Movatterモバイル変換


[0]ホーム

URL:


CN120560213A - A device control system and method for producing OPS computer motherboards - Google Patents

A device control system and method for producing OPS computer motherboards

Info

Publication number
CN120560213A
CN120560213ACN202511071885.9ACN202511071885ACN120560213ACN 120560213 ACN120560213 ACN 120560213ACN 202511071885 ACN202511071885 ACN 202511071885ACN 120560213 ACN120560213 ACN 120560213A
Authority
CN
China
Prior art keywords
parameter
module
production
context
parameters
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
CN202511071885.9A
Other languages
Chinese (zh)
Inventor
刘姜
刘应
徐维理
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Dingsheng Technology Co ltd
Original Assignee
Shenzhen Dingsheng Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Dingsheng Technology Co ltdfiledCriticalShenzhen Dingsheng Technology Co ltd
Priority to CN202511071885.9ApriorityCriticalpatent/CN120560213A/en
Publication of CN120560213ApublicationCriticalpatent/CN120560213A/en
Withdrawnlegal-statusCriticalCurrent

Links

Landscapes

Abstract

Translated fromChinese

本发明公开了一种用于OPS电脑主板生产的设备控制系统及方法,涉及设备控制技术领域,该方法包括:收集工艺参数,构建拓扑向量空间;记录换线过程中的关键信息,存入上下文记忆网络;当接收到新的OPS主板生产订单时,识别订单中产品型号,在拓扑向量空间和上下文记忆网络中进行相似性检索;以拓扑空间检索到的参数作为基础参数,将上下文记忆网络中提取的调整策略作为增量,生成初始工艺参数设定;进行小批量试产,将实际生产结果与预期结果进行对比;利用强化学习算法,生成优化参数。本发明通过拓扑向量空间与上下文记忆网络的协同检索,快速获取最优工艺参数,减少换线调试时间,完成更准确的控制。

The present invention discloses an equipment control system and method for the production of OPS computer motherboards, relating to the field of equipment control technology. The method comprises: collecting process parameters and constructing a topological vector space; recording key information during the line change process and storing it in a context memory network; when a new OPS motherboard production order is received, identifying the product model in the order, and performing similarity search in the topological vector space and the context memory network; using the parameters retrieved in the topological space as the basic parameters, and using the adjustment strategy extracted from the context memory network as the increment, generating initial process parameter settings; conducting a small-batch trial production, and comparing the actual production results with the expected results; and using a reinforcement learning algorithm to generate optimized parameters. The present invention quickly obtains optimal process parameters through the collaborative search of the topological vector space and the context memory network, reduces the line change debugging time, and achieves more accurate control.

