A vacuum adsorption system running parameter optimization method based on transfer learning

Abstract

This invention discloses a method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning. Specifically, the method includes: aligning historical and current adsorption condition data by time and dimension to form a sequence of operating condition features; extracting material, roughness, load, pressure, flow rate, leakage, and cycle time features to form a set of node vectors; constructing an initial structure of a semi-naive Bayesian network to generate a set of candidate parent sets; performing parent set competition explicitness calculations to determine the dominant parent set identifier; statistically analyzing the evidence quantity of historical, current, and conflicting samples to generate a local evidence concentration sequence and form a conditional probability table; performing parameter inference to generate vacuum pressure values, pressure holding time values, valve switching time values, and leakage compensation amplitude values; and performing constraint verification to output a set of operating parameters. This invention achieves adaptive optimization of the operating parameters of a vacuum adsorption system under the combined influence of historical and current operating conditions.

Description

Technical Field

[0001] This invention relates to the field of vacuum adsorption control technology, and in particular to a method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning. Background Technology

[0002] Vacuum adsorption systems are widely used in material handling, sorting, loading, positioning assembly, and automated production processes. They achieve stable gripping and transfer of workpieces through negative pressure adsorption. In actual operation, different workpieces exhibit significant differences in material type, surface roughness, size, load conditions, and leakage characteristics. A single set of vacuum pressure values, holding time values, valve switching times, and leakage compensation values ​​is often insufficient to simultaneously adapt to various adsorption conditions. Especially in scenarios with frequent changes in operating conditions, improper setting of operating parameters can easily lead to problems such as unstable adsorption, workpiece drop, high energy consumption, response lag, and cycle time fluctuations, thereby affecting the overall stability of the machine and the continuity of production.

[0003] In existing technologies, parameter setting for vacuum adsorption systems largely relies on empirical tuning, fixed rule matching, or establishing local optimization models for single operating conditions. These methods typically only utilize current operating condition data for analysis, lacking the inheritance and utilization of historical operating condition patterns, making it difficult to quickly generate targeted parameter configuration results when new operating conditions arrive. Furthermore, existing solutions insufficiently utilize the consistency, conflict, and transferability relationships between historical and current samples, easily leading to poor adaptability of parameter inference results and difficulty in balancing adsorption stability and parameter adjustment accuracy. Therefore, there is an urgent need for a vacuum adsorption system operating parameter optimization method that can combine historical and current operating conditions and utilize transfer learning to optimize operating parameters. Summary of the Invention

[0004] One objective of this invention is to propose a method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning. This invention introduces a transfer association table mechanism, a parent set competition explicitation calculation model, and a local evidence concentration adjustment mechanism. It performs transfer association analysis, conditional probability mass redistribution, and parameter inference calculations on historical and current adsorption condition data, constructing a continuous optimization process for vacuum adsorption operating parameters oriented towards multi-condition switching scenarios. This method possesses advantages such as strong adaptability to operating conditions, high parameter output accuracy, and good adsorption operation stability.

[0005] An embodiment of the present invention provides a method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning, comprising the following steps:

[0006] Obtain historical adsorption condition data sets and current adsorption condition data sets, and perform time alignment and dimension unification operations on the historical and current adsorption condition data sets to form a condition feature sequence;

[0007] Perform feature parsing operations on the working condition feature sequence to extract workpiece material features, surface roughness features, load features, pressure change features, flow change features, leakage change features, and cycle time features, forming a set of node vectors;

[0008] Construct the initial structure of a semi-naive Bayesian network, set up adsorption state nodes, parameter nodes, and result nodes in the node vector set, establish a migration association table from historical working condition nodes to current working condition nodes, and generate a set of candidate parent sets corresponding to each parameter node.

[0009] Perform parent set competition explicit calculation on the candidate parent set corresponding to each parameter node, read the working condition adaptation value, migration support value, and local correlation value corresponding to each candidate parent set, and generate the dominant parent set identifier.

[0010] At the corresponding position of the dominant parent set identifier, the amount of historical sample evidence, the amount of current sample evidence, and the amount of conflicting sample evidence are counted to generate a local evidence concentration sequence. Then, the conditional probability quality corresponding to the dominant parent set is partitioned and redistributed to form a conditional probability table.

[0011] Perform parameter inference calculations on the conditional probability table to generate vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude;

[0012] Perform constraint verification calculations on vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude, and output a set of operating parameters.

[0013] Optionally, extracting workpiece material features, surface roughness features, load features, pressure change features, flow rate change features, leakage change features, and cycle time features specifically includes:

[0014] Read the workpiece identification data, perform category-based calculations, and generate workpiece material characteristics;

[0015] Read the workpiece surface inspection data, perform roughness interval division calculation, and generate surface roughness features;

[0016] Read mass data and dimension data, perform numerical conversion calculations, extract mass components, area components, and center of gravity offset components, and generate load characteristics;

[0017] Read the vacuum pressure time series, perform calculations of the difference between adjacent time points, maximum value extraction calculation, minimum value extraction calculation, and duration calculation to generate pressure change characteristics;

[0018] Read the traffic time series, perform calculations for the difference between adjacent time points, maximum value extraction, minimum value extraction, and stage average value, and generate traffic change characteristics;

[0019] Read the leak monitoring sequence, perform unit time period change calculation, duration calculation, and adjacent time difference comparison calculation to generate leak change characteristics;

[0020] Read the task cycle data, perform single adsorption duration calculation, adjacent adsorption interval calculation, and task count per unit time period calculation, and generate cycle characteristics;

[0021] Vector assembly calculations are performed on the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics, and cycle time characteristics to form a set of node vectors.

[0022] Optionally, the construction of the initial structure of the semi-naive Bayesian network, setting adsorption state nodes, parameter nodes, and result nodes within the node vector set, establishing a migration association table from historical working condition nodes to the current working condition node, and generating a candidate parent set corresponding to each parameter node includes the following steps:

[0023] Perform node category classification calculation on the node vector set, and write the adsorption state node identifier, parameter node identifier and result node identifier at the corresponding position of each node;

[0024] Hierarchical calculations are performed on nodes with adsorption state node identifiers, nodes with parameter node identifiers, and nodes with result node identifiers. State layer, parameter layer, and result layer are established sequentially at the network level to form the initial hierarchical structure of the semi-naive Bayesian network.

[0025] Perform source tag calculation on the node vectors corresponding to the historical adsorption condition data set and the node vectors corresponding to the current adsorption condition data set, and write the historical condition identifier and the current condition identifier at the node source position respectively.

[0026] Perform similar feature matching calculations on nodes with historical operating condition identifiers and nodes with current operating condition identifiers. Establish node correspondence based on matching conditions of consistent feature categories, consistent parameter semantics, and corresponding value ranges, and form a migration association table from historical operating condition nodes to current operating condition nodes.

[0027] Perform parent node source screening calculations for each parameter node, screen nodes that have state influence relationships with the corresponding parameter node in the adsorption state nodes, and screen nodes that have migration association relationships with the corresponding parameter node in the historical working condition nodes according to the migration association table, forming the initial parent node sequence for each parameter node.

[0028] Perform node combination calculations on the initial parent node sequence corresponding to each parameter node, and generate parent node combination sequences according to single node combination, double node combination and multi node combination.

[0029] Perform constraint filtering calculations on the parent node combination sequence, delete combinations containing result nodes, delete combinations that do not satisfy hierarchical connection relationships, delete combinations with semantic conflicts between nodes, retain node combinations that satisfy parameter dependency requirements, and generate a set of candidate parent sets corresponding to each parameter node.

[0030] Optionally, the step of performing parent set competition explicit calculation on the candidate parent set corresponding to each parameter node specifically includes:

[0031] Read the candidate parent set set corresponding to each parameter node, perform sequential numbering calculation on each parent set combination in the candidate parent set set set, and write the parent set identifier at the corresponding position of each parent set combination.

