Product quality problem reason analysis method based on causal knowledge network
By building a causal knowledge network, matching product quality problems and causal knowledge nodes, finding the root cause and inferring the causal chain, the problem of insufficient utilization of causal knowledge in the existing technology is solved, and the effect of analysis of cause of product quality problems is improved.
Patent Information
- Application Number
- CN202411919719.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-04-18
AI Technical Summary
The prior art is difficult to make full use of product quality problems to solve the causal knowledge contained in the data for cause analysis, resulting in poor results in cause analysis.
Using a method based on causal knowledge network, by building a causal knowledge network, the actual quality problems of the product are matched with causal knowledge nodes, finding the root cause nodes, inferring the causal chain, and improving the causal knowledge network through expert verification.
It improves the effectiveness of analysis of causes of product quality problems, reduces the complexity of causal relationships and the limitations of personal experience, and enhances the accuracy and reliability of the analysis.
Smart Images
Figure CN119850012A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of causal analysis of product quality problems, and in particular to a method for analyzing the causes of product quality problems based on a causal knowledge network. Background Art
[0002] Different from the general production activities of enterprises, the solution of product quality problems is an activity with high degree of personnel participation and strong knowledge dependence, which mainly includes three links: clarifying the problem, causal analysis, and formulating solutions. Among them, causal analysis is an indispensable key link, and its effect directly affects the effectiveness of the solution and determines the benefits of solving product quality problems. However, causal analysis is very complicated, mainly manifested in: ① It is necessary to combine the specific situation of the problem so that specific problems can be analyzed specifically, which is convenient for formulating temporary containment measures; ② It is necessary to abstract and summarize the problem to make it general, so as to facilitate in-depth analysis of the causal relationship of such problems, find the root cause of the problem, and formulate long-term problem-solving solutions; ③ The causal relationship behind product quality problems is complicated and presents a network structure.
[0003] At present, in the process of solving product quality problems, product quality management personnel generally use fishbone diagrams to conduct horizontal cause-and-effect analysis from the aspects of people, machines, materials, methods, and environment, or use the five-question method (5-WHY) to conduct vertical cause-and-effect analysis. However, both the fishbone diagram and the five-question method are affected by the complexity and diversity of cause-and-effect relationships, as well as the limitations and bias of personal experience and cognitive differences, resulting in poor cause analysis results.
[0004] With the informatization of manufacturing enterprises, product quality management personnel can use information systems to record the process of solving product quality problems and form product quality problem solving data. This type of data contains experience and knowledge in solving product quality problems. This part of knowledge is not only the externalization of personal experience, but also the integration of many sources of knowledge, and the result of group knowledge externalization and sharing. If we can make full use of the causal knowledge contained in the product quality problem solving data to conduct cause analysis, we will definitely be able to improve the analysis effect. However, there is currently no method that can fully utilize the causal knowledge contained in the product quality problem solving data to conduct cause analysis. Summary of the invention
[0005] In order to solve the problem that there is currently no method that can fully utilize the causal knowledge contained in product quality problem solving data to perform cause analysis, the present invention provides a product quality problem cause analysis method based on a causal knowledge network.
[0006] The present invention is achieved by adopting the following technical solutions:
[0007] A method for analyzing the causes of product quality problems based on a causal knowledge network is implemented by the following steps:
[0008] Step 1: Construct a causal knowledge network;
[0009] Step 2: Match the actual product quality issues with the causal knowledge nodes;
[0010] Step 3: Find the root cause of the actual product quality problem;
[0011] Step 4: Infer a causal chain that leads to the actual quality problem of the product;
[0012] Step 5: Determine whether the inferred causal chain is correct;
[0013] Step 6: Improve the causal knowledge network based on the correct cause analysis results.
[0014] Furthermore, the step 1 specifically includes the following steps:
[0015] Step 1.1: Construction of causal knowledge nodes; specifically includes the following steps:
[0016] Step 1.1.1: Collect existing product quality problem solving data;
[0017] Step 1.1.2: Read all sentences recording product quality problems and their causes from the product quality problem resolution data;
[0018] Step 1.1.3: For each sentence, read all the words that describe the problem or cause.
