A rule and process fingerprint based process classification method

Abstract

The application provides a process classification method based on rules and process fingerprints, comprising an initial process category generation step and a process category incremental generation step. The initial process category generation step is used for classifying a first process set to be classified to establish an initial process category library. The process category incremental generation step is used for classifying a second process set to be classified based on the initial process category library and updating the initial process category library according to the classification result to obtain a final process category library. The beneficial effect is that the application can realize efficient and low-cost warehousing of new processes and ensure long-term stability and orderly evolution of the process category library.

Description

Technical Field

[0001] This invention relates to the technical field of process classification, and more specifically, to a process classification method based on rules and process fingerprints. Background Technology

[0002] In the field of computer-aided process design (CAPP), especially under the current hybrid CAPP system architecture empowered by artificial intelligence, the scientific and reasonable classification of the process knowledge base is the key foundation for determining the system's retrieval accuracy, reuse efficiency, and reasoning ability.

[0003] Currently, traditional methods of classifying processes mainly include two categories: The first category is group technology that relies on human experience. This type of method requires process engineers to manually develop a coding system and classify parts based on their geometry, material, machining precision, and other characteristics. This approach is not only inefficient and highly subjective, but also has extremely poor scalability when the company's product line is updated.

[0004] The second category comprises data-driven automatic clustering techniques that have emerged in recent years, such as similarity clustering methods based on process fingerprints. These methods perform batch merging by calculating the topological similarity between processes, improving the automation level of classification to some extent. However, existing fingerprint-based clustering schemes generally employ offline batch processing. In actual industrial production scenarios, process design data is continuously and dynamically growing, with a large number of newly compiled process documents needing to be entered into the database daily. Using existing batch clustering methods, whenever a new batch of processes is added, the system must merge all historical data with the new data and re-execute a complete clustering operation. This static calculation mode of "starting from scratch" leads to an exponential increase in time and space overhead as the process database grows larger, severely slowing down the knowledge base's update efficiency and failing to meet the real-time and lightweight requirements of digital transformation in the manufacturing industry.

[0005] More critically, existing classification methods suffer from fundamental architectural flaws. They typically treat process classification as a one-off, isolated operation, lacking a phased division between "initial database construction" and "incremental evolution." Under this straightforward classification architecture, once new processes cause category adjustments (such as subclass splitting or the creation of new classes), the entire category numbering system and hierarchical structure are highly susceptible to disordered drift. This prevents downstream process retrieval and derived reuse modules from operating based on a stable, continuous classification anchor point, significantly reducing the stability and precision of process matching, and consequently making it difficult to truly transform the previously built process knowledge base into reliable productivity.

[0006] In summary, there is an urgent need in the existing technology for a process classification method that can achieve efficient and low-cost entry of new processes into the database, while ensuring the long-term stability and orderly evolution of the process category database, so as to truly support the dual requirements of intelligent hybrid CAPP system for dynamic management and high-precision retrieval of knowledge base. Summary of the Invention

[0007] The technical problem this invention aims to solve is how to achieve both efficient and low-cost entry of new processes into the database, while ensuring the long-term stability and orderly evolution of the process category database. To overcome the shortcomings of the existing technology (or related technology), this invention provides a process classification method based on rules and process fingerprints.

[0008] This invention provides a process classification method based on rules and process fingerprints, including an initial process category generation step and a process category incremental generation step. The initial process category generation step is used to classify a first set of processes to be classified to establish an initial process category library. The process category incremental generation step is used to classify a second set of processes to be classified based on the initial process category library and update the initial process category library according to the classification results to obtain a final process category library.

[0009] The process classification method based on rules and process fingerprints of this invention has the following advantages compared with the prior art: This invention establishes an independent incremental generation step for process categories, enabling the efficient absorption of new processes without retraining or reprocessing all historical data, thus significantly reducing the computational cost of maintaining the process category library. Furthermore, dividing the process classification process into a stage of establishing an initial process category library from scratch and a stage of updating the initial process category library to obtain the final process category library from existing to optimized ensures the continuity and stability of the process classification system. This provides a reliable organizational foundation for subsequent process retrieval and reuse, enabling both efficient and low-cost entry of new processes into the library and ensuring the long-term stability and orderly evolution of the process category library.

