Classification method and device for quality problems in production process

Abstract

A method (100) for classifying quality problems in a production process comprises: for each of a plurality of quality problems to be classified, extracting a quality map corresponding to the quality problem from a production map based on a propagation pattern of the quality problem between nodes in the production map, wherein the production map includes all knowledge related to production (101); reducing the extracted multiple quality maps according to a preset rule (102); and determining the similarity of the multiple quality problems using the reduced multiple quality maps, and classifying the multiple quality problems according to the determined similarity (103). By classifying quality problems through quantitative measurement, it is possible to discover the hidden reasons that make two seemingly different problems similar. When a new quality problem occurs, the problem solving records of historical quality problems of the same category or with high similarity are used to quickly determine the cause and solve the quality problem, saving engineers' time and energy.

Description

Technical Field

[0001] The present disclosure relates to the technical field of industrial production, and more particularly, to a method, apparatus, computing device, computer-readable storage medium, and program product for classifying quality problems in a production process. Background Art

[0002] Quality is no longer just a synonym for product integrity. Instead, it is related to all departments, individuals and processes involved in the production of a product. Quality control and management have high priority throughout the production process of the product, especially in industries where high prices or defective products may threaten human or social safety, such as the automotive industry. In most factories in these industries, some quality-related systems are equipped, such as the Andon system and QMS (Quality Management System). Different systems focus on different production stages. The Anton system is used to notify workers of problems that occur in the process, while the QMS is used to record quality problems that occur in the process and provide analysis tools.

[0003] Common cause analysis methods include FMEA, 5 whys and Ishikawa method, etc. In these cause analysis methods, inspection rules are set in advance for each type of quality problem based on the potential cause of these quality problems. When a specific quality problem occurs in the production process, the inspection items are determined according to the inspection rules set in advance for this type of quality problem, and inspection tasks are dispatched or assigned to find out the cause of the quality problem. Summary of the invention

[0004] The above-mentioned cause analysis method in the prior art starts from a single problem and searches for the cause according to pre-set inspection rules, which works well and effectively for a single problem or quality problems of the same type. However, some quality problems, although clearly different in words, are actually caused by the same or similar causes. Therefore, if there are multiple quality problems of different types, it is difficult to generate and study the entire cause relationship diagram. If these problems are handled individually, overlapping causes may be missed, or these overlapping causes will not be considered to have potentially high risks. Moreover, the above-mentioned cause analysis method is also unable to use the problem-solving records of historical quality problems to solve new quality problems.

[0005] The first embodiment of the present disclosure proposes a method for classifying quality problems in a production process, including: for each quality problem among multiple quality problems to be classified, extracting a quality map corresponding to the quality problem from a production map based on a propagation pattern of the quality problem between nodes in the production map, wherein the production map includes all knowledge related to production; reducing the extracted multiple quality maps according to preset rules; and determining the similarity of the multiple quality problems using the reduced multiple quality maps, and classifying the multiple quality problems according to the determined similarity.

[0006] In this embodiment, the quality map extracted from the production map is used to determine the similarity between multiple quality problems, and the quality problems can be classified by quantitative measurement rather than literal type. Since the similarity is based on data behavior, the method of this embodiment helps to discover the hidden reasons that make two seemingly different problems similar. When a new quality problem occurs, historical quality problems that belong to the same category or have a higher similarity can be found, and the problem-solving records of the historical quality problems (such as actual inspection items, executed action items, and final causes, etc.) are used to prioritize potential causes, so that engineers can prioritize the executed action items, which helps engineers quickly determine the causes and solve quality problems, saving engineers time and energy.

[0007] A second embodiment of the present disclosure proposes a device for classifying quality problems in a production process, including: a quality map extraction unit, which is configured to extract a quality map corresponding to the quality problem from a production map based on a propagation pattern of the quality problem between nodes in the production map for each quality problem in a plurality of quality problems to be classified, wherein the production map includes all knowledge related to production; a quality map reduction unit, which is configured to reduce the extracted plurality of quality maps according to preset rules; and a quality problem classification unit, which is configured to determine the similarity of the plurality of quality problems based on the reduced plurality of quality maps, and classify the plurality of quality problems according to the determined similarity.

[0008] A third embodiment of the present disclosure provides a computing device, which includes: a processor; and a memory for storing computer-executable instructions, which enables the processor to execute the method in the first embodiment when the computer-executable instructions are executed.

[0009] A fourth embodiment of the present disclosure provides a computer-readable storage medium having computer-executable instructions stored thereon, wherein the computer-executable instructions are used to execute the method of the first embodiment.

