not to be affected.

[0238] In summary, the circuit board processing control method provided by the present invention for the industrial Internet of Things realizes precise control and optimization of the circuit board processing process through a series of steps such as collecting real-time device status data, generating an initial processing path, adjusting parameters, allocating control instructions, and jointly iteratively updating the model based on feedback signals. This method makes full use of the advantages of the industrial Internet of Things and combines machine learning models to be able to adjust the processing path and process parameters in real time according to the actual processing situation, improving the processing quality and production efficiency of the circuit board, and having important practical application value.

[0239] Please refer to Figure 2 , Figure 2 which is a schematic structural diagram of a circuit board processing control system provided by an embodiment of the present invention. This circuit board processing control system is, for example, a control device in a processing production line, such as a computer device, and at least includes a processor 101, a communication interface 102, and a memory 103. Among them, the processor 101, the communication interface 102, and the memory 103 can be connected through a bus or other means. Among them, the processor 101 (or Central Processing Unit (CPU)) is the computing core and control core of the circuit board processing control system, which can parse various instructions in the circuit board processing control system and process various data of the circuit board processing control system. The communication interface 102 can optionally include a standard wired interface, a wireless interface (such as WI-FI, a mobile communication interface, etc.), and can be controlled by the processor 101 to be used for sending and receiving data; the communication interface 102 can also be used for the transmission and interaction of internal data of the circuit board processing control system. The memory 103 (Memory) is a memory device in the circuit board processing control system for storing programs and data. It can be understood that the memory 103 here can include both the built-in memory of the circuit board processing control system and, of course, the extended memory supported by the circuit board processing control system. The memory 103 provides a storage space, and this storage space stores the operating system of the circuit board processing control system, which can include but is not limited to: Android system, iOS system, Windows Phone system, etc., and the present invention does not make any limitations in this regard. In one embodiment, the processor 101 executes the circuit board processing control method for the industrial Internet of Things provided by the above embodiments of the present invention by running the computer program in the memory 103.

Claims

1. A circuit board processing control method applied to industrial Internet of Things, characterized in that: The method comprises: Collecting real-time equipment status data of multiple processing equipment during the processing of the target circuit board, wherein the equipment status data includes a processing parameter set and a physical sensor data set; Based on the historical processing path data in the processing parameter set, calling the path generation model to generate an initial processing path, and inputting the initial processing path and the physical sensor data set into the parameter adjustment model for matching, and generating a path node adjustment instruction; Optimizing the parameters of the processing nodes in the initial processing path based on the path node adjustment instructions through the parameter adjustment model to generate a target processing path, and extracting the process constraint conditions of each processing node in the target processing path; Allocating real-time control instructions to the plurality of processing equipment according to the process constraints, and monitoring feedback signals after the processing equipment executes the control instructions; The path generation model and the parameter adjustment model are jointly iteratively updated based on the feedback signal until the machining error rate of the target machining path is lower than a preset threshold.

2. The method according to claim 1, characterized in that The processing parameter set includes text-based processing parameters and image-based processing parameters. Based on the historical processing path data in the processing parameter set, calling the path generation model to generate the initial processing path includes: Performing semantic segmentation on the text-based processing parameters to extract multiple process description fragments, and performing edge contour detection on the image-based processing parameters to generate a corresponding processing area positioning map; Aligning the process description fragment with the processing area positioning map in spatial coordinates, constructing a process feature sequence, and inputting the process feature sequence into the temporal coding layer of the path generation model; Topologically sorting the process dependencies in the process feature sequence through the temporal coding layer to generate a process node connection graph; The path prediction layer in the path generation model is called to generate the node arrangement sequence and the transmission time between nodes of the initial processing path according to the process weight of each node in the process node connection diagram.