[0019] Step 1.1.4: Standardize and unify all different descriptive terms for the same problem or cause. The standardization and unification methods include but are not limited to: semantic clustering of terms / sentences, manual screening, and appropriate classification algorithms;
[0020] Step 1.1.5: Use the standardized and unified vocabulary to describe the problem or cause as the causal knowledge node, thereby completing the construction of the causal knowledge node;
[0021] Step 1.2: Determine the causal relationship between knowledge nodes from the product quality problem solving data; the causal relationship includes but is not limited to: ① the causal relationship contained in the causal sentence; ② the causal relationship between data structures; ③ the mixture of the above two causal relationships; ④ other forms of causal relationships; the causal relationship determination methods for different sources are different, as follows:
[0022] The causal relationship contained in the causal sentence is determined by the causal grammar of the language recorded by the product quality problem solving data;
[0023] The causal relationship between data structures is determined by the special structural relationship of product quality problem solving data, such as: the node in the problem description field is the effect, and the node in the cause analysis field is the cause;
[0024] The confounding of the above two causal relationships can be determined by merging the same nodes in the above two causal relationships. If there are no common nodes, it can be determined with the help of employee experience;
[0025] Other forms of causation need to be determined based on the specific circumstances;
[0026] The above-mentioned causal relationship determination method includes but is not limited to: automatic completion by designing an algorithm and computer programming, manual completion;
[0027] Step 1.3: Construction of causal knowledge network; specifically includes the following steps:
[0028] Step 1.3.1: After determining the causal relationship between knowledge nodes, use the node-edge-node form to represent the causal knowledge, thereby forming a large number of causal knowledge groups;
[0029] Step 1.3.2: Merge the same nodes in a large number of causal knowledge groups to form a causal knowledge network;
[0030] Step 1.3.3: Count the fusion frequency of causal knowledge nodes and edges, and record and calculate the parameters of the causal knowledge network; the parameters of the causal knowledge network include but are not limited to the following parameters:
[0031] (1) Node frequency: This parameter indicates the frequency of a node appearing repeatedly in ML-CKN. i , and its frequency is recorded as f(n i ); f(n i ) value is larger, indicating that the node n i The more frequently the problem represented occurs, the more attention should be paid;
[0032] (2) Edge frequency: This parameter indicates the frequency of a certain edge appearing repeatedly in ML-CKN; for edge e ij , and its frequency is recorded as f(e ij ); f(e ij ) value is larger, indicating that the edge e ij The more frequently the causal relationship represented occurs, the more attention should be paid;
[0033] (3) Probability of finding the cause: This parameter represents the probability of finding the cause for a problem n p Determine the possibility of the problem being caused by various reasons when it occurs, that is, point to the result node n p The probability of each cause node occurring is shown in the following formula:
[0034]
[0035] Where: Indicates the problem p It is due to the reason i The probability of causing; f(e ip ) represents edge e ip The frequency of
[0036] (4) Cause-to-effect probability: This parameter refers to the probability of a certain problem n p When it is determined, the possibility that the problem may further lead to other problems, that is, node n p The probability of causing other result nodes to occur is shown in the following formula:
[0037]
[0038] Where: Indicates the problem p As a cause to further raise questions i The probability of f(e pi ) represents edge e pi The frequency of
[0039] (5) Out-degree of a node: This parameter refers to the sum of the frequencies of the edges with the node as the starting point in the causal knowledge network, as shown in the following formula:
[0040]
[0041] Where: D out (n p ) represents node n p The out-degree of the nodes in the causal knowledge network; f(e pi ) represents node n p The frequency of any side in the causal knowledge network as the starting point;
[0042] (6) In-degree of a node: This parameter refers to the sum of the frequencies of the edges that point to the node in the causal knowledge network, as shown in the following formula:
[0043]
[0044] Where: D in (n p ) represents node n p The in-degree of the nodes in the causal knowledge network; f(e ip ) represents node n p The frequency of any edge in the causal knowledge network as a pointing point;
[0045] (7) Number of steps for the reason:
[0046] Assume n i、n j are two nodes in the causal knowledge network, n u Yes j One reason is that n i Yes j Therefore, these two nodes are at least in the same causal chain. If node n is in a causal chain i With node n j There are m-1 nodes between them, that is, node n on this causal chain i With node n j If the step length between them is m, then we can say that n i Yes j The m-layer reasons;
[0047] In this way, a causal knowledge network G = (N, E, R) is constructed, where G represents the causal knowledge network; N represents the set of nodes in the causal knowledge network; E represents the set of edges in the causal knowledge network; and R represents the set of parameters in the causal knowledge network.
[0048] Furthermore, the step 2 specifically includes the following steps:
[0049] Step 2.1: The product quality management personnel shall record the actual product quality problems encountered on site; the recording methods include but are not limited to: text description records, photo records, and video records;
[0050] Step 2.2: The actual product quality issues recorded in the photographs and videos can be converted into text descriptions by technical means or manual recognition; the actual product quality issues recorded in the text descriptions do not need to be processed;
[0051] Step 2.3: Filter all problem description words from the text description;
[0052] Step 2.4: Compare all problem description words with all causal knowledge nodes in the causal knowledge network and calculate the semantic similarity;
[0053] Step 2.5: Set a semantic similarity threshold and filter out all causal knowledge nodes with semantic similarity greater than the threshold, so as to complete the matching of actual product quality problems with causal knowledge nodes.