[0032] Read the node combination corresponding to each parent set identifier, extract the adsorption state node, historical working condition node and current working condition node from the node combination, perform combination expansion calculation, and form the parent set composition sequence.

[0033] Perform working condition adaptation calculation on the sequence of parent sets, read the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics and cycle time characteristics corresponding to each node, calculate the feature matching degree of each parent set combination to the corresponding parameter node, and generate working condition adaptation values.

[0034] Perform migration support calculation on the parent set composition sequence, read the migration association relationship between historical working condition nodes and current working condition nodes in the parent set composition sequence, count the historical support sample size, current corresponding sample size and migration consistent sample size related to the corresponding parameter node, and generate migration support value;

[0035] Perform local association calculations on the sequence composed of the parent set, read the local conditional dependencies between each node in the sequence composed of the parent set and the corresponding parameter node, calculate the local association strength of the node combination to the corresponding parameter node, and generate local association values.

[0036] Competitive explicit calculations are performed on the working condition adaptation values, migration support values, and local correlation values, and competitive comparison results of each parent set combination are generated within the candidate parent set corresponding to the same parameter node.

[0037] Perform a selection and filtering calculation on the competition comparison results, extract the parent set combination with the highest competition result, and write the dominant parent set identifier at the corresponding parameter node position.

[0038] Optionally, generating the local evidence concentration sequence includes the following steps:

[0039] Read the dominant parent set identifier corresponding to each parameter node, and extract the parent set combination corresponding to the dominant parent set identifier;

[0040] Read the historical operating condition nodes, current operating condition nodes, and adsorption state nodes corresponding to the dominant parent set, and establish the sample collection range according to the parameter node category;

[0041] Within the sample collection range, retrieve historical adsorption condition data corresponding to the dominant parent set combination, count the number of historical samples that satisfy the parent set value combination and have consistent result values, and generate historical sample evidence quantity.

[0042] Within the sample collection range, retrieve the current adsorption condition data corresponding to the dominant parent set combination, count the number of current samples that satisfy the parent set value combination and have consistent parameter responses, and generate the current sample evidence quantity.

[0043] Conflict detection calculations are performed on historical adsorption condition data and current adsorption condition data. The number of samples that satisfy the combination of parent set values ​​but have inconsistent result values, inconsistent parameter response directions, or inconsistent state indications is counted, and the number of conflict sample evidence is generated.

[0044] The historical sample evidence volume, current sample evidence volume, and conflicting sample evidence volume are calculated and arranged according to their positions to form an evidence statistical sequence.

[0045] Perform concentration mapping calculations on the evidence statistical sequence, write the historical concentration component at the historical evidence position, write the current concentration component at the current evidence position, and write the conflict suppression component at the conflict evidence position to generate a local evidence concentration sequence;

[0046] Perform parameter node attribution writing calculation on the local evidence concentration sequence, and write the local evidence concentration sequence to the conditional probability quality adjustment position of the corresponding parameter node.

[0047] Optionally, the step of performing partitioning and redistribution calculations on the conditional probability quality corresponding to the dominant parent set to form a conditional probability table specifically includes:

[0048] Read the dominant parent set identifier, dominant parent set combination and local evidence concentration sequence corresponding to each parameter node, and extract the initial distribution of conditional probability mass corresponding to the dominant parent set combination;

[0049] The initial distribution of conditional probability mass is partitioned according to the different result values ​​corresponding to the parameter nodes, generating several probability mass partitions;

[0050] Map the historical concentration component, current concentration component, and conflict suppression component in the local evidence concentration sequence to the corresponding probability quality partitions, and perform probability quality increase / decrease calculations.

[0051] After the addition and subtraction calculations are completed, partition balancing and normalization calculations are performed on each probability quality partition to generate a conditional probability quality sequence that satisfies the probability distribution constraints.

[0052] Perform table entry writing calculations on the conditional probability quality sequence, and write the conditional probability quality values ​​corresponding to each result value at the corresponding position of the dominant parent set combination to form a conditional probability table.

[0053] Optionally, the parameter inference calculation for the conditional probability table includes the following steps:

[0054] Read the conditional probability table corresponding to each parameter node and extract the conditional probability mass value corresponding to the combination of dominant parent sets.

[0055] Read the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics and cycle time characteristics corresponding to the current adsorption condition data set, perform parent set matching calculation, and determine the dominant parent set combination corresponding to the current adsorption condition;

[0056] Read the conditional probability mass sequence corresponding to the vacuum pressure parameter node, pressure holding time parameter node, valve switching time parameter node and leakage compensation amplitude parameter node at the corresponding positions of the determined dominant parent set combination;

[0057] Perform probability comparison calculations on the conditional probability mass sequence corresponding to each parameter node, and extract the parameter value corresponding to the position with the highest probability mass.

[0058] The extracted parameter values ​​are processed and calculated to generate vacuum pressure, pressure holding time, valve switching time, and leakage compensation amplitude.

[0059] Optionally, the constraint verification calculation performed on the vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude, and the output operating parameter set specifically include:

[0060] Read the vacuum pressure value, pressure holding time value, valve switching time value and leakage compensation amplitude, and establish pressure constraint interval, time constraint interval, switching time constraint interval and compensation amplitude constraint interval respectively;

[0061] Perform interval comparison calculations on vacuum pressure value, pressure holding time value, valve switching time value and leakage compensation amplitude, and generate out-of-bounds flags for each parameter;

[0062] Perform joint matching calculations on vacuum pressure and pressure holding time, and perform timing coordination calculations on valve switching time and leakage compensation amplitude to generate parameter matching identifiers;

[0063] Perform a comprehensive judgment calculation on the boundary crossing indicator and the parameter coordination indicator, filter the parameter values ​​that satisfy the interval constraints and coordination constraints, and generate a parameter sequence that passes the verification.

[0064] Assemble and calculate the results of the validated parameter sequence, and output the set of running parameters.

[0065] The beneficial effects of this invention are:

[0066] (1) This invention incorporates the historical adsorption condition data set and the current adsorption condition data set into the analysis process, and establishes a migration association table from the historical condition node to the current condition node. This makes the operation parameter optimization process no longer limited to a single current condition matching, but can inherit the experience of historical available conditions, and improve the parameter adaptation capability under different workpieces, different surface conditions and different load conditions.

[0067] (2) By setting up a candidate parent set and performing parent set competition explicit calculation, the present invention uses the working condition adaptation value, migration support value and local correlation value together to screen the dominant parent set. It can determine the parameter dependence structure that is more in line with the current adsorption state from multiple node combination relationships, and improve the inference of vacuum pressure value, pressure holding time value, valve switching time value and leakage compensation amplitude.

[0068] (3) This invention generates a local evidence concentration sequence by statistically analyzing the amount of historical sample evidence, the amount of current sample evidence, and the amount of conflicting sample evidence. It further performs partition redistribution calculation on the conditional probability quality, which can suppress the inconsistency between historical experience and current response, reduce the interference of conflicting samples on parameter inference results, and improve the stability and reliability of parameter output results.

[0069] (4) By performing interval constraint verification and parameter matching verification on the inferred vacuum pressure value, pressure holding time value, valve switching time value and leakage compensation amplitude, the present invention can output a set of operating parameters that meet the operating constraint requirements, reduce the possibility of adsorption instability, parameter mismatch and cycle fluctuation, and improve the continuous operation stability of the vacuum adsorption system. Attached Figure Description

[0070] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0071] Figure 1 This is a flowchart of a method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning, as proposed in this invention.

[0072] Figure 2 This diagram illustrates the transfer association table mechanism and the parent set competition explicit calculation process of the vacuum adsorption system operating parameter optimization method based on transfer learning proposed in this invention.