3. The method according to claim 2, characterized in that The step of aligning the process description fragment with the processing area positioning map in spatial coordinates to construct a process feature sequence includes: Extracting a text paragraph containing a location identifier from the process description segment, generating a location keyword set, and identifying a coordinate region boundary corresponding to the location identifier in the processing region location map; According to the coordinates of the geometric center point of the coordinate area boundary, the regional anchor point coordinates are assigned to each position keyword, and a keyword list with coordinate labels is generated; Performing topological traversal on the contour segmentation lines in the processing area positioning map, determining the processing order priority of each contour area, and sorting the keyword list with coordinate labels based on the processing order priority; Performing distance matching between the coordinates of the regional anchor points in the sorted keyword list and the coordinates of the actual processing path starting point in the processing area positioning map, and screening out candidate coordinate pairs whose distance deviation is less than a tolerance threshold; According to the process description content of the keywords in the candidate coordinate pair, semantically annotate the contour area of ​​the processing area positioning map to generate an annotated process space distribution map; According to the distribution density of the marked areas in the process space distribution diagram, the text content in the process description segment is temporally bound to the corresponding marked areas to form a process feature sequence containing a spatial coordinate dependency relationship.

4. The method according to claim 2, characterized in that: The topological sorting of the process dependencies in the process feature sequence by the temporal coding layer to generate a process node connection graph includes: Extracting a node identifier and a corresponding processing action description of each process node from the process feature sequence, and generating a process node feature set including input-output relationships between nodes; Traversing the process node feature set, analyzing the predecessor process node and the successor process node of each process node, establishing a process dependency chain, and marking the process transmission direction between the nodes in the process dependency chain; Reconstructing the process dependency chain into a directed graph according to the process transmission direction, and determining the initial in-degree value and triggerable execution state of each process node; Based on the initial in-degree value, a candidate starting node with an in-degree of zero is selected from the process node feature set, and the candidate starting node is added to a queue of nodes to be processed; Sequentially taking out the current processing node from the queue of nodes to be processed, traversing the successor process nodes of the current processing node, updating the dynamic in-degree values ​​of the successor process nodes, and adding the successor process nodes whose dynamic in-degree values ​​are reduced to zero to the queue of nodes to be processed; Recording the connection sequence and process transmission direction between the current processing node and the subsequent process node, and generating a directed edge set with weight identification; A process node connection graph including a topological hierarchical relationship is generated according to the directed edge set and the arrangement order of the nodes in the process node feature set, wherein the topological hierarchical relationship is used to indicate the parallel execution interval and serial dependency depth of the process node.

5. The method according to claim 2, characterized in that: The step of generating the node arrangement sequence and the transmission time between nodes of the initial processing path according to the process weight of each node in the process node connection diagram includes: Obtaining the processing time weight, equipment switching weight and path continuity weight included in the process weight, and generating a weight type association table based on the weight type; Traversing all nodes in the process node connection graph, normalizing the processing time weight of each node according to the weight type association table, and generating a node comprehensive priority score; Extracting a set of predecessor nodes of a current node from the process node connection graph, and determining a triggerable execution condition of the current node according to a comprehensive priority score of the set of predecessor nodes; Based on the triggerable execution condition, candidate nodes that meet the path continuity weight requirement are screened, and the candidate nodes are arranged in descending order according to the node comprehensive priority scores to generate a candidate node execution queue; Selecting head nodes from the candidate node execution queue in turn to add to the initial processing path sequence, and calculating the device switching time between the head node and the previous node based on the device switching weight; According to the set of successor nodes of the head node in the process node connection graph, the node arrangement order in the candidate node execution queue is updated, and the transmission time increment from the current node to the successor node is recorded; The node selection, switching time consumption calculation and queue update steps are repeatedly performed until the initial processing path sequence includes all nodes in the process node connection diagram, and the accumulated value of the transmission time consumption between the nodes is synchronously output.