[0010] A fifth embodiment of the present disclosure proposes a computer program product, which is tangibly stored on a computer-readable storage medium and includes computer-executable instructions, which, when executed, cause at least one processor to perform the method of the first embodiment. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The features, advantages and other aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings, in which several embodiments of the present disclosure are shown in an exemplary and non-limiting manner. In the accompanying drawings:

[0012] Figure 1 A method for classifying quality problems in a production process according to some embodiments of the present disclosure is shown;

[0013] Figures 2(a)-(c) are schematic diagrams showing exemplary propagation modes, respectively;

[0014] Figure 3 A flow chart showing a method for classifying quality issues according to a specific embodiment of the present disclosure is shown;

[0015] Figure 4 Shown in Figure 3 A schematic diagram of a mass spectrum extracted in an embodiment of the present invention;

[0016] Figure 5(a)-(b) shows the Figure 3 Schematic diagram of two other mass spectra extracted in the embodiment of ;

[0017] Figure 6 A device for classifying quality problems in a production process according to an embodiment of the present disclosure is shown; and

[0018] Figure 7 A block diagram of a computing device for classifying quality issues in a production process according to one embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0019] The various exemplary embodiments of the present disclosure are described in detail below with reference to the accompanying drawings. Although the exemplary methods and devices described below include software and / or firmware executed on hardware among other components, it should be noted that these examples are merely illustrative and should not be regarded as restrictive. For example, it is considered that any or all hardware, software, and firmware components can be implemented exclusively in hardware, exclusively in software, or in any combination of hardware and software. Therefore, although exemplary methods and devices have been described below, it should be readily understood by those skilled in the art that the examples provided are not intended to limit the manner in which these methods and devices are implemented.

[0020] In addition, the flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions and operations of the methods and systems according to the various embodiments of the present disclosure. It should be noted that the functions marked in the boxes may also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession may actually be executed substantially in parallel, or they may sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of boxes in the flowchart and / or block diagram, may be implemented using a dedicated hardware-based system that performs a specified function or operation, or may be implemented using a combination of dedicated hardware and computer instructions.

[0021] The terms "include", "comprising" and similar terms used herein are open terms, i.e., "including / includes but not limited to", indicating that other contents may also be included. The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment", etc.

[0022] Figure 1 A method for classifying quality problems in a production process according to some embodiments of the present disclosure is shown. Figure 1 , method 100 starts from step 101. In step 101, for each of the multiple quality problems to be classified, a quality map corresponding to the quality problem is extracted from the production map based on the propagation mode of the quality problem between the nodes in the production map, wherein the production map includes all knowledge related to production. The pre-established ontology model is instantiated to obtain the production map. The ontology model describes all classes (abstract concepts) related to production and the relationships between them. Classes are embodied as nodes in the ontology model, and the relationships between them are embodied as edges between nodes. When establishing the ontology model, the production elements involved in all links of the production process are taken as nodes, such as materials, suppliers, purchase orders, warehouses, processing plans, work orders, production lines, machines, machine parts, products, quality tests, etc. The relationships between these production elements are taken as edges between nodes. Relationships may include whole-part relationships, genus relationships, attribute relationships, and the like.

[0023] In some embodiments, a propagation mode is defined for each type of quality problem, and the propagation mode is pre-added to the ontology model of the production map. The nodes in the production map involve the entire production process, and for a specific quality problem, only a part of the relevant nodes in the production map need to be considered, and these relevant nodes are the potential causes of the quality problem. Therefore, these relevant nodes can be filtered out from the entire production map by label propagation. For each type of quality problem, a propagation mode is defined for the edge between the nodes. The propagation mode represents the propagation relationship between the two nodes connected by the edge of this type of quality problem. The propagation relationship includes whether to propagate and the propagation direction. The defined propagation mode is added to the ontology model as an annotation. When the ontology model is instantiated to form a production map, the propagation mode is also reflected in the production map. For a specific quality problem, the nodes related to the quality problem are extracted from the production map through label propagation, and these nodes together with the propagation relationship between the nodes form a quality map corresponding to the quality problem.

[0024] In some embodiments, the propagation mode includes an inclusion mode, an upstream and downstream mode, and an equivalent mode. Figures 2(a)-(c) respectively show schematic diagrams of exemplary propagation modes. Figure 2(a) shows an inclusion mode. In the production map, the relationship between the parent node 201 and the child node 202 is a whole-part relationship. As shown by the arrow in Figure 2(a), the propagation direction of the quality problem between the parent node 201 and the child node 202 is from the child node 202 to the parent node 201, that is, a problem with the child node 202 causes a problem with its parent node 201. For example, a worn tool has a 90% chance of reducing the machining accuracy of a CNC machine tool. Therefore, as a child node, when a problem occurs with the tool, a problem also occurs with its parent node, the CNC machine tool. The propagation direction between the tool and the CNC machine tool is from the tool to the CNC machine tool. Generally, this mode can be applied to most hierarchical system structures.