[0054] Furthermore, the specific method of step 3 is as follows:
[0055] Step 3.1: Select any one of the causal knowledge nodes selected in step 2.5 and set it as The processing method of the remaining nodes is the same as same;
[0056] Step 3.2: Based on the causal knowledge network constructed in step 1, calculate the connectivity between nodes through the reachability matrix method to find the problem node All the root cause nodes of the problem are as follows: ① Find the problem node The node set corresponding to the non-zero elements in the row and column of the reachable matrix M This set is the problem node ② Find the set of nodes with in-degree 0 among all the cause nodes This set is the problem node All root cause nodes of
[0057] Through the above steps, the root cause of the actual product quality problems can be found.
[0058] Furthermore, the step 4 specifically includes the following steps:
[0059] Step 4.1: Calculate the cause probability as follows:
[0060] set up It is the problem node A root cause node, It is the problem node Root cause node A causal chain between is the set of nodes of the causal chain; is the edge set of the causal chain; is the parameter set of the causal chain, including the probability of each node to seek the cause and effect calculated according to the causal knowledge network Therefore, the causal chain Problem Node Root Cause Node The probability of The calculation method is as follows:
[0061]
[0062] set up It is the problem node Root cause node Since the nodes and parameters of each causal chain are different, the problem nodes on each causal chain are Root Cause Node The resulting probabilities are also different; therefore, the problem nodes on each causal chain are calculated Root Cause Node The probability caused by the accumulation can get the problem node Root Cause Node The probability of As shown below:
[0063]
[0064] Step 4.2: Calculate the node importance. The specific method is as follows:
[0065] set up Root cause node Located at the problem node Root cause node The number of cause steps in each causal chain between them; then for the problem node For example, the root cause node The node importance The calculation method is as follows:
[0066]
[0067] Where: g(x) is a reconciliation formula formulated according to the actual situation, its purpose is to prevent the sum of the cause steps from being too large and affecting the accuracy of the node importance;
[0068] Step 4.3: Infer a causal chain that leads to the actual quality problem of the product. The specific process is as follows: ① Select the root cause node with the greatest node importance; ② Select the nodes with the largest priority value layer by layer from bottom to top until a complete causal chain is inferred;
[0069] The priority value is calculated as follows:
[0070] Set the problem node Root cause node A problem node that may lead to a cause-effect probability is The problem node The priority value The calculation method is as follows:
[0071]
[0072] Where: Represents the problem node The node importance; h(x) is a reconciliation formula formulated according to the actual situation, its purpose is to prevent the node importance from being too large and affecting the accuracy of the priority value.
[0073] Furthermore, the step 5 specifically includes the following steps:
[0074] The causal chain inferred in step 4 is handed over to the product quality problem solving expert for analysis and judgment based on the actual situation; if the causal chain is the correct causal chain that leads to the actual quality problem of the product, the reasoning is completed; otherwise, the wrong node is found and corrected. At this time, two situations may occur: ① If the corrected node is on the causal knowledge network, the reasoning process of step 4 is repeated based on the corrected node to generate a new causal chain, which is again handed over to the product quality problem solving expert for analysis and judgment; ② If the corrected node is not on the causal knowledge network, the product quality problem solving expert will conduct subsequent analysis, find the new causal node and judge whether it is on the causal knowledge network, and then repeat the above steps until the correct causal chain that leads to the actual quality problem of the product is found.
[0075] Furthermore, the step 6 specifically includes the following steps:
[0076] Assume that the causal chain G γ =(N γ ,E γ ) is the correct cause and effect chain leading to the actual quality problem of the product; is the causal chain G γ The nodes of In the causal knowledge network, there are the same nodes, nodes There are no identical nodes in the causal knowledge network; is the causal chain G γ The edge of In the causal knowledge network, there are the same edges, edges There are no identical edges in the causal knowledge network;
[0077] set up Nodes in the causal knowledge network If the nodes are the same Frequency Updated to in For Node The previous frequency; set the edge and the edges in the causal knowledge network Same, then the side Frequency Updated to in For edge Previous frequency; traversal nodes With edge Complete the same edge and node updates as the causal knowledge network;
[0078] If it is a node that is not in the causal knowledge network, then add the node to the causal knowledge network and set the frequency of the node If it is an edge that is not in the causal knowledge network, then add the edge to the causal knowledge network and set the frequency of the edge All nodes that do not exist in the causal knowledge network With edge Add to the causal knowledge network and update the parameters of the causal knowledge network, including but not limited to: the frequency of nodes, the frequency of edges, the probability of seeking causes from results, the probability of predicting results from causes, the out-degree of nodes, the in-degree of nodes, and the number of cause steps, so as to improve the causal knowledge network.