[0073] Figure 3 This is a schematic diagram illustrating the implementation process of local evidence concentration adjustment and conditional probability mass partition redistribution in the vacuum adsorption system operating parameter optimization method based on transfer learning proposed in this invention. Detailed Implementation

[0074] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0075] refer to Figures 1-3 A method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning includes the following steps:

[0076] Obtain historical adsorption condition data sets and current adsorption condition data sets, and perform time alignment and dimension unification operations on the historical and current adsorption condition data sets to form a condition feature sequence;

[0077] Perform feature parsing operations on the working condition feature sequence to extract workpiece material features, surface roughness features, load features, pressure change features, flow change features, leakage change features, and cycle time features, forming a set of node vectors;

[0078] Construct the initial structure of a semi-naive Bayesian network, set up adsorption state nodes, parameter nodes, and result nodes in the node vector set, establish a migration association table from historical working condition nodes to current working condition nodes, and generate a set of candidate parent sets corresponding to each parameter node.

[0079] Perform parent set competition explicit calculation on the candidate parent set corresponding to each parameter node, read the working condition adaptation value, migration support value, and local correlation value corresponding to each candidate parent set, and generate the dominant parent set identifier.

[0080] At the corresponding position of the dominant parent set identifier, the amount of historical sample evidence, the amount of current sample evidence, and the amount of conflicting sample evidence are counted to generate a local evidence concentration sequence. Then, the conditional probability quality corresponding to the dominant parent set is partitioned and redistributed to form a conditional probability table.

[0081] Perform parameter inference calculations on the conditional probability table to generate vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude;

[0082] Perform constraint verification calculations on vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude, and output a set of operating parameters.

[0083] In this embodiment, the extraction of workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow rate change characteristics, leakage change characteristics, and cycle time characteristics specifically includes:

[0084] Read the workpiece identification data, perform category-based calculations, and generate workpiece material characteristics;

[0085] Read the workpiece surface inspection data, perform roughness interval division calculation, and generate surface roughness features;

[0086] Read mass data and dimension data, perform numerical conversion calculations, extract mass components, area components, and center of gravity offset components, and generate load characteristics;

[0087] Read the vacuum pressure time series, perform calculations of the difference between adjacent time points, maximum value extraction calculation, minimum value extraction calculation, and duration calculation to generate pressure change characteristics;

[0088] Read the traffic time series, perform calculations for the difference between adjacent time points, maximum value extraction, minimum value extraction, and stage average value, and generate traffic change characteristics;

[0089] Read the leak monitoring sequence, perform unit time period change calculation, duration calculation, and adjacent time difference comparison calculation to generate leak change characteristics;

[0090] Read the task cycle data, perform single adsorption duration calculation, adjacent adsorption interval calculation, and task count per unit time period calculation, and generate cycle characteristics;

[0091] Vector assembly calculations are performed on the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics, and cycle time characteristics to form a set of node vectors.

[0092] In this embodiment, the initial structure of a semi-naive Bayesian network is constructed, and adsorption state nodes, parameter nodes, and result nodes are set within the node vector set. A migration association table from historical working condition nodes to current working condition nodes is established, and the candidate parent set corresponding to each parameter node is generated, including the following steps:

[0093] Perform node category classification calculation on the node vector set, and write the adsorption state node identifier, parameter node identifier and result node identifier at the corresponding position of each node;

[0094] Hierarchical calculations are performed on nodes with adsorption state node identifiers, nodes with parameter node identifiers, and nodes with result node identifiers. State layer, parameter layer, and result layer are established sequentially at the network level to form the initial hierarchical structure of the semi-naive Bayesian network.

[0095] Perform source tag calculation on the node vectors corresponding to the historical adsorption condition data set and the node vectors corresponding to the current adsorption condition data set, and write the historical condition identifier and the current condition identifier at the node source position respectively.

[0096] Perform similar feature matching calculations on nodes with historical operating condition identifiers and nodes with current operating condition identifiers. Establish node correspondence based on matching conditions of consistent feature categories, consistent parameter semantics, and corresponding value ranges, and form a migration association table from historical operating condition nodes to current operating condition nodes.

[0097] Perform parent node source screening calculations for each parameter node, screen nodes that have state influence relationships with the corresponding parameter node in the adsorption state nodes, and screen nodes that have migration association relationships with the corresponding parameter node in the historical working condition nodes according to the migration association table, forming the initial parent node sequence for each parameter node.

[0098] Perform node combination calculations on the initial parent node sequence corresponding to each parameter node, and generate parent node combination sequences according to single node combination, double node combination and multi node combination.

[0099] Perform constraint filtering calculations on the parent node combination sequence, delete combinations containing result nodes, delete combinations that do not satisfy hierarchical connection relationships, delete combinations with semantic conflicts between nodes, retain node combinations that satisfy parameter dependency requirements, and generate a set of candidate parent sets corresponding to each parameter node.

[0100] The initial hierarchical structure of the semi-naive Bayesian network specifically includes:

[0101] The state layer contains nodes representing the adsorption conditions. These nodes include those corresponding to workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow rate change characteristics, leakage change characteristics, and cycle time characteristics. The parameter layer contains nodes representing the operating parameters to be optimized. These nodes include those for vacuum pressure, holding time, valve switching time, and leakage compensation amplitude. The result layer contains nodes representing the adsorption results. These nodes include nodes for successful adsorption, stable adsorption, or leakage response. The leakage compensation amplitude node indicates the amount of compensation used when adjusting the adsorption process after detecting leakage changes.

[0102] In terms of hierarchical connections, state layer nodes point to parameter layer nodes, representing the influence of adsorption conditions on the values ​​of operating parameters; parameter layer nodes point to result layer nodes, representing the effect of operating parameter configuration on adsorption results; directed edges are allowed to be established between parameter nodes within the same parameter layer, representing the local dependencies between different operating parameters; no result-pointing edges are established between state layer nodes, and result layer nodes do not point back to parameter layer nodes or state layer nodes, thus maintaining the consistency of network hierarchy and forming an initial hierarchical structure that meets the requirements of a semi-naive Bayesian network.

[0103] Specifically, performing feature matching calculations between nodes with historical operating condition identifiers and nodes with current operating condition identifiers includes:

[0104] Read the set of nodes with historical operating condition identifiers and the set of nodes with current operating condition identifiers, and extract the feature category identifier and value range identifier corresponding to each node;

[0105] Based on matching conditions of consistent workpiece material type, consistent surface roughness type, consistent load type, consistent pressure change type, consistent flow change type, consistent leakage change type, and consistent cycle time type, generate a sequence of matching nodes of the same type;

[0106] Perform node name consistency checks and value range correspondence checks on the same type of node pairing sequences, and filter node pairings with the same node name and corresponding value ranges; perform correspondence calculation on the filtered node pairings to form a migration association table from historical working condition nodes to current working condition nodes.

[0107] Specifically, the parent node source filtering calculation for each parameter node includes:

[0108] Read the node identifier corresponding to each parameter node, retrieve the state nodes that have an interaction relationship with the corresponding parameter nodes in the adsorption state node set, and form a state source node sequence;

[0109] Based on the migration association table, historical nodes corresponding to the corresponding parameter nodes are retrieved from the historical operating condition node set to form a historical source node sequence;

[0110] The state source node sequence and the historical source node sequence are merged and arranged. Nodes that do not match the node category, are written repeatedly, or have no relation to the corresponding parameter node are deleted. Nodes that meet the source requirements are retained to form the initial parent node sequence corresponding to each parameter node.

[0111] In this embodiment, the explicit calculation of parent set competition for the candidate parent set corresponding to each parameter node specifically includes:

[0112] Read the candidate parent set set corresponding to each parameter node, perform sequential numbering calculation on each parent set combination in the candidate parent set set set, and write the parent set identifier at the corresponding position of each parent set combination.

[0113] Read the node combination corresponding to each parent set identifier, extract the adsorption state node, historical working condition node and current working condition node from the node combination, perform combination expansion calculation, and form the parent set composition sequence.

[0114] Perform working condition adaptation calculation on the sequence of parent sets, read the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics and cycle time characteristics corresponding to each node, calculate the feature matching degree of each parent set combination to the corresponding parameter node, and generate working condition adaptation values.