[0025] FIG2(b) shows the upstream-downstream model. As shown by the arrow in FIG2(b), the quality problem propagates from the upstream node 203 to the downstream node 204, that is, a problem in the upstream node 203 causes a problem in its downstream node 204. For example, a workpiece is transferred to machine 2 for processing step 2 after processing step 1 of machine 1 (without considering the buffer between the two machines). Then the long processing time of machine 1 will cause a long waiting time of machine 2. Therefore, machine 1 is an upstream node, and when it has a problem, it will cause a problem in its downstream node machine 2. The propagation direction between machine 1 and machine 2 is from machine 1 to machine 2. For another example, a supplier's delayed delivery will cause a delay in the production plan. Therefore, when the supplier is an upstream node, when it has a problem, it will cause a problem in the production plan of its downstream node. The propagation direction between the supplier and the production plan is from the supplier to the production plan. This model can be used to describe the relationship between independent facts.

[0026] FIG2(c) shows an equivalent pattern, which indicates that the quality problem does not propagate or does not need to propagate between two nodes, that is, if a problem occurs in one node 205, the other node 206 also has a problem without propagation. For example, a product cannot pass the quality test if it does not meet the quality standards. Therefore, when a problem occurs in the product quality as a node, the quality test as another node also has a problem without propagation because the two are equivalent. There is no need to propagate between the product quality and the quality test. This pattern can be used to describe the fact of bundling.

[0027] Next, in step 102, the extracted multiple quality maps are reduced according to preset rules. In the quality map, there are usually many nodes with high node degrees (i.e., the number of edges connected to the node), so the quality map is a dense graph. The possibility of these nodes with high node degrees becoming the cause of quality problems is relatively low. In addition, since multiple quality maps share almost all manufacturing resources, such as machines, materials, workers, etc., some nodes appear in more than one or even all quality maps at the same time, and the possibility of these nodes becoming the cause of quality problems is also relatively low. The above-mentioned nodes are called low-information nodes, which are very likely not to become the final cause of the quality problem, but affect the scale of the quality map. Moreover, because the existence of low-information nodes makes the difference between the network structures of the quality map not so obvious, it will also affect the accuracy of quality problem classification to a certain extent. Therefore, it is necessary to remove these low-information nodes to reduce the extracted multiple quality maps.

[0028] In some embodiments, reducing the extracted multiple mass spectra according to a preset rule further includes: determining one or more low-information nodes in the multiple mass spectra; and reducing the multiple mass spectra by removing one or more low-information nodes from the multiple mass spectra. By reducing the mass spectra, on the one hand, the mass spectra are reduced in size, thereby speeding up the subsequent similarity calculation, and on the other hand, the multiple mass spectra are more distinguishable, thereby making the similarity between them more accurate.

[0029] In some embodiments, determining one or more low-information nodes in multiple quality maps further includes: calculating the node information entropy of multiple candidate nodes in the multiple quality maps respectively, the node information entropy indicating the possibility of the corresponding candidate node becoming the cause of multiple quality problems; and taking the candidate nodes whose node information entropy is lower than the preset threshold as low-information nodes. All other nodes in the quality map except the nodes that reflect the quality problem itself are taken as candidate nodes. For example, if the quality problem is that the product breaks, all other nodes in the quality map except the product node and the break test node (these two nodes reflect the quality problem) are candidate nodes. The node information entropy is used to evaluate the information contained in the candidate node. If the possibility that the candidate node becomes the cause of the quality problem is low, the node information entropy is small, indicating that the candidate node contains less information. On the contrary, if the possibility that the candidate node becomes the cause of the quality problem is high, the node information entropy is large, indicating that the candidate node contains more information. After calculating the node information entropy of all candidate nodes, the node information entropy is compared with the preset threshold respectively, and the candidate nodes whose node information entropy is lower than the preset threshold are taken as low-information nodes.

[0030] In some embodiments, the node information entropy is determined according to one or more of the following: the average information entropy of the corresponding candidate node relative to a single quality map in a plurality of quality maps, the information entropy of the corresponding candidate node relative to all quality maps in a plurality of quality maps, and the node type of the corresponding candidate node. As described above, the possibility that nodes with higher node degrees in each quality map and nodes appearing in a plurality of quality maps become the cause of quality problems is relatively low. The average information entropy of the candidate node relative to a single quality map in a plurality of quality maps and the information entropy relative to all quality maps in a plurality of quality maps can be calculated respectively, and the obtained results can be added to calculate the node information entropy. The average information entropy of the candidate node relative to a single quality map in a plurality of quality maps represents the average amount of information contained in the candidate node from the perspective of a single quality map. The information entropy of the candidate node relative to all quality maps in a plurality of quality maps represents the amount of information contained in the candidate node from the perspective of a plurality of quality maps. In addition, the possibility that different types of nodes (such as workers, machines, materials, etc.) become the cause of quality problems is also different. Therefore, when calculating the node information entropy of the candidate node, different weights can be set according to the node type.