Description

Equipment control system and method for OPS computer motherboard production
Technical Field
The invention relates to the technical field of equipment control, in particular to an equipment control system and method for OPS computer mainboard production.
Background
Currently, the production of OPS computer mainboards faces two major core challenges, namely insufficient suitability of multiple types and low wire replacement efficiency.
When the traditional production method faces line changing of different types of mainboards, the adjustment of process parameters is seriously dependent on the experience of engineers, and a systematic parameter retrieval and migration mechanism is lacked, so that line changing and debugging time is long, and the production requirements of small batches and multiple varieties are difficult to meet. In the prior art, the optimization process of the technological parameters is often carried out in isolation, and the associated memory of the parameters, the production scene and the equipment state is not established, so that the history production experience cannot be effectively reused. When the equipment state fluctuates or the material batch changes, the adaptability of parameter adjustment is poor, and the quality problems of high solder joint cold solder joint rate and exceeding component offset are easy to occur. In addition, the traditional method lacks a deep mining and closed-loop optimization mechanism for production data, and parameter deviation found in the trial production process cannot be automatically fed back to a knowledge base, so that repeated debugging is still needed when the same model is produced again, and an 'empirical fault' is formed.
Disclosure of Invention
The invention aims to provide a device control system and a method for OPS computer mainboard production, which are used for solving the problems in the prior art.
In order to achieve the above purpose, the invention provides a device control method for OPS computer motherboard production, comprising the following steps:
s1, aiming at OPS mainboards of different models, collecting technological parameters, compressing the technological parameters into high-dimensional topological feature vectors by utilizing a self-encoder, and constructing a topological vector space, wherein each production line is provided with a context memory network, and key information in the line changing process is recorded and stored in the context memory network after each line changing is completed;
S2, when a new OPS main board production order is received, identifying the product model in the order, and carrying out similarity retrieval in a topology vector space and a context memory network;
s3, dynamically distributing weights according to the confidence coefficient of the topological similarity and the context memory similarity by using a weighted fusion algorithm, taking the parameters retrieved by the topological space as basic parameters, taking the adjustment strategy extracted from the context memory network as an increment, and fusing to generate initial process parameter settings of new order production equipment;
S4, performing small-batch trial production according to initial parameter setting, detecting and analyzing production data and product quality in the trial production process, and comparing an actual production result with an expected result;
And S5, when the deviation between the actual trial production effect and the predicted result exceeds a dynamic threshold value based on historical data statistics, recording the current line changing task into a context memory network, covering or supplementing similar historical records, feeding back the technological parameters of the new model, the adjusted optimization parameters and the actual production result into a topology vector space, recalculating the position of the new model in the topology space and the similarity of the new model with other models, and updating the parameter distribution in the topology space.
According to the scheme, the technological parameters comprise a patch path, a welding curve, a PCB layer number, element density, a bonding pad size, element type distribution, signal layer characteristics and thermal design parameters;
The key information comprises a primary model, a target model, adjusted equipment parameters, production results after line replacement, equipment states, order priority, environment parameters, material batches and manual intervention records.
According to the above scheme, step S2 includes:
s21, analyzing the product model of the new OPS main board production order, and extracting the search keywords of the product model in the topological vector space;
S22, calculating the topological similarity of the novel number topological feature vector and the historical model vector, sorting according to the descending order of the topological similarity, selecting K historical models with the topological similarity exceeding a preset threshold value as topological neighbors, and acquiring an optimal process parameter set corresponding to the K neighbor models from a topological vector space, wherein K is a positive integer;
S23, taking the current line changing direction as a retrieval condition, matching historical records with the same or similar line changing directions in a memory bank, and screening a historical adjustment strategy with the production result score higher than a qualified threshold value, wherein the qualified threshold value is obtained according to the production quality standard and historical production result data;
s24, cross-verifying the topology retrieval result and the context retrieval result, and when the topology retrieval result and the context retrieval result point to the same group of historical model, improving the parameter confidence weight of the historical model;
s25, outputting a K neighbor technological parameter set obtained by topology space retrieval and a history adjustment strategy set matched with the context memory.
According to the above scheme, step S3 includes:
s31, extracting a process parameter set of each neighbor model in the topological space retrieval result, and calculating the occurrence frequency and numerical distribution characteristics of each parameter in the K neighbors;