[0115] Perform migration support calculation on the parent set composition sequence, read the migration association relationship between historical working condition nodes and current working condition nodes in the parent set composition sequence, count the historical support sample size, current corresponding sample size and migration consistent sample size related to the corresponding parameter node, and generate migration support value;

[0116] Perform local association calculations on the sequence composed of the parent set, read the local conditional dependencies between each node in the sequence composed of the parent set and the corresponding parameter node, calculate the local association strength of the node combination to the corresponding parameter node, and generate local association values.

[0117] Competitive explicit calculations are performed on the working condition adaptation values, migration support values, and local correlation values, and competitive comparison results of each parent set combination are generated within the candidate parent set corresponding to the same parameter node.

[0118] Perform a selection and filtering calculation on the competition comparison results, extract the parent set combination with the highest competition result, and write the dominant parent set identifier at the corresponding parameter node position.

[0119] Specifically, performing migration support calculations on the parent set composition sequence includes: reading the correspondence between each historical working condition node and each current working condition node in the parent set composition sequence, and extracting the node pairing records related to the corresponding parameter nodes from the migration association table;

[0120] Count the number of historical samples in each node pairing record that meet the requirements of consistent feature category, consistent node name and corresponding value range, and generate the historical support sample size.

[0121] The current sample quantity corresponding to the paired records of the above nodes in the current adsorption condition data is counted, and the current corresponding sample quantity is generated; a consistency test is performed on the historical supporting sample quantity and the current corresponding sample quantity, and the number of samples that simultaneously satisfy the historical correlation relationship and the current condition correspondence relationship is counted, and the transferable consistent sample quantity is generated.

[0122] Perform combined calculations on the historical supporting sample size, the current corresponding sample size, and the transferable consistent sample size to generate the transfer supporting value.

[0123] Specifically, performing local association calculation on the parent set composition sequence includes: reading the pairing relationship between each node and its corresponding parameter node in the parent set composition sequence, counting the number of times each node value and its corresponding parameter node value co-occur in the same working condition sample, and generating pairing statistics; and counting the number of times each value of the corresponding parameter node appears when each node value appears at the corresponding position in the parent set composition sequence, and generating node association statistics.

[0124] Perform a summary calculation on the node association statistics corresponding to each node to generate the local association strength of the parent set combination to the corresponding parameter nodes;

[0125] Perform numerical writing calculations on the local correlation strength to generate local correlation values.

[0126] Specifically, the competitive explicit calculation of the working condition adaptation values, migration support values, and local association values ​​includes: reading the working condition adaptation values, migration support values, and local association values ​​of each parent set combination corresponding to the same parameter node, arranging them according to the execution position of the parent set identifier, and forming a parent set comparison sequence.

[0127] The working condition adaptation values, migration support values ​​and local correlation values ​​corresponding to each parent set combination in the parent set comparison sequence are written and calculated in parallel to form the parent set evaluation record.

[0128] Within the range corresponding to the same parameter node, perform item-by-item comparison calculations on the evaluation records of each parent set, and count the number of times each parent set combination leads in terms of working condition adaptation value, migration support value, and local correlation value; the number of times a parent set combination leads is the cumulative number of times that a certain parent set combination has a higher value than other parent set combinations when comparing item by item.

[0129] Perform a summary sorting calculation on the number of comparisons to generate the competitive sorting results for each parent set combination; take the parent set combination with the highest competitive sorting result as the dominant parent set, and write the dominant parent set identifier at the corresponding parameter node position.

[0130] In this embodiment, generating a local evidence concentration sequence includes the following steps:

[0131] Read the dominant parent set identifier corresponding to each parameter node, and extract the parent set combination corresponding to the dominant parent set identifier;

[0132] Read the historical operating condition nodes, current operating condition nodes, and adsorption state nodes corresponding to the dominant parent set, and establish the sample collection range according to the parameter node category;

[0133] Within the sample collection range, retrieve historical adsorption condition data corresponding to the dominant parent set combination, count the number of historical samples that satisfy the parent set value combination and have consistent result values, and generate historical sample evidence quantity.

[0134] Within the sample collection range, retrieve the current adsorption condition data corresponding to the dominant parent set combination, count the number of current samples that satisfy the parent set value combination and have consistent parameter responses, and generate the current sample evidence quantity.

[0135] Conflict detection calculations are performed on historical adsorption condition data and current adsorption condition data. The number of samples that satisfy the combination of parent set values ​​but have inconsistent result values, inconsistent parameter response directions, or inconsistent state indications is counted, and the number of conflict sample evidence is generated.

[0136] The historical sample evidence volume, current sample evidence volume, and conflicting sample evidence volume are calculated and arranged according to their positions to form an evidence statistical sequence.

[0137] Perform concentration mapping calculations on the evidence statistical sequence, write the historical concentration component at the historical evidence position, write the current concentration component at the current evidence position, and write the conflict suppression component at the conflict evidence position to generate a local evidence concentration sequence;

[0138] Perform parameter node attribution writing calculation on the local evidence concentration sequence, and write the local evidence concentration sequence to the conditional probability quality adjustment position of the corresponding parameter node.

[0139] Specifically, counting the number of historical samples that satisfy the combination of parent set values ​​and have consistent result values ​​includes: reading the value range of each parent node and the value identifier of the result node corresponding to the dominant parent set combination, retrieving historical sample records corresponding to the value range of each parent node in the historical adsorption condition data set, and generating a parent set matching sample sequence.

[0140] Extract the result node value identifier corresponding to each historical sample record from the parent set matching sample sequence; perform a consistency comparison calculation between the result node value identifier corresponding to each historical sample record and the result node value identifier corresponding to the dominant parent set combination, and filter historical sample records with consistent result values.

[0141] Perform a statistical calculation on the filtered historical sample records to generate the number of historical samples that satisfy the combination of values ​​from the parent set and have the same result value.

[0142] Specifically, the conflict detection calculation between historical adsorption condition data and current adsorption condition data includes:

[0143] Read the sample records corresponding to the dominant parent set from the historical adsorption condition data set and the current adsorption condition data set, and extract the parent node value, parameter node value and adsorption result identifier corresponding to each sample record;

[0144] Sample pairing calculations are performed according to the value range of the parent node corresponding to the dominant parent set combination to form a historical-current comparison sample sequence; adsorption result consistency check, parameter response direction consistency check, and state indication consistency check are performed on the historical-current comparison sample sequence to filter out sample pairs with inconsistent adsorption results, inconsistent parameter response directions, or inconsistent state indications to form a conflict sample sequence.

[0145] Perform quantitative statistical calculations on the conflicting sample sequences to generate the conflicting sample evidence quantity.

[0146] Specifically, performing concentration mapping calculations on the evidence statistical sequence includes:

[0147] Read the arrangement results of historical sample evidence quantity, current sample evidence quantity, and conflicting sample evidence quantity in the evidence statistics sequence at the corresponding positions of each parameter node;

[0148] Perform a forward write calculation on the amount of historical sample evidence to generate historical concentration components at the locations of historical evidence.

[0149] Perform forward write calculation on the current sample evidence quantity to generate the current concentration component at the current evidence location; perform reverse suppression write calculation on the conflicting sample evidence quantity to generate the conflict suppression component at the conflicting evidence location;

[0150] The historical concentration component, the current concentration component, and the conflict suppression component are sequentially assembled and calculated to form a local evidence concentration sequence consistent with the corresponding parameter nodes.

[0151] Specifically, the parameter node attribution calculation for the local evidence concentration sequence includes: reading the parameter node identifier corresponding to the local evidence concentration sequence and determining the target parameter node to which the local evidence concentration sequence belongs; establishing a conditional probability quality adjustment region at the position corresponding to the target parameter node; and sequentially setting the historical concentration writing position, the current concentration writing position, and the conflict suppression writing position within the conditional probability quality adjustment region.

[0152] Write the historical concentration component from the local evidence concentration sequence to the historical concentration writing position, write the current concentration component to the current concentration writing position, and write the conflict suppression component to the conflict suppression writing position; perform node correspondence checks on the conditional probability quality adjustment region that has been written, and form a concentration writing result that corresponds one-to-one with the target parameter node.

