Process grouping method
By using process fingerprint generation and clustering algorithms, process knowledge is managed automatically, solving the problems of low efficiency, strong subjectivity and insufficient flexibility of traditional process grouping technology, and realizing efficient and accurate process grouping analysis and management.
Patent Information
- Application Number
- CN202511461485.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-14
- Publication Date
- 2026-03-03
AI Technical Summary
Traditional process grouping technology relies on manual coding, resulting in low efficiency, strong subjectivity, poor consistency, and insufficient flexibility. It cannot adapt to the digital and intelligent needs of modern manufacturing and is difficult to achieve high-efficiency, high-precision, and highly adaptive process knowledge management.
By generating process fingerprints from process diagrams and using clustering algorithms and similarity calculations, process diagrams are automatically classified and updated, creating a process category library, thus achieving automated organization and rapid response of process knowledge.
It improves the accuracy of process retrieval and similar process recommendation, enhances the scalability and adaptability of the process category library, and can quickly respond to the dynamic growth of process data, realizing automatic, intelligent and efficient process group analysis.
Smart Images
Figure CN121598129A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of process analysis, and more specifically, to a process grouping method. Background Technology
[0002] In manufacturing, Computer-Aided Process Design (CAPP) serves as a bridge connecting product design and production. One of its core objectives is to efficiently manage and utilize existing process knowledge. Group Technology (GT), a key method for achieving this goal, aims to classify and group components or process procedures with similar structures or processes into process families. This facilitates process retrieval, reuse, and standardized production, thereby reducing repetitive work and improving process design efficiency and consistency.
[0003] Traditional process grouping technologies generally rely on manual coding systems. In such systems, process engineers need to manually assign a classification code to each process or part based on factors such as the geometry, size, and processing characteristics of the parts, referring to a complex set of classification coding rules. Subsequently, the manual coding system performs similarity grouping and retrieval based on these manually assigned classification codes.
[0004] However, this traditional method based on manual coding has many inherent flaws and is no longer suitable for the development needs of digitalization and intelligentization in modern manufacturing. The flaws are as follows: Inefficient and reliant on expert experience: The coding process heavily depends on the professional knowledge and experience of process engineers, which is not only time-consuming and labor-intensive, but also, in the context of a large-scale process library, manual coding becomes a bottleneck for process data entry, making it difficult to achieve rapid accumulation and digital management of process knowledge. High subjectivity and poor consistency: Different engineers may have different understandings and coding of the same process, resulting in highly subjective classification results and inconsistent standards, which greatly reduces the accuracy and reusability of process retrieval. Insufficient flexibility and inability to adapt to dynamic changes: Once a traditional classification and coding system is established, it is relatively fixed. When new process types emerge or existing processes evolve, the original coding system may not be able to effectively cover and represent them. It lacks adaptability and scalability. For newly generated processes, the system cannot automatically and intelligently classify them into existing categories or create new categories. Manual intervention and adjustment of coding rules are required, which is cumbersome and slow to respond. Insufficient depth makes accurate matching difficult: Simple encoding based on surface features cannot deeply and comprehensively reflect the inherent and complex topological sequence logic of the process (such as the arrangement and dependency of processes and steps), resulting in coarse similarity calculations and making it difficult to achieve truly accurate clustering and intelligent recommendation.
[0005] In summary, there is an urgent need for a process grouping method that can automatically, intelligently, and efficiently perform group analysis of processes. This method can overcome the drawbacks of manual coding, directly extract essential features from the process data itself, and realize the automated initialization and construction of the process library and online real-time incremental updates, so as to meet the requirements of modern intelligent manufacturing for high efficiency, high precision, and high adaptability in process knowledge management. Summary of the Invention
[0006] The technical problem to be solved by this invention is how to achieve automatic, intelligent and efficient group analysis of processes to meet the requirements of high efficiency, high precision and high adaptability of process knowledge management in modern intelligent manufacturing. In order to overcome the defects of the above-mentioned prior art (or related technology), this invention provides a process grouping method.
