A production quality directional detection method and system for 200-level direct-weld enameled wire
By constructing a defect-triggered timing matrix and confidence-driven spread map screening mechanism, the problem of cross-process defect transmission chain tracking in the manufacturing of 200-level direct welding enameled wire is solved, and efficient defect detection and quality control are achieved.
Patent Information
- Application Number
- CN202510855400.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-25
- Publication Date
- 2025-09-02
- Estimated Expiration
- 2045-06-25
AI Technical Summary
The prior art cannot effectively track the cross-process defect transmission chain in the manufacturing process of 200-level direct welding enameled wire, and the detection efficiency is inefficient, resulting in waste of resources and missed potential defects.
By constructing a defect-triggered timing matrix and confidence-driven spread map screening mechanism, dynamically simulate defect delivery paths, identify high-risk defect clusters and perform targeted quality detection.
It significantly improves the targeting nature of defect detection and the full process quality prevention and control capabilities, reduces the cost of invalid screening, realizes early identification of cross-process defect transmission chains, and improves detection efficiency.
Smart Images
Figure CN120372223B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of enameled wire quality detection, and in particular to a production quality directional detection method and system for 200-level direct-weldable enameled wire. Background Art
[0002] In the manufacturing of Class 200 solderable enameled wire, the stringent temperature resistance and solderability requirements mean that even minor defects can lead to performance failure. The complex, multi-step process chain further amplifies the risk of defect transmission. Therefore, quality inspection is crucial to ensure that the wire's electrical performance, mechanical strength, and solderability meet standards. Existing technologies typically rely on random sampling for quality control. This involves selecting samples from a production batch and then comprehensively testing each sample for numerous defect types, including but not limited to uneven insulation thickness, coating defects, and thermal aging instabilities. While this comprehensive inspection approach can uncover a variety of potential issues, it inherently leads to significant inefficiencies. Firstly, all defect types, regardless of their frequency, must be examined equally, consuming significant testing resources and time. Secondly, with the interconnected nature of the production process, from pay-out, annealing, painting, baking, and take-up, minor deviations in upstream processes can propagate through defects downstream, forming a complex temporal relationship. Traditional sampling methods, which isolate defects at each node, fail to capture this dynamic propagation pattern, resulting in poor targeting and redundant resources. In addition, when faced with large-scale production, efficiency bottlenecks are particularly prominent. Insufficient frequency of spot checks may miss key defects, and simply increasing the amount of testing will increase the cost burden and be inefficient.
[0003] In the existing technology, the problem of being unable to track the cross-process defect transmission chain and low detection efficiency in the manufacturing field of 200-level direct-weld enameled wire needs to be solved urgently. Summary of the Invention
[0004] The present application provides a method and system for directional detection of the production quality of 200-level direct-weld enameled wire, which is used to solve the technical problems in the prior art of being unable to track the cross-process defect transmission chain and having low detection efficiency.
[0005] In view of the above problems, the present application provides a method and system for directional detection of the production quality of 200-level direct-weld enameled wire.
[0006] In a first aspect, the present application provides a method for directional detection of production quality of 200-level direct-weld enameled wire, the method comprising: performing a production simulation based on a k-th production control deviation matrix of a k-th process node, combined with a k-th node enameled wire defect parameter, to obtain the k-th node enameled wire defect parameter, where k is an integer, N ≥ k ≥ 1, N represents the total number of enameled wire process nodes, and the k-th node is an upstream process node of the k-th process node;
[0007] Among them, when k=1, the enameled wire defect parameter of the k-1th node is empty;
[0008] When k is equal to N, the defect parameters of the enameled wire at the first node are integrated to the defect parameters of the enameled wire at the Nth node to construct a defect triggering timing matrix;
[0009] Combining defect propagation types on the defect trigger timing matrix to generate a plurality of defect propagation maps;
[0010] Traversing the plurality of defect propagation maps to perform confidence evaluation and obtain a plurality of confidence levels, wherein the confidence levels are positively correlated with the triggering frequencies of the defect propagation maps;
[0011] According to the plurality of confidence levels, the defect types of the union of the defect propagation maps with confidence levels greater than or equal to a confidence threshold are extracted and sent to a quality inspection end for performing directional quality inspection.
