Product defect analysis method, device and storage medium based on data modeling

Abstract

The present application relates to the technical field of product defect analysis, and specifically to a product defect analysis method, device and storage medium based on data modeling. The method includes the following steps: obtaining a product manufacturing process database; the product manufacturing process database includes multiple data tables, each of which is an influencing factor affecting the probability of occurrence of product defects; constructing an optimal decision tree model; using the optimal decision tree model, respectively calculating the impact value of each of the data tables on the probability of occurrence of product defects; sorting the data tables according to the size of the impact value calculated by the optimal decision tree model to form a first sequence; wherein the impact value of each of the data tables indicates the degree of influence of the influencing factors represented by the data table on the probability of occurrence of product defects. The product defect analysis method, device and storage medium based on data modeling can solve the problems of low efficiency of the analysis process and limited analysis effect in the related technology.

Description

Technical Field

[0001] The present application relates to the technical field of product defect analysis, and in particular to a product defect analysis method, device and storage medium based on data modeling. Background Art

[0002] In the manufacturing of parts and components, it is inevitable that there will be a certain amount of defective parts. If the defect rate is high, the production cost of the product will increase significantly, and the after-sales quality will be difficult to guarantee, and even lead to customer complaints. Therefore, it is necessary to analyze the factors that affect the product defect rate and transform the corresponding process to reduce the product manufacturing defect rate, further reduce production costs and ensure product quality.

[0003] When analyzing the impact of production factors on the probability of product defects, related technologies not only need to conduct a large number of simulations and control experiments on different factors, wasting a lot of manpower and material resources, but also only conduct independent analysis and statistics on the impact of each production factor, and the analysis effect is limited. In addition, related technologies cannot quantify the impact of a certain factor on the result, which is not conducive to the adjustment of process parameters. Summary of the invention

[0004] The present application provides a product defect analysis method, device and storage medium based on data modeling, which can solve the problems of low efficiency of the analysis process and limited analysis effect in the related technology.

[0005] In order to solve the technical problem described in the background technology, the first aspect of the present application provides a product defect analysis method based on data modeling, and the product defect analysis method based on data modeling comprises the following steps:

[0006] Acquire a product manufacturing process database; the product manufacturing process database includes a plurality of data tables, each of which is an influencing factor affecting the probability of occurrence of product defects;

[0007] Construct the optimal decision tree model;

[0008] Using the optimal decision tree model, respectively calculating the impact value of each of the data tables on the probability of occurrence of product defects;

[0009] According to the magnitude of the influence value calculated by the optimal decision tree model, the data table is sorted to form a first sequence;

[0010] The influence value of each of the data tables indicates the degree of influence of the influencing factors represented by the data table on the probability of occurrence of product defects.

[0011] Optionally, after forming the first sequence, the following steps are further performed:

[0012] Obtaining constraint relationship information between the data tables;

[0013] Constructing a Bayesian network model based on the product manufacturing process database and the constraint relationship information;

[0014] Using the Bayesian network model, traversing the combinations of different data tables in the product manufacturing process database, and respectively calculating the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects;

[0015] According to the magnitude of the influence value calculated by the Bayesian network model, the data table is sorted to form a second sequence;

[0016] Based on the sorting position of each data table in the second sequence and whether the sorting position of the corresponding data table in the first sequence is consistent, it is determined whether the degree of influence of the influencing factors represented by the data table on the probability of product defect occurrence is credible.

[0017] Optionally, it is characterized in that after forming the second sequence, the following steps are further performed:

[0018] Using a probability statistics method, traversing the combinations of different data tables in the product manufacturing process database, and respectively calculating the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects;

[0019] According to the magnitude of the influence value calculated by the probability statistics method, the data table is sorted to form a third sequence;

[0020] Based on whether the sorting position of each data table in the third sequence corresponds to the sorting position of the data table in the first sequence, or,

[0021] Based on whether the sorting position of each data table in the third sequence is consistent with the sorting position of the corresponding data table in the second sequence, or,

[0022] Based on whether the sorting position of each data table in the third sequence, the sorting position of the corresponding data table in the second sequence, and the sorting position of the corresponding data table in the first sequence are consistent,

[0023] Determine whether the degree of influence of the influencing factors represented by the data table on the probability of product defects is credible.

[0024] Optionally, after forming the first sequence, the following steps are further performed:

[0025] Using a probability statistics method, traversing the combinations of different data tables in the product manufacturing process database, and respectively calculating the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects;

[0026] According to the magnitude of the influence value calculated by the probability statistics method, the data table is sorted to form a third sequence;

[0027] Based on whether the sorting position of each data table in the third sequence is consistent with the sorting position of the corresponding data table in the first sequence, it is determined whether the degree of influence of the influencing factors represented by the data table on the probability of product defect occurrence is credible.