[0153] In this embodiment, performing partitioning and redistribution calculations on the conditional probability quality corresponding to the dominant parent set to form a conditional probability table specifically includes:

[0154] Read the dominant parent set identifier, dominant parent set combination and local evidence concentration sequence corresponding to each parameter node, and extract the initial distribution of conditional probability mass corresponding to the dominant parent set combination;

[0155] The initial distribution of conditional probability mass is partitioned according to the different result values ​​corresponding to the parameter nodes, generating several probability mass partitions;

[0156] Map the historical concentration component, current concentration component, and conflict suppression component in the local evidence concentration sequence to the corresponding probability quality partitions, and perform probability quality increase / decrease calculations.

[0157] After the addition and subtraction calculations are completed, partition balancing and normalization calculations are performed on each probability quality partition to generate a conditional probability quality sequence that satisfies the probability distribution constraints.

[0158] Perform table entry writing calculations on the conditional probability quality sequence, and write the conditional probability quality values ​​corresponding to each result value at the corresponding position of the dominant parent set combination to form a conditional probability table.

[0159] Specifically, the historical concentration component, current concentration component, and conflict suppression component in the local evidence concentration sequence are mapped to corresponding probability quality partitions, and the calculation of probability quality increases or decreases includes:

[0160] Read the probability quality partition and local evidence concentration sequence corresponding to each parameter node, and determine the historical concentration component writing area, the current concentration component writing area and the conflict suppression component writing area at the corresponding positions of each probability quality partition.

[0161] Write the values ​​corresponding to the historical concentration components as historical increments into the corresponding probability mass partitions, and add the same probability mass as the historical concentration component values ​​to the original probability mass values.

[0162] Write the value corresponding to the current concentration component as the current increase to the corresponding probability mass partition, and add the same probability mass as the current concentration component value to the original probability mass value.

[0163] Write the value corresponding to the conflict suppression component as the conflict deduction amount into the corresponding probability quality partition, and deduct the probability quality with the same value as the conflict suppression component from the original probability quality value.

[0164] After completing the addition and subtraction calculations, the partitioning results are recorded for each probability quality partition, forming a sequence of probability quality increase and decrease results.

[0165] Specifically, performing partition balancing and normalization calculations on each probability quality partition after the addition and subtraction calculations include:

[0166] Read the results of each probability quality partition after the addition and subtraction calculations are completed, and calculate the adjusted probability quality values ​​corresponding to each probability quality partition.

[0167] Perform partition balancing calculations on each probability quality partition, check whether the adjusted probability quality values ​​corresponding to each probability quality partition are less than zero, perform zero value replacement calculations on probability quality values ​​less than zero, and perform upper limit truncation calculations on probability quality values ​​exceeding the preset upper limit to form a balanced probability quality sequence.

[0168] Perform total statistical calculations on the balanced probability mass sequence to generate a total probability mass value; use the total probability mass value as a unified conversion base value, perform proportional conversion calculations on the balanced probability mass values ​​corresponding to each probability mass partition to generate normalized probability mass values ​​corresponding to each probability mass partition.

[0169] The normalized probability mass values ​​are sequentially written into the calculation to form a conditional probability mass sequence that meets the requirement of consistent total probability mass.

[0170] Specifically, the calculation of writing entries to the conditional probability mass sequence includes:

[0171] Read the parameter node identifier, dominant parent set identifier, and result value identifier corresponding to the conditional probability quality sequence, and locate the table entry position corresponding to the combination of parameter node, dominant parent set, and result value in the conditional probability table.

[0172] Write the normalized probability mass values ​​corresponding to each result value into the corresponding table entry position according to the order of arrangement in the conditional probability mass sequence.

[0173] Perform a position correspondence check on the completed entries to confirm that the parameter node identifier, dominant parent set identifier, and result value identifier are consistent; perform sequential assembly calculation on all entries that pass the check to form a conditional probability table corresponding to the dominant parent set combination.

[0174] In this embodiment, the parameter inference calculation for the conditional probability table includes the following steps:

[0175] Read the conditional probability table corresponding to each parameter node and extract the conditional probability mass value corresponding to the combination of dominant parent sets.

[0176] Read the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics and cycle time characteristics corresponding to the current adsorption condition data set, perform parent set matching calculation, and determine the dominant parent set combination corresponding to the current adsorption condition;

[0177] Read the conditional probability mass sequence corresponding to the vacuum pressure parameter node, pressure holding time parameter node, valve switching time parameter node and leakage compensation amplitude parameter node at the corresponding positions of the determined dominant parent set combination;

[0178] Perform probability comparison calculations on the conditional probability mass sequence corresponding to each parameter node, and extract the parameter value corresponding to the position with the highest probability mass.

[0179] The extracted parameter values ​​are processed and calculated to generate vacuum pressure, pressure holding time, valve switching time, and leakage compensation amplitude.

[0180] Specifically, performing parent set matching calculation to determine the dominant parent set combination corresponding to the current adsorption condition includes: reading the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics and cycle time characteristics corresponding to the current adsorption condition data set, as well as the dominant parent set combination corresponding to each parameter node;

[0181] Extract the feature category identifier and value range identifier corresponding to each parent node in each dominant parent set combination; perform a one-by-one correspondence check between each feature corresponding to the current adsorption condition data set and the feature category identifier and value range identifier corresponding to each parent node in each dominant parent set combination, and filter the parent set combination with consistent feature categories and corresponding value ranges;

[0182] Perform statistical calculations on the matching results of the selected parent set combinations to determine the dominant parent set combination corresponding to the current adsorption conditions.

[0183] In this embodiment, constraint verification calculations are performed on the vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude. The output set of operating parameters specifically includes:

[0184] Read the vacuum pressure value, pressure holding time value, valve switching time value and leakage compensation amplitude, and establish pressure constraint interval, time constraint interval, switching time constraint interval and compensation amplitude constraint interval respectively;

[0185] Perform interval comparison calculations on vacuum pressure value, pressure holding time value, valve switching time value and leakage compensation amplitude, and generate out-of-bounds flags for each parameter;

[0186] Perform joint matching calculations on vacuum pressure and pressure holding time, and perform timing coordination calculations on valve switching time and leakage compensation amplitude to generate parameter matching identifiers;

[0187] Perform a comprehensive judgment calculation on the boundary crossing indicator and the parameter coordination indicator, filter the parameter values ​​that satisfy the interval constraints and coordination constraints, and generate a parameter sequence that passes the verification.

[0188] Assemble and calculate the results of the validated parameter sequence, and output the set of running parameters.

[0189] Specifically, a joint matching calculation is performed on the vacuum pressure value and the pressure holding time value, including:

[0190] Read the vacuum pressure value and the holding time value, and extract the pressure range identifier corresponding to the vacuum pressure value and the time range identifier corresponding to the holding time value;

[0191] Retrieve the set of allowable duration intervals corresponding to the pressure interval identifier in the preset parameter matching table, and perform a corresponding check between the duration interval identifier corresponding to the pressure holding duration value and the set of allowable duration intervals;

[0192] When the time interval identifier corresponding to the pressure holding time value falls into the set of allowable time intervals, it is determined that the vacuum pressure value and the pressure holding time value match.

[0193] When the time interval identifier corresponding to the pressure holding time value does not fall into the set of allowable time intervals, it is determined that the vacuum pressure value and the pressure holding time value do not match.

[0194] For matching results that are true or false, perform an identifier writing calculation to generate parameter matching identifiers corresponding to the vacuum pressure value and the pressure holding time value.

[0195] Specifically, the timing coordination calculation for valve switching time values ​​and leakage compensation amplitude includes:

[0196] Read the valve switching time value and leakage compensation amplitude, and extract the time interval identifier corresponding to the valve switching time value and the compensation interval identifier corresponding to the leakage compensation amplitude;

[0197] Retrieve the set of allowable compensation intervals corresponding to the time interval identifier in the preset timing coordination relationship table, and perform a corresponding check between the compensation interval identifier corresponding to the leakage compensation amplitude and the set of allowable compensation intervals.

