Server manufacturability intelligent evaluation MES system based on machine learning

Abstract

The invention relates to the technical field of server manufacturing, and discloses a server manufacturability intelligent evaluation MES system based on machine learning, and the system comprises a data collection module which is used for obtaining, classifying and marking the operation time data of each process in real time; the standard library management module is used for storing a standard library and supporting data management; the man-hour comparison module is used for calling standard man-hours and comparing the standard man-hours with real-time data; the problem processing module triggers a feedback process and generates a standardized problem description; the evaluation scoring module is used for quantitatively scoring the manufacturability of the model machine based on a standard; and the standard updating module updates a design standard through a machine learning model. According to the system, intelligent evaluation of the manufacturability of the server is realized, the data acquisition accuracy, the standard library dynamic updating capability, the problem processing normalization and the evaluation scoring objectivity are improved, and the manufacturing efficiency and the manufacturability level of the server are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of server manufacturing, specifically to a server manufacturability intelligent evaluation MES system based on machine learning. BACKGROUND

[0002] In the server manufacturing industry, with the continuous change of market demand and the rapid development of technology, the manufacturability requirements of servers are becoming higher and higher. The server prototype production process is a key link to verify its manufacturability, but the traditional evaluation method has many shortcomings. The traditional server manufacturability evaluation mainly relies on manual work. In terms of data collection, it is difficult to accurately obtain the operation time data of each process in real time, and manual recording is prone to errors and omissions, making it difficult to effectively classify and mark the data, which makes the subsequent analysis and evaluation lack reliable data support.

[0003] In terms of standard library management, the traditional system stores relatively fixed server manufacturability design standard library and standard time library of each process, which is not updated in time and is difficult to adapt to the changing production process and technical requirements. When new problems or process improvements occur during production, the standard library cannot be updated in time, resulting in a disconnection between the evaluation standard and the actual production.

[0004] In the process of time comparison, the traditional method often uses simple numerical comparison without considering the differences in process characteristics of different processes, making it difficult to accurately find the standard time corresponding to the current process and set a reasonable error range, resulting in inaccurate comparison results and making it difficult to accurately judge whether there are problems in the production process.

[0005] When the operation time exceeds the allowed error range, the traditional problem handling method lacks effective feedback processes and standardized problem description generation mechanisms. The production staff's input of problem types and specific descriptions is often not standardized and uniform, which is not conducive to problem analysis and solution, and it is also difficult to accumulate problem data for subsequent improvement.

[0006] In the evaluation scoring link, the traditional evaluation method usually uses simple qualitative evaluation or single-index quantitative evaluation, which cannot comprehensively and objectively quantify the manufacturability of the prototype. Lack of scientific weight coefficient setting and multi-index comprehensive evaluation, cannot accurately reflect the actual manufacturability level of the server in the trial production process.

[0007] In addition, the traditional system cannot use historical data and machine learning technology to optimize and update the standard library. It cannot dynamically adjust and optimize the server manufacturability design standards according to the problem list formed during the trial production process and historical evaluation data, resulting in poor adaptability and intelligence of the system, making it difficult to continuously improve the manufacturability level of the server.

[0008] Traditional server manufacturability evaluation methods have problems such as non-real-time data collection, untimely standard library updates, inaccurate labor time comparison, irregular problem handling, incomplete evaluation and scoring, and lack of intelligent optimization. There is an urgent need for an intelligent evaluation MES system based on machine learning to solve these problems and improve the manufacturability and production efficiency of servers. Summary of the Invention

[0009] The purpose of the present invention is to provide a server manufacturability intelligent evaluation MES system based on machine learning to solve the problems raised in the above background technology.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a machine learning-based server manufacturability intelligent evaluation MES system, the system comprising:

[0011] The data acquisition module is used to obtain the operation time data of each process in real time during the trial production of the server prototype, and classify and mark the operation time data according to the trial production process;

[0012] The standard library management module is used to store the server's manufacturability design standard library and the standard working time library for each process, and supports the addition, deletion, modification and query operations of the data in the standard library;

[0013] The working time comparison module is used to retrieve the standard working time and allowable error range corresponding to the current process from the standard library management module, and compare the real-time collected operation time data with the standard working time interval;

[0014] The problem handling module is used to trigger the problem feedback process when the operation time exceeds the allowable error range. It receives the problem type and specific description input by the production staff and generates a standardized problem description through the natural language processing model;

[0015] The evaluation and scoring module is used to quantitatively score the manufacturability of the prototype based on the manufacturability design standards in the standard library management module after the entire trial production process is completed;

[0016] The standard update module is used to generate updated server manufacturability design standards through machine learning models based on the problem list and historical evaluation data generated during the trial production process.

[0017] Preferably, the data acquisition module is implemented as follows:

[0018] For any process in the trial production process, the start and end timestamps of the process are obtained through sensors deployed on the production equipment;

[0019] Calculate the difference between the start timestamp and the end timestamp to get the actual operation time of the process;

[0020] Perform noise filtering on the actual operation time, eliminate abnormal values ​​caused by equipment abnormalities or operational errors, and form effective operation time data;

[0021] The effective operation time data is classified and marked according to the process number, prototype model, and production batch, and passed to the working time comparison module as input data.

[0022] Preferably, the implementation method of performing noise filtering on the actual operation time further includes:

[0023] Obtain the operating time distribution range of the same process in historical trial production data;

[0024] Calculate the deviation between the current actual operation time and the mean of the historical distribution range. If the deviation exceeds the preset threshold, it is determined to be an outlier.

