Methods and apparatus for processing yield information

CN123076880APending Publication Date: 2026-11-10BOE TECHNOLOGY GROUP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202580000245.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-02-28
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing yield screening tools rely on machine learning algorithms to analyze massive amounts of data, which is time-consuming and inefficient. Offline surveys rely on manual knowledge sorting, which is labor-intensive and the accuracy is difficult to guarantee.

Method used

By employing a decoupled analysis module and a question-answering module, query requests are generated for semantic retrieval, yield analysis dimensions are filtered, computational load is reduced, and business experience is combined to form an adaptive knowledge base, thereby improving the efficiency and accuracy of troubleshooting.

Benefits of technology

It enables rapid identification of the causes of yield anomalies, reduces machine learning computation time, improves the efficiency and accuracy of yield troubleshooting, and reduces the cost of manual knowledge processing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123076880A_ABST
    Figure CN123076880A_ABST
Patent Text Reader

Abstract

A method and apparatus for processing yield information are provided, wherein the method includes: determining yield anomaly information of industrial products by an analysis module based on a first database (301); generating a query request by the analysis module based on the yield anomaly information and sending the query request to a question-and-answer module (302); performing semantic retrieval in a second database based on the query request and generating multiple yield analysis dimensions and sorted data for the multiple yield analysis dimensions based on the results of the semantic retrieval (303); sending the sorted data for the yield analysis dimensions to the analysis module by the question-and-answer module (304); selecting one or more yield analysis dimensions from the multiple yield analysis dimensions by the analysis module based on the sorted data for the yield analysis dimensions (305); and performing data analysis by the analysis module based at least on data in the first database to generate yield influencing factors based on the selected one or more yield analysis dimensions (306).
Need to check novelty before this filing date? Find Prior Art

Description

Methods and apparatus for processing yield information Technical Field

[0001] This disclosure relates to the field of artificial intelligence services, and more specifically to a method for processing yield information, an apparatus for processing yield information, an analysis apparatus for processing yield information, a question-and-answer apparatus for processing yield information, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Yield rate assessment is a core task in the manufacturing industry. Its purpose is to identify and analyze the specific causes of product yield abnormalities and formulate corresponding improvement measures accordingly. In practice, production and quality inspection data are used to initially screen for anomalies. Business personnel identify suspicious points by analyzing the correlation between multiple factors such as the location of the defect, equipment parameters, time points, and operators. Then, through offline investigation methods such as design drawing comparison, equipment inspection, material testing, process verification, and operation comparison, the root cause of the problem is finally confirmed and improvement measures are formulated.

[0003] Currently, traditional yield assessment tools mainly rely on machine learning algorithms to analyze massive amounts of production data, which is time-consuming and inefficient. Meanwhile, experience summaries derived from offline investigations still require business personnel to regularly compile documents and preserve the knowledge, which is both labor-intensive and difficult to guarantee in terms of comprehensiveness and accuracy.

[0004] Therefore, the current yield screening tools need further improvement. Summary of the Invention

[0005] This disclosure provides a method for processing yield information, an apparatus for processing yield information, an analysis apparatus for processing yield information, a question-and-answer apparatus for processing yield information, an electronic device, and a computer-readable storage medium.

[0006] This disclosure provides a method for processing yield information, comprising: determining yield anomaly information of industrial products by an analysis module based on a first database; generating a query request and sending the query request to a question-and-answer module based on the yield anomaly information; performing semantic retrieval in a second database based on the query request and generating multiple yield analysis dimensions and ranking data for the multiple yield analysis dimensions based on the semantic retrieval results; sending the ranking data for the yield analysis dimensions to the analysis module by the question-and-answer module; selecting one or more yield analysis dimensions from the multiple yield analysis dimensions based on the ranking data for the yield analysis dimensions; and performing data analysis by the analysis module based at least on data in the first database to generate yield influencing factors based on the selected one or more yield analysis dimensions.

[0007] This disclosure provides an apparatus for processing yield information. The apparatus includes a decoupled analysis module and a question-answering module, as well as a decoupled first database and a second database. The analysis module is configured to: determine yield anomaly information of industrial products based on the first database; and generate a query request based on the yield anomaly information and send the query request to the question-answering module. The question-answering module is configured to: perform semantic retrieval in the second database based on the query request and generate multiple yield analysis dimensions and ranking data for the multiple yield analysis dimensions based on the semantic retrieval results; and send the ranking data for the yield analysis dimensions to the analysis module. The analysis module is further configured to: select one or more yield analysis dimensions from the multiple yield analysis dimensions based on the ranking data for the yield analysis dimensions; and perform data analysis based on the selected one or more yield analysis dimensions, at least based on data in the first database, to generate yield influencing factors.