[0198] When the compensation interval identifier corresponding to the leakage compensation amplitude falls into the set of allowable compensation intervals, it is determined that the valve switching time value and the timing coordination of the leakage compensation amplitude are valid.

[0199] When the compensation interval identifier corresponding to the leakage compensation amplitude does not fall into the set of allowable compensation intervals, it is determined that the valve switching time value and the timing coordination of the leakage compensation amplitude are not valid.

[0200] For the results of successful or unsuccessful timing coordination, an identifier is written into the calculation to generate a parameter coordination identifier corresponding to the valve switching time value and the leakage compensation amplitude.

[0201] The process of selecting parameter values ​​that satisfy both interval constraints and coordination constraints, and generating a parameter sequence that passes verification, specifically includes:

[0202] Read the out-of-bounds flags corresponding to the vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude, as well as the parameter matching flags corresponding to the vacuum pressure value and pressure holding time value, and the parameter matching flags corresponding to the valve switching time value and leakage compensation amplitude;

[0203] Perform a validity check on the out-of-bounds flags for each parameter, and filter out parameter values ​​that are not out of bounds; perform a consistency check on the parameter coordination flags for each parameter, and filter out parameter values ​​that are coordinated.

[0204] The parameter values ​​that simultaneously satisfy both the non-boundary requirement and the fit requirement are grouped together to generate parameter value combinations that satisfy both interval constraints and fit constraints.

[0205] Perform sequential calculations on the combinations of parameter values ​​to form a sequence of parameters that pass the verification.

[0206] Example 1:

[0207] To verify the feasibility of this invention in practice, it was applied to an automated adsorption and handling production line. This line needs to continuously grasp, transfer, and place plate-shaped workpieces. During continuous material intake, the workpieces exhibit changes in material, surface condition, and load. Common examples include relatively smooth metal plates, coated plates with slight textures, and composite plates with localized small leakage channels. Current practices typically rely on manual experience to preset vacuum pressure, holding time, and valve switching timing, supplemented by a fixed compensation amplitude to maintain the adsorption action. While this approach is usable under single operating conditions, it becomes problematic when workpiece types change frequently. A set of parameters may be applicable to one batch of workpieces but not to the next. The most obvious manifestations are: sometimes the vacuum pressure is too high, which, while able to grasp the workpiece, increases energy consumption and causes sluggish response; sometimes the holding time is too short, causing corners to loosen immediately after grasping; sometimes the valve switches too early, resulting in insufficient vacuum buildup; and sometimes the leakage compensation amplitude is too small, causing slow recovery after micro-leakage, affecting the overall line cycle time.

[0208] In this embodiment, historical adsorption condition data sets and current adsorption condition data sets are first collected. The historical adsorption condition data set comes from stable and abnormal adsorption records saved during previous continuous operation, while the current adsorption condition data set comes from the real-time detection results of the current batch to be executed. The collected content includes workpiece identification data, surface inspection data, quality data, dimensional data, vacuum pressure time series, flow rate time series, leakage monitoring sequence, and task cycle time data. After aligning the execution time and unifying the dimensions of the above data, a directly comparable operating condition feature sequence is formed. Subsequently, feature parsing is performed on the operating condition feature sequence to extract workpiece material features, surface roughness features, load features, pressure change features, flow rate change features, leakage change features, and cycle time features. These features are then assembled into a set of node vectors. After this processing, information that was originally scattered across different acquisition ports, different sampling frequencies, and different units of measurement is consolidated into a unified structure, providing a stable input for subsequent network modeling.

[0209] In the modeling phase, a semi-naive Bayesian network initial structure is constructed based on the node vector set. Adsorption state nodes, parameter nodes, and result nodes are set within the network. Adsorption state nodes characterize the current adsorbed object and its operating condition. Parameter nodes correspond to vacuum pressure, holding time, valve switching time, and leakage compensation amplitude. Result nodes characterize the adsorption results. Simultaneously, the system establishes a migration correlation table based on the similar characteristic relationships between historical and current operating condition nodes. Then, based on the influence relationship of adsorption state and the migration correlation, a candidate parent set is generated for each parameter node. Next, parent set competition explicitness calculation is performed on the candidate parent set, obtaining operating condition adaptation values, migration support values, and local correlation values, from which the dominant parent set identifier is determined. The purpose of this process is to move away from simply relying on a single empirical rule to determine the parameter source, and instead filter out combinations from multiple possible parent set combinations that are closer to the current adsorption state while also inheriting effective historical experience.

[0210] After determining the dominant parent set, the system continues to statistically analyze the amount of evidence from historical samples, current samples, and conflicting samples, forming a local evidence concentration sequence. If a certain combination of parent sets is stable in historical data and also receives corresponding support in the current samples, the historical and current concentration components corresponding to that combination will be increased. If there is a significant inconsistency between historical data and current samples in terms of adsorption results, parameter response direction, or state indication, a conflict suppression component will be formed, and the corresponding conditional probability mass will be reduced. Subsequently, the system partitions and redistributes the conditional probability mass corresponding to the dominant parent set, forming a conditional probability table. Based on the conditional probability table, parameter inference is performed, and the vacuum pressure value, holding time value, valve switching time value, and leakage compensation amplitude are output. Finally, the system performs verification by combining the pressure constraint interval, duration constraint interval, switching time constraint interval, and compensation amplitude constraint interval, while checking whether the pressure and holding time match and whether the valve switching time and leakage compensation amplitude are coordinated, to obtain the final set of operational parameters that can be issued.

[0211] This embodiment uses 360 consecutive operating cycles as a verification sample, with the first 240 cycles serving as the historical adsorption data source and the last 120 cycles serving as the current adsorption data source. Before using the method of this invention, the production line relied on empirically tuned parameters, keeping the vacuum pressure setting within a relatively constant range. The holding time and valve switching timing were also executed in a fixed pattern. When encountering workpieces of different materials and roughness, only a large safety margin could cover the risks. As a result, while the adsorption success rate was acceptable under some conditions, energy consumption was high, and the risk of part dropping could not be completely avoided. After adopting the method of this invention, the system can provide more refined parameter combinations based on workpiece differences. For example, for workpieces with relatively flat surfaces and light loads, the output vacuum pressure value is slightly lower, and the valve switching timing is advanced, ensuring adsorption stability while avoiding unnecessary vacuum maintenance. For workpieces with rough surfaces and a slight leakage tendency, the output holding time value is appropriately increased, and the leakage compensation amplitude is correspondingly increased, thereby maintaining the stability of the adsorption edge region. The entire process does not require frequent manual adjustments by operators, and the equipment maintains a relatively smooth operating state during multi-condition switching.

[0212] The verification results show that after adopting the method of this invention, the adsorption success rate, continuous stable operation rate, and effective handling times per unit time are all improved, while the part drop rate, adsorption setup time fluctuation value, and energy consumption per hundred cycles are all reduced. Especially under conditions of frequent workpiece material switching, large surface roughness range, and micro-leakage disturbances, this invention utilizes the migration correlation between historical and current conditions to avoid directly copying old parameters that are unsuitable for the current state, while not completely discarding proven historical experience. In other words, this invention does not simply stack historical samples onto current samples, but after competitive manifestation and evidence concentration adjustment, only retains data relationships that truly support the current adsorption task. Therefore, the set of operating parameters output by the system is both targeted and stable, making it more suitable for use in continuous production environments.