[0007] This invention provides a process grouping method, comprising the following steps: Step S1: For the pre-established initial process library, generate corresponding initial process fingerprints based on multiple initial process diagrams stored in the initial process library, perform clustering processing on each initial process fingerprint to obtain initial process categories and store them in the process category library; Step S2: Obtain at least one real-time process diagram, extract features from the real-time process diagram, and generate a real-time process fingerprint; Step S3: For each initial process category, calculate the corresponding category similarity based on the real-time process fingerprint and any initial process fingerprint in the initial process category; Step S4: Determine whether the category similarity is greater than a preset threshold corresponding to the initial process category. If so, the real-time process fingerprint is assigned to the category of the initial process category; If not, a new process category, distinct from each of the initial process categories, is created in the process category library, and the real-time process fingerprint is assigned to the new process category.
[0008] Compared with the prior art, the process grouping method of the present invention has the following advantages: In this invention, step S1 converts process diagrams into process fingerprints and performs automated clustering, overcoming the inefficiencies and subjectivity of traditional methods that rely on manual coding and classification. This achieves automated organization and archiving of process knowledge. Steps S2-S4 update the process category library, enabling not only rapid classification and archiving of newly acquired real-time process diagrams but also automatic creation of new process categories upon discovery of entirely new processes. This significantly enhances the scalability and adaptability of the process category library, meeting the needs of dynamic growth and rapid response of process data in modern manufacturing. Clustering and classification based on process fingerprint similarity calculations allow for matching from deep process features, which is more accurate than keyword-based or simple coding-based matching methods. This significantly improves the accuracy of process retrieval and similar process recommendation, thereby achieving automatic, intelligent, and efficient group analysis of processes.
[0009] In one possible implementation, in step S1, for each initial process drawing, feature extraction is performed on the initial process drawing to obtain initial topology features, initial shape features, initial material type features, and initial material state features, and the initial process fingerprint is formed by processing the initial topology features, initial shape features, initial material type features, and initial material state features according to the initial topology features, initial shape features, initial material type features, and initial material state features.
[0010] In one possible implementation, the initial topology features include an initial process number, an initial step length, an initial step number, an initial process length, and an initial process number sequence composed of each initial process number; the initial shape features include an initial component shape number; the initial material type features include an initial component material type number; and the initial material state features include an initial component material state number. In step S1, the initial directory is formed based on the initial component shape number, the initial component material type number, the initial component material state number, and the initial process length; the initial process information is formed based on the initial process number and the initial step length; the initial step information is formed based on the initial step number; and the initial process number sequence is subjected to fuzzy hashing to obtain the initial process hash.
[0011] In one possible implementation, in step S2, feature extraction is performed on the real-time process diagram to obtain real-time topology features, real-time shape features, real-time material type features, and real-time material state features. Based on the real-time topology features, real-time shape features, real-time material type features, and real-time material state features, real-time catalog, real-time process hash, real-time process information, and real-time process step information are processed to form the real-time process fingerprint.
[0012] Compared with existing technologies, by adopting the above technical solution, a multi-dimensional process fingerprint can be constructed by comprehensively extracting topological features, shape features, material type features, and material state features. This can comprehensively and profoundly depict the essence of a process and provide a reliable data foundation for subsequent high-precision similarity comparison.
[0013] In one possible implementation, the real-time topology features include a real-time process number, a real-time step length, a real-time step number, a real-time process length, and a real-time process number sequence composed of the real-time process numbers for each real-time process. The real-time shape features include real-time component shape numbers. The real-time material type features include real-time component material type numbers. The real-time material status features include real-time component material status numbers. In step S2, the real-time directory is composed based on the real-time component shape number, the real-time component material type number, the real-time component material status number, and the real-time process length. The real-time process information is composed based on the real-time process number and the real-time step length. The real-time step information is composed based on the real-time step number. The real-time process number sequence is then subjected to fuzzy hashing to obtain the real-time process hash.
[0014] Compared with existing technologies, the above-mentioned technical solution can further refine how various features are specifically transformed into the components of the process fingerprint, making the technical solution clear, explicit and easy to implement. Furthermore, by generating process hashes through fuzzy hashing of the process number sequence, even if two processes have slight differences in individual processes, their process hashes can still maintain a high degree of similarity. This greatly improves the robustness and practicality of the clustering algorithm in practical applications and avoids the situation where minor changes are misjudged as completely different processes.