[0031] In step 103, the similarity of the multiple quality problems is determined using the reduced multiple quality maps, and the multiple quality problems are classified according to the determined similarity. As described above, each quality problem corresponds to a quality map, and the nodes in the quality map represent the potential causes of the quality problem. Therefore, by comparing the quality maps, the similarity between the multiple quality problems can be obtained. Afterwards, several quality problems with similarity higher than a preset threshold are regarded as quality problems of the same category, and they have the same or similar potential causes.

[0032] In some embodiments, determining the similarity of multiple quality problems using the reduced multiple quality maps, and classifying the multiple quality problems according to the determined similarity further includes: obtaining multiple vectors representing one quality problem in the multiple quality problems according to the reduced multiple quality maps; determining the similarity between the multiple quality problems by calculating the distance between the multiple vectors; and classifying the multiple quality problems according to the determined similarity. The node2vec algorithm can be applied to the multiple quality maps to obtain multiple vectors. The node2vec algorithm can learn the node features in the graph of the network structure and represent the features with vectors (i.e., map the node features to a low-dimensional space), which are called graph embedding vectors. Node features include the connection relationship between the node and other nodes in the graph. When the node2vec algorithm is applied to the quality map, the multiple quality maps are regarded as a whole, the node features of the nodes in the quality map that reflect the quality problem itself are learned, and a vector representing the node features is generated. Each vector is used to represent a quality problem. Therefore, through the node2vec algorithm, multiple quality problems can be quantitatively represented by a set of corresponding vectors, which is convenient for more accurate comparison and classification.

[0033] After obtaining a set of vectors representing multiple quality issues, the similarity between the quality issues can be determined by calculating the distance between the vectors. Any distance calculation method (such as Euclidean distance, Manhattan distance, cosine distance, etc.) can be used to determine the similarity between the quality issues. The smaller the distance between two vectors, the higher their similarity. The distance calculation method to be used can be selected and determined according to actual needs (such as according to the node type). Several quality issues with similarities higher than a preset threshold are classified into the same category. Any existing clustering method can also be used to classify multiple quality issues according to the calculated distance.

[0034] In the above embodiment, the similarity of multiple quality issues is determined by using the quality map extracted from the production map, and the quality issues can be classified by quantitative measurement rather than literal type. Since the similarity is based on data behavior, the method of this embodiment helps to discover the hidden reasons that make two seemingly different problems similar.

[0035] In some embodiments, the above method 100 can be used to assist engineers in determining the causes of newly emerging quality problems. New quality problems and all or selected parts (e.g., of the same type) of historical quality problems can be classified, and the problem-solving records of historical quality problems that belong to the same category (or have the highest similarity) as the new quality problem (such as actual inspection items, executed action items, and final causes, etc.) can be viewed to prioritize potential causes. In addition, when the quality map corresponding to the new quality problem is reduced, the nodes in the reduced quality map (i.e., the potential causes of the new quality problem) can also be prioritized according to the level of node information entropy, and the priority can be prioritized again in combination with the problem-solving records of historical quality problems. Through the above method, when facing new quality problems, engineers can prioritize the executed action items according to the priority of potential causes, which helps engineers quickly determine the causes and solve new quality problems, saving engineers time and energy.

[0036] The following is a specific example to illustrate Figure 1 Classification method of quality problems in. Figure 3 A flow chart of a method for classifying quality problems according to a specific embodiment of the present disclosure is shown. In method 300, step 301 includes defining a propagation mode for each type of edge in an ontology model related to production for each type of quality problem (such as excessive processing time of a certain production line, cracking of a certain product, etc.). In this embodiment, the three propagation modes shown in Figures 2(a)-(c) are used. For different types of quality problems, the defined propagation modes may be different. The defined propagation modes are added to the ontology model as annotations. In this way, when the ontology model is instantiated to form a production map, the propagation mode is also reflected in the production map.