[0012] In a second aspect, the present application provides a production quality directional detection system for 200-level direct-weld enameled wire, comprising:
[0013] A control deviation matrix module is used to perform production simulation based on the kth production control deviation matrix of the kth process node and the enameled wire defect parameter of the k-1th node to obtain the enameled wire defect parameter of the kth node, where k is an integer, N ≥ k ≥ 1, N represents the total number of enameled wire process nodes, and the k-1th node is the upstream process node of the kth process node;
[0014] Among them, when k=1, the enameled wire defect parameter of the k-1th node is empty;
[0015] When k is equal to N, the defect parameters of the enameled wire at the first node are integrated to the defect parameters of the enameled wire at the Nth node to construct a defect triggering timing matrix;
[0016] A defect propagation map module is used to combine defect propagation types on the defect trigger timing matrix to generate a plurality of defect propagation maps;
[0017] A confidence evaluation module is used to traverse the plurality of defect propagation maps to perform confidence evaluation and obtain a plurality of confidence levels, wherein the confidence level is positively correlated with the triggering frequency of the defect propagation map;
[0018] The defect detection summary module is used to extract the union defect types of the defect propagation maps with confidence greater than or equal to the confidence threshold according to the multiple confidence levels and send them to the quality inspection end to perform directional quality inspection.
[0019] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0020] This application proposes a method and system for targeted production quality inspection of Class 200 direct-weld enameled wire. By constructing a multi-node defect transmission time series model and a confidence-driven propagation map screening mechanism, it significantly improves the targeting of defect detection and the quality control capabilities of the entire process. Compared with traditional methods, the technical solution provided by this application significantly overcomes the blind spots in tracking the cross-process defect transmission chain. By dynamically simulating the cascading impact of upstream defects on downstream processes, it achieves early identification of defect initiation nodes. At the same time, based on a confidence evaluation system weighted by historical frequency, it accurately eliminates low-probability defect combinations, focusing detection resources on high-incidence defect clusters and significantly reducing the cost of ineffective screening.
[0021] This application achieves the technical effect of covering the maximum quality risk area with the minimum detection cost, tracking the cross-process defect transmission chain, and improving detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0023] Figure 1 A flow chart of a method for directional detection of production quality of a Class 200 direct-weld enameled wire provided in an embodiment of the present application;
[0024] Figure 2 This is a structural schematic diagram of a production quality directional detection system for 200-level direct-weld enameled wire provided in an embodiment of the present application.
[0025] In the accompanying drawings, the components represented by the reference numerals are described as follows:
[0026] Control deviation matrix module 100, defect propagation map module 200, confidence evaluation module 300, defect detection summary module 400. DETAILED DESCRIPTION
[0027] This application provides a production quality directional detection method and system for 200-level direct-weld enameled wire, which is used to solve the technical problems in the existing technology of the inability to track the cross-process defect transmission chain and low detection efficiency in the field of 200-level direct-weld enameled wire manufacturing.
[0028] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0029] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.
[0030] Example 1, as Figure 1 As shown, the present application provides a method for directional detection of production quality of 200-level direct-weld enameled wire, wherein the method comprises:
[0031] S10: Based on the kth production control deviation matrix of the kth process node and in combination with the enameled wire defect parameter of the k-1th node, a production simulation is performed to obtain the enameled wire defect parameter of the kth node, where k is an integer, N ≥ k ≥ 1, N represents the total number of enameled wire process nodes, and the k-1th node is the upstream process node of the kth process node;
[0032] Among them, when k=1, the enameled wire defect parameter of the k-1th node is empty;
[0033] When k is equal to N, the defect parameters of the enameled wire at the first node are integrated to the defect parameters of the enameled wire at the Nth node to construct a defect triggering timing matrix.
[0034] In the multi-step continuous production of enameled wire, traditional methods, due to the isolation of inspection data between processes, are unable to establish a quantitative model of the impact of upstream defects on downstream processes. This is especially true when defects originate in the first process, lacking an initial parameter baseline, leading to simulation interruptions. Downstream nodes, on the other hand, rely solely on control deviation analysis within their own process, ignoring upstream defects, such as the cascading effects of annealing and oxidation caused by abnormal pay-off tension. This can lead to missed root cause detection and disrupted transmission paths. Furthermore, discrete node inspection data is difficult to integrate into a global defect evolution view, hindering risk prediction throughout the entire process and resulting in low inspection efficiency.
[0035] Step S10 in the method provided in the embodiment of the present application includes:
[0036] The kth production control deviation record matrix set of the kth process node and the k-1th enameled wire defect record parameters are collected as input data, the kth enameled wire defect record parameters are collected as true value data, and the kth process node production simulator training data is constructed;
[0037] Wherein, the kth process node includes:
[0038] Pay-off process node, annealing process node, painting process node, baking process node, cooling process node, take-up process node;
[0039] Based on the k-th process node production simulator training data, a plurality of k-th process node simulators are trained, wherein when k=1, the k-1-th node enameled wire defect recording parameter is empty;
[0040] The outputs of the plurality of k-th process node simulators are integrated to obtain a k-th process node production simulator, and a production simulation is performed based on a k-th production control deviation matrix of the k-th process node and a k-1-th node enameled wire defect parameter to obtain a k-th node enameled wire defect parameter.