[0213] Table 1: Comparison of the Implementation Effects of Vacuum Adsorption Operating Parameter Optimization

[0214] Operating conditions Validate the loop number Parameter method Vacuum pressure value / kPa Holding time / ms Valve switching time / ms Leakage compensation amplitude / % Adsorption success rate / % Drop rate / % Energy consumption per 100 cycles / kWh Average beat / s flat metal plate 40 Experience tuning 61 240 82 6 97.2 1.8 3.84 4.31 flat metal plate 40 Method of the present invention 56 218 76 5 99.1 0.4 3.46 4.07 Laminated panels 40 Experience tuning 63 255 85 7 96.8 2.1 3.97 4.39 Laminated panels 40 Method of the present invention 58 229 79 6 98.9 0.6 3.58 4.12 Textured composite board 40 Experience tuning 65 268 88 8 95.9 2.6 4.12 4.48 Textured composite board 40 Method of the present invention 60 241 82 7 98.4 0.8 3.69 4.18 Minor Leakage Conditions 40 Experience tuning 67 281 91 9 94.7 3.3 4.26 4.56 Minor Leakage Conditions 40 Method of the present invention 62 257 86 10 97.9 1.1 3.83 4.26 Mixed switching conditions 40 Experience tuning 66 276 89 8 95.1 3.0 4.21 4.53 Mixed switching conditions 40 Method of the present invention 59 246 83 7 98.2 0.9 3.71 4.20 Overall Average 200 Experience tuning 64.4 264.0 87.0 7.6 95.94 2.56 4.08 4.45 Overall Average 200 Method of the present invention 59.0 238.2 81.2 7.0 98.50 0.76 3.65 4.17

[0215] As shown in Table 1, the method of this invention exhibits good adaptability under various working conditions. Taking a flat metal plate as an example, under the premise of 40 cycles for verification, the adsorption success rate under the empirical tuning method was 97.2%, the part drop rate was 1.8%, the energy consumption per 100 cycles was 3.84 kWh, and the average cycle time was 4.31 s. After adopting the method of this invention, the adsorption success rate increased to 99.1%, the part drop rate decreased to 0.4%, the energy consumption per 100 cycles decreased to 3.46 kWh, and the average cycle time was shortened to 4.07 s. In this type of working condition alone, the adsorption success rate increased by 1.9 percentage points, the part drop rate decreased by 1.4 percentage points, and the cycle time was shortened by 0.24 s. For coated panels, the adsorption success rate under empirical tuning was 96.8%, while the method of this invention reached 98.9%, an improvement of 2.1 percentage points. Energy consumption per 100 cycles decreased from 3.97 kWh to 3.58 kWh, a reduction of 0.39 kWh, indicating that the parameter output did not simply increase vacuum pressure to achieve stability, but rather achieved better adsorption under lower energy consumption conditions. Textured composite panels represent a more challenging type of application. Under empirical tuning, the adsorption success rate was 95.9%, with a component drop rate of 2.6%. Using the method of this invention, the adsorption success rate increased to 98.4%, and the component drop rate decreased to 0.8%. This means that the number of abnormal detachments per 100 cycles decreased from approximately 2.6 to 0.8, demonstrating a significant improvement in stability. The effects of the migration correlation table and local evidence concentration adjustment in this invention are best demonstrated in the case of minor leakage. Under this operating condition, the adsorption success rate of the empirical tuning method was 94.7%, the part drop rate was 3.3%, and the average cycle time was 4.56s. With the method of this invention, the adsorption success rate increased to 97.9%, the part drop rate decreased to 1.1%, and the average cycle time was shortened to 4.26s. Although the leakage compensation amplitude increased from 9% to 10%, the energy consumption per 100 cycles still decreased from 4.26kWh to 3.83kWh, indicating that the compensation was not blindly increased but coordinated with the valve switching time and pressure holding time, resulting in higher overall efficiency. The mixed switching condition more closely resembles the actual continuous production state. In this case, the adsorption success rate of the empirical tuning method was 95.1%, while the method of this invention reached 98.2%, an increase of 3.1 percentage points; the part drop rate decreased from 3.0% to 0.9%, a decrease of 2.1 percentage points; the energy consumption per 100 cycles decreased from 4.21kWh to 3.71kWh, and the average cycle time was shortened from 4.53s to 4.20s. The comprehensive average results further illustrate the overall effectiveness of this invention. Compared with the empirical tuning method, the method of the present invention reduces the overall average vacuum pressure value from 64.4 kPa to 59.0 kPa, a decrease of 5.4 kPa; the pressure holding time is reduced from 264.0 ms to 238.2 ms, a reduction of 25.8 ms; the valve switching time is advanced from 87.0 ms to 81.2 ms; and the leakage compensation amplitude is adjusted from 7.6% to 7.0%.With more refined overall parameters, the comprehensive average adsorption success rate increased from 95.94% to 98.50%, an improvement of 2.56 percentage points; the component drop rate decreased from 2.56% to 0.76%, a decrease of 1.80 percentage points; the energy consumption per 100 cycles decreased from 4.08 kWh to 3.65 kWh, a decrease of approximately 10.54%; and the average cycle time was shortened from 4.45 s to 4.17 s, a reduction of 0.28 s. Based on an estimated 1000 consecutive cycles, the method of this invention can reduce the number of abnormal component drops from approximately 26 to approximately 8, while simultaneously reducing cumulative energy consumption by approximately 4.3 kWh. Therefore, this invention not only improves the adsorption success rate but also reduces energy consumption and cycle time fluctuations, effectively addressing the problems of insufficient parameter adaptation, weak operational stability, and strong reliance on manual tuning in vacuum adsorption systems under multi-condition switching conditions.

[0216] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing operating parameters of a vacuum adsorption system based on transfer learning, characterized in that, Includes the following steps: Obtain historical adsorption condition data sets and current adsorption condition data sets, and perform time alignment and dimension unification operations on the historical and current adsorption condition data sets to form a condition feature sequence; Perform feature parsing operations on the working condition feature sequence to extract workpiece material features, surface roughness features, load features, pressure change features, flow change features, leakage change features, and cycle time features, forming a set of node vectors; Construct the initial structure of a semi-naive Bayesian network, set up adsorption state nodes, parameter nodes, and result nodes in the node vector set, establish a migration association table from historical working condition nodes to current working condition nodes, and generate a set of candidate parent sets corresponding to each parameter node. Perform parent set competition explicit calculation on the candidate parent set corresponding to each parameter node, read the working condition adaptation value, migration support value, and local correlation value corresponding to each candidate parent set, and generate the dominant parent set identifier. At the corresponding position of the dominant parent set identifier, the amount of historical sample evidence, the amount of current sample evidence, and the amount of conflicting sample evidence are counted to generate a local evidence concentration sequence. Then, the conditional probability quality corresponding to the dominant parent set is partitioned and redistributed to form a conditional probability table. Perform parameter inference calculations on the conditional probability table to generate vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude; Perform constraint verification calculations on vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude, and output a set of operating parameters.

2. The method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning according to claim 1, characterized in that, The extraction of workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow rate change characteristics, leakage change characteristics, and cycle time characteristics specifically includes: Read the workpiece identification data, perform category-based calculations, and generate workpiece material characteristics; Read the workpiece surface inspection data, perform roughness interval division calculation, and generate surface roughness features; Read mass data and dimension data, perform numerical conversion calculations, extract mass components, area components, and center of gravity offset components, and generate load characteristics; Read the vacuum pressure time series, perform calculations of the difference between adjacent time points, maximum value extraction calculation, minimum value extraction calculation, and duration calculation to generate pressure change characteristics; Read the traffic time series, perform calculations for the difference between adjacent time points, maximum value extraction, minimum value extraction, and stage average value, and generate traffic change characteristics; Read the leak monitoring sequence, perform unit time period change calculation, duration calculation, and adjacent time difference comparison calculation to generate leak change characteristics; Read the task cycle data, perform single adsorption duration calculation, adjacent adsorption interval calculation, and task count per unit time period calculation, and generate cycle characteristics; Vector assembly calculations are performed on the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics, and cycle time characteristics to form a set of node vectors.