[0028] Optionally, before the step of obtaining the product manufacturing process database, the following steps are further performed:

[0029] Obtain the original database of product manufacturing process;

[0030] The product manufacturing process original database is cleaned to obtain the product manufacturing process data.

[0031] Optionally, the step of performing data cleaning on the product manufacturing process original database to obtain the product manufacturing process data includes:

[0032] Remove abnormal data in the original database of the product manufacturing process;

[0033] Removing incomplete data from the original database of the product manufacturing process;

[0034] Encoding each type feature of the type data table in the manufacturing process database;

[0035] The continuous variables in the data table in the manufacturing process database are discretized.

[0036] Optionally, the step of constructing an optimal decision tree model includes:

[0037] Setting a training set and a validation set based on the product manufacturing process database;

[0038] Setting different maximum depths of decision trees, and establishing decision tree models respectively according to the maximum depths of the decision trees;

[0039] Using the training set and the validation set to train and validate each of the decision tree models respectively;

[0040] The maximum depth of the decision tree that makes the decision tree model perform the same on the training set and the validation set is determined as the optimal maximum depth of the decision tree, and the decision tree model with the optimal maximum depth of the decision tree is determined as the optimal decision tree model.

[0041] Optionally, the continuous variables in the data table in the manufacturing process database are discretized by the following formula 1, where the formula 1 is:

[0042] If X min is the minimum value of the continuous variable, X max is the maximum value of the continuous variable, N is the number of categories after discretization, and the interval of each discrete feature is:

[0043]

[0044] Then the continuous variable X can be transformed by the following formula:

[0045]

[0046] Optionally, the product manufacturing process database includes a type data table, and the type data table includes a plurality of different type features, and different type features have different effects on the probability of occurrence of product defects.

[0047] Optionally, when the data table is a type data table, the step of using the optimal decision tree model to calculate the impact value of each data table affecting the probability of occurrence of product defects includes:

[0048] The optimal decision tree model is used to calculate the impact value of each type feature in the type data table on the probability of product defect occurrence.

[0049] Optionally, the step of sorting the data table according to the influence values ​​calculated by the optimal decision tree model to form a first sequence includes:

[0050] According to the size of the influence value calculated by the optimal decision tree model, different type features in the data table and the type data table are sorted to form the first sequence.

[0051] Optionally, the optimal decision tree model calculates the impact value of each data table on the probability of occurrence of product defects based on the following formula 2, and the formula 2 is:

[0052]

[0053] N is the total number of samples, N t is the number of samples at the current node, impurity is the Gini coefficient at the current node, N tR is the number of samples belonging to the right tree after splitting, N tL is the number of samples belonging to the left tree after splitting, impurity right is the Gini coefficient of the right tree, impurity left is the Gini coefficient of the left tree.

[0054] Optionally, the Bayesian network model calculates the impact value of the combination of different data tables on the probability of occurrence of product defects based on the following formula three, and the formula three is:

[0055]

[0056] In the formula, Condition is the data table combination event, NG represents the event of product defects, and P BN (Condition) represents the impact value of the data table combination event on the probability of product defect occurrence, P(Condition|NG) represents the probability of product defect occurrence in the data table combination event inferred by the Bayesian network model, and P(Condition) represents the probability of occurrence of the data table combination event inferred by the Bayesian network model, P BN The larger the value, the more likely defects will occur under this condition. BN The smaller the value, the less likely defects will occur under this condition. BN When the value is 1, it indicates that the condition has no significant effect on the product results.

[0057] Optionally, the probability statistics method calculates the impact value of the combination of different data tables on the probability of occurrence of product defects based on the following formula 4, and the formula 4 is:

[0058]

[0059] Wherein Condition is a data table combination event, P(NG, Condition) represents the probability of a product defect occurring in the data table combination event calculated by the probability statistics method, P(NG) represents the probability of a product defect occurring in all events, and P(Condition) represents the probability of a data table combination event occurring in all events. P Stat The larger the value, the more likely defects will occur under this condition. Stat The smaller the value, the less likely defects will occur under this condition. Stat When the value is 1, it indicates that the condition has no significant effect on the product results.

[0060] In order to solve the technical problems described in the background technology, the second aspect of the present application provides a device for product defect analysis based on data modeling, and the product defect analysis device based on data modeling is used to execute the product defect analysis method based on data modeling described in the first aspect of the present application.

[0061] In order to solve the technical problems described in the background technology, the third aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes at least one program or instruction, and the program or instruction is used to execute the product defect analysis method based on data modeling described in the first aspect of the present application.