[0041] When k<N, transmitting the k-th node enameled wire defect parameter to the k+1-th process node production simulator;
[0042] When k=N, the first-node enameled wire defect parameter is output until the N-th node enameled wire defect parameter is output.
[0043] Exemplarily, the kth production control deviation record matrix set and the enameled wire defect record parameters of the k-1th node, included in the kth process node, are collected as input data. The kth process node includes the pay-off process node, annealing process node, painting process node, baking process node, cooling process node, and take-up process node, with N=6. The production control deviation record matrix set is a two-dimensional matrix representing the deviations of the production control parameters (such as temperature, speed, and tension) at the kth process node from their standard values. The rows of the matrix represent time batch sequences, such as the batch at 10:00 AM on June 5th, and the columns represent the absolute values of the differences between specific production control parameters and their standard values, such as paint liquid temperature deviation and pay-off length deviation. The data units are specific production control parameters, such as °C or mm. Exemplarily, paint thickness deviation is measured using a measuring tape to measure paint liquid temperature. The production control parameter for the painting process node is |actual thickness value - standard value|. For example, if the standard paint thickness value is 5 mm and the measured value is 6 mm, and the batch is at 10:00 AM on June 5th, then the production control parameter for the painting process node is |6-5| = 1. This production control parameter is represented in the matrix by (6.5.10:00AM, 1).
[0044] The enameled wire at the kth node is collected as the true value data. For example, whether the paint thickness does not meet the standard is collected as the true value data of the painting process node.
[0045] Construct several k-th process node production simulators. For example, a neural network is used to construct the k-th process node production simulator. The process node production simulator adopts a three-layer structure, namely the input layer, the hidden layer, and the output layer. The input layer has 2 nodes, which input the k-th production control deviation record matrix set of the k-th process node and the k-1-th node enameled wire defect record. The hidden layer has a three-layer structure. The first fully connected layer has 128 neurons, and the activation function uses ReLU. The purpose is to fuse the defect and control deviation features. The dropout layer dropout rate parameter is set to 0.3 to prevent overfitting. The second fully connected layer has 64 neurons, and the activation function uses ReLU. The purpose is to extract high-order nonlinear relationships. The output layer has 1 node, which outputs the simulated k-th process node defect record parameters. Among them, when k=1, the k-1-th node enameled wire defect record parameter is empty.
[0046] Train several k-th process node production simulators, and integrate the output data of several process node production simulators. For example, train 5 painting process node production simulators, combine them with the annealing process node enameled wire defect parameters, perform production simulation, obtain the painting process node enameled wire defect parameters, and take the union of the defect record parameters output by the painting process node production simulator for output, such as the paint thickness is too thick, the paint thickness is insufficient, etc.
[0047] When k<N, the defect parameters of the enameled wire at the kth node are transmitted to the k+1th process node production simulator. For example, the defect parameters of the annealing process node are transmitted to the painting process node.
[0048] When k = N, output the enameled wire defect parameters for the first node up to the Nth node. For example, when k = N = 6, output the enameled wire defect parameters for the first node up to the sixth node, such as the defect parameters from the payout process node to the takeup process node. Construct a defect trigger timing matrix, where the rows of the matrix represent the k process nodes and the columns represent the defect types.
[0049] This embodiment of the application constructs a closed-loop simulation chain for defect transmission between processes by dynamically coupling the k-th node production control deviation matrix with the k-1 node defect parameters. When k=1, the parameters are automatically initialized to null to ensure the integrity of the starting point of the entire chain. Node-by-node simulation tracks defects from initiation to evolution, such as how scratches on annealing nodes can cause bubbles in paint nodes. Finally, the time series matrix generated by integrating N node data presents a structured representation of the defect transmission trajectory across processes, laying the core data foundation for cross-process correlation analysis.
[0050] S20: performing defect propagation type combination on the defect trigger timing matrix to generate a plurality of defect propagation maps;
[0051] While the defect time series matrix records the defect type at each node, the massive amount of discrete data makes it difficult to automatically identify highly correlated cross-process combinations. Manually enumerating potential transmission paths, such as when annealing temperature deviation leads to uneven baking and curing, is inefficient and prone to missing low-frequency but highly harmful hidden links, such as the weak correlation between cooling microcracks and early wire-laying scratches. Existing methods are unable to transform the spatiotemporal relationships of defects in the matrix into a computable topological network, resulting in insufficient depth in the discovery of key defect clusters.