[0025] Outliers are marked as invalid data and stored in isolation, and only the actual operation time with deviation within the threshold range is retained as valid data.

[0026] Preferably, the implementation method of retrieving the standard working hours corresponding to the current process also includes:

[0027] Extract the process feature tags of the current process, including process name, operation type, and number of parts involved;

[0028] In the process standard working time library of the standard library management module, search for completely identical standard processes by matching feature tags;

[0029] If no completely identical standard process is found, a similar process with the highest similarity in process feature tags is searched, and its standard working hours are retrieved as a comparison benchmark.

[0030] Preferably, the problem handling module is implemented as follows:

[0031] When the operation time exceeds the allowable error range, a problem type selection menu is displayed on the production terminal interface. The menu options include assembly complexity issues, component compatibility issues, and missing operation instructions issues.

[0032] If the production employee selects a custom problem type, a text input box is triggered to receive a free description;

[0033] The pre-trained BERT natural language processing model is used to extract keywords and perform semantic analysis on free description texts, generating standardized problem descriptions that include the root causes and impact aspects of the problem.

[0034] The standardized problem description is associated with the original input content and stored, and pushed to the process engineer terminal for confirmation.

[0035] Preferably, the implementation method of generating a standardized problem description includes:

[0036] Extract core problem words, process-related words, and result-affecting words from free description texts;

[0037] Match the core problem words with the problem keyword library in the manufacturability design standard library to determine the problem classification label;

[0038] Combining the words related to the process and the words affecting the results, a standardized description is generated according to the structured template of "[Problem classification label]: [Involved process] leads to [Impact result]".

[0039] Preferably, the implementation of the evaluation and scoring module includes:

[0040] Obtain the weight coefficients of each evaluation index in the manufacturability design standard from the standard library management module. The evaluation indexes include the man-hour compliance rate, the problem occurrence rate, and the operation fluency.

[0041] Calculate the percentage of processes in the entire trial production process whose operation time is within the allowable error range as the labor time compliance score;

[0042] The ratio of the number of problems confirmed to have passed in the problem handling module to the total number of processes is calculated as the problem occurrence rate score;

[0043] Analyze the number of operation interruptions and recovery time through production staff operation logs to calculate the operation fluency score;

[0044] The three scores are weighted and summed according to the weight coefficient of each indicator to obtain the final score of the prototype manufacturability.

[0045] Preferably, the implementation of calculating the operation fluency score includes:

[0046] Extract the operation start time, operation interruption time, and operation recovery time from the production staff's operation log;

[0047] Calculate the ratio of the interruption time of a single operation to the total operation time to obtain the fluency of the single operation;

[0048] The average of the single operation fluency of the entire process is calculated to obtain the operation fluency score.

[0049] Preferably, the standard update module is implemented as follows:

[0050] Collect all the problem lists confirmed by design engineers during the historical trial production process and the corresponding design improvement plans;

[0051] Associate and label the problem types, standardized descriptions, and improvement plans in the problem list to form a training dataset;

[0052] Use the training dataset to supervise the initial machine learning model and optimize the model's feature extraction and pattern recognition capabilities;

[0053] The latest problem list from the pilot production was input into the optimized model to generate an updated DFM standard that included new evaluation metrics, new work-hour standards, and new problem classifications.

[0054] Preferably, the method for generating an updated version of the design for manufacturability standard further includes:

[0055] Verify the effectiveness of the newly generated evaluation indicators by selecting prototypes with completed design improvements from historical trial production data, re-scoring them using the new indicators, and comparing them with the actual improvement results.

[0056] If the verification pass rate exceeds the preset threshold, the new evaluation indicators, new working time standards, and new problem classifications will be merged into the manufacturability design standard library of the standard library management module.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] In terms of data collection, sensors deployed on production equipment are used to obtain the start and end timestamps of each process in real time, calculate the actual operation time and perform noise filtering. This can accurately eliminate abnormal values ​​caused by equipment abnormalities or operational errors, and also mark valid data by categories such as process numbers, providing real-time, accurate and standardized data support for subsequent analysis, solving the problem of large errors and untimely collection of traditional manual data.

[0059] The standard library management module supports the addition, deletion, modification and query of the server manufacturability design standard library and the standard working time library of each process. Combined with the standard update module, it uses machine learning models to generate updated design standards based on the trial production problem list and historical evaluation data. The new indicators are merged into the standard library after validity verification, realizing the dynamic update of the standard library, so that the evaluation standards can keep pace with production reality, and overcoming the defects of the traditional standard library that is fixed and lagging.

[0060] The labor time comparison module extracts the process feature labels of the current process and first matches the completely identical standard process. If no match is found, it looks for the similar process with the highest similarity to retrieve the standard labor time. This intelligent matching method based on feature labels fully considers the differences in processes, ensures the accuracy of the comparison benchmark, and is more scientific and reasonable than traditional simple numerical comparison.

[0061] The problem handling module displays a problem type selection menu when the operation time exceeds the tolerance. When customizing problems, the BERT model is called to generate a standardized description and store it in association with the original content for push confirmation. This standardizes the problem feedback process and description, facilitates problem analysis and resolution, and data accumulation, providing a good foundation for subsequent improvements and changing the non-standard situation of traditional problem handling.