[0079] The method for analyzing the causes of product quality problems based on a causal knowledge network described in the present invention makes full use of the causal knowledge contained in the product quality problem solving data. This knowledge is a collection of the wisdom of the masses and the externalization of group knowledge. Therefore, it can greatly reduce or even eliminate the complexity and diversity of causal relationships, as well as the limitations and bias of personal experience, cognitive differences and other factors that affect the causal analysis of quality problems, thereby improving the effect of quality problem cause analysis and contributing to improving the efficiency of product quality problem solving.
[0080] The present invention effectively solves the problem that there is currently no method that can fully utilize the causal knowledge contained in product quality problem solving data to perform cause analysis. It is not only suitable for the causal analysis of product quality problems, but also provides a reference for the causal analysis of other similar problems. BRIEF DESCRIPTION OF THE DRAWINGS
[0081] Figure 1 It is a schematic diagram of the process of the present invention.
[0082] Figure 2 It is a schematic flow diagram of step 5 in the present invention. DETAILED DESCRIPTION
[0083] A method for analyzing the causes of product quality problems based on a causal knowledge network is implemented by the following steps:
[0084] Step 1: Construct a causal knowledge network;
[0085] Step 2: Match the actual product quality issues with the causal knowledge nodes;
[0086] Step 3: Find the root cause of the actual product quality problem;
[0087] Step 4: Infer a causal chain that leads to the actual quality problem of the product;
[0088] Step 5: Determine whether the inferred causal chain is correct;
[0089] Step 6: Improve the causal knowledge network based on the correct cause analysis results.
[0090] The step 1 specifically includes the following steps:
[0091] Step 1.1: Construction of causal knowledge nodes; specifically includes the following steps:
[0092] Step 1.1.1: Collect existing product quality problem solving data;
[0093] Step 1.1.2: Read all sentences recording product quality problems and their causes from the product quality problem resolution data;
[0094] Step 1.1.3: For each sentence, read all the words that describe the problem or cause.
[0095] Step 1.1.4: Standardize and unify all different descriptive terms for the same problem or cause. The standardization and unification methods include but are not limited to: semantic clustering of terms / sentences, manual screening, and appropriate classification algorithms;
[0096] Step 1.1.5: Use the standardized and unified vocabulary to describe the problem or cause as the causal knowledge node, thereby completing the construction of the causal knowledge node;
[0097] Step 1.2: Determine the causal relationship between knowledge nodes from the product quality problem solving data; the causal relationship includes but is not limited to: ① the causal relationship contained in the causal sentence; ② the causal relationship between data structures; ③ the mixture of the above two causal relationships; ④ other forms of causal relationships; the causal relationship determination methods for different sources are different, as follows:
[0098] The causal relationship contained in the causal sentence is determined by the causal grammar of the language recorded by the product quality problem solving data;
[0099] The causal relationship between data structures is determined by the special structural relationship of product quality problem solving data, such as: the node in the problem description field is the effect, and the node in the cause analysis field is the cause;
[0100] The confounding of the above two causal relationships can be determined by merging the same nodes in the above two causal relationships. If there are no common nodes, it can be determined with the help of employee experience;
[0101] Other forms of causation need to be determined based on the specific circumstances;
[0102] The above-mentioned causal relationship determination method includes but is not limited to: automatic completion by designing an algorithm and computer programming, manual completion;
[0103] Step 1.3: Construction of causal knowledge network; specifically includes the following steps:
[0104] Step 1.3.1: After determining the causal relationship between knowledge nodes, use the node-edge-node form to represent the causal knowledge, thereby forming a large number of causal knowledge groups;
[0105] Step 1.3.2: Merge the same nodes in a large number of causal knowledge groups to form a causal knowledge network;
[0106] Step 1.3.3: Count the fusion frequency of causal knowledge nodes and edges, and record and calculate the parameters of the causal knowledge network; the parameters of the causal knowledge network include but are not limited to the following parameters:
[0107] (1) Node frequency: This parameter indicates the frequency of a node appearing repeatedly in ML-CKN. i , and its frequency is recorded as f(n i ); f(n i ) value is larger, indicating that the node n i The more frequently the problem represented occurs, the more attention should be paid;
[0108] (2) Edge frequency: This parameter indicates the frequency of a certain edge appearing repeatedly in ML-CKN; for edge e ij , and its frequency is recorded as f(e ij ); f(e ij ) value is larger, indicating that the edge e ij The more frequently the causal relationship represented occurs, the more attention should be paid;
[0109] (3) Probability of finding the cause: This parameter represents the probability of finding the cause for a problem n p Determine the possibility of the problem being caused by various reasons when it occurs, that is, point to the result node n p The probability of each cause node occurring is shown in the following formula:
[0110]
[0111] Where: Indicates the problem p It is due to the reason i The probability of causing; f(e ip ) represents edge e ip The frequency of
[0112] (4) Cause-to-effect probability: This parameter refers to the probability of a certain problem n p When it is determined, the possibility that the problem may further lead to other problems, that is, node n p The probability of causing other result nodes to occur is shown in the following formula:
[0113]
[0114] Where: Indicates the problem p As a cause to further raise questions i The probability of f(e pi ) represents edge e pi The frequency of
[0115] (5) Out-degree of a node: This parameter refers to the sum of the frequencies of the edges with the node as the starting point in the causal knowledge network, as shown in the following formula:
[0116]
[0117] Where: D out (n p ) represents node n p The out-degree of the nodes in the causal knowledge network; f(e pi ) represents node n p The frequency of any side in the causal knowledge network as the starting point;
[0118] (6) In-degree of a node: This parameter refers to the sum of the frequencies of the edges that point to the node in the causal knowledge network, as shown in the following formula:
[0119]
[0120] Where: D in (n p ) represents node n p The in-degree of the nodes in the causal knowledge network; f(e ip ) represents node n p The frequency of any edge in the causal knowledge network as a pointing point;
[0121] (7) Number of steps for the reason:
[0122] Assume n i 、n j are two nodes in the causal knowledge network, n u Yes j One reason is that n i Yes j Therefore, these two nodes are at least in the same causal chain. If node n is in a causal chain i With node n j There are m-1 nodes between them, that is, node n on this causal chain i With node n j If the step length between them is m, then we can say that n i Yes j The m-layer reasons;
[0123] In this way, a causal knowledge network G = (N, E, R) is constructed, where G represents the causal knowledge network; N represents the set of nodes in the causal knowledge network; E represents the set of edges in the causal knowledge network; and R represents the set of parameters in the causal knowledge network.
[0124] The step 2 specifically includes the following steps:
[0125] Step 2.1: The product quality management personnel shall record the actual product quality problems encountered on site; the recording methods include but are not limited to: text description records, photo records, and video records;
[0126] Step 2.2: The actual product quality issues recorded in the photographs and videos can be converted into text descriptions by technical means or manual recognition; the actual product quality issues recorded in the text descriptions do not need to be processed;
[0127] Step 2.3: Filter all problem description words from the text description;
[0128] Step 2.4: Compare all problem description words with all causal knowledge nodes in the causal knowledge network and calculate the semantic similarity;
[0129] Step 2.5: Set a semantic similarity threshold and filter out all causal knowledge nodes with semantic similarity greater than the threshold, so as to complete the matching of actual product quality problems with causal knowledge nodes.
[0130] The specific method of step 3 is as follows:
[0131] Step 3.1: Select any one of the causal knowledge nodes selected in step 2.5 and set it as The processing method of the remaining nodes is the same as same;
[0132] Step 3.2: Based on the causal knowledge network constructed in step 1, calculate the connectivity between nodes through the reachability matrix method to find the problem node All the root cause nodes of the problem are as follows: ① Find the problem node The node set corresponding to the non-zero elements in the row and column of the reachable matrix M This set is the problem node ② Find the set of nodes with in-degree 0 among all the cause nodes This set is the problem node All root cause nodes of
[0133] Through the above steps, the root cause of the actual product quality problems can be found.