[0052] Step S20 in the method provided in the embodiment of the present application includes:
[0053] Extracting a first node defect type set from the defect trigger timing matrix, performing i-item combination enumeration to obtain a first node defect combination set, where 1≤i≤Q, i is an integer, and Q is the total number of first node defect types;
[0054] The first node defect type set is extracted from the defect trigger timing matrix, and i-item combination enumeration is performed to obtain the first node defect combination set, including:
[0055] Enumerate i items of the first node defect type set to obtain an i-item initial combination set of the first node;
[0056] Traversing the i-item initial combination set of the first node, retrieving the detection frequency percentage set of each i-item initial combination in the historical production samples of the first node;
[0057] According to the detection frequency ratio set, i initial combinations whose detection frequency ratio is greater than or equal to the detection frequency ratio threshold are extracted and set as the first node i combination set, where the initial value of i is 1;
[0058] If i=Q, add the first node one item combination set to the first node Q item combination set to the first node defect combination set;
[0059] If i<Q, i is incremented by one and the loop is executed.
[0060] Until the Nth node defect type set is extracted from the defect trigger timing matrix, j-item combination enumeration is performed to obtain the Nth node defect combination set, where 1≤j≤Y, j is an integer, and Y is the total number of Nth node defect types;
[0061] For the first node defect combination set up to the Nth node defect type set, randomly select a combination for each node, build a first defect propagation graph according to the node time sequence connection, and add it to the plurality of defect propagation graphs;
[0062] After the defect propagation maps are enumerated, the plurality of defect propagation maps are output.
[0063] In this embodiment, a set of first-node defect types is extracted from the defect trigger timing matrix. An i-item enumeration of the first-node defect types is performed to obtain an initial set of i-items of combinations for the first node, where 1≤i≤Q, i being an integer and Q being the total number of first-node defect types. For example, if Q is 3, then 1≤i≤3. When i=1, all single defects, such as scratches, are enumerated. Enumeration is implemented using the itertools.combinations function, and conditional filtering is performed using filter.
[0064] Traverse the initial combination set of i items at the first node, and retrieve the detection frequency percentage set of each i-item initial combination in the historical production samples of the first node. For example, if scratches appear 200 times in 1000 productions, then the frequency = number of occurrences ÷ total number of productions = 200 ÷ 1000 = 0.2.
[0065] According to the detection frequency ratio set, i initial combinations whose detection frequency ratio is greater than or equal to the detection frequency ratio threshold are extracted and set as the first node i combination set, where the initial value of i is equal to 1; the detection frequency threshold is pre-set data reflecting the frequency of occurrence of the defect type. For example, the detection frequency threshold is set to 0.05, and all initial combinations whose detection frequency is greater than or equal to 0.05 are extracted.
[0066] If i=Q, add the first node item combination set to the first node Q item combination set into the first node defect combination set to obtain the defect type set whose extracted detection frequency ratio is greater than or equal to the detection frequency ratio threshold. If i<Q, i is increased by one and the loop is executed again until i=Q.
[0067] Repeat the enumeration process until the Nth node defect type set is extracted from the defect trigger timing matrix. Perform the j-item combination enumeration to obtain the Nth node defect combination set. Where 1 ≤ j ≤ Y, j is an integer, and Y is the total number of defect types at the Nth node. Traverse the Nth node j-item initial combination set and retrieve the detection frequency percentage set for each j-item initial combination in the historical production samples at the Nth node. For example, if excessive paint thickness occurs 100 times in 1000 production runs, then the frequency = number of occurrences ÷ total number of production runs = 200 ÷ 1000 = 0.1. Based on the detection frequency percentage set, extract the j-item initial combinations whose detection frequency percentage is greater than or equal to the detection frequency percentage threshold. Set these as the Nth node j-item combination set, with the initial value of j equal to 1 and the detection frequency threshold set to 0.05, to obtain the Nth node defect combination set.
[0068] For the first node defect combination set through the Nth node defect type set, a combination is randomly selected for each node and connected sequentially to construct a first defect propagation map. For example, for the first node, wire bend is selected; for the second node, oxidation spot is selected; for the third node, excessive paint thickness is selected; for the fourth node, uneven curing is selected; for the fifth node, insufficient cooling is selected; and for the sixth node, excessive diameter is selected. These six defects are then connected sequentially to construct a first defect propagation map. Directed defect propagation maps are generated repeatedly until all combinations are exhausted, resulting in several defect propagation maps. Random selection is implemented using the random.choice function.
[0069] When the defect propagation map enumeration is completed, several defect propagation maps are output.
[0070] In this embodiment, a set of defect propagation maps covering the entire process chain is generated through automated combinational enumeration based on the node defect type set of the timing matrix. Each map represents a possible defect transmission path, such as oxidation spots as the second node, excessive paint thickness as the third node, and uneven curing as the fourth node. This transforms discrete defect data into a visual directed relationship network, visually presenting potential defect transmission paths. This mechanism transcends the limitations of manual experience, systematically presenting the association between explicit and implicit defects, and providing a complete path space for subsequent confidence evaluation.