[0062] The evaluation scoring module obtains the weight coefficients of each evaluation index from the standard library, calculates the work hour compliance rate, problem occurrence rate and operation fluency score, and weightedly sums to obtain the final score. The operation fluency score is calculated by analyzing the operation log, which comprehensively and objectively quantifies the prototype manufacturability, avoids the one-sidedness of traditional single index evaluation, and can more accurately reflect the actual manufacturability level.

[0063] The standard updating module collects the problem list and improvement scheme confirmed by the design engineer, associates and labels to form a training data set for supervised learning of the machine learning model, and then inputs the latest problem list to generate an updated version of the standard, realizes intelligent optimization of the standard, enables the system to have self-learning and continuous improvement capability, continuously improves the accuracy and adaptability of the server manufacturability evaluation, and improves the intelligent level and production efficiency of server manufacturing. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 The working principle diagram of the server manufacturability intelligent evaluation MES system based on machine learning described in the application;

[0065] Figure 2 The working principle diagram of the data acquisition module;

[0066] Figure 3 The working principle diagram of the noise filtering process;

[0067] Figure 4 The working principle diagram of calculating the operation fluency score;

[0068] Figure 5 The working principle diagram of the standard updating module. DETAILED DESCRIPTION

[0069] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0070] Please refer to Figures 1-5 The application relates to a server manufacturability intelligent evaluation MES system based on machine learning, which comprises a data acquisition module, a standard library management module, a work hour comparison module, a problem processing module, an evaluation scoring module and a standard updating module. The specific implementation is as follows:

[0071] The data acquisition module is used to obtain real-time operating time data for each process during the trial production of server prototypes and to classify and label the operating time data according to the trial production process. For any process in the trial production process, the start and end timestamps of the process are obtained through sensors deployed on the production equipment. The difference between the start and end timestamps is calculated to obtain the actual operating time of the process. The actual operating time is then subjected to noise filtering. Specifically, the operating time distribution range of the same process in the historical trial production data is obtained, and the deviation between the current actual operating time and the mean of the historical distribution range is calculated. If the deviation exceeds the preset threshold, it is determined to be an outlier. The outlier is marked as invalid data and stored in isolation. Only the actual operating time with a deviation within the threshold range is retained as valid data. The valid operating time data is then classified and labeled according to the process number, prototype model, and production batch, and passed as input data to the working time comparison module.

[0072] The standard library management module is used to store the server's manufacturability design standard library and the standard working time library for each process, and supports the addition, deletion, modification and query operations on the data in the standard library.

[0073] The work time comparison module retrieves the standard work time and tolerance range corresponding to the current process from the standard library management module, and compares the real-time collected operation time data with the standard work time interval. This is achieved by extracting the process feature tags of the current process, including the process name, operation type, and number of parts involved. The feature tag matching is used to search for an identical standard process in the process standard work time library of the standard library management module. If no identical standard process is found, a similar process with the highest similarity in process feature tags is searched for and its standard work time is used as the comparison benchmark.

[0074] The problem handling module triggers the problem feedback process when the operation time exceeds the allowable error range. It receives the problem type and detailed description input by the production employee and generates a standardized problem description using a natural language processing model. When the operation time exceeds the allowable error range, a problem type selection menu appears on the production terminal interface. The menu options include assembly complexity issues, component compatibility issues, and missing operating instructions issues. If the production employee selects a custom problem type, a text input box is triggered to receive a free description. The pre-trained BERT natural language processing model is called to perform keyword extraction and semantic analysis on the free description text. Specifically, the core problem words, process-related words, and result-affecting words in the free description text are extracted. The core problem words are matched with the problem keyword library in the manufacturability design standard library to determine the problem classification label. Combined with the process-related words and result-affecting words, a standardized description is generated according to the structured template "[Problem classification label]: [Process-related] leads to [Result-affecting]". The standardized problem description is associated with the original input content and stored and pushed to the process engineer terminal for confirmation.

[0075] After the pilot production process is complete, the evaluation and scoring module quantitatively scores the prototype's manufacturability based on the design for manufacturability standards in the standard library management module. The module obtains the weight coefficients for each evaluation metric in the design for manufacturability standards, including man-hour compliance, problem occurrence rate, and operational fluency. The module calculates the proportion of processes whose operating time falls within the allowable error range as the man-hour compliance score. The problem occurrence rate score is calculated by calculating the ratio of the number of confirmed problems to the total number of processes in the problem handling module. The operational fluency score is calculated by analyzing the number of interruptions and recovery times from production employee operation logs. Specifically, the module extracts the operation start time, interruption time, and recovery time from the production employee operation logs and calculates the ratio of the interruption time to the total operation time to obtain the single operation fluency score. The single operation fluency scores for all processes are then averaged to obtain the operational fluency score. Finally, the three scores are weighted and summed according to the weight coefficients of each metric to obtain the final prototype manufacturability score.

[0076] The standard updating module is used to generate updated server design for manufacturability standards based on the problem list formed during the trial production process and historical evaluation data. The problem list confirmed by the designed engineer and the corresponding design improvement scheme in the historical trial production process are collected, the problem type, standardized description and improvement scheme in the problem list are associated and labeled to form a training data set, the initial machine learning model is supervised learning using the training data set, the feature extraction and pattern recognition ability of the model is optimized, the latest trial production problem list is input into the optimized model to generate an updated manufacturability design standard including new evaluation indicators, new work time standards and new problem classification. After generating the updated manufacturability design standard, the effectiveness of the new evaluation indicators is verified, the verification method is to select the prototype that has completed design improvement in the historical trial production data, re-score using the new indicators and compare with the actual improvement effect, if the verification rate exceeds the preset threshold, the new evaluation indicators, new work time standards and new problem classification are merged into the manufacturability design standard library of the standard library management module.