[0008] This disclosure provides an analysis apparatus for processing yield information, comprising: one or more memories; and one or more processors coupled to the one or more memories, the one or more processors being configured to: determine yield anomaly information of industrial products based on a first database; generate a query request based on the yield anomaly information and send the query request to a question-and-answer device for processing yield information; receive sorted data for yield analysis dimensions from the question-and-answer device for processing yield information; select one or more yield analysis dimensions from multiple yield analysis dimensions based on the sorted data for yield analysis dimensions; and perform data analysis based on the selected one or more yield analysis dimensions, at least based on data in the first database, to generate yield influencing factors.

[0009] This disclosure provides a question-and-answer apparatus for processing yield information, comprising: one or more memories; and one or more processors coupled to the one or more memories, the one or more processors being configured to: receive a query request based on yield anomaly information from an analysis apparatus for processing yield information; perform semantic retrieval in a second database based on the query request and generate multiple yield analysis dimensions and sorted data for the multiple yield analysis dimensions based on the results of the semantic retrieval; and send the sorted data for the yield analysis dimensions to the analysis apparatus for processing yield information.

[0010] This disclosure provides an electronic device, including: one or more processors; and one or more memories, wherein the memories store a computer-executable program, and when the processor executes the computer-executable program, the above-described method is performed.

[0011] This disclosure provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the above-described method.

[0012] According to another aspect of this disclosure, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable medium and executes the computer instructions, causing the computer device to perform the methods provided in the foregoing aspects or various alternative implementations of the foregoing aspects. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments will be briefly described below. Obviously, the drawings described below only relate to some embodiments of this disclosure, and are not intended to limit this disclosure.

[0014] Figure 1 is an example schematic diagram illustrating a scenario according to an embodiment of the present disclosure.

[0015] Figure 2 illustrates a schematic diagram of an example yield screening task according to an embodiment of the present disclosure.

[0016] Figure 3 shows a flowchart of a method for processing yield information according to an embodiment of the present disclosure.

[0017] Figure 4 illustrates a schematic diagram of the interaction between the yield inspection auxiliary tool and the business process according to an embodiment of the present disclosure.

[0018] Figure 5 shows a schematic diagram of a question-and-answer module according to an embodiment of the present disclosure.

[0019] Figure 6 illustrates a schematic diagram of the process by which the question-and-answer module prepares and applies data according to an embodiment of the present disclosure.

[0020] Figure 7 shows a schematic diagram of the data structure of the data stored in the second database according to an embodiment of the present disclosure.

[0021] Figure 8 shows a schematic diagram of an electronic device according to an embodiment of the present disclosure.

[0022] Figure 9 illustrates a schematic diagram of the architecture of an exemplary computing device according to an embodiment of the present disclosure.

[0023] Figure 10 shows a schematic diagram of a storage medium according to an embodiment of the present disclosure. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0025] In this specification and accompanying drawings, operations and elements that are substantially the same or similar are indicated by the same or similar reference numerals, and repeated descriptions of these operations and elements are omitted. Furthermore, in the description of this disclosure, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance or order.

[0026] To facilitate the description of this disclosure, the following concepts related to this disclosure are introduced.

[0027] Optionally, the models used in embodiments of this disclosure as described below can all be artificial intelligence models, especially artificial intelligence-based neural network models. Typically, artificial intelligence-based neural network models are implemented as acyclic graphs, where neurons are arranged in different layers. Generally, a neural network model includes an input layer and an output layer, separated by at least one hidden layer. The hidden layer transforms the input received by the input layer into a representation useful for generating the output in the output layer. Nodes are fully connected to nodes in adjacent layers via edges, and there are no edges between nodes within each layer. Data received at nodes in the input layer of the neural network is propagated to nodes in the output layer via any of the hidden layers, activation layers, pooling layers, convolutional layers, etc. The input and output of the neural network model can take various forms, and this disclosure does not limit this.

[0028] The solutions provided in this disclosure involve technologies such as artificial intelligence and / or machine learning, and are specifically illustrated through the following embodiments.

[0029] First, with reference to FIG1, an application scenario of the method and corresponding apparatus for processing yield information according to an embodiment of the present disclosure will be described. FIG1 shows a schematic diagram of an application scenario 100 according to an embodiment of the present disclosure, wherein a server 110 and a plurality of terminals 120 are schematically shown.