[0062] The technical solution of the present application includes at least the following advantages: based on the product manufacturing process database, the influence value of each influencing factor on the probability of product defects is analyzed through the optimal decision tree model, and the degree of influence is quantified, which is conducive to correcting the process parameters according to the quantified influence value, thereby improving the analysis efficiency and analysis effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0064] Figure 1 A flow chart of a product defect analysis method based on data modeling provided by an embodiment of the present application is shown;

[0065] Figure 1a Shown in Figure 1 Based on the flowchart shown, a flowchart of a product defect analysis method based on data modeling provided in other embodiments of the present application;

[0066] Figure 1b Shown in Figure 1a Based on the flowchart shown, a flowchart of a product defect analysis method based on data modeling is provided in another embodiment of the present application;

[0067] Figure 1c Shown in Figure 1a Based on the flowchart shown, a flowchart of a product defect analysis method based on data modeling is provided in yet another embodiment of the present application;

[0068] Figure 2 A schematic diagram of the HCD glue manufacturing process database is shown;

[0069] Figure 2a A schematic diagram of the raw material data table is shown;

[0070] Figure 2b A table showing the types of raw material components required in the HCD glue manufacturing process is shown;

[0071] Figure 3It shows that, taking the glue cracking analysis project as an example, Figure 1 Based on the product defect analysis method based on data modeling shown, a sorting diagram of each of the data tables in the first sequence in the embodiment of the adhesive cracking analysis project;

[0072] Figure 4 Taking the glue cracking analysis project as an example, Figure 1a or Figure 1b The schematic diagram of the constructed Bayesian network model is shown on the basis of the product defect analysis method based on data modeling. DETAILED DESCRIPTION

[0073] The following will be combined with the accompanying drawings to clearly and completely describe the technical solutions in this application. Obviously, the described embodiments are 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 creative work are within the scope of protection of this application.

[0074] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0075] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, it can also be the internal connection of two components, it can be a wireless connection, or it can be a wired connection. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0076] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0077] Figure 1 A flow chart of a product defect analysis method based on data modeling provided by an embodiment of the present application is shown. Figure 1It can be seen that the product defect analysis method based on data modeling includes the following steps S11 to S14:

[0078] Step S11: Acquire a product manufacturing process database. The product manufacturing process database includes a plurality of data tables, each of which is an influencing factor affecting the probability of occurrence of product defects.

[0079] In order to make the data in each data table of the product manufacturing process database meet the subsequent modeling and data analysis requirements, the present embodiment may further perform the following steps before performing S11: obtaining the product manufacturing process database:

[0080] Step S10: Acquire the original database of product manufacturing process.

[0081] Among them, the original database of the product manufacturing process may contain data that does not meet the requirements of subsequent modeling and data analysis, such as abnormal data and incomplete data.

[0082] Step S20: performing data cleaning on the product manufacturing process original database to obtain the product manufacturing process data.

[0083] Wherein, data cleaning of the product manufacturing process original database includes:

[0084] Remove abnormal data in the original database of the product manufacturing process.

[0085] Remove incomplete data in the original database of the product manufacturing process.

[0086] Each type feature of the type data table in the manufacturing process database is encoded to facilitate the establishment of a decision tree model in subsequent steps.

[0087] The continuous variables in the data table in the manufacturing process database are discretized to facilitate the establishment of Bayesian network modeling and multi-factor correlation analysis in subsequent steps.

[0088] Exemplarily, the continuous variables in the data table in the manufacturing process database may be discretized by the following formula 1.

[0089] Formula 1 is:

[0090] If X min is the minimum value of the continuous variable, X max is the maximum value of the continuous variable, N is the number of categories after discretization, and the interval of each discrete feature is:

[0091]

[0092] Then the continuous variable X can be transformed by the following formula:

[0093]

[0094] Step S12: construct an optimal decision tree model.

[0095] Exemplarily, in this embodiment, the optimal decision tree model can be constructed by the following steps:

[0096] Step S121: setting a training set and a validation set based on the product manufacturing process database.

[0097] Optionally, the data in the training set accounts for 85% of the total data, and the data in the test set accounts for 15% of the total data.

[0098] Step S122: setting different maximum depths of decision trees, and establishing decision tree models respectively according to the maximum depths of the decision trees.

[0099] Step S123: using the training set and the validation set to train and validate each decision tree model respectively.

[0100] Step S124: determining the maximum depth of the decision tree that makes the decision tree model perform the same on the training set and the validation set as the optimal maximum depth of the decision tree, and determining the decision tree model with the optimal maximum depth of the decision tree as the optimal decision tree model.