s32, establishing a dynamic weight distribution model, inputting topological similarity, context matching degree scores, statistical distribution characteristics of neighbor parameters and verification times and success rate of a history adjustment strategy, wherein the formula is as follows:
;
Wherein wtopo is represented as a topological similarity weight, Stopo is represented as a topological similarity, C is represented as a statistical distribution feature of a neighbor parameter, alpha and beta are represented as adjustment coefficients for balancing the influence of each factor on the weight, Scontext is represented as a context matching degree score, N is represented as the verification times of a history adjustment strategy, Psuccess is represented as the verification success rate of the history adjustment strategy, the weight satisfies wtopo+wcontext =1 through normalization processing, and the context matching degree weight wcontext=1-wtopo;
The context similarity is based on the matching degree scores of the line changing direction and the production scene;
S33, according to the output result of the dynamic weight distribution model, taking topology retrieval parameters as a basic parameter frame, taking a context adjustment strategy as an increment parameter item, and carrying out normalization superposition according to weight proportion to generate a fused initial parameter set;
The topology retrieval parameter set is T= { T1,t2,…,th,…,ts }, wherein Th is represented as an index of the topology retrieval parameter, the context adjustment policy parameter set is Ca={c1,c2,…,ch,…,cs }, wherein Ch is represented as an index of the topology retrieval parameter, s is represented as the parameter number, the fused initial parameter set is I= { I1,i2,…,ih,…,is }, wherein Ih is represented as an index of the fused initial parameter, Ih=wtopo×th+wcontext×ch;
s34, selecting a party with higher confidence as a final parameter value for the parameter item with conflict based on an arbitration mechanism of the confidence;
S35, simulating and verifying the fused initial parameter set based on historical production data, evaluating the stability of parameter combinations under different equipment states and material batches, generating a parameter confidence level report, outputting initial process parameter setting of new order production equipment when the confidence level is higher than a preset threshold, re-executing the step S31 when the confidence level is lower than the preset threshold, and triggering manual intervention when the confidence level is still lower than the threshold after 3 times of continuous re-execution.
According to the above scheme, step S4 includes:
S41, performing small-batch trial production according to initial parameter setting, and collecting multidimensional production data in the trial production process in real time;
S42, constructing a trial production data analysis matrix based on a topological similarity search result and a context memory strategy, and comparing acquired data with an expected target value item by item to calculate the deviation degree of each index, wherein the indexes comprise parameter items exceeding an expected value range, defect types which do not reach quality indexes and production efficiency deviation points;
S43, constructing a reinforcement learning optimization model, wherein the reinforcement learning optimization model comprises a state space, an action space and a reward function, the state space defines the mapping relation between equipment parameters and product quality, the action space sets the adjustment range and step length of each technological parameter, and the reward function constructs a multi-objective optimization function based on the deviation degree of each index;
S44, sorting the priority of the difference parameter items, generating an initial adjustment strategy based on a history adjustment experience library, selecting an optimal adjustment strategy through a reinforcement learning model iteration optimization strategy by simulating the production effect after each adjustment, and generating an optimized parameter set, wherein the optimized parameter set comprises a correction value of a basic parameter, an adjustment amplitude of an increment parameter and a confidence score of each parameter.
According to the scheme, the multidimensional production data comprise equipment operation parameters, product production indexes and environmental state data, wherein the equipment operation parameters comprise positioning accuracy of a chip mounter and actual measurement temperature of a reflow soldering temperature zone, the product production indexes comprise a welding spot virtual soldering rate, element offset and electrical test passing rate, and the environmental state data comprise workshop temperature and humidity and voltage fluctuation amplitude.
According to the above scheme, step S5 includes:
S51, calculating the comprehensive deviation degree of the actual effect of trial production and the predicted result, wherein the comprehensive deviation degree comprises the steps of extracting an actual measurement value of a key quality index, comparing the actual measurement value with the predicted value to calculate a single deviation rate, determining a dynamic threshold value of each index based on the statistical distribution of historical data, and weighting to calculate the comprehensive deviation degree, and the formula is as follows:
;
Wherein Etotal is expressed as comprehensive deviation degree, Ei is expressed as single deviation rate, i is expressed as index of index, wi is expressed as weight of each index, and is determined according to statistical distribution of historical data, u is expressed as total number of indexes;
S52, when the comprehensive deviation degree exceeds the dynamic deviation threshold, judging that the method is a valid deviation case, and when the comprehensive deviation degree does not exceed the dynamic deviation threshold, recording production data and not updating a context memory network and a topology vector space;
The dynamic bias threshold, comprising:
Setting a basic threshold value, wherein the formula isWherein Ti is represented as a base threshold, mui is represented as a historical mean of the index i, sigmai is represented as a standard deviation, and k is represented as a confidence coefficient;
Combining the order priority P and the equipment state S to dynamically correct the threshold, wherein the formula is Tadjusted=Ti X (1+τxP+ψxS), Tadjusted is represented as a dynamically adjusted threshold, and τ and ψ are represented as adjustment coefficients;