3. The method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning according to claim 2, characterized in that, The construction of the initial structure of the semi-naive Bayesian network, which involves setting up adsorption state nodes, parameter nodes, and result nodes within the node vector set, establishing a migration association table from historical operating condition nodes to the current operating condition node, and generating a candidate parent set corresponding to each parameter node, includes the following steps: Perform node category classification calculation on the node vector set, and write the adsorption state node identifier, parameter node identifier and result node identifier at the corresponding position of each node; Hierarchical calculations are performed on nodes with adsorption state node identifiers, nodes with parameter node identifiers, and nodes with result node identifiers. State layer, parameter layer, and result layer are established sequentially at the network level to form the initial hierarchical structure of the semi-naive Bayesian network. Perform source tag calculation on the node vectors corresponding to the historical adsorption condition data set and the node vectors corresponding to the current adsorption condition data set, and write the historical condition identifier and the current condition identifier at the node source position respectively. Perform similar feature matching calculations on nodes with historical operating condition identifiers and nodes with current operating condition identifiers. Establish node correspondence based on matching conditions of consistent feature categories, consistent parameter semantics, and corresponding value ranges, and form a migration association table from historical operating condition nodes to current operating condition nodes. Perform parent node source screening calculations for each parameter node, screen nodes that have state influence relationships with the corresponding parameter node in the adsorption state nodes, and screen nodes that have migration association relationships with the corresponding parameter node in the historical working condition nodes according to the migration association table, forming the initial parent node sequence for each parameter node. Perform node combination calculations on the initial parent node sequence corresponding to each parameter node, and generate parent node combination sequences according to single node combination, double node combination and multi node combination. Perform constraint filtering calculations on the parent node combination sequence, delete combinations containing result nodes, delete combinations that do not satisfy hierarchical connection relationships, delete combinations with semantic conflicts between nodes, retain node combinations that satisfy parameter dependency requirements, and generate a set of candidate parent sets corresponding to each parameter node.

4. The method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning according to claim 3, characterized in that, The specific steps of performing parent set competition explicit calculation on the candidate parent set corresponding to each parameter node include: Read the candidate parent set set corresponding to each parameter node, perform sequential numbering calculation on each parent set combination in the candidate parent set set set, and write the parent set identifier at the corresponding position of each parent set combination. Read the node combination corresponding to each parent set identifier, extract the adsorption state node, historical working condition node and current working condition node from the node combination, perform combination expansion calculation, and form the parent set composition sequence. Perform working condition adaptation calculation on the sequence of parent sets, read the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics and cycle time characteristics corresponding to each node, calculate the feature matching degree of each parent set combination to the corresponding parameter node, and generate working condition adaptation values. Perform migration support calculation on the parent set composition sequence, read the migration association relationship between historical working condition nodes and current working condition nodes in the parent set composition sequence, count the historical support sample size, current corresponding sample size and migration consistent sample size related to the corresponding parameter node, and generate migration support value; Perform local association calculations on the sequence composed of the parent set, read the local conditional dependencies between each node in the sequence composed of the parent set and the corresponding parameter node, calculate the local association strength of the node combination to the corresponding parameter node, and generate local association values. Competitive explicit calculations are performed on the working condition adaptation values, migration support values, and local correlation values, and competitive comparison results of each parent set combination are generated within the candidate parent set corresponding to the same parameter node. Perform a selection and filtering calculation on the competition comparison results, extract the parent set combination with the highest competition result, and write the dominant parent set identifier at the corresponding parameter node position.

5. The method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning according to claim 4, characterized in that, The generation of the local evidence concentration sequence includes the following steps: Read the dominant parent set identifier corresponding to each parameter node, and extract the parent set combination corresponding to the dominant parent set identifier; Read the historical operating condition nodes, current operating condition nodes, and adsorption state nodes corresponding to the dominant parent set, and establish the sample collection range according to the parameter node category; Within the sample collection range, retrieve historical adsorption condition data corresponding to the dominant parent set combination, count the number of historical samples that satisfy the parent set value combination and have consistent result values, and generate historical sample evidence quantity. Within the sample collection range, retrieve the current adsorption condition data corresponding to the dominant parent set combination, count the number of current samples that satisfy the parent set value combination and have consistent parameter responses, and generate the current sample evidence quantity. Conflict detection calculations are performed on historical adsorption condition data and current adsorption condition data. The number of samples that satisfy the combination of parent set values ​​but have inconsistent result values, inconsistent parameter response directions, or inconsistent state indications is counted, and the number of conflict sample evidence is generated. The historical sample evidence volume, current sample evidence volume, and conflicting sample evidence volume are calculated and arranged according to their positions to form an evidence statistical sequence. Perform concentration mapping calculations on the evidence statistical sequence, write the historical concentration component at the historical evidence position, write the current concentration component at the current evidence position, and write the conflict suppression component at the conflict evidence position to generate a local evidence concentration sequence; Perform parameter node attribution writing calculation on the local evidence concentration sequence, and write the local evidence concentration sequence to the conditional probability quality adjustment position of the corresponding parameter node.

6. The method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning according to claim 5, characterized in that, The step of performing partitioning and redistribution calculations on the conditional probability quality corresponding to the dominant parent set to form a conditional probability table specifically includes: Read the dominant parent set identifier, dominant parent set combination and local evidence concentration sequence corresponding to each parameter node, and extract the initial distribution of conditional probability mass corresponding to the dominant parent set combination; The initial distribution of conditional probability mass is partitioned according to the different result values ​​corresponding to the parameter nodes, generating several probability mass partitions; Map the historical concentration component, current concentration component, and conflict suppression component in the local evidence concentration sequence to the corresponding probability quality partitions, and perform probability quality increase / decrease calculations. After the addition and subtraction calculations are completed, partition balancing and normalization calculations are performed on each probability quality partition to generate a conditional probability quality sequence that satisfies the probability distribution constraints. Perform table entry writing calculations on the conditional probability quality sequence, and write the conditional probability quality values ​​corresponding to each result value at the corresponding position of the dominant parent set combination to form a conditional probability table.

7. The method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning according to claim 6, characterized in that, The parameter inference calculation for the conditional probability table includes the following steps: Read the conditional probability table corresponding to each parameter node and extract the conditional probability mass value corresponding to the combination of dominant parent sets. Read the workpiece material characteristics, surface roughness characteristics, load characteristics, pressure change characteristics, flow change characteristics, leakage change characteristics and cycle time characteristics corresponding to the current adsorption condition data set, perform parent set matching calculation, and determine the dominant parent set combination corresponding to the current adsorption condition; Read the conditional probability mass sequence corresponding to the vacuum pressure parameter node, pressure holding time parameter node, valve switching time parameter node and leakage compensation amplitude parameter node at the corresponding positions of the determined dominant parent set combination; Perform probability comparison calculations on the conditional probability mass sequence corresponding to each parameter node, and extract the parameter value corresponding to the position with the highest probability mass. The extracted parameter values ​​are processed and calculated to generate vacuum pressure, pressure holding time, valve switching time, and leakage compensation amplitude.

8. The method for optimizing the operating parameters of a vacuum adsorption system based on transfer learning according to claim 7, characterized in that, The constraint verification calculation performed on the vacuum pressure value, pressure holding time value, valve switching time value, and leakage compensation amplitude, and the output set of operating parameters specifically include: Read the vacuum pressure value, pressure holding time value, valve switching time value and leakage compensation amplitude, and establish pressure constraint interval, time constraint interval, switching time constraint interval and compensation amplitude constraint interval respectively; Perform interval comparison calculations on vacuum pressure value, pressure holding time value, valve switching time value and leakage compensation amplitude, and generate out-of-bounds flags for each parameter; Perform joint matching calculations on vacuum pressure and pressure holding time, and perform timing coordination calculations on valve switching time and leakage compensation amplitude to generate parameter matching identifiers; Perform a comprehensive judgment calculation on the boundary crossing indicator and the parameter coordination indicator, filter the parameter values ​​that satisfy the interval constraints and coordination constraints, and generate a parameter sequence that passes the verification. Assemble and calculate the results of the validated parameter sequence, and output the set of running parameters.

Images