[0071] S30: traversing the plurality of defect propagation maps to perform confidence evaluation and obtain a plurality of confidence levels, wherein the confidence level is positively correlated with the triggering frequency of the defect propagation map;
[0072] The authenticity assessment of defect propagation maps has long relied on expert experience and lacked objective quantitative evidence. High-frequency, low-risk paths, such as wire scraping, can overconsume inspection resources, while low-frequency, high-risk hidden links, such as annealing oxidation-induced cooling fractures, can be overlooked due to insufficient samples. Existing methods fail to balance the weights of path frequency and criticality, and short-chain maps, such as those containing only two nodes, have a high probability of misjudging reliability due to data sparsity.
[0073] Step S30 in the method provided in the embodiment of the present application includes:
[0074] extracting a first defect propagation map from the plurality of defect propagation maps;
[0075] Splitting the first defect propagation graph to obtain propagation graphs of the first two nodes and then the first N nodes;
[0076] Retrieve the first detection frequency ratio of the first two node propagation maps in the 200-level direct solderability enameled wire quality inspection log;
[0077] Until the N-1th detection frequency ratio of the first N node propagation graph in the 200-level direct solderability enameled wire quality inspection log is retrieved;
[0078] Calculating an average value of the first detection frequency ratio up to the N-1th detection frequency ratio, setting the average value as a first defect propagation map confidence level, and adding the confidence level to the plurality of defect propagation maps;
[0079] Calculating the average of the first detection frequency ratio up to the N-1th detection frequency ratio includes:
[0080] The ratio of 2 to N is the weight of the first detection frequency, and the ratio of N to N-1 is the weight of the N-1th detection frequency.
[0081] Based on the first detection frequency proportion weight up to the N-1th detection frequency proportion weight, an average of the first detection frequency proportion up to the N-1th detection frequency proportion is calculated.
[0082] In the embodiment of the present application, a first defect propagation map is randomly extracted from a plurality of propagation maps, and the random extraction is implemented using the random.choice function.
[0083] The first defect propagation graph is split to obtain the propagation graphs of the first two nodes, the first three nodes, and finally the first N nodes.
[0084] Retrieve the first detection frequency ratio of the first two node spread graphs in the quality inspection log of 200-level direct-weldable enameled wire. The first detection frequency ratio = the number of times the first two node graphs appear in the quality inspection log divided by the total number of quality inspection log entries. For example, if the first two node graphs appear 300 times in the quality inspection log and the total number of quality inspection log entries is 1000, then the first detection frequency ratio = 300 ÷ 1000 = 0.3.
[0085] The search is repeated until the N-1th detection frequency ratio of the propagation graph of the first N nodes in the 200-level direct solderability enameled wire quality inspection log is reached.
[0086] The ratio of 2 to N is used as the first detection frequency weight, the ratio of 3 to N is used as the second detection frequency weight, and so on until the ratio of N-1 to N is used as the N-1th detection frequency weight. For example, if N=6, the first detection frequency weight = 2÷6 = 0.33, the second detection frequency weight = 3÷6 = 0.5, and the N-1th detection frequency weight = 5÷6 = 0.83.
[0087] Based on the weights of the first detection frequency share through the N-1th detection frequency share, calculate the average of the first detection frequency share through the N-1th detection frequency share. The first detection frequency average = Σ(first to Nth frequency share × first to Nth weight) / Σfirst to Nth weight. Set the calculated first detection frequency as the first confidence level and add it to the defect propagation map. Repeat the same calculation method to obtain the first to N-1th confidence levels.
[0088] This embodiment of the application uses frequency statistics from historical quality inspection logs to assign a confidence value to each propagation graph that is positively correlated with the trigger frequency, replacing subjective experience with objective data. A time-series weighting mechanism, such as increasing the weight of subsequent nodes, strengthens the rigor of evaluating long-chain graphs and avoids the phenomenon of falsely high confidence in short-chain paths. For example, a four-node "scratch → oxidation → bubble → microcrack" graph receives a higher weight due to its complete transmission chain, while a two-node "bubble → microcrack" subpath receives a lower weight, ensuring that the evaluation results truly reflect the global reliability of the defect chain.
[0089] S40: extracting, based on the plurality of confidence levels, the defect types of the union of the defect propagation maps with confidence levels greater than or equal to a confidence threshold, and sending the union of the defect types to a quality inspection end for performing directional quality inspection;
[0090] The traditional full-process inspection model rigidly allocates resources, failing to flexibly focus on dynamically changing high-risk defect chains. Ineffective screening of low-confidence patterns, such as low-frequency, isolated defects, results in wasted screening costs and manpower, while high-risk defects, due to their sparse distribution, may go undetected. Existing technologies lack a closed-loop mechanism to translate pattern analysis conclusions into inspection instructions, hindering the effective integration of prediction and execution to achieve effective early warning results.