The dynamic thresholds of all the quality indexes are weighted and fused to form a dynamic deviation threshold, and the formula is as follows: Wherein Ttotal is represented as a dynamic deviation threshold, wi is the weight of the ith index;
S53, extracting complete context information of the current line change, searching similar historical records in a memory network, comparing production effects and then preferentially covering when the highly similar records exist, and adding independent memory nodes when the similar records do not exist;
S54, combining and encoding the technological parameters of the new model and the optimized parameters, recalculating the feature vectors of the new model in the topological space, updating the similarity relation matrix with the historical model, and adjusting the parameter distribution density of the affected area;
S55, after the context memory network and the topology vector space are updated, consistency verification is carried out on the updated content, whether the newly added or modified record in the memory network is matched with the change of the medium-size parameter distribution in the topology space is checked, when the verification is passed, the whole knowledge updating flow is judged to be completed, when the verification is not passed, the data record and the calculation step in the updating process are checked back, and the updating operation is carried out again after the error is corrected.
The equipment control system for OPS computer mainboard production comprises a data management module, a similarity retrieval module, a parameter fusion module, a parameter optimization module and a knowledge evolution module;
the data management module comprises a process parameter module and a line changing information module, wherein the process parameter module is used for collecting process parameters of OPS mainboards with different types, compressing the process parameters into high-dimensional topological feature vectors and constructing a topological vector space;
The similarity retrieval module comprises a topology vector space retrieval module, a context memory network retrieval module and a cross verification module; the system comprises a topology vector space retrieval module, a context memory network retrieval module, a cross verification module, a context memory network retrieval module, a topology vector space retrieval module, a cross verification module and a context retrieval module, wherein the topology vector space retrieval module analyzes the product model of a new order, extracts retrieval keywords and acquires optimal technological parameter sets corresponding to K neighbor models;
The parameter fusion module comprises a weight calculation module, a conflict arbitration module and a simulation verification module; the system comprises a weight calculation module, a conflict arbitration module, a simulation verification module, a calculation module and a calculation module, wherein the weight calculation module establishes a dynamic weight distribution model and calculates topological similarity weight and context matching degree weight;
The parameter optimization module comprises a trial production data module, a data deviation analysis module and a reinforcement learning module; the trial production data module is used for carrying out small-batch trial production according to initial parameter setting and collecting multidimensional production data in the trial production process in real time; the data deviation analysis module is used for constructing a trial production data analysis matrix based on a topological similarity retrieval result and a context memory strategy, comparing acquired data with an expected target value item by item, and calculating the deviation degree of each index;
The knowledge evolution module comprises a deviation judging module, a knowledge updating module and a consistency verification module, wherein the deviation judging module calculates the comprehensive deviation degree of the actual effect of trial production and the predicted result, judges whether the comprehensive deviation degree exceeds a dynamic deviation threshold value, updates a topology vector space and a context memory network, and the consistency verification module performs consistency verification on updated contents of the context memory network and the topology vector space and checks whether a record newly added or modified in the memory network is matched with the change of the medium-size parameter distribution in the topology space.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the invention, through collaborative retrieval of the topology vector space and the context memory network, the optimal process parameters of similar models are rapidly obtained, and the line changing and debugging time is reduced;
2. the invention adopts an incremental parameter superposition mechanism to ensure that parameter adjustment is more accurate and accurate by dynamically calculating the joint confidence coefficient of the topological similarity and the context matching degree;
3. According to the invention, through the reinforcement learning model, dynamic fine adjustment is performed aiming at the process difference of the mainboards of different models, so that the debugging time of small-batch custom orders is reduced.
Drawings
FIG. 1 is a flow chart showing the steps of an apparatus control method for OPS computer motherboard production according to the invention;
Fig. 2 is a schematic structural diagram of an equipment control system for OPS computer motherboard production according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Embodiment as shown in fig. 1-2, the present invention provides a technical solution, an apparatus control method for OPS computer motherboard production, the method comprising the steps of:
s1, aiming at OPS mainboards of different models, collecting technological parameters, compressing the technological parameters into high-dimensional topological feature vectors by using a self-encoder to construct a topological vector space, deploying a context memory network for each production line, and recording key information in the line changing process and storing the key information into the context memory network after each line changing is completed;