[0101] The trained optimal decision tree model can calculate the impact value of each data table on the probability of product defect occurrence based on the following formula 2.

[0102] Formula 2 is:

[0103]

[0104] N is the total number of samples, N t is the number of samples at the current node, impurity is the Gini coefficient at the current node, N tR is the number of samples belonging to the right tree after splitting, N tL is the number of samples belonging to the left tree after splitting, impurity right is the Gini coefficient of the right tree, impurity left is the Gini coefficient of the left tree.

[0105] Step S13: using the optimal decision tree model, respectively calculating the impact value of each data table on the probability of occurrence of product defects.

[0106] Among them, the impact value of each data table on the probability of product defect occurrence can be calculated according to the above formula 2.

[0107] Step S14: sorting the data table according to the influence values ​​calculated by the optimal decision tree model to form a first sequence.

[0108] The influence value of each of the data tables indicates the degree of influence of the influencing factors represented by the data tables on the probability of occurrence of product defects, that is, indicates the importance of the influence of each influencing factor on product quality.

[0109] This embodiment is based on the product manufacturing process database, and uses the optimal decision tree model to model and analyze the influence value of each influencing factor on the probability of product defects, quantify the degree of influence, and facilitate the correction of process parameters according to the quantified influence value, thereby improving the analysis efficiency and analysis effect.

[0110] Figure 1a Shown in Figure 1 Based on the flowchart shown in FIG. 1 , a flowchart of a product defect analysis method based on data modeling is provided in another embodiment of the present application. Figure 1a It can be seen that after step S14 is completed and the first sequence is formed, the following steps S15 to S19 are further performed, wherein:

[0111] Step S15: Obtain constraint relationship information between the data tables.

[0112] Among them, some data tables are not independent, but have constraint relationships with other data tables. The constraint relationship information is used to characterize the constraint relationship between data tables. That is, it is necessary to first determine that the execution conditions meet the constraint relationship information, and then perform corresponding operations on the data tables with mutual constraint relationships.

[0113] Step S16: constructing a Bayesian network model based on the product manufacturing process database and the constraint relationship information.

[0114] Exemplarily, the constraint relationship information may include a feature execution sequence of different process flows, and the feature execution sequence is used to constrain the execution sequence between different process flows.

[0115] Step S17: using the Bayesian network model, traversing the combinations of different data tables in the product manufacturing process database, and respectively calculating the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects.

[0116] The Bayesian network model is used to calculate the impact value of the combination of different data tables on the probability of product defect occurrence based on the following formula three.

[0117] Formula three is:

[0118]

[0119] In the formula, Condition is the data table combination event, NG represents the event of product defects, and P BN (Condition) represents the impact value of the data table combination event on the probability of product defects, P(Condition|NG) represents the probability of product defects occurring in the data table combination event inferred by the Bayesian network model, and P(Condition) represents the probability of occurrence of the data table combination event inferred by the Bayesian network model. P BN The larger the value, the more likely defects will occur under this condition. BN The smaller the value, the less likely defects will occur under this condition. BN When the value is 1, it indicates that the condition has no significant effect on the product results.

[0120] Step S18: sorting the data table according to the influence values ​​calculated by the Bayesian network model to form a second sequence.

[0121] Step S19: Based on whether the sorting position P2 of each data table in the second sequence is consistent with the sorting position P1 of the corresponding data table in the first sequence, determine whether the degree of influence of the influencing factors represented by the data table on the probability of product defect occurrence is credible.

[0122] When the sorting position P2 of a data table in the second sequence is consistent with the sorting position P1 of the data table in the first sequence, it is determined that the degree of influence (influence value) of the influencing factors represented by the data table on the probability of occurrence of product defects calculated by the above method is credible.

[0123] Figure 1b Shown in Figure 1a Based on the flowchart shown in FIG. 1 , a flowchart of a product defect analysis method based on data modeling is provided in another embodiment of the present application. Figure 1b It can be seen that after step S18 is completed and the second sequence is formed, the following steps S21 to S231 are further performed, wherein:

[0124] Step S21: using a probability statistics method, traverse the combinations of different data tables in the product manufacturing process database, and respectively calculate the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects.

[0125] Among them, the probability statistics method can calculate the impact value of the combination of different data tables on the probability of product defect occurrence based on the following formula 4.

[0126] Formula 4 is:

[0127]

[0128] Wherein Condition is a data table combination event, P(NG, Condition) represents the probability of a product defect occurring in the data table combination event calculated by the probability statistics method, P(NG) represents the probability of a product defect occurring in all events, and P(Condition) represents the probability of a data table combination event occurring in all events. P Stat The larger the value, the more likely defects will occur under this condition. Stat The smaller the value, the less likely defects will occur under this condition. Stat When the value is 1, it indicates that the condition has no significant effect on the product results.