[0010] In one possible implementation, the initial process category generation step includes: Step A1: Based on a set of static feature classification rules, each first process to be classified in the first process set is classified into one of a plurality of first coarse process categories, and the first process to be classified that is not covered by any of the static feature classification rules is marked as a first missed process. Step A2: For each of the first coarse-classified process categories, based on the relationship between the process fingerprint similarity and the preset similarity threshold between each of the first processes to be classified within the first coarse-classified process category, the first coarse-classified process category is split into at least one fission subclass to form a fission subclass set. Step A3: For each of the first missed detection processes, calculate the process fingerprint similarity between the first missed detection process and each of the fission subclasses in the fission subclass set, and classify the first missed detection process into the corresponding fission subclass or create a new subclass for the first missed detection process based on the process fingerprint similarity, so as to generate the initial process category library.

[0011] Compared with existing technologies, the above technical solution can first use the expert experience of static feature classification rules for coarse classification, and then use the inherent features of the data, namely fingerprint similarity, for further subdivision, which takes into account both the interpretability of domain knowledge and the objectivity of data-driven approaches. Furthermore, step A3 can actively capture abnormal or novel processes that are not covered by static feature classification rules, avoiding the problem of processes being missed or forcibly misclassified due to imperfect rules in traditional methods, and significantly improving the coverage of process classification.

[0012] In one possible implementation, in step A1, each of the static feature classification rules is obtained by combining one or more sub-rules through logical operators, wherein the sub-rules are based on the shape features, material features, state features, or keyword features of the part.

[0013] Compared with existing technologies, the above technical solution enables engineers to flexibly combine sub-rules to construct static feature classification rules based on the characteristics of different industries or different parts, and adapt to new business scenarios without modifying the underlying algorithm.

[0014] In one possible implementation, in step A2, each of the first processes to be classified in the first coarse process category is regarded as an independent vertex. If the process fingerprint similarity between any two processes to be classified corresponding to the first vertex is greater than the preset similarity threshold, the two vertices are connected. Then, at least one connected graph is formed according to the connection relationship of all the vertices. The set of vertices contained in each connected graph constitutes a fission subclass.

[0015] Compared with existing technologies, the above technical solution can refine the first coarse process category through a connected graph approach, preserving the integrity of the process family to the maximum extent. It is particularly suitable for process types with gradual feature changes and continuous spectrum characteristics, effectively avoiding information loss caused by hard boundary cutting.

[0016] In one possible implementation, in step A2, each of the first processes to be classified in the first coarse process category is regarded as an independent vertex. If the process fingerprint similarity between any two processes to be classified corresponding to the first vertex is greater than the preset similarity threshold, the two vertices are connected. Then, at least one connected graph is formed according to the connection relationship of all the vertices, and the formed connected graph is split into several unconnected complete graphs. The set of vertices contained in each complete graph constitutes a fission subclass.

[0017] Compared with existing technologies, the above-mentioned technical solution can refine the first coarse-division process category through a complete graph approach. The complete graph requires that any two processes within the fission subclass must be highly similar. The fission subclass generated by this stringent standard has highly standardized and unified internal processes. Such extremely pure fission subclasses are very suitable as a standard template library. When performing derived CAPP searches, the most accurate reference process can be matched very quickly, greatly improving the accuracy of process reuse.

[0018] In one possible implementation, in step A3, it is determined whether the process fingerprint similarity between the first missed detection process and any of the first unclassified processes in the target fission subclass within the fission subclass set is greater than the preset similarity threshold. If so, the first missed detection process is classified into the target fission subclass; If not, then create a new subclass for the first missed detection process.

[0019] Compared with existing technologies, the above technical solution can ensure that the new process can quickly find the target fission subclass to which it belongs, which is in line with the actual situation of the many variations of parts in the manufacturing industry.

[0020] In one possible implementation, in step A3, it is determined whether the process fingerprint similarity between the first missed detection process and each of the first processes to be classified in the target fission subclass within the fission subclass set is greater than the preset similarity threshold: If so, the first missed detection process is classified into the target fission subclass; If not, then create a new subclass for the first missed detection process.

[0021] Compared with existing technologies, the above technical solution can effectively prevent low-quality or ambiguous process contamination of the high-precision standard library, ensuring that the core process categories remain highly stable and reliable in the long term.