[0134] The step 4 specifically includes the following steps:
[0135] Step 4.1: Calculate the cause probability as follows:
[0136] set up It is the problem node A root cause node, It is the problem node Root cause node A causal chain between is the set of nodes of the causal chain; is the edge set of the causal chain; is the parameter set of the causal chain, including the probability of each node to seek the cause and effect calculated according to the causal knowledge network Therefore, the causal chain Problem Node Root Cause Node The probability of The calculation method is as follows:
[0137]
[0138] set up It is the problem node Root cause node Since the nodes and parameters of each causal chain are different, the problem nodes on each causal chain are Root Cause Node The resulting probabilities are also different; therefore, the problem nodes on each causal chain are calculated Root Cause Node The probability caused by the accumulation can get the problem node Root Cause Node The probability of As shown below:
[0139]
[0140] Step 4.2: Calculate the node importance. The specific method is as follows:
[0141] set up Root cause node Located at the problem node Root cause node The number of cause steps in each causal chain between them; then for the problem node For example, the root cause node The node importance The calculation method is as follows:
[0142]
[0143] Where: g(x) is a reconciliation formula formulated according to the actual situation, its purpose is to prevent the sum of the cause steps from being too large and affecting the accuracy of the node importance;
[0144] Step 4.3: Infer a causal chain that leads to the actual quality problem of the product. The specific process is as follows: ① Select the root cause node with the greatest node importance; ② Select the nodes with the largest priority value layer by layer from bottom to top until a complete causal chain is inferred;
[0145] The priority value is calculated as follows:
[0146] Set the problem node Root cause node A problem node that may lead to a cause-effect probability is The problem node The priority value The calculation method is as follows:
[0147]
[0148] Where: Represents the problem node The node importance; h(x) is a reconciliation formula formulated according to the actual situation, its purpose is to prevent the node importance from being too large and affecting the accuracy of the priority value.
[0149] The step 5 specifically includes the following steps:
[0150] The causal chain inferred in step 4 is handed over to the product quality problem solving expert for analysis and judgment based on the actual situation; if the causal chain is the correct causal chain that leads to the actual quality problem of the product, the reasoning is completed; otherwise, the wrong node is found and corrected. At this time, two situations may occur: ① If the corrected node is on the causal knowledge network, the reasoning process of step 4 is repeated based on the corrected node to generate a new causal chain, which is again handed over to the product quality problem solving expert for analysis and judgment; ② If the corrected node is not on the causal knowledge network, the product quality problem solving expert will conduct subsequent analysis, find the new causal node and judge whether it is on the causal knowledge network, and then repeat the above steps until the correct causal chain that leads to the actual quality problem of the product is found.
[0151] The step 6 specifically includes the following steps:
[0152] Assume that the causal chain G γ =(N γ ,E γ ) is the correct cause and effect chain leading to the actual quality problem of the product; is the causal chain G γ The nodes of In the causal knowledge network, there are the same nodes, nodes There are no identical nodes in the causal knowledge network; is the causal chain G γ The edge of In the causal knowledge network, there are the same edges, edges There are no identical edges in the causal knowledge network;
[0153] set up Nodes in the causal knowledge network If the nodes are the same Frequency Updated to in For Node The previous frequency; set the edge and the edges in the causal knowledge network Same, then the side Frequency Updated to in For edge Previous frequency; traversal nodes With edge Complete the same edge and node updates as the causal knowledge network;
[0154] If it is a node that is not in the causal knowledge network, then add the node to the causal knowledge network and set the frequency of the node If it is an edge that is not in the causal knowledge network, then add the edge to the causal knowledge network and set the frequency of the edge All nodes that do not exist in the causal knowledge network With edge Add to the causal knowledge network and update the parameters of the causal knowledge network, including but not limited to: the frequency of nodes, the frequency of edges, the probability of seeking causes from results, the probability of predicting results from causes, the out-degree of nodes, the in-degree of nodes, and the number of cause steps, so as to improve the causal knowledge network.
[0155] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that these are only examples, and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.
Claims
1. A method for analyzing the causes of product quality problems based on a causal knowledge network, characterized by: This method is implemented by the following steps: Step 1: Construct a causal knowledge network; Step 2: Match the actual product quality issues with the causal knowledge nodes; Step 3: Find the root cause of the actual product quality problem; Step 4: Infer a causal chain that leads to the actual quality problem of the product; Step 5: Determine whether the inferred causal chain is correct; Step 6: Improve the causal knowledge network based on the correct cause analysis results.