[0037] In step 302, for each of the multiple quality problems to be classified, a quality graph corresponding to the quality problem is extracted from the production graph based on the propagation mode of the quality problem between nodes in the production graph. In this step, nodes related to the quality problem are extracted from the production graph using label propagation, and these nodes together with the propagation relationship between the nodes form a quality graph corresponding to the quality problem. Figure 4 Shown in Figure 3Schematic diagram of a quality spectrum extracted in an embodiment of . A quality spectrum 400 is extracted for the quality problem of "the processing time of production line A is too long". The quality spectrum 400 includes nodes related to the quality problem and the propagation relationship between the nodes, wherein arrows indicate the propagation direction, and lines without arrows indicate no propagation or no need for propagation. For example, in the quality spectrum 400, the node "processing time" is the node that reflects the quality problem itself, and the propagation mode between it and the node "production line A" is equivalent propagation, that is, no propagation is required; the node "unit 1" is a child node of the node "production line A", and the propagation mode between the two nodes is inclusive propagation, and the propagation direction is from the node "unit 1" to the node "production line A"; the node "machine 1" is the upstream node of the node "machine 2", and the propagation mode between the two nodes is upstream and downstream propagation, and the propagation direction is from the node "machine 1" to the node "machine 2", and so on.

[0038] Figure 5(a)-(b) shows the Figure 3 Schematic diagram of two other quality maps extracted in the embodiment of FIG. Quality map 500 is extracted for the quality problem of "the leakage test of product A fails", and quality map 501 is extracted for the quality problem of "the rupture test of product B fails". Similar to the quality map 400 shown in FIG. (4), the quality maps 500 and 501 respectively include nodes related to the corresponding quality problems and the propagation relationship between the nodes, wherein the arrow indicates the propagation direction, and the line without an arrow indicates no propagation or no need for propagation.

[0039] Then, in step 303, node information entropy of candidate nodes in the extracted multiple quality graphs is calculated respectively. All nodes except nodes that reflect quality problems are considered as candidate nodes. The following formulas (1)-(6) show the calculation process of node information entropy.

[0040] H(x)=H in (x)+H across (x) (1)

[0041]

[0042] H across (x) = -p 2 (x=1)logp 2 (x=1)-p 2 (x=0)logp 2 (x=0) (3)

[0043] In formulas (1)-(3), x is a node in the mass spectrum X, and G represents a mass spectrum set, which includes multiple mass spectra. in(x) evaluates the information of node x from the perspective of a single quality graph, which represents the average information entropy of node x relative to a single quality graph in multiple quality graphs. across (x) evaluates the information of x from the perspective of multiple quality graphs, which represents the information entropy of node x relative to all quality graphs in multiple quality graphs. count(x in G) represents the number of times node x appears in multiple quality graphs. logp 1 (x) is the natural logarithm function.

[0044]

[0045]

[0046] p 2 (x=0)=1-p 2 (x=1) (6)

[0047] In formula (4), degree(x) is the node degree of node x in the quality graph X, and node count is the total number of nodes in the quality graph X. 1 (x) represents the importance of node x in the quality graph X. The higher the importance, the lower the possibility that node x will become the cause of the quality problem. In formula (5), count(G) is the total number of quality graphs in the quality graph set. 2 (x=1) represents the ratio of the number of quality graphs containing node x to the number of all quality graphs in the quality graph set. Similarly, p 2 (x=0) represents the ratio of the number of quality maps that do not contain node x to the number of all quality maps in the quality map set. In other embodiments, other methods may be used to calculate the node information entropy of the candidate node. For example, different weights may be set according to the type of the candidate node based on the above formulas (1)-(6).

[0048] The following uses the quality graphs 500 and 501 in Figures 5(a)-(b) as examples to illustrate the process of calculating node information entropy. The nodes "leakage test", "product A", "crack test", and "product B" in the quality graphs 500 and 501 reflect the quality problem itself. Therefore, the candidate nodes are "production line A", "machine 1", "machine 2", "worker 1", "bolt M60", "supplier 1", "worker 2", and "bolt M40". Table 1 shows the p of each candidate node relative to the quality graphs 500 and 501. 1 (x) value, count(x in G) value and p 2 (x=1) value.

[0049]

[0050]

[0051] Table 1

[0052] H(x) Production Line A 0.35 Machine 1 0.36 Machine 2 0.27 Bolt M40 1.05 Bolt M60 1.05 Supplier 1 0.30 Worker 1 0.99 Worker 2 0.99

[0053] Table 2

[0054] As shown in Table 2, the nodes "Production Line A", "Machine 1", "Machine 2", and "Supplier 1" have smaller node information entropy, that is, these nodes contain less information. The nodes "Bolt M40", "Bolt M60", "Worker 1", and "Worker 2" have larger node information entropy, indicating that these nodes contain more information. This means that for quality graphs 500 and 501, the nodes "Production Line A", "Machine 1", "Machine 2", and "Supplier 1" are less likely to be the cause of the quality problems "Failed leakage test of product A" and "Failed rupture test of product B", while the nodes "Bolt M40", "Bolt M60", "Worker 1", and "Worker 2" are more likely to be the cause of these two quality problems.