[0030] The neural network model of this disclosure can be integrated into various electronic devices, such as any electronic device among the server 110 and multiple terminals 120 in Figure 1. For example, the neural network model can be integrated into terminal 120. Terminal 120 can be a mobile phone, tablet computer, laptop computer, desktop computer, personal computer (PC), smart speaker, or smartwatch, but is not limited to these. Alternatively, the neural network model can also be integrated into server 110. Server 110 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence modules. Terminals and servers can be directly or indirectly connected via wired or wireless communication, and this disclosure does not impose any limitations.

[0031] It is understood that the apparatus for inference using the neural network model of this disclosure can be a terminal, a server, or a system composed of a terminal and a server. The method for processing yield information in this disclosure can be executed on a terminal, on a server, or jointly by a terminal and a server.

[0032] The artificial intelligence model provided in this disclosure can also relate to artificial intelligence cloud services in the field of cloud technology. It is worth noting that both the terminal 120 and server 110 according to the embodiments of this disclosure adhere to data protection principles, respect users' data rights, and ensure user data security and privacy. The terminal 120 and server 110 according to the embodiments of this disclosure will clearly inform users of the purpose, method, and scope of collecting, using, storing, transmitting, and deleting user data, and obtain user consent. The terminal 120 and server 110 according to the embodiments of this disclosure will take reasonable technical and management measures to prevent user data from being leaked, tampered with, damaged, or lost. The providers of the terminal 120 and server 110 according to the embodiments of this disclosure will regularly review and update user data, and promptly delete expired or useless data. Furthermore, cloud service providers using the embodiments of this disclosure respect users' rights to data access, correction, deletion, withdrawal of consent, complaints, and claims, and provide convenient channels and procedures to enable users to effectively exercise these rights.

[0033] Furthermore, the process of data analysis using artificial intelligence technology in terminal 120 or server 110 is conducted based on the principles of legality, rationality, and transparency. The data collected and processed by the artificial intelligence model according to embodiments of this disclosure is relevant, necessary, and appropriate for the predictive purpose, and does not contain any personally identifiable or sensitive information. The neural network model according to embodiments of this disclosure employs appropriate techniques and organizational measures to protect the security and integrity of data, preventing unauthorized access, use, or disclosure.

[0034] The artificial intelligence-based neural network model according to embodiments of this disclosure will comply with relevant data protection regulations and ethical principles. This neural network model is trained on a large amount of anonymized and de-identified data, and does not infringe on the privacy rights of any individual or group. The artificial intelligence model has also undergone rigorous testing and evaluation to ensure that its output results are accurate and reliable, and will not cause any misleading or discriminatory behavior. The artificial intelligence model is designed solely to improve service quality and customer satisfaction and will not be used for any illegal or unethical purposes. Furthermore, the neural network model will be regularly reviewed and updated to adapt to changes in the data environment and legal regulations.

[0035] In the manufacturing sector, the core objective of yield assessment is to accurately identify the specific reasons affecting product yield and to formulate and implement corresponding improvement measures based on these reasons. In the actual production process, business personnel first use data analysis tools to conduct in-depth mining and detailed analysis of production data to discover yield anomalies and initially identify yield-influencing factors with high correlation. These yield-influencing factors cover many aspects, including the specific location of defects, various parameters of equipment during operation, product production time points, and personnel involved in production operations. After initially identifying the relevant factors affecting yield, business personnel need to conduct a series of offline investigations, including at least one of the following: reviewing product design drawings to check for design flaws or unreasonable aspects; conducting a comprehensive inspection of production equipment to check for malfunctions, aging, or improper parameter settings; analyzing the materials used in production to determine if the material quality meets standards and production requirements; verifying production on different production lines through process switching to determine if the problem lies on a specific production line; and conducting comparative operational experiments to observe the operational effects of different operators under the same conditions to determine if operational techniques affect yield. Through this series of online and offline investigations, the specific causes affecting yield are finally identified, and corresponding improvement measures are formulated.

[0036] However, for complex manufacturing processes, yield is influenced by numerous and intertwined factors, making yield troubleshooting exceptionally complex. Throughout the process, sales personnel need extensive experience to keenly detect subtle anomalies and accurately assess their impact on yield. Therefore, a high level of expertise is required; they must be familiar with production processes, equipment performance, material properties, and possess strong analytical and problem-solving skills. Furthermore, the rich experience accumulated by sales personnel over long-term practice needs to be systematically reviewed and summarized regularly to form a complete and practical knowledge system, enabling knowledge sharing with colleagues. This aims to reduce training costs for new sales personnel, allowing them to quickly familiarize themselves with their work, master methods, and improve efficiency.