[0015] In one possible implementation, in step S2, the real-time process number sequence is composed of n real-time process numbers in sequence, and a corresponding hash function is constructed based on each three adjacent real-time process numbers. Then, the hash functions are combined to form the real-time process hash.
[0016] Compared with existing technologies, the above technical solution can transform the real-time process number sequence into a hash set by using a sliding window hashing method. This method can effectively capture the local pattern and sequence features of the real-time process number sequence, and the generated hash set is very suitable for calculating set similarity indicators such as Jaccard similarity, thereby quickly and accurately measuring the similarity between two process fingerprints.
[0017] In one possible implementation, in step S2, each real-time step has a length of 1 character, each real-time process number has a length of 2 characters, and the real-time process information composed of each real-time step length and each real-time process number is a 3n-byte string, where n represents the total number of real-time processes.
[0018] In one possible implementation, in step S2, the real-time step information is composed of a sequence of real-time step numbers for each real-time process, and the real-time step information is 2(L1+L2+...L...). n A string of length L, where L n This indicates the length of the nth real-time process step number sequence.
[0019] In one possible implementation, in step S2, the real-time component shape number is 2 characters, the real-time component material type number is 2 characters, the real-time component material status number is 2 characters, the real-time process length is 2 characters, and the real-time directory is an 8-byte string.
[0020] Compared with existing technologies, the above technical solution can limit the character length of each component, making each process fingerprint a compact string of fixed or predictable length, which greatly reduces the occupation of computer storage space and has higher efficiency in network transmission and memory calculation. Attached Figure Description
[0021] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation
[0022] 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.
[0023] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0024] See Figure 1 This invention discloses a process grouping method, comprising: Step S1: For the pre-established initial process library, generate corresponding initial process fingerprints based on multiple initial process diagrams stored in the initial process library, perform clustering processing on each initial process fingerprint to obtain initial process categories and store them in the process category library; Step S2: Obtain at least one real-time process diagram, extract features from the real-time process diagram, and generate a real-time process fingerprint. Step S3: For each initial process category, calculate the corresponding category similarity based on the real-time process fingerprint and any initial process fingerprint in the initial process category. Step S4: Determine whether the category similarity is greater than the preset threshold corresponding to the initial process category. If so, the real-time process fingerprint will be assigned to the category of the initial process category; If not, a new process category, distinct from each initial process category, is created in the process category library, and the real-time process fingerprint is assigned to the new process category.
[0025] In this embodiment of the invention, in step S1, for each initial process drawing, feature extraction is performed on the initial process drawing to obtain initial topology features, initial shape features, initial material type features, and initial material state features. The initial topology features, initial shape features, initial material type features, and initial material state features are then processed to obtain an initial directory, initial process hash, initial process information, and initial process step information to form the initial process fingerprint. The initial topology features include the initial process number, initial process step length, initial process number, initial process length, and an initial process number sequence composed of each initial process number. The initial shape features include the initial component shape number. The initial material type features include the initial component material type number. The initial material state features include the initial component material state number. In step S1, the initial directory is formed based on the initial component shape number, the initial component material type number, the initial component material state number, and the initial process length. The initial process information is formed based on the initial process number and the initial process step length. The initial process step information is formed based on the initial process step number. The initial process hash is obtained by performing fuzzy hashing on the initial process number sequence.
[0026] In this embodiment of the invention, in step S2, feature extraction is performed on the real-time process diagram to obtain real-time topology features, real-time shape features, real-time material type features, and real-time material state features. These features are then processed to obtain a real-time directory, a real-time process hash, real-time process information, and real-time step information, forming the real-time process fingerprint. The real-time topology features include the real-time process number, real-time step length, real-time step number, real-time process length, and a real-time process number sequence composed of all the real-time process numbers. The real-time shape features include real-time component shape numbers. The real-time material type features include real-time component material type numbers. The real-time material state features include real-time component material state numbers. In step S2, the real-time directory is formed based on the real-time component shape numbers, real-time component material type numbers, real-time component material state numbers, and real-time process lengths. The real-time process information is formed based on the real-time process numbers and real-time step lengths. The real-time step information is formed based on the real-time step numbers. Finally, the real-time process hash is obtained by performing fuzzy hashing on the real-time process number sequence.