2. The method for analyzing the causes of product quality problems based on a causal knowledge network according to claim 1 is characterized by: The step 1 specifically includes the following steps: Step 1.1: Construction of causal knowledge nodes; specifically includes the following steps: Step 1.1.1: Collect existing product quality problem solving data; Step 1.1.2: Read all sentences recording product quality problems and their causes from the product quality problem resolution data; Step 1.1.3: For each sentence, read all the words that describe the problem or cause. Step 1.1.4: Standardize and unify all different descriptive terms for the same problem or cause. The standardization and unification methods include but are not limited to: semantic clustering of terms / sentences, manual screening, and appropriate classification algorithms; Step 1.1.5: Use the standardized and unified vocabulary to describe the problem or cause as the causal knowledge node, thereby completing the construction of the causal knowledge node; Step 1.2: Determine the causal relationship between knowledge nodes from the product quality problem solving data; the causal relationship includes but is not limited to: ① the causal relationship contained in the causal sentence; ② the causal relationship between data structures; ③ the mixture of the above two causal relationships; ④ other forms of causal relationships; the causal relationship determination methods for different sources are different, as follows: The causal relationship contained in the causal sentence is determined by the causal grammar of the language recorded by the product quality problem solving data; The causal relationship between data structures is determined by the special structural relationship of product quality problem solving data, such as: the node in the problem description field is the effect, and the node in the cause analysis field is the cause; The confounding of the above two causal relationships can be determined by merging the same nodes in the above two causal relationships. If there are no common nodes, it can be determined with the help of employee experience; Other forms of causation need to be determined based on the specific circumstances; The above-mentioned causal relationship determination method includes but is not limited to: automatic completion by designing an algorithm and computer programming, manual completion; Step 1.3: Construction of causal knowledge network; specifically includes the following steps: Step 1.3.1: After determining the causal relationship between knowledge nodes, use the node-edge-node form to represent the causal knowledge, thereby forming a large number of causal knowledge groups; Step 1.3.2: Merge the same nodes in a large number of causal knowledge groups to form a causal knowledge network; Step 1.3.3: Count the fusion frequency of causal knowledge nodes and edges, and record and calculate the parameters of the causal knowledge network; the parameters of the causal knowledge network include but are not limited to the following parameters: (1) Node frequency: This parameter indicates the frequency of a node appearing repeatedly in ML-CKN. i , and its frequency is recorded as f(n i ); f(n i ) value is larger, indicating that the node n i The more frequently the problem represented occurs, the more attention should be paid; (2) Edge frequency: This parameter indicates the frequency of a certain edge appearing repeatedly in ML-CKN; for edge e ij , and its frequency is recorded as f(e ij ); f(e ij ) value is larger, indicating that the edge e ij The more frequently the causal relationship represented occurs, the more attention should be paid; (3) Probability of finding the cause: This parameter represents the probability of finding the cause for a problem n p Determine the possibility of the problem being caused by various reasons when it occurs, that is, point to the result node n p The probability of each cause node occurring is shown in the following formula: Where: Indicates the problem p It is due to the reason i The probability of causing; f(e ip ) represents edge e ip The frequency of (4) Cause-to-effect probability: This parameter refers to the probability of a certain problem n p When it is determined, the possibility that the problem may further lead to other problems, that is, node n p The probability of causing other result nodes to occur is shown in the following formula: Where: Indicates the problem p As a cause to further raise questions i The probability of f(e pi ) represents edge e pi The frequency of (5) Out-degree of a node: This parameter refers to the sum of the frequencies of the edges with the node as the starting point in the causal knowledge network, as shown in the following formula: Where: D out (n p ) represents node n p The out-degree of the nodes in the causal knowledge network; f(e pi ) represents node n p The frequency of any side in the causal knowledge network as the starting point; (6) In-degree of a node: This parameter refers to the sum of the frequencies of the edges that point to the node in the causal knowledge network, as shown in the following formula: Where: D in (n p ) represents node n p The in-degree of the nodes in the causal knowledge network; f(e ip ) represents node n p The frequency of any edge in the causal knowledge network as a pointing point; (7) Number of steps for the reason: Assume n i 、n j are two nodes in the causal knowledge network, n i Yes j One reason is that n i Yes j Therefore, these two nodes are at least in the same causal chain. If node n is in a causal chain i With node n j There are m-1 nodes between them, that is, node n on this causal chain i With node n j If the step length between them is m, then we can say that n i Yes j The m-layer reasons; In this way, a causal knowledge network G = (N, E, R) is constructed, where G represents the causal knowledge network; N represents the set of nodes in the causal knowledge network; E represents the set of edges in the causal knowledge network; and R represents the set of parameters in the causal knowledge network.
3. The method for analyzing the causes of product quality problems based on a causal knowledge network according to claim 2 is characterized by: The step 2 specifically includes the following steps: Step 2.1: The product quality management personnel shall record the actual product quality problems encountered on site; the recording methods include but are not limited to: text description records, photo records, and video records; Step 2.2: The actual product quality issues recorded in the photographs and videos can be converted into text descriptions by technical means or manual recognition; the actual product quality issues recorded in the text descriptions do not need to be processed; Step 2.3: Filter all problem description words from the text description; Step 2.4: Compare all problem description words with all causal knowledge nodes in the causal knowledge network and calculate the semantic similarity; Step 2.5: Set a semantic similarity threshold and filter out all causal knowledge nodes with semantic similarity greater than the threshold, so as to complete the matching of actual product quality problems with causal knowledge nodes.