[0077] Embodiment 1:

[0078] The implementation of the data collection module is as follows: In the server prototype trial production process, the core function of this module is to obtain the working time data of each process in real time, and to classify and label these data according to the trial production process. For any process in the trial production, first, the start time stamp and the end time stamp of the process are obtained through the sensors deployed on the production equipment. The sensors mentioned here need to have accurate time recording capability, which can accurately capture the time information at the moment of process start and end, to ensure the accuracy of subsequent time calculation.

[0079] After obtaining the start time stamp and the end time stamp, the difference between the two time stamps needs to be calculated to obtain the actual working time of the process. This calculation process needs to use the corresponding time processing algorithm to ensure the accuracy of the time difference calculation and avoid the inaccuracy of subsequent data caused by calculation error.

[0080] In order to ensure the effectiveness of the obtained working time data, noise filtering processing needs to be performed on the actual working time. First, the working time distribution range of the same process in the historical trial production data needs to be obtained. The historical trial production data needs to cover enough samples to ensure that the obtained working time distribution range is representative. By collecting and organizing historical data, the fluctuation range, mean, standard deviation and other parameters of the working time of this process in the past trial production process can be calculated to determine a reasonable working time distribution range.

[0081] Next, the deviation of the current actual operating time from the mean of the historical distribution range is calculated. This deviation can be calculated by dividing the difference between the current actual operating time and the mean by the standard deviation. This standardizes the deviation and facilitates comparison with a preset threshold. If the deviation exceeds the preset threshold, the actual operating time is considered an outlier. This preset threshold needs to be appropriately set based on actual production conditions and historical data. It should be able to filter out outliers caused by equipment anomalies or operational errors, while preventing normal operating time from being misclassified as an outlier.

[0082] When an actual operation time is determined to be an outlier, it is marked as invalid data and stored separately. This separate storage prevents these outliers from interfering with subsequent data analysis and processing. Furthermore, only those actual operation times that fall within the threshold deviation are retained as valid data. This valid data serves as the basis for subsequent process analysis and processing, so its accuracy and reliability are crucial.

[0083] After obtaining valid operation time data, it is necessary to categorize and label this data by process number, prototype model, and production batch. The process number identifies the specific production process to which the data corresponds, facilitating subsequent process-by-process data analysis and comparison. The prototype model distinguishes production data for different server prototype models, enabling targeted analysis and improvements. The production batch records the production batch to which the data belongs, facilitating tracing production process issues and comparing batches.

[0084] After classification and labeling, this valid work time data is passed as input to the work time comparison module. During data transmission, data integrity and accuracy must be ensured to avoid data loss or errors during transmission. Reliable data transmission protocols and storage methods can be used to ensure that data is accurately transmitted to the work time comparison module, providing reliable data support for subsequent work time comparison and analysis.

[0085] Through the above series of operation steps, the data acquisition module can accurately and effectively obtain the operation time data of each process in the trial production process, and clean and organize the data, providing high-quality data input for the entire machine learning-based server manufacturability intelligent evaluation MES system, ensuring that subsequent modules can analyze and process based on reliable data, thereby achieving accurate evaluation and optimization of server manufacturability.

[0086] Example 2:

[0087] The implementation of the work time comparison module calling the standard work time corresponding to the current process is as follows: in the server prototype production process, the work time comparison module needs to accurately call the standard work time and the allowable error range matching the current process from the standard library management module, so as to realize the interval comparison of the real-time collected work time data. The core of this process is to ensure the high applicability and accuracy of the called standard work time through the extraction and matching of process feature tags.

[0088] For the current process being executed, the system needs to extract its process feature tags. These tags mainly include process name, operation type and involved component quantity. Among them, the process name needs to accurately correspond to the standard process name defined in the production process, such as "mainboard assembly", "cable connection" and the like, so as to ensure the accuracy of subsequent matching; the operation type can be divided into assembly, welding, testing and other different categories, which is used to refine the operation characteristics of the process; the involved component quantity refers to the number of parts involved in the execution of the process, which is of great significance to the evaluation of the complexity and work time consumption of the process. When extracting these feature tags, the system needs to obtain the relevant information of the current process from the production management database in real time, so as to ensure the accuracy and timeliness of the tags.

[0089] After extracting the process feature tags, the system will search in the process standard work time library of the standard library management module. The process standard work time library stores a large number of standard processes and their corresponding work time standards determined based on historical production data and process design, which have been accumulated and verified for a long time and have high authority and reference value. The first step of the search is to find a standard process whose process feature tags are exactly the same as those of the current process. For example, if the process feature tags of the current process are "mainboard assembly", "assembly" and "involved component quantity 5", the system will search the library to see if there is a completely identical tag combination.

[0090] If a standard process whose process feature tags are exactly the same as those of the current process is found in the process standard work time library, the system will directly call the standard work time and the allowable error range corresponding to the standard process. At this time, the called standard work time can be directly used as the comparison benchmark of the work time of the current process, and the allowable error range is used to judge whether the real-time collected work time is within a reasonable range.

[0091] If no standard process whose process feature tags are exactly the same as those of the current process is found, the system will enter the similar process searching process. At this time, the system will use the feature tag similarity algorithm to calculate the similarity of the process feature tags of all standard processes in the process standard work time library and the tags of the current process. The process of similarity calculation needs to consider the weight of each tag, for example, the weight of the process name may be higher than that of the involved component quantity, because the process name can better reflect the essential characteristics of the process. After calculating the similarity values of each standard process and the current process by the algorithm, the system will select the similar process with the highest similarity.