Specifically, the process parameters comprise a patch path, a welding curve, a PCB layer number, element density, a bonding pad size, element type distribution, signal layer characteristics and thermal design parameters, and the key information comprises a primary model, a target model, an adjusted equipment parameter, a production result after line replacement, an equipment state, order priority, an environment parameter, a material batch and a manual intervention record.
For example, the OPS motherboard of the model A is required to be switched to the OPS motherboard of the model B, the number of layers of the PCB of the model A is 12, the element density is 80 per square inch, the high-precision BGA chip is contained, the number of layers of the PCB of the model B is 8, the element density is 60 per square inch, and the conventional SMT element is mainly used;
collecting technological parameters of model A, namely, patch path density 150 pieces per square inch, welding curve peak temperature 245 ℃ and pad size 0.3mm multiplied by 0.3mm, compressing the parameters into 128-dimensional topological vectors [0.23,0.41..0.17 ], [0.31,0.35..0.24 ];
The historical line change record shows that when the model A is switched to the similar model C, the reflow soldering temperature rising rate is adjusted by +10% (for example, from 2.5 ℃ per second to 2.75 ℃ per second), the patch pressure is reduced from 50g to 45g, the first inspection qualification rate is increased from 85% to 95%, and the record is stored in a context memory network, and the record is only used for illustration and not limitation.
S2, when a new OPS main board production order is received, identifying the product model in the order, and carrying out similarity retrieval in a topology vector space and a context memory network;
Specifically, step S2 includes:
s21, analyzing the product model of the new OPS main board production order, and extracting the search keywords of the product model in the topology vector space;
S22, calculating the topological similarity of a novel number topological feature vector and a historical model vector, sorting according to a topological similarity descending order, selecting K historical models with the topological similarity exceeding a preset threshold value as topological neighbors, acquiring an optimal process parameter set corresponding to the K neighbor models from a topological vector space, wherein K is a positive integer, for example, calculating the cosine similarity of the topological vector of model B and the historical model vector, screening out K=3 neighbors, namely model C, similarity 0.85, optimal parameter set comprises a paster speed of 18000 points/hour, reflow soldering constant temperature area temperature of 180 ℃, model D, similarity 0.82, parameter set comprises a paster pressure of 45g, soldering time of 25S, model E, similarity 0.79, and parameter set comprises soldering paste printing thickness of 80 mu m.
S23, using the current line changing direction as a search condition, matching historical records with the same or similar line changing directions in a memory bank, screening a historical adjustment strategy with the production result score higher than a qualified threshold value, wherein the qualified threshold value is obtained according to the production quality standard and historical production result data, for example, the line changing direction is A-B, the current line changing direction is matched with the line changing strategy of A-C in the historical records, the adjustment strategy comprises reflow soldering heating rate +10%, patch pressure-10%, and the production result score is 92 (80 minutes of the qualified threshold value);
The production result score is obtained based on core quality and efficiency indexes in trial production or mass production, a target value is set for each index, a single score is calculated according to the deviation degree of an actual value and the target value, weights are distributed according to the importance of the indexes, and a comprehensive score is obtained through weighted summation, wherein for example, the target virtual welding rate is 1 percent, the single score is 90 percent, if the actual virtual welding rate exceeds 1.5 percent, the single score is lower than 60 percent, the production result score is 92 percent in a history record with the line changing direction of A-C, the welding point virtual welding rate is reduced to 1.5 percent from 5 percent before adjustment, the single score is 85 percent, the weight is 30 percent, the electrical test passing rate is increased to 95 percent, the single score is 95 percent, the weight is 40 percent, the line changing debugging time is shortened to 6 hours from 10 hours, the single score is 90 percent, the weight is 30 percent, and the comprehensive score is 0.3X185+0.4X105+0.3X10=92 percent;
S24, cross-verifying the topology retrieval result and the context retrieval result, and when the topology retrieval result and the context retrieval result point to the same group of historical models, improving the parameter confidence weight of the historical models, for example, the model C of the topology neighbor model C and the model C of the context memory record point to the same group of parameters, and improving the confidence weight by 15%;
s25, outputting a K neighbor technological parameter set obtained by topology space retrieval and a history adjustment strategy set matched with the context memory.
S3, dynamically distributing weights according to the confidence coefficient of the topological similarity and the context memory similarity by using a weighted fusion algorithm, taking the parameters retrieved by the topological space as basic parameters, taking the adjustment strategy extracted from the context memory network as an increment, and fusing to generate initial process parameter settings of new order production equipment;
specifically, step S3 includes:
s31, extracting a process parameter set of each neighbor model in the topological space retrieval result, and calculating the occurrence frequency and numerical distribution characteristics of each parameter in the K neighbors;
s32, establishing a dynamic weight distribution model, inputting topological similarity, context matching degree scores, statistical distribution characteristics of neighbor parameters and verification times and success rate of a history adjustment strategy, wherein the formula is as follows:
;