[0022] In one possible implementation, the process category increment generation step includes: Step B1: Based on a set of static feature classification rules, each second process to be classified in the second process set is classified into one of a plurality of second coarse process categories, and the second process to be classified that is not covered by any of the static feature classification rules is marked as a second missed process. Step B2: For each second process to be classified into any second coarse process category, calculate the process fingerprint similarity between the second process to be classified and each first process to be classified under the corresponding first coarse process category in the initial process category library, and classify the second process to be classified into the corresponding fission subclass or create a new subclass for the second process to be classified according to the relationship between the process fingerprint similarity and the preset similarity threshold, so as to generate an updated process category library; Step B3: For each of the second missed processes, calculate the process fingerprint similarity between the second missed process and each fission subclass in the updated process category library, and classify the second missed process into the corresponding fission subclass or create a new subclass for the second missed process according to the relationship between the process fingerprint similarity and the preset similarity threshold, so as to obtain the final process category library.

[0023] Compared with existing technologies, the above technical solution can compare new processes with their corresponding subclasses under the coarse classification in step B2, rather than with the entire database. This local matching strategy significantly reduces time complexity. Even if the process database grows exponentially, the incremental update speed remains extremely fast, perfectly supporting the high-frequency data inflow in the industrial internet scenario. Step B3 still retains the global scanning mechanism, ensuring that even new processes not covered by static feature classification rules can find the most similar destination or be identified as the starting point of a new group in the global scope, guaranteeing the robustness and self-improvement capability of the process category database during dynamic growth. Attached Figure Description

[0024] Figure 1 A flowchart illustrating the steps for generating the initial process category of this invention; Figure 2 This is a flowchart illustrating the incremental generation steps of the process category in this invention. Detailed Implementation

[0025] First, those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art can make adjustments as needed to adapt to specific application scenarios.

[0026] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0027] This invention discloses a process classification method based on rules and process fingerprints, including an initial process category generation step and a process category incremental generation step. The initial process category generation step is used to classify a first set of processes to be classified in order to establish an initial process category library. The process category incremental generation step is used to classify a second set of processes to be classified based on the initial process category library and update the initial process category library according to the classification results to obtain a final process category library.

[0028] See Figure 1 In this embodiment of the invention, the initial process category generation step includes: Step A1: Based on a set of static feature classification rules, each first process to be classified in the first process set is classified into one of a plurality of first coarse process categories, and the first process to be classified that is not covered by any static feature classification rules is marked as the first missed process. Step A2: For each first coarse process category, based on the relationship between the process fingerprint similarity and the preset similarity threshold between each first process to be classified within the first coarse process category, the first coarse process category is split into at least one fission subclass to form a fission subclass set. Step A3: For each first missed process, calculate the process fingerprint similarity between the first missed process and each fission subclass in the fission subclass set, and classify the first missed process into the corresponding fission subclass or create a new subclass for the first missed process based on the process fingerprint similarity to generate an initial process category library.

[0029] See Figure 2 In this embodiment of the invention, the process category increment generation step includes: Step B1: Based on a set of static feature classification rules, each second process to be classified in the second process set is classified into one of a plurality of second coarse process categories, and the second process to be classified that is not covered by any static feature classification rules is marked as a second missed process. Step B2: For each second process to be classified into any second coarse process category, calculate the process fingerprint similarity between the second process to be classified and each first process to be classified under the corresponding first coarse process category in the initial process category library, and classify the second process to be classified into the corresponding fission subclass or create a new subclass for the second process to be classified according to the relationship between the process fingerprint similarity and the preset similarity threshold, so as to generate an updated process category library. Step B3: For each second missed process, calculate the process fingerprint similarity between the second missed process and each fission subclass in the updated process category library, and classify the second missed process into the corresponding fission subclass or create a new subclass for the second missed process according to the relationship between the process fingerprint similarity and the preset similarity threshold, so as to obtain the final process category library.

[0030] In this embodiment of the invention, static feature classification rules are expressed and stored in the form of production rules. Each static feature classification rule is obtained by combining one or more sub-rules through logical operators AND, OR, and NOT. The sub-rules are based on the shape features, material features, state features, or keyword features of the parts. The above-mentioned static feature classification rules are stored in the rule base in structured text or JSON format. The rule engine uses the Rete algorithm or forward chain inference algorithm for efficient matching. The static feature vector of each first process to be classified is matched with each static feature classification rule in the rule base one by one. If a match is found, it is assigned to the corresponding first coarse process category. If a first process to be classified matches multiple static feature classification rules at the same time, the final assignment is determined according to the priority value of the static feature classification rules or the conflict resolution strategy (such as first matching first, specificity priority, etc.). The rule base supports dynamic addition, deletion, modification, and query operations. Process engineers can maintain and adjust the rule base through a graphical interface or configuration file according to the actual part types and process characteristics of the enterprise without modifying the underlying algorithm code.