[0092] After identifying the most similar process, the system retrieves the standard working hours of that similar process as a comparison benchmark for the current process. It's important to note that since similar processes aren't exactly identical to the current process, their standard working hours may need to be adjusted based on actual conditions. For example, if the similar process involves two more parts than the current process, the system can adjust the retrieved standard working hours based on the relationship between part count and working hours in historical data to better align them with the actual conditions of the current process.

[0093] After accessing the standard working hours and the tolerance range, the working hours comparison module compares the real-time collected working time data with the standard working hours. Specifically, the system determines whether the real-time working time falls within the range of the standard working hours plus or minus the tolerance range. If it does, the working time for the process meets the standard. If it does, the problem handling process is triggered, and the problem handling module provides subsequent feedback and resolution.

[0094] Throughout the entire retrieval and comparison process, the system must ensure real-time and accurate data. The standard library management module must be updated in real time to reflect the latest process improvements and production data. Furthermore, the labor time comparison module must maintain real-time communication with the data acquisition module to ensure timely access to the latest work time data. Furthermore, the system must possess a certain level of fault tolerance, enabling appropriate processing and notification when data transmission delays or anomalies occur, to ensure the stability and reliability of the entire MES system.

[0095] Through the above detailed implementation methods, the labor time comparison module can accurately retrieve the standard labor time that matches the current process in a complex production environment, provide strong support for labor time monitoring and manufacturability evaluation during the trial production of server prototypes, and ensure the efficiency of the production process and the stability of product quality.

[0096] Example 3:

[0097] The implementation method of the problem handling module is as follows: During the trial production of the server prototype, when the working time comparison module determines that the operation time exceeds the allowable error range of the standard working hours, the problem handling module will be triggered. Its core function is to start the problem feedback process, collect the problem types and specific descriptions input by production employees, and generate standardized problem descriptions through the natural language processing model to achieve standardized management and subsequent processing of problems.

[0098] When the operation time exceeds the allowable error range, the system displays a problem type selection menu in real time on the production terminal interface. This menu presets several common problem types, including assembly complexity issues, component compatibility issues, and missing operating instructions. These preset options are based on a summary of common problem types during historical pilot production and cover most possible problem scenarios. Production employees can select the type from the menu that best matches the current problem based on actual production conditions.

[0099] If production staff feel the pre-set problem types don't accurately describe the issue they're facing, they can choose a custom problem type. The system then triggers a text input box for the staff to freely describe the issue. When designing the text input box, the system provides prompts to guide the staff in describing the issue, the process in which it occurred, and the extent of the impact, ensuring accurate analysis and resolution.

[0100] After obtaining the production employee's free description text, the system calls the pre-trained BERT natural language processing model to process the text. The BERT model has powerful semantic understanding and feature extraction capabilities, enabling it to accurately extract key information from free description text. First, the model performs pre-processing on the text, including word segmentation and part-of-speech tagging. Then, using a multi-layer Transformer architecture, it deeply understands the text's semantics, extracting key problem words, process-related words, and words that influence results.

[0101] Core problem words refer to words that can directly reflect the nature of the problem, such as "stuck", "loose", "incompatible", etc.; process-related words are used to locate the specific process where the problem occurs, such as "motherboard installation", "cable connection", etc.; impact words describe the impact of the problem on the production process or product quality, such as "extended operation time", "component damage", etc.

[0102] After extracting these keywords, the system matches the core problem words with the problem keyword library in the DFM standard library. The problem keyword library stores the core keywords of various standard problems and their corresponding classification labels. This matching process can determine the classification label of the current problem. For example, if the core problem word is "component incompatibility" and it matches the "compatibility issue" label in the keyword library, the problem classification label is determined to be "component compatibility issue."

[0103] After determining the problem classification label, the system combines the involved process terms and the impact result terms to generate a standardized problem description using the structured template "[Problem classification label]: [Involved process] leads to [Impact result]." This structured description clearly and accurately expresses the core elements of the problem, making it easier for subsequent process engineers to understand and address it. For example, if the involved process term is "mainboard assembly," the impact result term is "extended operation time," and the problem classification label is "assembly complexity issue," the generated standardized description will be "Assembly complexity issue: mainboard assembly leads to extended operation time."

[0104] After generating a standardized problem description, the system associates it with the production employee's original input and stores it. This association allows process engineers to access the original description during subsequent processing, enabling a more comprehensive understanding of the problem context. The system encrypts the data in the database to ensure its security and integrity.

[0105] After storage, the system pushes the standardized problem description and original input to the process engineer's terminal for confirmation. Upon receiving this information, the process engineer reviews the problem description for accuracy and completeness. If the description is deemed accurate, it is approved. If any deviations are identified or additional information is needed, the engineer can make modifications or add comments on the terminal before returning it to the system.

[0106] Throughout the entire problem-solving process, the system must ensure real-time and accurate data transmission. Data exchange between production terminals and servers, and between servers and process engineers' terminals, must occur via secure and reliable network protocols to prevent data loss or tampering. Furthermore, the system must feature a user-friendly interface to facilitate operation and interaction between production staff and process engineers, improving problem-solving efficiency.