[0027] In this embodiment of the invention, the essence of process grouping based on process fingerprints is to automatically cluster / classify the initial process diagrams in the initial process library according to the similarity of process fingerprints. This process grouping technology consists of two parts: the initialization of the process category library and the incrementalization of the process category library. The initialization of the process category library involves generating a process category library from the initial process library, while the incrementalization of the process category library involves assigning new process fingerprints to existing initial process categories or generating new process categories. Both parts use two algorithms: a process fingerprint similarity calculation algorithm and a process fingerprint clustering algorithm. The calculation result of the process fingerprint similarity calculation algorithm is 0% to 100%, and the logic of the process fingerprint clustering algorithm is that if the similarity between two process fingerprints is greater than a preset threshold, they are classified into the same category; otherwise, they are classified into different categories.
[0028] In this embodiment of the invention, there are many algorithms for calculating process fingerprint similarity. One implementation method is listed here: The process fingerprint consists of four parts: <directory, process hash, process information, and process step information>. Although the directory, process information, and process step information can also be used to calculate the similarity between two process fingerprints, the essence of this invention is not to calculate similarity, but to use process similarity to cluster and obtain process grouping technology. Therefore, this embodiment can provide a very simple method for calculating process fingerprint similarity without affecting the overall idea of process grouping technology. This algorithm only uses process hashing to calculate similarity, assuming the process number sequence is C1C2C3...Cn Its process hash is a string H(C1C2C3) H(C2C3C4)…H(C n-2 C n-1 C n H(s) is a hash function that converts 6 characters into 4 characters, transforming the process hash into a string set where each member is a four-byte character derived from H(C). k C k+1 C k+2 ), where the size of the string set is n-2, and the calculation formula is: in, Indicates similarity. and This represents two process fingerprints: a real-time process fingerprint and an initial process fingerprint. and This represents the set of strings obtained from the hash transformations of the two processes.
[0029] In this embodiment of the invention, a K-Medoids clustering algorithm that can be used for clustering is listed. K process fingerprints are randomly selected as initial Medoids. Each process fingerprint is assigned to the cluster containing the Medoid that is most similar to it (with the largest SIM value). For each cluster, the Medoid and another process fingerprint are swapped. The total similarity after the swap is calculated. If the total similarity increases, the swap is retained; if it does not increase, it is deleted. The above actions are repeated until the Medoids no longer change.
[0030] In this embodiment of the invention, each process is numbered, and the number consists of two bytes; each step is also numbered, and the number also consists of two bytes. The process sequence is the process route, for example, blanking -> rough turning -> heat treatment -> finish turning -> rough milling -> finish milling -> grinding the outer diameter -> inspection. The step sequence is the process. Shape characteristics include shafts, discs, housings, brackets, gears, sheet metal parts, castings, forgings, injection molded parts, etc.; material type characteristics include steel (carbon steel, alloy steel), cast iron, aluminum alloy, copper alloy, plastics, composite materials, etc.; material state characteristics include bars, forgings, castings, profiles, plates, etc.
[0031] In this embodiment of the invention, the process hash is a fuzzy hash, generated from the process number sequence. A hash function H(s) that can generate 4 bytes from 6 bytes is constructed or selected. Assuming the process number sequence is C1C2C3...C n Its fuzzy hash is a string H(C1C2C3) H(C2C3C4)…H(C n-2 C n-1 Cn The length of the fuzzy hash is 4(n-2) bytes.
[0032] In this embodiment of the invention, the process information is also a string of characters, composed of a sequence of <process number, step length>, where the step length is 1 character, and each <process number, step length> is 3 characters. Since the process consists of n processes, the process information is a string of 3n bytes. The step information is also a string of characters, composed of a sequence of step numbers for each process, that is, <step number sequence>1, <step number sequence>2, ..., <step number sequence> n Among them, <work step number sequence> k Let L be the sequence of steps for the k-th process, and the string length of the step information is 2(L1+L2+...L). n ), where L k The length of the process step sequence for the k-th process; the directory is also composed of the part shape number, material type number, material status number, and process length n (2 characters), for a total of 8 characters; the total length of the process fingerprint string is 7n+2(L1+L2+...L n )=8+4(n-2) + 3n+2(L1+L2+...L n ).