[0091] In an embodiment of the present application, based on the calculated confidence level, the union defect types of the defect propagation maps with a confidence level greater than or equal to a confidence threshold are extracted. The confidence threshold is a pre-set value that reflects the accuracy of the confidence level. For example, if the confidence level is set to 0.85, the defect propagation maps with a confidence level greater than or equal to 0.85 are screened, and the union defect types of all defect propagation maps with a confidence level greater than or equal to 0.85 are taken and sent to the quality inspection end. For example, the Internet of Things is used to send the union defect types. After receiving the union defect types, the quality inspection end displays text and plays a ringtone to remind the user to conduct a targeted quality inspection.
[0092] In this embodiment, this step automatically filters high-reliability maps using confidence thresholds, extracting and summing defect types such as oxidation spots, paint bubbles, and uneven curing, and forwarding these to the quality inspection end. This mechanism achieves three breakthroughs: first, it dynamically targets high-frequency, high-risk defect clusters, precisely directing inspection resources to the highest risk points; second, it avoids duplicate inspections through the union operation, compressing redundant monitoring; and third, it establishes a closed "prediction-execution" loop, enabling cross-process defect analysis conclusions to directly drive quality inspection actions, ultimately achieving the core goal of accurate inspection with minimal inspection investment covering the highest risk areas.
[0093] Example 2, as Figure 2 As shown, based on the same inventive concept as the method for directional detection of production quality of a 200-level direct-weld enameled wire provided in Example 1, an embodiment of the present invention further provides a directional detection system for production quality of a 200-level direct-weld enameled wire, comprising:
[0094] The control deviation matrix module 100 is configured to perform a production simulation based on the kth production control deviation matrix of the kth process node and the enameled wire defect parameter of the k-1th node to obtain the enameled wire defect parameter of the kth node, where k is an integer, N ≥ k ≥ 1, N represents the total number of enameled wire process nodes, and the k-1th node is the upstream process node of the kth process node;
[0095] Among them, when k=1, the enameled wire defect parameter of the k-1th node is empty;
[0096] When k is equal to N, the defect parameters of the first node enameled wire are integrated to the Nth node enameled wire defect parameters to construct the defect triggering timing matrix
[0097] The defect propagation map module 200 is used to combine the defect propagation types of the defect trigger timing matrix to generate a plurality of defect propagation maps;
[0098] A confidence evaluation module 300 is configured to traverse the plurality of defect propagation maps to perform confidence evaluation and obtain a plurality of confidence levels, wherein the confidence level is positively correlated with the triggering frequency of the defect propagation map;
[0099] The defect detection summary module 400 is used to extract the union defect types of the defect propagation maps with confidence greater than or equal to the confidence threshold according to the plurality of confidence levels and send them to the quality inspection end for performing directional quality inspection.
[0100] In one embodiment, the control deviation matrix module 100 is further configured to:
[0101] The kth production control deviation record matrix set of the kth process node and the k-1th enameled wire defect record parameters are collected as input data, the kth enameled wire defect record parameters are collected as true value data, and the kth process node production simulator training data is constructed;
[0102] The kth process node includes a pay-off process node, an annealing process node, a painting process node, a baking process node, a cooling process node, and a take-up process node;
[0103] Based on the k-th process node production simulator training data, a plurality of k-th process node simulators are trained, wherein when k=1, the k-1-th node enameled wire defect recording parameter is empty;
[0104] The outputs of the plurality of k-th process node simulators are integrated to obtain a k-th process node production simulator, and a production simulation is performed based on a k-th production control deviation matrix of the k-th process node and a k-1-th node enameled wire defect parameter to obtain a k-th node enameled wire defect parameter.
[0105] When k<N, transmitting the k-th node enameled wire defect parameter to the k+1-th process node production simulator;
[0106] When k=N, the first-node enameled wire defect parameter is output until the N-th node enameled wire defect parameter is output.
[0107] In one embodiment, the defect propagation map module 200 is further configured to:
[0108] Extracting a first node defect type set from the defect trigger timing matrix, performing i-item combination enumeration to obtain a first node defect combination set, where 1≤i≤Q, i is an integer, and Q is the total number of first node defect types;
[0109] The first node defect type set is extracted from the defect trigger timing matrix, and i-item combination enumeration is performed to obtain the first node defect combination set, including:
[0110] Enumerate i items of the first node defect type set to obtain an i-item initial combination set of the first node;
[0111] Traversing the i-item initial combination set of the first node, retrieving the detection frequency percentage set of each i-item initial combination in the historical production samples of the first node;
[0112] According to the detection frequency ratio set, i initial combinations whose detection frequency ratio is greater than or equal to the detection frequency ratio threshold are extracted and set as the first node i combination set, where the initial value of i is 1;
[0113] If i=Q, add the first node one item combination set to the first node Q item combination set to the first node defect combination set;
[0114] If i<Q, i is incremented by one and the loop is executed.