[0031] In this embodiment of the invention, the process fingerprint of each process to be classified is a multidimensional numerical vector or structured descriptor used to characterize the topological features of the process. The process fingerprint generation process is as follows: First, the process route is analyzed into a series of ordered machining operations. Each machining operation includes elements such as equipment type, machining method, tool information, and process parameters (such as cutting speed, feed rate, and depth of cut). Secondly, the above processing sequence is converted into a standardized feature vector or graph structure representation. In a preferred embodiment, an undirected feature dependency graph (UFDG) is used to express the process topology: the nodes of the undirected feature dependency graph represent each processing operation or step, and the edges of the undirected feature dependency graph represent the temporal dependency relationship, resource constraint relationship or process association relationship between processing operations. Each node and edge is attached with attribute labels (such as operation type, equipment number, etc.). After obtaining the process fingerprint (i.e., feature vector or graph structure representation) of each process to be classified, the two processes to be classified... and The similarity of process fingerprints between them is calculated as follows: When process fingerprints are represented in feature vector form, the process fingerprint similarity can be calculated using the cosine similarity of the angle, i.e.: in, and The processes to be classified are as follows and fingerprint vector, The Euclidean norm of the fingerprint vector is denoted by , and the fingerprint similarity of this process ranges from [0, 1]. Alternatively, the normalized form of the Euclidean distance can be used for calculation, i.e.: in, Indicates the calculation of Euclidean distance; When process fingerprints are represented using a graph structure, the similarity of process fingerprints can be calculated using a normalized form of graph edit distance (GED) or maximum common subgraph (MCS), i.e.: in, and The processes to be classified are as follows and fingerprint image, This indicates the number of nodes in the fingerprint graph. This represents the number of nodes in the maximum common subgraph. In this embodiment, it is preferable to use the cosine of the included angle combined with the weight of the process type to calculate the process fingerprint similarity, because it has high computational efficiency and can better reflect the overall structural similarity of the processing process sequence. Of course, the above method of calculating process fingerprint similarity is only one of the methods shown in this embodiment. When calculating process fingerprint similarity, it can also be assumed that the fingerprint sets of the two process fingerprints are... and Using Jaccard similarity | intersection | / | union The similarity of the process fingerprints is calculated.

[0032] In this embodiment of the invention, a preset similarity threshold is used. It is an important parameter for controlling the classification granularity. Its value ranges from 0% to 100%. The higher the preset similarity threshold is set, the more refined the classification (the more fission subclasses there are and the more consistent the internal processes of each fission subclass); the lower the preset similarity threshold is set, the coarser the classification (the fewer the fission subclasses there are and the more diverse the internal processes of each fission subclass).

[0033] In this embodiment of the invention, the preset similarity threshold is determined by one of the following methods: Method 1 (Determined by experience): Process experts directly set a fixed threshold based on the actual process characteristics of the enterprise and the requirements for classification granularity. For example, the preset similarity threshold is set to 75%. This method is suitable for application scenarios where the process type is relatively stable and the requirements for classification granularity are clear. Method 2 (Adaptive Calculation): The elbow method or silhouette coefficient is used to conduct multiple clustering experiments on a set of sampling processes, and the clustering effect curves under different thresholds are plotted. The threshold corresponding to the inflection point of the curve is selected as the value of the preset similarity threshold. This method is suitable for application scenarios with diverse process types and lack of prior experience. Method 3 (Dynamic Adjustment): In actual operation, if it is found that the process differences within a certain fission subclass are too large (such as the average process fingerprint similarity within the fission subclass being lower than the preset similarity threshold), the preset similarity threshold will be automatically increased to re-split the fission subclass. If it is found that the number of fission subclasses is too large, resulting in a decrease in retrieval efficiency, the preset similarity threshold will be appropriately decreased to merge them. This method realizes adaptive dynamic optimization of the preset similarity threshold.