[0033] In the description of this invention, the references to "one embodiment," "some embodiments," "in this embodiment," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is 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, structures, 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.
[0034] 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. Step S1: For the pre-established initial process library, generate corresponding initial process fingerprints based on multiple initial process diagrams stored in the initial process library, perform clustering processing on each initial process fingerprint to obtain initial process categories and store them in the process category library; Step S2: Obtain at least one real-time process diagram, extract features from the real-time process diagram, and generate a real-time process fingerprint; Step S3: For each initial process category, calculate the corresponding category similarity based on the real-time process fingerprint and any initial process fingerprint in the initial process category; Step S4: Determine whether the category similarity is greater than a preset threshold corresponding to the initial process category. If so, the real-time process fingerprint is assigned to the category of the initial process category; If not, a new process category, distinct from each of the initial process categories, is created in the process category library, and the real-time process fingerprint is assigned to the new process category.
2. The process grouping method according to claim 1, characterized in that, In step S1, for each initial process diagram, feature extraction is performed on the initial process diagram to obtain initial topology features, initial shape features, initial material type features, and initial material state features. Based on the initial topology features, initial shape features, initial material type features, and initial material state features, the initial directory, initial process hash, initial process information, and initial process step information are processed to form the initial process fingerprint.
3. The process grouping method according to claim 2, characterized in that, The initial topology features include the initial process number, initial step length, initial step number, initial process length, and an initial process number sequence composed of the initial process numbers for each initial process. The initial shape features include the initial component shape number. The initial material type features include the initial component material type number. The initial material state features include the initial component material state number. In step S1, the initial directory is formed based on the initial component shape number, the initial component material type number, the initial component material state number, and the initial process length. The initial process information is formed based on the initial process number and the initial step length. The initial step information is formed based on the initial step number. The initial process number sequence is then subjected to fuzzy hashing to obtain the initial process hash.
4. The process grouping method according to claim 1, characterized in that, In step S2, feature extraction is performed on the real-time process diagram to obtain real-time topology features, real-time shape features, real-time material type features, and real-time material state features. Based on the real-time topology features, real-time shape features, real-time material type features, and real-time material state features, the real-time catalog, real-time process hash, real-time process information, and real-time process step information are processed to form the real-time process fingerprint.
5. The process grouping method according to claim 4, characterized in that, The real-time topology features include the real-time process number, real-time step length, real-time step number, real-time process length, and a real-time process number sequence composed of the real-time process numbers for each real-time process. The real-time shape features include real-time component shape numbers. The real-time material type features include real-time component material type numbers. The real-time material status features include real-time component material status numbers. In step S2, the real-time directory is composed based on the real-time component shape number, the real-time component material type number, the real-time component material status number, and the real-time process length. The real-time process information is composed based on the real-time process number and the real-time step length. The real-time step information is composed based on the real-time step number. The real-time process number sequence is then subjected to fuzzy hashing to obtain the real-time process hash.
6. The process grouping method according to claim 5, characterized in that, In step S2, the real-time process number sequence is composed of n real-time process numbers in sequence. A corresponding hash function is constructed based on each of three adjacent real-time process numbers, and then the hash functions are combined to form the real-time process hash.
7. The process grouping method according to claim 5, characterized in that, In step S2, each real-time step has a length of 1 character, each real-time process number has a length of 2 characters, and the real-time process information composed of the length of each real-time step and the number of each real-time process is a string of 3n bytes, where n represents the total number of real-time processes.
8. The process grouping method according to claim 5, characterized in that, In step S2, the real-time step information is composed of a sequence of real-time step numbers for each real-time process, and the real-time step information is 2(L1+L2+...L...). n A string of length L, where L n This indicates the length of the nth real-time process step number sequence.
9. The process grouping method according to claim 5, characterized in that, In step S2, the real-time component shape number is 2 characters, the real-time component material type number is 2 characters, the real-time component material status number is 2 characters, the real-time process length is 2 characters, and the real-time directory is an 8-byte string.