[0129] Step S22: sorting the data table according to the magnitude of the influence value calculated by the probability statistics method to form a third sequence.

[0130] Step S231: based on whether the sorting position P3 of each data table in the third sequence corresponds to the sorting position P1 of the data table in the first sequence, or,

[0131] Based on whether the sorting position P3 of each data table in the third sequence is consistent with the sorting position P2 of the corresponding data table in the second sequence, or,

[0132] Based on whether the sorting position P3 of each data table in the third sequence, the sorting position P2 of the corresponding data table in the second sequence, and the sorting position P1 of the corresponding data table in the first sequence are consistent, determine whether the degree of influence of the influencing factors represented by the data table on the probability of product defect occurrence is credible.

[0133] Figure 1c Shown in Figure 1a Based on the flowchart shown in FIG. 1 , a flowchart of a product defect analysis method based on data modeling is provided in another embodiment of the present application. Figure 1c It can be seen that after step S14 is completed and the first sequence is formed, the following steps S21 to S232 are further performed, wherein:

[0134] Step S21: using a probability statistics method, traverse the combinations of different data tables in the product manufacturing process database, and respectively calculate the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects.

[0135] Step S22: sorting the data table according to the magnitude of the influence value calculated by the probability statistics method to form a third sequence.

[0136] Step S232: Determine whether the influence degree of the influencing factors represented by the data tables on the probability of product defect occurrence is credible based on whether the sorting position P3 of each data table in the third sequence corresponds to the sorting position P1 of the data table in the first sequence.

[0137] Figure 1c In the embodiment shown, step S21 and step S23 are Figure 1b The embodiments shown are identical.

[0138] This embodiment uses a decision tree model to measure the impact of a single factor on the result, and uses a Bayesian network and probability statistics method to determine the impact of multiple factors on the result, that is, two factors that have no significant impact when viewed alone may have a significant impact on the result when combined. Finally, a cross-validation is performed based on the conclusions of the three methods to determine whether the impact values ​​of the influencing factors determined above are credible, so as to improve the efficiency and effect of the analysis process.

[0139] for Figure 1 to Figure 1c In the product defect analysis method based on data modeling shown, the product manufacturing process database may also include a type data table, and the type data table includes multiple different type features. Different type features have different effects on the probability of product defect occurrence. Therefore, the type features in the type data table also serve as influencing factors that affect the probability of product defect occurrence.

[0140] When the data table is a type data table, in the step S13: using the optimal decision tree model to calculate the impact value of each data table on the probability of occurrence of product defects, when calculating the impact value of the type data table on the probability of occurrence of product defects, the calculation can be performed according to the following steps:

[0141] The optimal decision tree model is used to calculate the impact value of each type feature in the type data table on the probability of product defect occurrence.

[0142] The step S14: sorting the data table according to the influence values ​​calculated by the optimal decision tree model to form a first sequence comprises the following steps:

[0143] According to the size of the influence value calculated by the optimal decision tree model, the different type features in the data table and the type data table are sorted to form the first sequence, that is, the first sequence also includes the different type features in the type data table.

[0144] The step S17: using the Bayesian network model to traverse the combinations of different data tables in the product manufacturing process database and respectively calculating the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects, comprises the following steps:

[0145] The Bayesian network model is used to traverse the combination of different data tables in the product manufacturing process database and the combination of different types of features in the type data table, and the impact values ​​of the combination of different data tables and the combination of different types of features in the type data table on the probability of product defect occurrence are calculated respectively.

[0146] The step S18: sorting the data table according to the influence values ​​calculated by the Bayesian network model to form a second sequence may include the following steps:

[0147] According to the magnitude of the influence value calculated by the Bayesian network model, the data table and the different type features in the type data table are sorted to form a second sequence, that is, the second sequence also includes the different type features in the type data table.

[0148] The step S19: determining whether the influence degree of the influencing factors represented by the data tables on the probability of occurrence of product defects is credible based on whether the sorting position P2 of each data table in the second sequence is consistent with the sorting position P1 of the corresponding data table in the first sequence, may include the following steps:

[0149] Based on the different type features in each of the data tables and the type data table, whether the sorting position P2 in the second sequence is consistent with the sorting position P1 in the first sequence corresponding to the different type features in the data table and the type data table, it is determined whether the degree of influence of the influencing factors represented by the data table and the type features on the probability of product defects is credible.