[0107] In addition, the question processing module accumulates and analyzes historical question data, continuously optimizing question type options and keyword libraries. By learning from a large amount of historical question data, the system can more accurately pre-set question types and improve the accuracy of keyword matching, thereby improving the efficiency and quality of the entire question processing process.

[0108] Through the above detailed implementation methods, the problem handling module can efficiently collect and process problem information when the operation time is abnormal, generate standardized problem descriptions, provide strong support for subsequent problem analysis and design improvements, and ensure that problems in the server prototype trial production process can be handled in a timely and accurate manner, thereby improving the product's manufacturability and production efficiency.

[0109] Example 4:

[0110] The implementation of the operation fluency score calculation of the evaluation scoring module is as follows: after the trial production of the server prototype is completed, the evaluation scoring module needs to quantitatively score the manufacturability of the prototype based on the manufacturability design standards in the standard library management module, and the calculation of the operation fluency score is realized by analyzing the production staff operation log. Taking the trial production of a certain type of server prototype as an example, assuming that the trial production process includes 20 processes such as motherboard assembly, cable connection, component testing, etc., and each process has corresponding operation log records.

[0111] The system extracts the production staff operation log of all processes in the trial production process of the prototype from the production management database. The operation log needs to include key information such as the start time, interruption time, and recovery time of each operation. For example, in the motherboard assembly process, the operation log of a certain employee may be recorded as: the operation start time is June 21, 2025 09:00:00, the first interruption time is 09:15:30, the recovery time is 09:18:45, the second interruption time is 09:25:10, the recovery time is 09:30:20, and the operation end time is 09:45:00. These time data need to be accurate to seconds to ensure the accuracy of subsequent calculations.

[0112] For each operation record, the system needs to calculate the ratio of the interruption duration of a single operation to the total operation duration, thereby obtaining the single operation fluency. Taking the operation record of the motherboard assembly process just now as an example, the total operation duration is the difference between the start time and the end time, i.e. 45 minutes (2700 seconds). The first interruption duration is the recovery time minus the interruption time, i.e. 3 minutes and 15 seconds (195 seconds); the second interruption duration is 5 minutes and 10 seconds (310 seconds), and the total interruption duration is 195+310=505 seconds. Then the single operation fluency is (total operation duration-total interruption duration) ÷ total operation duration × 100%, i.e. (2700-505) ÷ 2700 × 100% ≈ 81.3%.

[0113] During the calculation process, the system needs to handle various special cases. If there is no interruption time in a certain operation record, the total interruption duration is 0, and the single operation fluency is 100%; if the interruption time is earlier than the start time or the recovery time is later than the end time, it is considered as invalid record and needs to be marked and excluded from the calculation. In addition, for the case of continuous interruption, such as interruption without recovery and then interruption again, the system needs to combine and calculate the multiple interruption times to ensure the accuracy of the interruption duration.

[0114] After calculating the fluency of all individual operations, the system averages the fluency of all individual operations across all processes to create a fluency score. Continuing with the prototype, let's assume each of the 20 processes has an average of 5 operation records, for a total of 100 operation records. The fluency of each operation record is calculated to be 81.3%, 92.5%, 78.6%, and so on. The system then adds these 100 values ​​and divides them by 100 to obtain the average fluency score.

[0115] When calculating the mean, the system needs to consider whether the weights of different processes are the same. In the design for manufacturability standard, the importance of each process may be different. For example, the weight of the core component assembly process may be higher than that of the appearance processing process. Therefore, the system will obtain the weight coefficient of each process from the standard library management module and perform a weighted average of the single operation fluency of different processes. Assume that the weight of the motherboard assembly process is 0.1, the weight of the cable connection process is 0.08, the weight of the component testing process is 0.12, and the weights of other processes are similar. The operation fluency score is calculated as follows: (81.3% × 0.1 + 92.5% × 0.08 + 78.6% × 0.12 + ...) ÷ (0.1 + 0.08 + 0.12 + ...), and the final weighted operation fluency score is obtained.

[0116] In addition to the operational fluency score, the assessment and scoring module also calculates the time compliance score and the problem occurrence score. The time compliance score represents the percentage of processes in the trial production process whose operation time is within the allowable error range. For example, if 16 of 20 processes have operation times within the allowable error range, the time compliance score is 16 ÷ 20 × 100% = 80%. The problem occurrence score is the ratio of the number of confirmed problems in the problem handling module to the total number of processes. Assuming that there are 3 confirmed problems during the trial production process, the problem occurrence score is 3 ÷ 20 × 100% = 15%.

[0117] The evaluation and scoring module obtains the weight coefficients for each evaluation indicator from the standard library management module. Assuming that the weight of the man-hour compliance rate is 0.4, the weight of the problem occurrence rate is 0.3, and the weight of the operation fluency is 0.3, the final score of the prototype manufacturability is 80% × 0.4 + 15% × 0.3 + the operation fluency score × 0.3. For example, if the operation fluency score is 85%, the final score is 80% × 0.4 + 15% × 0.3 + 85% × 0.3 = 32% + 4.5% + 25.5% = 62%.

[0118] During the entire evaluation scoring process, the system needs to ensure the accuracy and integrity of the data. The operation log data needs to be synchronized to the database in real time to avoid omissions or errors; the weight coefficient and evaluation standard need to be consistent with the standard library management module to ensure the fairness and comparability of the score. At the same time, the system also needs to provide query and export functions of the scoring results, which is convenient for production managers and design engineers to view and analyze, and provides the basis for subsequent process improvement and design optimization.