Wherein wtopo is represented as a topology similarity weight, Stopo is represented as a topology similarity, C is represented as a statistical distribution feature of a neighbor parameter, alpha and beta are represented as adjustment coefficients for balancing the influence of each factor on the weight, Scontext is represented as a context matching degree score, N is represented as the verification times of a history adjustment strategy, Psuccess is represented as the verification success rate of the history adjustment strategy, and the context matching degree weight wcontext=1-wtopo;
For example, the input topology similarity Stopo =0.85, the context matching degree Scontext =0.92, the neighbor parameter statistical distribution characteristic c=0.8, the history policy verification times n=15, the verification success rate Psuccess =0.95 of the history adjustment policy, the adjustment coefficient α=0.8, β=0.1, and the wtopo=0.3,wcontext=1-wtopo =0.7 obtained by calculation;
When the parameters recommended by topology retrieval exceed the physical limit of the current equipment, for example, the recommended welding temperature is 260 ℃, but the limit of the equipment B is 250 ℃, the equipment B is automatically cut off to the maximum value allowed by the equipment, and an alarm is recorded;
s33, according to the output result of the dynamic weight distribution model, taking topology retrieval parameters as a basic parameter frame, taking a context adjustment strategy as an increment parameter item, and carrying out normalization superposition according to weight proportion to generate a fused initial parameter set;
The topology retrieval parameter set is T= { T1,t2,…,th,…,ts }, wherein Th is represented as an index of the topology retrieval parameter, the context adjustment policy parameter set is Ca={c1,c2,…,ch,…,cs }, wherein Ch is represented as an index of the topology retrieval parameter, s is represented as the parameter number, the fused initial parameter set is I= { I1,i2,…,ih,…,is }, wherein Ih is represented as an index of the fused initial parameter, Ih=wtopo×th+wcontext×ch;
For example, topology parameters, namely reflow peak temperature 245 ℃ (model C), a context strategy, namely heating rate +10%, corresponding to theoretical increment of peak temperature 245×10% =24.5 ℃, and peak temperature after fusion, namely 0.3×245+0.7× (245+24.5) = 262.15 ℃, which is 262 ℃;
S34, selecting a party with higher confidence as a final parameter value for parameter items with conflicts based on an arbitration mechanism of the confidence, wherein for example, the topology parameter suggests a patch speed of 20000 points/hour, the context policy suggests 18000 points/hour, and 18000 points/hour are selected due to more verification times of the context policy;
S35, simulating and verifying the fused initial parameter set based on historical production data, evaluating the stability of parameter combinations under different equipment states and material batches, generating a parameter confidence level report, outputting initial process parameter setting of new order production equipment when the confidence level is higher than a preset threshold, re-executing the step S31 when the confidence level is lower than the preset threshold, and triggering manual intervention when the confidence level is still lower than the threshold after 3 times of continuous re-execution.
S4, performing small-batch trial production according to initial parameter setting, detecting and analyzing production data and product quality in the trial production process, and comparing an actual production result with an expected result;
specifically, step S4 includes:
S41, performing small-batch trial production according to initial parameter setting, and collecting multidimensional production data in the trial production process in real time, wherein for example, 5 model B mainboards are subjected to trial production, the collected data comprise equipment operation parameters, namely actual measurement temperature 256 ℃ in a reflow soldering temperature area (set value 256 ℃), positioning accuracy of a chip mounter is +/-0.03 mm, and product indexes, namely a welding spot virtual welding rate of 3% (less than or equal to 1% of a target) and element offset of 0.08mm (less than or equal to 0.05mm of a target);
s42, constructing a trial production data analysis matrix based on a topological similarity search result and a contextual memory strategy, and comparing acquired data with an expected target value item by item to calculate the deviation degree of each index, wherein the indexes comprise parameter items exceeding an expected value range, defect types which do not reach quality indexes and production efficiency deviation points;
S43, constructing a reinforcement learning optimization model, wherein the reinforcement learning optimization model comprises a state space, an action space and a reward function, wherein the state space defines the mapping relation between equipment parameters and product quality, the action space sets the adjustment range and step length of each technological parameter, and the reward function constructs a multi-objective optimization function based on the deviation degree of each index;
for example, construct a bonus function of R=0.6× (1-3%) +0.4× (1-0.08/0.05) =0.82 (full scale 1);
S44, sorting the priority of the difference parameter items, generating an initial adjustment strategy based on a history adjustment experience library, selecting an optimal adjustment strategy through a reinforcement learning model iteration optimization strategy by simulating the production effect after each adjustment, and generating an optimized parameter set, wherein the optimized parameter set comprises a correction value of a basic parameter, an adjustment amplitude of an increment parameter and a confidence score of each parameter.
And S5, when the deviation between the actual trial production effect and the predicted result exceeds a dynamic threshold value based on historical data statistics, recording the current line changing task into a context memory network, covering or supplementing similar historical records, feeding back the technological parameters of the new model, the adjusted optimization parameters and the actual production result into a topology vector space, recalculating the position of the new model in the topology space and the similarity of the new model with other models, and updating the parameter distribution in the topology space.