[0150] The step S21: using a probability statistics method to traverse the combinations of different data tables in the product manufacturing process database and respectively calculate the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects, includes the following steps:

[0151] By using a probability statistics method, the combinations of different data tables in the product manufacturing process database and the combinations of different types of features in the type data table are traversed, and the impact values ​​of the combinations of different data tables on the probability of product defect occurrence are calculated respectively.

[0152] The step S22: sorting the data table according to the magnitude of the influence value calculated by the probability statistics method to form a third sequence comprises:

[0153] According to the magnitude of the influence value calculated by the probability statistics method, the data table and the different type features in the type data table are sorted to form a third sequence, that is, the third sequence also includes different type features in the type data table.

[0154] The step S23: determining whether the influence degree of the influencing factors represented by the data tables on the probability of occurrence of product defects is credible based on whether the sorting position P3 of each data table in the third sequence corresponds to the sorting position P1 of the data table in the first sequence. The step includes:

[0155] Based on the different type features in each of the data tables and the type data table, whether the sorting position P3 in the third sequence is consistent with the sorting position P1 in the first sequence corresponding to the different type features in the data table and the type data table, it is determined whether the degree of influence of the influencing factors represented by the data table and the type features on the probability of product defect occurrence is credible.

[0156] The process of this embodiment is described below by taking the adhesive cracking analysis project as an example. The main purpose of the adhesive cracking analysis project is to analyze the influence of various influencing factors in the process of making HCD adhesive on the probability of HCD adhesive cracking defects.

[0157] Taking the glue cracking analysis project as an example, Figure 1 Based on the product defect analysis method based on data modeling shown, step S11 in the embodiment of the glue cracking analysis project includes:

[0158] Obtaining a HCD adhesive manufacturing process database, wherein the HCD adhesive manufacturing process database includes a plurality of data tables, each of which is an influencing factor affecting the probability of occurrence of HCD adhesive cracking defects.

[0159] Figure 2 The schematic diagram of the HCD glue manufacturing process database is shown. Figure 2 It can be seen that the HCD glue manufacturing process database 200 includes a first process data table 210, a second process data table 220, a third process data table 230, a fourth process data table 240 and a raw material data table 250.

[0160] Taking the first process data table 210 as an example, the first process data table 210 stores process parameter information required during the first process of the HCD adhesive manufacturing process, such as process temperature information, process time information, etc.

[0161] The raw material data table 250 stores the raw material information required in the HCD glue manufacturing process.

[0162] Figure 2a The schematic diagram of the raw material data table is shown. Figure 2aIt can be seen that the raw materials required in the HCD glue manufacturing process include raw material A and raw material B. There are two different types of raw material A to choose from, namely A1 and A2, and there are two different types of raw material B to choose from, namely B1 and B2. That is, the raw materials required in the HCD glue manufacturing process are as follows: Figure 2b Any one of the four situations shown, namely, first, using A2 type raw material A in combination with B1 type raw material B; second, using A2 type raw material A in combination with B2 type raw material B; third, using A1 type raw material A in combination with B1 type raw material B; fourth, using A1 type raw material A in combination with B2 type raw material B.

[0163] It is understandable that, for the HCD adhesive produced by the HCD adhesive manufacturing process, the probability of cracking defects is closely related to the HCD adhesive manufacturing process, that is, each data table in the HCD adhesive manufacturing process database 200 can be used as an influencing factor affecting the probability of HCD adhesive cracking defects. Therefore, by finding the influencing factor with the greatest influence on the probability of HCD adhesive cracking defects, and by adjusting the execution parameters of the data table corresponding to the influencing factor, a solution to the HCD adhesive cracking defect problem is found.

[0164] Figure 3 It shows that, taking the glue cracking analysis project as an example, Figure 1 Based on the product defect analysis method based on data modeling shown, a sorting diagram of each of the data tables in the first sequence in the glue cracking analysis project embodiment.

[0165] from Figure 3 It can be seen that the influence values ​​of raw materials A and raw material B in the first process data table 210, the second process data table 220, the third process data table 230, the fourth process data table 240, and the raw material data table 250 on the probability of product defects are sorted in the first sequence.

[0166] Figure 4 Taking the glue cracking analysis project as an example, Figure 1a or Figure 1b The schematic diagram of the constructed Bayesian network model is shown on the basis of the product defect analysis method based on data modeling.

[0167] The information on the timing constraint relationship between the first process data table 210, the second process data table 220, the third process data table 230 and the fourth process data table 240 is obtained in advance, that is, the first process data table 210 is used to execute the first process, then the second process data table 220 is used to execute the second process, then the third process data table 230 is used to execute the third process, and finally the fourth process data table 240 is used to execute the fourth process. The raw material data table 250 is independent.