[0055] When the quality problem classification method of this embodiment is used for newly emerged quality problems, the multiple quality problems to be classified include new quality problems and several historical quality problems. At this time, it is necessary to calculate the average information entropy H of each candidate node in the quality spectrum corresponding to these quality problems relative to a single quality spectrum. in (x) and the information entropy H relative to all mass spectra across (x) to obtain the node information entropy H(x).

[0056] return Figure 3 In step 304, the extracted multiple quality maps are reduced according to the node information entropy of the candidate nodes. Specifically, the candidate nodes whose node information entropy is lower than the preset threshold are regarded as low-information nodes, and these low-information nodes are removed from the multiple quality maps to reduce the quality maps. Still taking the quality maps 500 and 501 in Figures 5(a)-(b) as an example, if the preset threshold of the node information entropy is set to 0.3, the nodes "machine 2" and "supplier 1" are removed from the quality maps 500 and 501.

[0057] In step 305, the node2Vec algorithm is applied to the reduced multiple quality graphs to learn the node features of the nodes reflecting the quality problems in the reduced multiple quality graphs, and each of them is represented by a vector. 1 , Q 2 ...Q k , indicating the quality problem Q 1 The vector can be (v 11 , v 12, …, v 1m ), indicating quality problem Q 2 The vector can be (v 21 , v 22 , …, v 2m ), indicating quality problem Q k The vector can be (v k1 , v k2 , …, v km ) and so on. Still taking the quality graphs 500 and 501 in FIG. 5(a)-(b) as an example, the nodes "Product A" and "Product B" respectively reflect the quality problem itself. Therefore, by applying the node2Vec algorithm to the reduced quality graphs 500 and 501, two vectors representing the node features of the nodes "Product A" and "Product B" are obtained. These two vectors can respectively represent the network structures of the two reduced quality graphs.

[0058] Next, in step 306, the similarities between the multiple quality issues are determined by calculating the distances between the multiple vectors obtained in step 305, and the multiple quality issues are classified according to the similarities. As described above, the smaller the distance between the two vectors, the higher the similarity between the corresponding two quality issues. Accordingly, in the production map (high-dimensional space), the nodes representing the two quality issues are closer to each other. The classification of quality issues can be achieved by treating several quality issues with a similarity higher than a preset threshold as the same category.

[0059] In the above embodiment, the quality map extracted from the production map is used to determine the similarity of multiple quality problems, and the quality problems can be classified by quantitative measurement rather than literal type. Since the similarity is based on data behavior, the method of this embodiment helps to discover the hidden reasons that make two seemingly different problems similar. When a new quality problem occurs, historical quality problems that belong to the same category or have a higher similarity can be found, and the problem-solving records of the historical quality problems (such as actual inspection items, executed action items, and final causes, etc.) are used to prioritize potential causes, so that engineers can prioritize the executed action items, which helps engineers quickly determine the causes and solve quality problems, saving engineers time and energy.

[0060] Figure 6 A device for classifying quality problems in a production process according to an embodiment of the present disclosure is shown. Figure 6, the device 600 includes a quality map extraction unit 601, a quality map reduction unit 602 and a quality problem classification unit 604. The quality map extraction unit 601 is configured to extract a quality map corresponding to the quality problem from the production map based on the propagation mode of the quality problem between nodes in the production map for each quality problem in the multiple quality problems to be classified, wherein the production map includes all knowledge related to production. The quality map reduction unit 602 is configured to reduce the extracted multiple quality maps according to preset rules. The quality problem classification unit 603 is configured to determine the similarity of multiple quality problems using the reduced multiple quality maps, and classify the multiple quality problems according to the determined similarity. Figure 6 Each unit in the system can be implemented by software, hardware (such as integrated circuit, FPGA, etc.) or a combination of software and hardware.

[0061] In some embodiments, the quality spectrum reduction unit 602 further includes ( Figure 6 (not shown) a low information node determination unit and a low information node removal unit. The low information node determination unit is configured to determine one or more low information nodes in a plurality of quality spectra, and the low information node removal unit is configured to reduce the plurality of quality spectra by removing the one or more low information nodes from the plurality of quality spectra, respectively.

[0062] In some embodiments, the low information node determination unit is further configured to: respectively calculate the node information entropy of multiple candidate nodes in multiple quality graphs, the node information entropy indicating the possibility that the corresponding candidate nodes will become the cause of multiple quality problems; and classify the candidate nodes whose node information entropy is lower than a preset threshold as low information nodes.