Specifically, step S5 includes:
s51, calculating the comprehensive deviation degree of the actual effect of trial production and the predicted result, wherein the comprehensive deviation degree comprises the steps of extracting actual measured values of key quality indexes, calculating a single deviation rate by comparing the predicted values, determining dynamic thresholds of all indexes based on historical data statistical distribution, and weighting to calculate the comprehensive deviation degree, wherein the formula is as follows:
;
Wherein Etotal is expressed as comprehensive deviation degree, Ei is expressed as single deviation rate, i is expressed as index of index, wi is expressed as weight of each index, and is determined according to statistical distribution of historical data, u is expressed as total number of indexes;
S52, when the comprehensive deviation degree exceeds the dynamic deviation threshold, judging that the method is a valid deviation case, and when the comprehensive deviation degree does not exceed the dynamic deviation threshold, recording production data and not updating a context memory network and a topology vector space;
a dynamic bias threshold, comprising:
Setting a basic threshold value, wherein the formula isWherein Ti is represented as a base threshold, mui is represented as a historical mean of the index i, sigmai is represented as a standard deviation, and k is represented as a confidence coefficient;
Combining the order priority P and the equipment state S to dynamically correct the threshold, wherein the formula is Tadjusted=Ti X (1+τxP+ψxS), Tadjusted is represented as a dynamically adjusted threshold, and τ and ψ are represented as adjustment coefficients;
The dynamic thresholds of all the quality indexes are weighted and fused to form a dynamic deviation threshold, and the formula is as follows: Wherein Ttotal is represented as a dynamic deviation threshold, wi is the weight of the ith index;
For example, Etotal =0.6x (3% -1%)/1% + 0.4x (0.08-0.05)/0.05=1.44, a dynamic threshold value Ttotal =1.2, and the comprehensive deviation degree exceeds the dynamic deviation threshold value, and the effective deviation case is judged to trigger updating;
s53, extracting complete context information of the current line change, searching similar historical records in a memory network, comparing production effect and then preferentially covering when high similar records exist, and adding independent memory nodes when no similar records exist;
S54, combining and encoding the technological parameters of the new model and the optimized parameters, recalculating the characteristic vector of the new model in the topological space, updating the similarity relation matrix between the new model and the historical model, and adjusting the parameter distribution density of an affected area, wherein for example, the vector of the model B in the topological space is updated to be [0.33,0.37, the number of the required values is 0.26], the similarity between the model B and the model C is improved to be 0.88, and the parameter distribution density is increased by 12% in an X main board area;
S55, after the context memory network and the topology vector space are updated, consistency verification is carried out on the updated content, whether the newly added or modified record in the memory network is matched with the change of the medium-size parameter distribution in the topology space is checked, when the verification is passed, the whole knowledge updating flow is judged to be completed, when the verification is not passed, the data record and the calculation step in the updating process are checked back, and the updating operation is carried out again after the error is corrected.
The invention provides another technical scheme, an equipment control system for OPS computer mainboard production, which comprises a data management module, a similarity retrieval module, a parameter fusion module, a parameter optimization module and a knowledge evolution module;
The data management module comprises a process parameter module and a line changing information module, wherein the process parameter module is used for collecting process parameters of OPS mainboards with different types, compressing the process parameters into high-dimensional topological feature vectors and constructing a topological vector space;
the system comprises a similarity retrieval module, a topology vector space retrieval module, a context memory network retrieval module and a cross verification module, wherein the topology vector space retrieval module analyzes the product model of a new order, extracts retrieval keywords and acquires the optimal technological parameter sets corresponding to K neighbor models;
The parameter fusion module comprises a weight calculation module, a conflict arbitration module and a simulation verification module; the system comprises a weight calculation module, a conflict arbitration module, a simulation verification module, a calculation module and a calculation module, wherein the weight calculation module establishes a dynamic weight distribution model and calculates topological similarity weight and context matching degree weight;
The parameter optimization module comprises a trial production data module, a data deviation analysis module and a reinforcement learning module, wherein the trial production data module is used for carrying out small-batch trial production according to initial parameter setting and collecting multidimensional production data in the trial production process in real time;
The knowledge evolution module comprises a deviation judging module, a knowledge updating module and a consistency verification module, wherein the deviation judging module calculates the comprehensive deviation degree of the actual effect of trial production and the predicted result, judges whether the comprehensive deviation degree exceeds a dynamic deviation threshold value, the knowledge updating module updates a topology vector space and a context memory network, and the consistency verification module performs consistency verification on the updated contents of the context memory network and the topology vector space and checks whether a record newly added or modified in the memory network is matched with the change of the medium-size parameter distribution in the topology space.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims (9)