4. The method for analyzing the causes of product quality problems based on a causal knowledge network according to claim 3 is characterized by: The specific method of step 3 is as follows: Step 3.1: Select any one of the causal knowledge nodes selected in step 2.5 and set it as The processing method of the remaining nodes is the same as same; Step 3.2: Based on the causal knowledge network constructed in step 1, calculate the connectivity between nodes through the reachability matrix method to find the problem node All the root cause nodes of the problem are as follows: ① Find the problem node The node set corresponding to the non-zero elements in the row and column of the reachable matrix M This set is the problem node ② Find the set of nodes with in-degree 0 among all the cause nodes This set is the problem node All root cause nodes of Through the above steps, the root cause of the actual product quality problems can be found.
5. The method for analyzing the causes of product quality problems based on a causal knowledge network according to claim 4 is characterized by: The step 4 specifically includes the following steps: Step 4.1: Calculate the cause probability as follows: set up It is the problem node A root cause node, It is the problem node Root cause node A causal chain between is the set of nodes of the causal chain; is the edge set of the causal chain; is the parameter set of the causal chain, including the probability of each node to seek the cause and effect calculated according to the causal knowledge network Therefore, the causal chain Problem Node Root Cause Node The probability of The calculation method is as follows: set up It is the problem node Root cause node Since the nodes and parameters of each causal chain are different, the problem nodes on each causal chain are Root Cause Node The resulting probabilities are also different; therefore, the problem nodes on each causal chain are calculated Root Cause Node The probability caused by the accumulation can get the problem node Root Cause Node The probability of As shown below: Step 4.2: Calculate the node importance. The specific method is as follows: set up Root cause node Located at the problem node Root cause node The number of cause steps in each causal chain between them; then for the problem node For example, the root cause node The node importance The calculation method is as follows: Where: g(x) is a reconciliation formula formulated according to the actual situation, its purpose is to prevent the sum of the cause steps from being too large and affecting the accuracy of the node importance; Step 4.3: Infer a causal chain that leads to the actual quality problem of the product. The specific process is as follows: ① Select the root cause node with the greatest node importance; ② Select the nodes with the largest priority value layer by layer from bottom to top until a complete causal chain is inferred; The priority value is calculated as follows: Set the problem node Root cause node A problem node that may lead to a cause-effect probability is The problem node The priority value The calculation method is as follows: Where: Represents the problem node The node importance; h(x) is a reconciliation formula formulated according to the actual situation, its purpose is to prevent the node importance from being too large and affecting the accuracy of the priority value.
6. The method for analyzing the causes of product quality problems based on a causal knowledge network according to claim 5 is characterized by: The step 5 specifically includes the following steps: The causal chain inferred in step 4 is handed over to the product quality problem solving expert for analysis and judgment based on the actual situation; if the causal chain is the correct causal chain that leads to the actual quality problem of the product, the reasoning is completed; otherwise, the wrong node is found and corrected. At this time, two situations may occur: ① If the corrected node is on the causal knowledge network, the reasoning process of step 4 is repeated based on the corrected node to generate a new causal chain, which is again handed over to the product quality problem solving expert for analysis and judgment; ② If the corrected node is not on the causal knowledge network, the product quality problem solving expert will conduct subsequent analysis, find the new causal node and judge whether it is on the causal knowledge network, and then repeat the above steps until the correct causal chain that leads to the actual quality problem of the product is found.
7. The method for analyzing the causes of product quality problems based on a causal knowledge network according to claim 6 is characterized by: The step 6 specifically includes the following steps: Assume that the causal chain G γ =(N γ ,E γ ) is the correct cause and effect chain leading to the actual quality problem of the product; is the causal chain G γ The nodes of In the causal knowledge network, there are the same nodes, nodes There are no identical nodes in the causal knowledge network; is the causal chain G γ The edge of In the causal knowledge network, there are the same edges, edges There are no identical edges in the causal knowledge network; set up Nodes in the causal knowledge network If the nodes are the same Frequency Updated to in For Node The previous frequency; set the edge and the edges in the causal knowledge network Same, then the side Frequency Updated to in For edge Previous frequency; Traversing nodes With edge Complete the same edge and node updates as the causal knowledge network; If it is a node that is not in the causal knowledge network, then add the node to the causal knowledge network and set the frequency of the node If it is an edge that is not in the causal knowledge network, then add the edge to the causal knowledge network and set the frequency of the edge All nodes that do not exist in the causal knowledge network With edge Add to the causal knowledge network and update the parameters of the causal knowledge network, including but not limited to: the frequency of nodes, the frequency of edges, the probability of seeking causes from results, the probability of predicting results from causes, the out-degree of nodes, the in-degree of nodes, and the number of cause steps, so as to improve the causal knowledge network.