[0168] Table 1 shows the Figure 4 The Bayesian network model shown is combined with Formula 3 to calculate the impact value of the combination of different types of characteristics of raw materials A and raw materials B on the probability of product defects.

[0169] Table 1:

[0170] Raw material A Raw material B <![CDATA[Influence value (P BN )]]> Type A1 Type B1 2.123355263 Type A1 Type B2 1.791463415 Type A2 Type B1 0.92781771 Type A2 Type B2 0.591642259

[0171] From Table 1, it can be seen that when A2 type raw material A is combined with B2 type raw material B, P BN The value is the smallest, so HCD glue is least prone to cracking defects.

[0172] Table 2 shows the Figure 4 The Bayesian network model shown is combined with Formula 3 to calculate the impact value of the combination of different types of characteristics of raw materials A and raw materials B on the probability of product defects.

[0173] Table 2:

[0174] Raw material A Raw material B <![CDATA[Influence value (P Stat )]]> Type A1 Type B1 1.10166143 Type A1 Type B2 1.090248066 Type A2 Type B1 0.122286049 Type A2 Type B2 0

[0175] From Table 2, it can be seen that when A2 type raw material A is combined with B2 type raw material B, P Stat The value is the smallest, so HCD glue is least prone to cracking defects.

[0176] By combining Table 1 and Table 2 to verify each other, it is confirmed that when A2 type raw material A is combined with B2 type raw material B, HCD glue is least likely to crack.

[0177] The present application also provides a device for product defect analysis based on data modeling, which is used to perform the following steps: Figure 1 to Figure 1c The product defect analysis method based on data modeling is shown.

[0178] The present application also provides a computer-readable storage medium, which includes at least one program or instruction for executing the following Figure 1 to Figure 1c The product defect analysis method based on data modeling is shown.

[0179] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection created by this application.

Claims

1. A product defect analysis method based on data modeling, characterized in that: The product defect analysis method based on data modeling comprises the following steps: Acquire a product manufacturing process database; the product manufacturing process database includes a plurality of data tables, each of which is an influencing factor affecting the probability of occurrence of product defects; Construct the optimal decision tree model; Using the optimal decision tree model, respectively calculating the impact value of each of the data tables on the probability of occurrence of product defects; According to the magnitude of the influence value calculated by the optimal decision tree model, the data table is sorted to form a first sequence; Obtaining constraint relationship information between the data tables; Constructing a Bayesian network model based on the product manufacturing process database and the constraint relationship information; Using the Bayesian network model, traversing the combinations of different data tables in the product manufacturing process database, and respectively calculating the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects; According to the magnitude of the influence value calculated by the Bayesian network model, the data table is sorted to form a second sequence; Determining whether the degree of influence of the influencing factors represented by the data tables on the probability of occurrence of product defects is credible based on whether the sorting position of each of the data tables in the second sequence is consistent with the sorting position of the corresponding data tables in the first sequence; The influence value of each of the data tables indicates the degree of influence of the influencing factors represented by the data table on the probability of occurrence of product defects.

2. The product defect analysis method based on data modeling according to claim 1, characterized in that: After forming the second sequence, the following steps are further performed: Using a probability statistics method, traversing the combinations of different data tables in the product manufacturing process database, and respectively calculating the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects; According to the magnitude of the influence value calculated by the probability statistics method, the data table is sorted to form a third sequence; Based on whether the sorting position of each data table in the third sequence corresponds to the sorting position of the data table in the first sequence, or, Based on whether the sorting position of each data table in the third sequence is consistent with the sorting position of the corresponding data table in the second sequence, or, Based on whether the sorting position of each data table in the third sequence, the sorting position of the corresponding data table in the second sequence, and the sorting position of the corresponding data table in the first sequence are consistent, Determine whether the degree of influence of the influencing factors represented by the data table on the probability of product defects is credible.

3. The product defect analysis method based on data modeling according to claim 1, characterized in that: After forming the first sequence, the following steps are further performed: Using a probability statistics method, traversing the combinations of different data tables in the product manufacturing process database, and respectively calculating the impact values ​​of the combinations of different data tables on the probability of occurrence of product defects; According to the magnitude of the influence value calculated by the probability statistics method, the data table is sorted to form a third sequence; Based on whether the sorting position of each data table in the third sequence is consistent with the sorting position of the corresponding data table in the first sequence, it is determined whether the degree of influence of the influencing factors represented by the data table on the probability of product defect occurrence is credible.