CN202511071885.9A2025-08-012025-08-01 A device control system and method for producing OPS computer motherboardsWithdrawnCN120560213A (en)

Priority Applications (1)

Application NumberPriority DateFiling DateTitle
CN202511071885.9ACN120560213A (en)2025-08-012025-08-01 A device control system and method for producing OPS computer motherboards

Applications Claiming Priority (1)

Application NumberPriority DateFiling DateTitle
CN202511071885.9ACN120560213A (en)2025-08-012025-08-01 A device control system and method for producing OPS computer motherboards

Publications (1)

Publication NumberPublication Date
CN120560213Atrue CN120560213A (en)2025-08-29

Family

ID=96815832

Family Applications (1)

Application NumberTitlePriority DateFiling Date
CN202511071885.9AWithdrawnCN120560213A (en)2025-08-012025-08-01 A device control system and method for producing OPS computer motherboards

Country Status (1)

CountryLink
CN (1)CN120560213A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN120469342A (en)*2025-07-162025-08-12四川普什宁江机床有限公司Processing method and system for machining precision prediction data of gear hobbing machine tool

Cited By (1)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN120469342A (en)*2025-07-162025-08-12四川普什宁江机床有限公司Processing method and system for machining precision prediction data of gear hobbing machine tool

Similar Documents

PublicationPublication DateTitle
CN120560213A (en) A device control system and method for producing OPS computer motherboards
TsaiThermal parameters optimization of a reflow soldering profile in printed circuit board assembly: A comparative study
CN119417189B (en) Intelligent scheduling method of construction equipment based on AI algorithm
CN118915677B (en) A method for monitoring the production process of CNC machine tool castings
CN109214426A (en)A kind of method and deep neural network model of the detection of object appearance
CN119399518A (en) An intelligent detection method for hot spots of photovoltaic cells by integrating multimodal data
CN118468040A (en) A small sample quality prediction method based on meta-learning
CN115408949B (en)Load model parameter identification method System, apparatus and medium
Tang et al.A nonparametric control chart for monitoring count data mean
CN119624000A (en) Carbon emission monitoring and management system based on industrial Internet of Things
CN119294270A (en) A parameter optimization method for powder laser 3D printing molding process
CN113656238A (en)AI industrial application capability test method and system of intelligent terminal
CN101477582B (en)Model modification method for a semiconductor device
CN118426514A (en)High-precision wave soldering temperature control system
CN118742895A (en) Component size classification method and production equipment
CN119647398A (en) A solder joint simulation life prediction method and system based on random forest
CN115422840A (en)Ridge-scale runoff estimation method based on physical model mixed deep learning model
CN115983113A (en)Screen printing process optimization method under full-connection neural network model
CN113515845A (en)Method and device for predicting reliability of atomic clock
CN120030974A (en) An interconnection system for printed circuit boards
CN114815494B (en)Optical proximity correction method and system, mask plate, equipment and storage medium
CN119150705B (en) Virtual power plant gateway data processing method and system based on edge computing
CN120181685B (en) Data annotation method and system for gold foil product testing
CN118232318B (en) Short-term wind power prediction method, system, medium and equipment under typhoon scenario
CN118395931B (en)Optimization design method of amplifier

Legal Events

DateCodeTitleDescription
PB01Publication
PB01Publication
WW01Invention patent application withdrawn after publication
WW01Invention patent application withdrawn after publication

Application publication date:20250829


[8]ページ先頭

©2009-2025 Movatter.jp