[0034] In this embodiment of the invention, a specific first coarse separation process category is used. For example, suppose it contains 5 processes to be classified first. With a preset similarity threshold of 70%, the process fingerprint similarity matrix between each pair of the first processes to be classified is as follows: process to ( The process fingerprint similarity is 85%, and the process pair ( The process fingerprint similarity of () is 72%, and the process pair () The process fingerprint similarity of () is 78%, and the process pair () The process fingerprint similarity of () is 82%, and the process pair () The similarity of the process fingerprints is 75%. Each first process to be classified is considered an independent vertex. Vertex pairs with a process fingerprint similarity greater than 70% are connected by edges. Based on the above data, we can obtain... and , and , and They are interconnected to form a connected graph. ; and connect, and Connection, but due to Already Therefore and pass also with Connect all 5 vertices to form a connected graph. At this point, the entire first coarse separation process category Become a fission subclass ; If the internal processes within this fission category differ significantly, requiring more refined classification, the preset similarity threshold can be increased to 80%. In this case, only process pairs ( The similarity of process fingerprints and process pairs () The similarity of process fingerprints and process pairs () The process fingerprint similarity of the two graphs meets the conditions, forming two connected graphs. and , Isolated and without connections, it becomes a separate fission subclass, thus the entire first coarse-grained process category Split into three fission subclasses: , , .

[0035] In this embodiment of the invention, continuing from the above example, a connected graph is formed under the condition of a preset similarity threshold of 70%. In the middle, check the edges between each vertex: although and There is no direct connection between them, but through... , Once the intermediate vertices are connected, if a complete graph approach is used, the connected graph needs to be split into several unconnected complete graphs. The splitting strategy can be a greedy algorithm: select a complete subgraph (i.e., any two vertices are connected by an edge) from the connected graph, strip it, and repeat this operation on the remaining vertices until all vertices have been assigned. Taking the above data as an example, under the condition of a preset similarity threshold of 70%, and , and , and The three are connected in pairs to form a complete graph. Remaining vertices and The process fingerprint similarity between them is greater than a preset similarity threshold, and and Connected, and Connected, but and , Not connected and , , Not connected, therefore To form a complete graph Ultimately, the entire first coarse separation process category Split into two fission subclasses: and .

[0036] In this embodiment of the invention, the fission subclasses generated by the connected graph method allow "chain similarity" (A and B need to be similar, B and C need to be similar, A and C do not need to be directly similar), which is suitable for situations where the process features are a continuous and gradual spectrum; the complete graph method requires that any two processes within the fission subclass are similar to each other, and the generated fission subclasses have higher internal consistency and purity, making them suitable as a high-precision standard template library; users can choose between the two methods according to their actual needs, or they can preset the default method in the system configuration.

[0037] In this embodiment of the invention, it is assumed that the fission subclass set contains a fission subclass generated through a connected graph. With a preset similarity threshold of 70%, there is now a first-stage leak detection process. It is necessary to determine whether it should be classified as ,calculate and Process fingerprint similarity between each process to be classified: because and In The process fingerprint similarity is greater than the preset similarity threshold, therefore Classified ;like and If the process fingerprint similarity of all processes to be classified is less than 70%, then... Cannot be classified Create a new subclass for it. .

[0038] In this embodiment of the invention, it is assumed that the fission subclass set contains a fission subclass generated through a complete graph method. With a preset similarity threshold of 70%, there is now a first-stage leak detection process. To determine whether it should be classified as ,calculate and Process fingerprint similarity between each process to be classified: Both are greater than the preset similarity threshold, therefore Classified After being included Updated to And the updated version needs to be verified. It is still a complete graph; like but ,but Cannot be classified Create a new subclass for it. .

[0039] In this embodiment of the invention, in the incremental generation step of process categories, it is assumed that a first coarse-classified process category already exists in the initial process category library. It contains two fission subclasses: and ; There is a new second classification process. After coarse separation in step B1, it is classified into the first coarse separation process category. In step B2, it is necessary to... and Matching is performed on each fission subclass: First calculate and Process fingerprint similarity of each process to be classified: because It was generated using a connected graph approach, as long as and Any process fingerprint with a similarity greater than a preset similarity threshold can be classified as such. The similarity exceeds the preset threshold, therefore Classified ; like and If none of the fission subclasses can be matched, then it is... exist Create a new fission subclass. This allows for incremental expansion of the initial process category library.

[0040] In this embodiment of the invention, after step B2, the second process to be classified that has been successfully coarsely divided has been assigned to the corresponding fission subclass or created as a new subclass. The updated process category library is denoted as... Step B3 specifically addresses the second set of missed detection processes that were not covered by any static feature classification rules in step B1. ; For the second set of missed detection processes Each of the second missed detection processes Calculate its relationship with The process fingerprint similarity of each fission subclass is used. If a fission subclass is generated through a connected graph, then the similarity is determined by the process fingerprint. and The comparison method is used for classification and judgment. If a certain fission subclass is generated through a complete graph method, then the comparison method is used. and The comparison method is used to classify and judge; if and If none of the fission subclasses can be matched, then it is... Create a completely new subclass and append it to Finally, a final process category library is formed. .