4. The product defect analysis method based on data modeling according to claim 1, characterized in that: Before the step of obtaining the product manufacturing process database, the following steps are also performed: Obtain the original database of product manufacturing process; The product manufacturing process original database is cleaned to obtain the product manufacturing process data.

5. The product defect analysis method based on data modeling according to claim 4, characterized in that: The step of performing data cleaning on the product manufacturing process original database to obtain the product manufacturing process data comprises: Remove abnormal data in the original database of the product manufacturing process; Removing incomplete data from the original database of the product manufacturing process; Encoding each type feature of the type data table in the manufacturing process database; The continuous variables in the data table in the manufacturing process database are discretized.

6. The product defect analysis method based on data modeling according to claim 1, characterized in that: The step of constructing the optimal decision tree model comprises: Setting a training set and a validation set based on the product manufacturing process database; Setting different maximum depths of decision trees, and establishing decision tree models respectively according to the maximum depths of the decision trees; Use the training set and the validation set to train and validate each decision tree model respectively; The maximum depth of the decision tree that makes the decision tree model perform the same on the training set and the validation set is determined as the optimal maximum depth of the decision tree, and the decision tree model with the optimal maximum depth of the decision tree is determined as the optimal decision tree model.

7. The product defect analysis method based on data modeling according to claim 6, characterized in that: The continuous variables in the data table in the manufacturing process database are discretized by the following formula 1, which is: If X min is the minimum value of the continuous variable, X max is the maximum value of the continuous variable, N is the number of categories after discretization, and the interval of each discrete feature is: Then the continuous variable X can be transformed by the following formula:

8. The product defect analysis method based on data modeling according to claim 1, characterized in that: The product manufacturing process database includes a type data table, and the type data table includes a plurality of different type features. Different type features have different effects on the probability of occurrence of product defects.

9. The product defect analysis method based on data modeling according to claim 8, characterized in that: When the data table is a type data table, the step of using the optimal decision tree model to calculate the impact value of each data table affecting the probability of occurrence of product defects includes: The optimal decision tree model is used to calculate the impact value of each type feature in the type data table on the probability of product defect occurrence.

10. The product defect analysis method based on data modeling according to claim 8, characterized in that: The step of sorting the data table according to the influence values ​​calculated by the optimal decision tree model to form a first sequence includes: According to the size of the influence value calculated by the optimal decision tree model, different type features in the data table and the type data table are sorted to form the first sequence.

11. The product defect analysis method based on data modeling according to any one of claims 1 to 10, characterized in that: The optimal decision tree model calculates the impact value of each data table on the probability of product defect occurrence based on the following formula 2, and the formula 2 is: N is the total number of samples, N t is the number of samples at the current node, impurity is the Gini coefficient at the current node, N tR is the number of samples belonging to the right tree after splitting, N tL is the number of samples belonging to the left tree after splitting, impurity right is the Gini coefficient of the right tree, impurity left is the Gini coefficient of the left tree.

12. The product defect analysis method based on data modeling according to claim 1 or 2, characterized in that: The Bayesian network model is used to calculate the impact value of the combination of different data tables on the probability of product defects based on the following formula 3, where the formula 3 is: In the formula, Condition is the data table combination event, NG represents the event of product defects, and P BN (Condition) represents the impact value of the data table combination event on the probability of product defect occurrence, P(Condition|NG) represents the probability of product defect occurrence in the data table combination event inferred by the Bayesian network model, and P(Condition) represents the probability of occurrence of the data table combination event inferred by the Bayesian network model, P BN The larger the value, the more likely defects will occur under this condition. BN The smaller the value, the less likely defects will occur under this condition. BN When the value is 1, it indicates that the condition has no significant effect on the product results.

13. The product defect analysis method based on data modeling according to claim 2 or 3, characterized in that: The probability statistics method is based on the following formula 4 to calculate the impact value of the combination of different data tables on the probability of product defects, and the formula 4 is: Wherein Condition is a data table combination event, P(NG, Condition) represents the probability of a product defect occurring in the data table combination event calculated by the probability statistics method, P(NG) represents the probability of a product defect occurring in all events, and P(Condition) represents the probability of a data table combination event occurring in all events. P Stat The larger the value, the more likely defects will occur under this condition. Stat The smaller the value, the less likely defects will occur under this condition. Stat When the value is 1, it indicates that the condition has no significant effect on the product results.

14. A device for product defect analysis based on data modeling, characterized in that: The product defect analysis device based on data modeling is used to execute the product defect analysis method based on data modeling as described in any one of claims 1 to 13.

15. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes at least one program or instruction, and the program or instruction is used to execute the product defect analysis method based on data modeling as described in any one of claims 1 to 13.

Images