[0115] Until the Nth node defect type set is extracted from the defect trigger timing matrix, j-item combination enumeration is performed to obtain the Nth node defect combination set, where 1≤j≤Y, j is an integer, and Y is the total number of Nth node defect types;
[0116] For the first node defect combination set up to the Nth node defect type set, randomly select a combination for each node, build a first defect propagation graph according to the node time sequence connection, and add it to the plurality of defect propagation graphs;
[0117] After the defect propagation maps are enumerated, the plurality of defect propagation maps are output.
[0118] In one embodiment, the confidence evaluation module 300 is further configured to:
[0119] extracting a first defect propagation map from the plurality of defect propagation maps;
[0120] Splitting the first defect propagation graph to obtain propagation graphs of the first two nodes and then the first N nodes;
[0121] Retrieve the first detection frequency ratio of the first two node propagation maps in the 200-level direct solderability enameled wire quality inspection log;
[0122] Until the N-1th detection frequency ratio of the first N node propagation graph in the 200-level direct solderability enameled wire quality inspection log is retrieved;
[0123] Calculating an average value of the first detection frequency ratio up to the N-1th detection frequency ratio, setting the average value as a first defect propagation map confidence level, and adding the confidence level to the plurality of defect propagation maps;
[0124] Calculating the average of the first detection frequency ratio up to the N-1th detection frequency ratio includes:
[0125] The ratio of 2 to N is the weight of the first detection frequency, and the ratio of N to N-1 is the weight of the N-1th detection frequency.
[0126] Based on the first detection frequency proportion weight up to the N-1th detection frequency proportion weight, an average of the first detection frequency proportion up to the N-1th detection frequency proportion is calculated.
[0127] In summary, the embodiments of the present application have at least the following technical effects:
[0128] This application proposes a method and system for targeted production quality inspection of Class 200 direct-weld enameled wire. By constructing a multi-node defect transmission time series model and a confidence-driven propagation map screening mechanism, it significantly improves the targeting of defect detection and the quality control capabilities of the entire process. Compared with traditional methods, the technical solution provided by this application significantly overcomes the blind spots in tracking the cross-process defect transmission chain. By dynamically simulating the cascading impact of upstream defects on downstream processes, it achieves early identification of defect initiation nodes. At the same time, based on a confidence evaluation system weighted by historical frequency, it accurately eliminates low-probability defect combinations, focusing detection resources on high-incidence defect clusters and significantly reducing the cost of ineffective screening.
[0129] This application achieves the technical effect of covering the maximum quality risk area with the minimum detection cost, tracking the cross-process defect transmission chain, and improving detection efficiency.
[0130] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0131] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.
[0132] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.
Claims
1. A method for directional detection of production quality of 200-grade direct-weld enameled wire, characterized in that: include: Based on the kth production control deviation matrix of the kth process node and the enameled wire defect parameter of the k-1th node, a production simulation is performed to obtain the enameled wire defect parameter of the kth node, where k is an integer, N ≥ k ≥ 1, N represents the total number of enameled wire process nodes, and the k-1th node is the upstream process node of the kth process node; Among them, when k=1, the enameled wire defect parameter of the k-1th node is empty; When k is equal to N, the defect parameters of the enameled wire at the first node are integrated to the defect parameters of the enameled wire at the Nth node to construct a defect triggering timing matrix; Combining defect propagation types on the defect trigger timing matrix to generate a plurality of defect propagation maps; Traversing the plurality of defect propagation maps to perform confidence evaluation and obtain a plurality of confidence levels, wherein the confidence levels are positively correlated with the triggering frequencies of the defect propagation maps; Extracting, based on the plurality of confidence levels, the defect types of the union of the defect propagation maps with confidence levels greater than or equal to a confidence threshold, and sending the results to a quality inspection end for performing directional quality inspection; The defect propagation type combination is performed on the defect trigger timing matrix to generate several defect propagation maps, including: Extracting a first node defect type set from the defect trigger timing matrix, performing i-item combination enumeration to obtain a first node defect combination set, where 1≤i≤Q, i is an integer, and Q is the total number of first node defect types; Until the Nth node defect type set is extracted from the defect trigger timing matrix, j-item combination enumeration is performed to obtain the Nth node defect combination set, where 1≤j≤Y, j is an integer, and Y is the total number of Nth node defect types; For the first node defect combination set up to the Nth node defect type set, randomly select a combination for each node, build a first defect propagation graph according to the node time sequence connection, and add it to the plurality of defect propagation graphs; After the defect propagation maps are enumerated, the plurality of defect propagation maps are output.