[0119] Through the above implementation examples combined with specific examples, the evaluation scoring module can comprehensively and objectively quantify the manufacturability of the server prototype, and the calculation of the operation fluency score accurately reflects the continuity and efficiency of the employee's operation in the production process through detailed analysis of the operation log data, providing an important reference index for manufacturability evaluation.

[0120] Example 5:

[0121] The implementation of the standard updating module to generate the updated manufacturability design standard is as follows: During the trial production of the server prototype, the standard updating module needs to generate the updated server manufacturability design standard based on the problem list formed during trial production and historical evaluation data through a machine learning model. Taking the trial production of a new server prototype developed by a certain enterprise as an example, it accumulated 50 problems confirmed by design engineers in the first round of trial production, involving 12 processes such as motherboard assembly and heat dissipation module installation, and subsequent updates to the standard based on these data.

[0122] The system collects all the problem lists confirmed by design engineers and corresponding design improvement schemes during the historical trial production process. For example, in the first round of trial production, the problem list shows that the motherboard assembly process has the problem of "screw hole deviation leading to increased assembly time", and the design improvement scheme is "adjusting the design tolerance of the motherboard hole from ±0.5mm to ±0.2mm"; the heat dissipation module installation process has the problem of "insufficient fan cable length leading to connection difficulty", and the improvement scheme is "extending the cable length from 30cm to 35cm". The system needs to associate and label the problem type, standardized description and improvement scheme in these problem lists to form a training data set.

[0123] During the association and labeling process, the problem type needs to be divided according to the classification in the manufacturability design standard library, such as "assembly complexity problem" and "component compatibility problem"; the standardized description needs to use a structured template of "[problem classification label]: [involved process] leads to [impact result]", such as "assembly complexity problem: motherboard assembly leads to extended operation time" and "component compatibility problem: heat dissipation module installation leads to cable connection difficulty". The improvement scheme needs to describe the specific content, implementation steps and expected effect of the design change in detail to ensure that the mapping relationship between the problem and the improvement can be accurately learned during subsequent model training.

[0124] After generating a training dataset, the system uses it to perform supervised learning on the initial machine learning model. This initial model can employ common supervised learning algorithms, such as random forests and support vector machines. Its core function is to optimize the model's feature extraction and pattern recognition capabilities by learning the correlation between problem features and improvement solutions in historical data. During training, the system uses problem type and standardized descriptions as input features and design improvement solutions as output labels. Through iterative training, the model parameters are adjusted to enable the model to accurately predict appropriate improvement directions and standard updates from a new list of problems.

[0125] For example, when training on the problem of "motherboard hole deviation," the model learns the associations between features like "hole deviation" and "increased assembly time" and the improvement solution of "adjusting tolerances." For the problem of "insufficient cable length," the model learns the associations between features like "cable length" and "connection difficulty" and the improvement solution of "extending the cable." By training on a large amount of historical data, the model gradually grasps the characteristic patterns of manufacturability problems and the corresponding improvement strategies.

[0126] After training is complete, the system feeds the latest pilot production problem list into the optimized model, generating an updated DFM standard with new evaluation metrics, new work time standards, and new problem classifications. For example, if the latest pilot production encountered a problem with "improper power interface placement resulting in difficulty plugging and unplugging," the model, based on the knowledge gained from training, might generate an improvement plan: "relocate the power interface to the side of the chassis." Based on this, it would propose new evaluation metrics, such as "reasonable interface placement," and simultaneously update the corresponding work time standards, adjusting the standard work time for the power interface installation process from 10 minutes to 12 minutes to accommodate the new design requirements.

[0127] After generating the updated DFM standards, the system needs to verify the effectiveness of the newly generated evaluation indicators. This verification method involves selecting prototypes from historical trial production data that have undergone design improvements, rescoring them using the new indicators, and comparing them with the actual improvement results. For example, a prototype that previously underwent design improvements due to a "motherboard hole position deviation" issue is selected, and its improved production data is scored using the newly generated "hole position tolerance evaluation indicator." This score is then compared with the actual reduction in operating time after the improvements. Assuming that the original issue caused the operating time to exceed the standard working hours by 20%, and that the improved operating time meets the standard, scoring using the new indicators should reflect this improvement.

[0128] If the verification pass rate exceeds a preset threshold (e.g., 80%), the new evaluation indicators, new work-hour standards, and new problem classifications are merged into the DFM standard library in the standard library management module. For example, after verifying 10 historical improvement cases, if the new indicator scores of 8 cases are consistent with the actual results, the new indicator is considered to have passed verification and is added to the standard library. During the merger, the system must ensure the compatibility of the new data with the original standards, update or delete conflicting or duplicate content, and maintain the consistency and accuracy of the standard library.

[0129] Throughout the standards update process, the system must possess data management and model optimization capabilities. For data management, historical problem lists, improvement plans, and training data sets must be categorized, stored, and versioned for easy traceability and reuse. For model optimization, models must be regularly retrained based on new pilot production data to ensure they can adapt to changing production needs and design improvements.

[0130] Furthermore, the standards update module must maintain data interaction with other modules. For example, it must obtain the latest problem list from the problem handling module, historical assessment data from the assessment and scoring module, and existing standards from the standards library management module as the basis for updates, ensuring the continuity of the update process and the consistency of the data.

[0131] Through the above implementation methods combined with specific examples, the standard update module can use machine learning technology to continuously optimize and update the manufacturability design standards based on historical trial production data and problem feedback, so that the standards can better adapt to production practices, improve the efficiency and quality of server manufacturing, and realize the intelligent evolution and continuous improvement of the MES system.