[0041] In the description of this invention, the terms "one embodiment," "some embodiments," "in this embodiment," "specific example," or "some examples," etc., refer to specific features, mechanisms, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, mechanisms, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A process classification method based on rules and process fingerprints, characterized in that, The process includes an initial process category generation step and a process category incremental generation step. The initial process category generation step is used to classify a first set of processes to be classified in order to establish an initial process category library. The process category incremental generation step is used to classify a second set of processes to be classified based on the initial process category library and update the initial process category library according to the classification results to obtain a final process category library.

2. The process classification method according to claim 1, characterized in that, The initial process category generation step includes: Step A1: Based on a set of static feature classification rules, each first process to be classified in the first process set is classified into one of a plurality of first coarse process categories, and the first process to be classified that is not covered by any of the static feature classification rules is marked as a first missed process. Step A2: For each of the first coarse-classified process categories, based on the relationship between the process fingerprint similarity and the preset similarity threshold between each of the first processes to be classified within the first coarse-classified process category, the first coarse-classified process category is split into at least one fission subclass to form a fission subclass set. Step A3: For each of the first missed detection processes, calculate the process fingerprint similarity between the first missed detection process and each of the fission subclasses in the fission subclass set, and classify the first missed detection process into the corresponding fission subclass or create a new subclass for the first missed detection process based on the process fingerprint similarity, so as to generate the initial process category library.

3. The process classification method according to claim 2, characterized in that, In step A1, each static feature classification rule is obtained by combining one or more sub-rules through logical operators. The sub-rules are based on the shape features, material features, state features, or keyword features of the part.

4. The process classification method according to claim 2, characterized in that, In step A2, each of the first processes to be classified in the first coarse process category is regarded as an independent vertex. If the process fingerprint similarity between any two processes to be classified corresponding to the first vertex is greater than the preset similarity threshold, the two vertices are connected. Then, at least one connected graph is formed according to the connection relationship of all the vertices. The set of vertices contained in each connected graph constitutes a fission subclass.

5. The process classification method according to claim 2, characterized in that, In step A2, each of the first processes to be classified in the first coarse process category is regarded as an independent vertex. If the process fingerprint similarity between any two processes to be classified corresponding to the first vertex is greater than the preset similarity threshold, the two vertices are connected. Then, at least one connected graph is formed according to the connection relationship of all the vertices, and the formed connected graph is split into several unconnected complete graphs. The set of vertices contained in each complete graph constitutes a fission subclass.

6. The process classification method according to claim 4, characterized in that, In step A3, it is determined whether the process fingerprint similarity between the first missed detection process and any of the first unclassified processes in the target fission subclass within the fission subclass set is greater than the preset similarity threshold. If so, the first missed detection process is classified into the target fission subclass; If not, then create a new subclass for the first missed detection process.

7. The process classification method according to claim 5, characterized in that, In step A3, it is determined whether the process fingerprint similarity between the first missed detection process and each of the first processes to be classified in the target fission subclass within the fission subclass set is greater than the preset similarity threshold: If so, the first missed detection process is classified into the target fission subclass; If not, then create a new subclass for the first missed detection process.

8. The process classification method according to claim 2, characterized in that, The process category increment generation step includes: Step B1: Based on a set of static feature classification rules, each second process to be classified in the second process set is classified into one of a plurality of second coarse process categories, and the second process to be classified that is not covered by any of the static feature classification rules is marked as a second missed process. Step B2: For each second process to be classified into any second coarse process category, calculate the process fingerprint similarity between the second process to be classified and each first process to be classified under the corresponding first coarse process category in the initial process category library, and classify the second process to be classified into the corresponding fission subclass or create a new subclass for the second process to be classified according to the relationship between the process fingerprint similarity and the preset similarity threshold, so as to generate an updated process category library; Step B3: For each of the second missed processes, calculate the process fingerprint similarity between the second missed process and each fission subclass in the updated process category library, and classify the second missed process into the corresponding fission subclass or create a new subclass for the second missed process according to the relationship between the process fingerprint similarity and the preset similarity threshold, so as to obtain the final process category library.

Images