6. The method according to claim 1, characterized in that The parameter adjustment model is obtained through multi-stage training, and the multi-stage training includes: Acquire a first-stage training data set, where the first-stage training data set includes multiple sets of equipment status sample data and their corresponding standard processing path labels; Inputting the equipment status sample data into an initial parameter adjustment model to generate a predicted processing path, and calculating a path coincidence loss value between the predicted processing path and the standard processing path label; Acquire a second-stage training data set, where the second-stage training data set includes multiple sets of abnormal process sample data and their corresponding path correction labels; Inputting the abnormal process sample data into the initial parameter adjustment model trained in the first stage, generating a revised predicted processing path, and calculating an error compensation loss value between the revised predicted processing path and the path correction label; The path coincidence loss value and the error compensation loss value are fused in a preset ratio to obtain a comprehensive training loss value, and the initial parameter adjustment model is back-propagated and optimized according to the comprehensive training loss value until the comprehensive training loss value converges.

7. The method according to claim 6, characterized in that The calculating of the path overlap loss value between the predicted processing path and the standard processing path label includes: Extracting a position coordinate sequence of each processing node in the predicted processing path, and matching the position coordinate sequence with a reference coordinate sequence in the standard processing path label point by point; Performing Euclidean distance calculation on unmatched coordinate points to generate a first local loss component, and performing processing time difference calculation on matched coordinate points to generate a second local loss component; Performing a weighted summation on the first local loss component and the second local loss component, and generating a dynamic adjustment coefficient according to the weighted summation result; The loss weights of the unmatched coordinate points and the matched coordinate points are redistributed based on the dynamic adjustment coefficient to obtain the path coincidence loss value.

8. The method according to claim 1, characterized in that The allocating real-time control instructions to the plurality of processing equipment according to the process constraints comprises: Analyzing the temperature threshold range, pressure threshold range and processing accuracy level in the process constraint conditions, and generating a temperature control instruction set for the heating device based on the temperature threshold range; Performing segmented calibration on the pressure parameters of the stamping equipment according to the pressure threshold range, and generating a pressure gradient adjustment instruction; Modulating the output power of the laser engraving device based on the processing accuracy level to generate a power adaptation instruction sequence; Verify the synchronization of the temperature control instruction set, the pressure gradient adjustment instruction and the power adaptation instruction sequence according to the processing time sequence, and delete the instruction fragments with time conflicts; Distribute the verified instruction set to the corresponding processing equipment, and record the instruction receiving timestamp and execution status code of each processing equipment; collecting in real time the current fluctuation data and the vibration frequency data of the processing equipment when executing the instruction set; Performing frequency domain transformation on the current fluctuation data, extracting energy distribution characteristics within a preset frequency band, and judging whether the device is in an overload state according to the energy distribution characteristics; Performing time domain segmentation on the vibration frequency data, obtaining peak amplitudes in multiple time windows, and calculating a device stability index based on the peak amplitudes; When the overload state is detected or the equipment stability index is lower than a safety threshold, a path interruption signal is sent to the parameter adjustment model, and the path generation model is triggered to regenerate an alternative processing path.

9. The method according to claim 1, characterized in that: The jointly iteratively updating the path generation model and the parameter adjustment model based on the feedback signal includes: Extracting actual execution time data and equipment energy consumption data of the processing path from the feedback signal, and calculating the deviation ratio between the actual execution time and the predicted time; Compensating and correcting the node transmission time-consuming parameter in the path generation model according to the deviation ratio to generate a first model update gradient; Extracting actual machining accuracy data of the machining node from the feedback signal, and comparing it with the expected machining accuracy in the target machining path, to generate an accuracy error distribution map; Iteratively optimize the process constraint condition generation rule in the parameter adjustment model according to the precision error distribution map to generate a second model update gradient; An alternating training strategy is adopted to fuse the first model update gradient and the second model update gradient, and the network parameters of the path generation model and the parameter adjustment model are synchronously updated.

10. A circuit board processing control system, characterized in that: include: a memory, wherein a computer program is stored in the memory; A processor is used to load the computer program to implement the circuit board processing control method applied to the industrial Internet of Things as described in any one of claims 1 to 9.

Images