[0063] In some embodiments, the node information entropy is determined based on one or more of: the information entropy of the corresponding candidate node relative to each quality spectrum in a plurality of quality spectra, the information entropy of the corresponding candidate node relative to all quality spectra in a plurality of quality spectra, and the node type of the corresponding candidate node.

[0064] In some embodiments, the quality problem classification unit further includes a vector obtaining unit, a similarity determination unit and a classification determination unit ( Figure 6 The vector obtaining unit is configured to obtain a plurality of vectors representing one of the plurality of quality problems respectively according to the reduced plurality of quality maps. The similarity determining unit is configured to determine the similarity between the plurality of quality problems by calculating the distance between the plurality of vectors. The classification determining unit is configured to classify the plurality of quality problems according to the determined similarity.

[0065] In some embodiments, the vector obtaining unit is further configured to obtain a plurality of vectors by applying a node2vec algorithm to the plurality of quality maps.

[0066] In some embodiments, the propagation mode includes an inclusive mode, an upstream and downstream mode, and an equivalent mode. The inclusive mode indicates propagation from a child node to a parent node, the upstream and downstream mode indicates propagation from an upstream node to a downstream node, and the equivalent mode indicates no propagation.

[0067] In some embodiments, a propagation pattern is defined for each type of quality issue, and the propagation pattern is pre-added to the ontology model of the production graph.

[0068] Figure 7 A block diagram of a computing device for classifying quality issues in a production process according to an embodiment of the present disclosure is shown. Figure 7 As can be seen in FIG. 7 , the computing device 700 for classifying quality problems in a production process includes a processor 701 and a memory 702 coupled to the processor 701. The memory 702 is used to store computer executable instructions, which, when executed, enable the processor 701 to execute the method in the above embodiment.

[0069] In addition, alternatively, the above method can be implemented by a computer-readable storage medium. Computer-readable storage media are loaded with computer-readable program instructions for executing various embodiments of the present disclosure. Computer-readable storage media can be a tangible device that can hold and store instructions used by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the above. More specific examples (non-exhaustive lists) of computer-readable storage media include: portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), static random access memories (SRAM), portable compact disk read-only memories (CD-ROM), digital versatile disks (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or protruding structures in grooves on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., light pulses through a fiber optic cable), or an electrical signal transmitted through wires.

[0070] Therefore, in another embodiment, the present disclosure provides a computer-readable storage medium having computer-executable instructions stored thereon, where the computer-executable instructions are used to execute the methods in various embodiments of the present disclosure.

[0071] In another embodiment, the present disclosure proposes a computer program product, which is tangibly stored on a computer-readable storage medium and includes computer-executable instructions that, when executed, cause at least one processor to perform the methods in various embodiments of the present disclosure.

[0072] In general, various example embodiments of the present disclosure may be implemented in hardware or dedicated circuits, software, firmware, logic, or any combination thereof. Certain aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. When various aspects of the embodiments of the present disclosure are illustrated or described as block diagrams, flow charts, or using some other graphical representation, it will be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuits or logic, general hardware or controllers or other computing devices, or some combination thereof as non-limiting examples.

[0073] Computer-readable program instructions or computer program products for executing the various embodiments of the present disclosure can also be stored in the cloud. When needed, users can access the computer-readable program instructions for executing an embodiment of the present disclosure stored in the cloud through mobile Internet, fixed network or other networks, thereby implementing the technical solutions disclosed in accordance with the various embodiments of the present disclosure.

[0074] Although the embodiments of the present disclosure have been described with reference to several specific embodiments, it should be understood that the embodiments of the present disclosure are not limited to the specific embodiments disclosed. The embodiments of the present disclosure are intended to cover various modifications and equivalent arrangements included in the spirit and scope of the appended claims. The scope of the claims is consistent with the broadest interpretation, thereby including all such modifications and equivalent structures and functions.