[0132] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0133] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A server manufacturability intelligent evaluation MES system based on machine learning, characterized by: include: The data acquisition module is used to obtain the operation time data of each process in real time during the trial production of the server prototype, and classify and mark the operation time data according to the trial production process; The standard library management module is used to store the server's manufacturability design standard library and the standard working time library for each process, and supports the addition, deletion, modification and query operations of the data in the standard library; The working time comparison module is used to retrieve the standard working time and allowable error range corresponding to the current process from the standard library management module, and compare the real-time collected operation time data with the standard working time interval; The problem handling module is used to trigger the problem feedback process when the operation time exceeds the allowable error range. It receives the problem type and specific description input by the production staff and generates a standardized problem description through the natural language processing model; The evaluation and scoring module is used to quantitatively score the manufacturability of the prototype based on the manufacturability design standards in the standard library management module after the entire trial production process is completed; The standard update module is used to generate updated server manufacturability design standards through machine learning models based on the problem list and historical evaluation data generated during the trial production process.

2. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 1 is characterized in that: The implementation of the data acquisition module includes: For any process in the trial production process, the start and end timestamps of the process are obtained through sensors deployed on the production equipment; Calculate the difference between the start timestamp and the end timestamp to get the actual operation time of the process; Perform noise filtering on the actual operation time, eliminate abnormal values ​​caused by equipment abnormalities or operational errors, and form effective operation time data; The effective operation time data is classified and marked according to the process number, prototype model, and production batch, and passed to the working time comparison module as input data.

3. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 2 is characterized in that: The implementation of noise filtering for actual operation time also includes: Obtain the operating time distribution range of the same process in historical trial production data; Calculate the deviation between the current actual operation time and the mean of the historical distribution range. If the deviation exceeds the preset threshold, it is determined to be an outlier. Outliers are marked as invalid data and stored in isolation, and only the actual operation time with deviation within the threshold range is retained as valid data.

4. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 1 is characterized in that: The implementation methods for retrieving the standard working hours corresponding to the current process also include: Extract the process feature tags of the current process, including process name, operation type, and number of parts involved; In the process standard working time library of the standard library management module, search for completely identical standard processes by matching feature tags; If no completely identical standard process is found, a similar process with the highest similarity in process feature tags is searched, and its standard working hours are retrieved as a comparison benchmark.

5. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 1 is characterized in that: The implementation of the problem handling module includes: When the operation time exceeds the allowable error range, a problem type selection menu is displayed on the production terminal interface. The menu options include assembly complexity issues, component compatibility issues, and missing operation instructions issues. If the production employee selects a custom problem type, a text input box is triggered to receive a free description; The pre-trained BERT natural language processing model is used to extract keywords and perform semantic analysis on free description texts, generating standardized problem descriptions that include the root causes and impact aspects of the problem. The standardized problem description is associated with the original input content and stored, and pushed to the process engineer terminal for confirmation.

6. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 3 is characterized in that: The implementation methods for generating standardized problem descriptions include: Extract core problem words, process-related words, and result-affecting words from free description texts; Match the core problem words with the problem keyword library in the manufacturability design standard library to determine the problem classification label; Combine the words related to the process and the words affecting the results, and generate a standardized description according to the structured template of "[problem classification label]: [related process] leads to [impact result]".

7. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 1 is characterized in that: The implementation methods of the evaluation and scoring module include: Obtain the weight coefficients of each evaluation index in the manufacturability design standard from the standard library management module. The evaluation indexes include the man-hour compliance rate, the problem occurrence rate, and the operation fluency. Calculate the percentage of processes in the entire trial production process whose operation time is within the allowable error range as the labor time compliance score; The ratio of the number of problems confirmed to have passed in the problem handling module to the total number of processes is calculated as the problem occurrence rate score; Analyze the number of operation interruptions and recovery time through production staff operation logs to calculate the operation fluency score; The three scores are weighted and summed according to the weight coefficient of each indicator to obtain the final score of the prototype manufacturability.

8. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 7 is characterized in that: The implementation methods for calculating the operation fluency score include: Extract the operation start time, operation interruption time, and operation recovery time from the production staff's operation log; Calculate the ratio of the interruption time of a single operation to the total operation time to obtain the fluency of the single operation; The average of the single operation fluency of the entire process is calculated to obtain the operation fluency score.

9. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 1 is characterized in that: The standard update module is implemented as follows: Collect all the problem lists confirmed by design engineers during the historical trial production process and the corresponding design improvement plans; Associate and label the problem types, standardized descriptions, and improvement plans in the problem list to form a training dataset; Use the training dataset to supervise the initial machine learning model and optimize the model's feature extraction and pattern recognition capabilities; The latest problem list from the pilot production was input into the optimized model to generate an updated DFM standard that included new evaluation metrics, new work-hour standards, and new problem classifications.

10. The server manufacturability intelligent evaluation MES system based on machine learning according to claim 9 is characterized in that: The generation of updated DFM standards can also be accomplished by: Verify the effectiveness of the newly generated evaluation indicators by selecting prototypes with completed design improvements from historical trial production data, re-scoring them using the new indicators, and comparing them with the actual improvement results. If the verification pass rate exceeds the preset threshold, the new evaluation indicators, new working time standards, and new problem classifications will be merged into the manufacturability design standard library of the standard library management module.

Images