2. The method according to claim 1, wherein Based on the kth production control deviation matrix of the kth process node and the k-1th node enameled wire defect parameter, a production simulation is performed to obtain the kth node enameled wire defect parameter, including: The kth production control deviation record matrix set of the kth process node and the k-1th enameled wire defect record parameters are collected as input data, the kth enameled wire defect record parameters are collected as true value data, and the kth process node production simulator training data is constructed; Based on the k-th process node production simulator training data, a plurality of k-th process node simulators are trained, wherein when k=1, the k-1-th node enameled wire defect recording parameter is empty; The outputs of the plurality of k-th process node simulators are integrated to obtain a k-th process node production simulator, and a production simulation is performed based on a k-th production control deviation matrix of the k-th process node and a k-1-th node enameled wire defect parameter to obtain a k-th node enameled wire defect parameter. When k<N, transmitting the k-th node enameled wire defect parameter to the k+1-th process node production simulator; When k=N, the first-node enameled wire defect parameter is output until the N-th node enameled wire defect parameter is output.
3. The method according to claim 2, wherein The kth process node includes a pay-off process node, an annealing process node, a painting process node, a baking process node, a cooling process node, and a take-up process node.
4. The method according to claim 1, wherein Extracting a first node defect type set from the defect trigger timing matrix, performing i-item combination enumeration, and obtaining a first node defect combination set, including: Enumerate i items of the first node defect type set to obtain an i-item initial combination set of the first node; Traversing the i-item initial combination set of the first node, retrieving the detection frequency percentage set of each i-item initial combination in the historical production samples of the first node; According to the detection frequency ratio set, i initial combinations whose detection frequency ratio is greater than or equal to the detection frequency ratio threshold are extracted and set as the first node i combination set, where the initial value of i is 1; If i=Q, add the first node one item combination set to the first node Q item combination set to the first node defect combination set; If i<Q, i is incremented by one and the loop is executed.
5. The method according to claim 1, wherein Traversing the plurality of defect propagation maps to perform confidence evaluation, and obtaining a plurality of confidence levels, including: extracting a first defect propagation map from the plurality of defect propagation maps; Splitting the first defect propagation graph to obtain propagation graphs of the first two nodes and then the first N nodes; Retrieve the first detection frequency ratio of the first two node propagation maps in the 200-level direct solderability enameled wire quality inspection log; Until the N-1th detection frequency ratio of the first N node propagation graph in the 200-level direct solderability enameled wire quality inspection log is retrieved; An average value of the first detection frequency ratio up to the N-1th detection frequency ratio is calculated, set as a first defect propagation map confidence level, and added to the plurality of defect propagation maps.
6. The method according to claim 5, wherein Calculating the average of the first detection frequency proportion up to the N-1th detection frequency proportion includes: The ratio of 2 to N is the weight of the first detection frequency, and the ratio of N-1 to N is the weight of the N-1th detection frequency. Based on the first detection frequency proportion weight up to the N-1th detection frequency proportion weight, an average of the first detection frequency proportion up to the N-1th detection frequency proportion is calculated.
7. A production quality directional detection system for 200-level direct-weld enameled wire, characterized in that: A system for implementing the production quality directional detection method of the 200-level direct-weldable enameled wire according to any one of claims 1 to 6, comprising: A control deviation matrix module is used to perform production simulation based on the kth production control deviation matrix of the kth process node and the enameled wire defect parameter of the k-1th node to obtain the enameled wire defect parameter of the kth node, where k is an integer, N ≥ k ≥ 1, N represents the total number of enameled wire process nodes, and the k-1th node is the upstream process node of the kth process node; When k=1, the enameled wire defect parameter of the k-1th node is empty; When k is equal to N, the defect parameters of the enameled wire at the first node are integrated to the defect parameters of the enameled wire at the Nth node to construct a defect triggering timing matrix; A defect propagation map module is used to combine defect propagation types on the defect trigger timing matrix to generate a plurality of defect propagation maps; A confidence evaluation module is used to traverse the plurality of defect propagation maps to perform confidence evaluation and obtain a plurality of confidence levels, wherein the confidence level is positively correlated with the triggering frequency of the defect propagation map; The defect detection summary module is used to extract the union defect types of the defect propagation maps with confidence greater than or equal to the confidence threshold according to the multiple confidence levels and send them to the quality inspection end to perform directional quality inspection.
Citation Information
Patent Citations
Water pump shell image processing method and system for quality detection
CN120259307A