Claims

1. Classification methods for quality problems in the production process, including: For each of the multiple quality problems to be classified, respectively based on the propagation mode of the quality problem between nodes in the production map, extracting a quality map corresponding to the quality problem from the production map, wherein the production map includes all knowledge related to production; wherein the propagation mode is defined for each type of quality problem, and the propagation mode is pre-added to the ontology model of the production map; reducing the extracted multiple mass spectra according to preset rules; and Determining similarities of the multiple quality issues using the reduced multiple quality maps, and classifying the multiple quality issues according to the determined similarities; Wherein, reducing the extracted multiple mass spectra according to the preset rules further includes: determining one or more low information nodes in the plurality of quality spectra; and reducing the plurality of quality spectra by removing the one or more low-information nodes from the plurality of quality spectra, respectively; Wherein, determining one or more low information nodes in the plurality of quality spectra further comprises: respectively calculating node information entropies of a plurality of candidate nodes in the plurality of quality graphs, the node information entropies representing the likelihood that the corresponding candidate nodes will become causes of the plurality of quality issues; and The candidate nodes whose node information entropy is lower than a preset threshold are regarded as the low-information nodes; all other nodes in the quality map except the nodes reflecting the quality problem themselves are regarded as candidate nodes; Wherein, using the reduced multiple quality maps to determine the similarity of the multiple quality issues, and classifying the multiple quality issues according to the determined similarity further includes: Obtaining a plurality of vectors respectively representing one of the plurality of quality problems according to the reduced plurality of quality maps; Determine the similarities between the plurality of quality issues by calculating the distances between the plurality of vectors; and The plurality of quality issues are classified according to the determined similarities.

2. The method according to claim 1, wherein: The node information entropy is determined according to one or more of the following items: the average information entropy of the corresponding candidate node relative to a single quality spectrum in the multiple quality spectra, the information entropy of the corresponding candidate node relative to all quality spectra in the multiple quality spectra, and the node type of the corresponding candidate node.

3. The method according to claim 1, wherein: Obtaining a plurality of vectors representing one of the plurality of quality problems respectively according to the reduced plurality of quality maps further comprises: The plurality of vectors are obtained by applying a node2vec algorithm to the plurality of mass spectra.

4. The method according to claim 1, wherein: The propagation mode includes an inclusion mode, an upstream and downstream mode, and an equivalent mode. The inclusion mode indicates propagation from a child node to a parent node, the upstream and downstream mode indicates propagation from an upstream node to a downstream node, and the equivalent mode indicates no propagation.

5. Classification device for quality problems in the production process, including: A quality graph extraction unit, configured to extract, for each of the multiple quality problems to be classified, a quality graph corresponding to the quality problem from the production graph based on a propagation pattern of the quality problem between nodes in the production graph, wherein the production graph includes all knowledge related to production; wherein the propagation pattern is defined for each type of quality problem, and the propagation pattern is pre-added to an ontology model of the production graph; a mass spectrum reduction unit, configured to reduce the extracted multiple mass spectra according to a preset rule; and a quality problem classification unit configured to determine similarities of the plurality of quality problems using the reduced plurality of quality graphs, and classify the plurality of quality problems according to the determined similarities; The mass spectrum reduction unit further comprises: a low information content node determination unit, configured to determine one or more low information content nodes in the plurality of quality graphs; and a low-information node removal unit, configured to reduce the plurality of quality spectra by removing the one or more low-information nodes from the plurality of quality spectra respectively; Wherein, the low information node determination unit is further configured to: respectively calculating node information entropies of a plurality of candidate nodes in the plurality of quality graphs, the node information entropies representing the likelihood that the corresponding candidate nodes will become causes of the plurality of quality issues; and The candidate nodes whose node information entropy is lower than a preset threshold are regarded as the low-information nodes; all other nodes in the quality map except the nodes reflecting the quality problem themselves are regarded as candidate nodes; Wherein, the quality problem classification unit further includes: a vector obtaining unit configured to obtain a plurality of vectors respectively representing one of the plurality of quality problems according to the reduced plurality of quality maps; a similarity determination unit configured to determine similarities between the plurality of quality issues by calculating distances between the plurality of vectors; and A classification determination unit is configured to classify the plurality of quality issues according to the determined similarities.

6. The device according to claim 5, wherein: The node information entropy is determined according to one or more of the following items: the average information entropy of the corresponding candidate node relative to a single quality spectrum in the multiple quality spectra, the information entropy of the corresponding candidate node relative to all quality spectra in the multiple quality spectra, and the node type of the corresponding candidate node.

7. The device according to claim 5, wherein: The vector obtaining unit is further configured to: The plurality of vectors are obtained by applying a node2vec algorithm to the plurality of mass spectra.

8. The device according to claim 5, wherein: The propagation mode includes an inclusion mode, an upstream and downstream mode, and an equivalent mode. The inclusion mode indicates propagation from a child node to a parent node, the upstream and downstream mode indicates propagation from an upstream node to a downstream node, and the equivalent mode indicates no propagation.

9. Computing equipment, including: processor; as well as A memory for storing computer executable instructions, which, when executed, causes the processor to perform the method according to any one of claims 1 to 4.

10. A computer-readable storage medium having computer-executable instructions stored thereon, the computer-executable instructions being used to execute the method according to any one of claims 1-4.

11. A computer program product tangibly stored on a computer-readable storage medium and comprising computer-executable instructions which, when executed, cause at least one processor to perform the method according to any one of claims 1-4.

Images