[0037] Currently, the functionality of yield assessment tools in the industry is relatively limited. Specifically, these tools can only use machine learning algorithms to identify anomalies and analyze correlations based on data generated during the production and quality inspection processes. In this way, these tools can help business personnel quickly identify data anomalies and conduct preliminary analysis of related factors. However, they cannot completely replace offline investigation work. Furthermore, machine learning methods require extracting massive amounts of information from databases and performing comprehensive analysis across numerous data dimensions. This approach is computationally intensive, highly complex, and time-consuming.

[0038] Furthermore, business personnel need to invest significant time and effort to regularly document the conclusions and accumulated experience from offline surveys for knowledge sharing. This traditional method of manually compiling and sharing knowledge continuously consumes substantial human resources. Moreover, due to the influence of human factors, the accuracy of the content is often difficult to guarantee, which significantly limits the effectiveness of knowledge sharing and the improvement of business personnel's work efficiency.

[0039] Therefore, the current yield assessment tools need further improvement.

[0040] In one aspect of this disclosure, embodiments of this disclosure achieve efficient localization and analysis of yield anomalies through the interaction between a question-and-answer module and an analysis module. In a big data environment, embodiments of this disclosure effectively reduce the computation time required for machine learning by reducing invalid dimensions, and quickly locate the causes of yield anomalies (i.e., yield-influencing factors).

[0041] In one optional aspect, embodiments of this disclosure utilize the interaction between a decoupled analysis module and a question-and-answer module to generate a query request based on yield anomaly information. Based on this query request, multiple yield analysis dimensions and ranked data for these dimensions are generated. Subsequently, the yield analysis dimensions are filtered according to the ranked data, thereby reducing the number of dimensions to be analyzed and determining the causes affecting yield with less computation. Furthermore, since the ranked data incorporates, to some extent, the experience and data of business personnel, embodiments of this disclosure can improve the efficiency and accuracy of yield troubleshooting.

[0042] In one optional aspect, embodiments of this disclosure store and mine the conclusions formed by business personnel during yield rate checks, forming an adaptive iterative second database (i.e., a knowledge base) to improve the accuracy of the large model's answers to yield rate check questions. Specifically, the knowledge base uses the business system as the data acquisition engine, continuously updating the knowledge base with the conclusions generated during yield rate checks to improve its timeliness and reduce the cost of manual knowledge organization. By using the business system as the data acquisition engine, the knowledge base can obtain the latest yield rate check conclusions in real time, thereby maintaining its timeliness and accuracy while reducing manual maintenance costs.

[0043] This disclosure proposes an improved yield troubleshooting tool. The tool includes a decoupled analysis module and a question-and-answer module. The analysis module is responsible for periodically updating yield statistics, automatically pushing yield anomalies to business personnel, and assisting business personnel in data-driven influencing factor analysis. For example, the analysis module can periodically collect data from the production line and analyze the impact of equipment parameters, material batches, etc., on yield. The question-and-answer module is responsible for generating and updating a second database (i.e., a knowledge base), using a large model based on RAG (Retrieval Augmented Generation), providing question-and-answer services to business personnel during the influencing factor analysis phase to help them quickly locate the root cause of defects. For example, business personnel can use the question-and-answer module to inquire about the impact of specific equipment parameter anomalies on yield, and the module will provide qualitative answers based on the knowledge base. The question-and-answer module automatically stores the conclusions of the influencing factor analysis, solutions to defect problems, and quality confirmation results, and periodically summarizes this data into text, forming and updating the knowledge base, thereby creating a closed loop in the business system process. Of course, this disclosure is not limited to this.

[0044] This disclosure provides a method for processing yield information to improve the yield investigation system, enabling yield investigation based on an improved yield investigation tool. The method includes: determining yield anomaly information of industrial products using an analysis module based on a first database; generating a query request and sending it to a question-and-answer module based on the yield anomaly information; performing semantic retrieval in a second database using the question-and-answer module based on the query request, and generating multiple yield analysis dimensions and ranking data for the multiple yield analysis dimensions based on the semantic retrieval results; sending the ranking data for the yield analysis dimensions to the analysis module using the question-and-answer module; selecting one or more yield analysis dimensions from the multiple yield analysis dimensions based on the ranking data for the yield analysis dimensions; and performing data analysis, at least based on data in the first database, to generate yield influencing factors based on the selected one or more yield analysis dimensions.

[0045] In the process of using the improved yield inspection tool for yield inspection, the embodiments of this disclosure reduce the amount of computation required by the analysis module by using sorted data from the question and answer module, while maintaining the accuracy of the yield influencing factor analysis.

[0046] The method for processing yield information, the method for training a neural network, and the apparatus according to embodiments of the present disclosure are described below with reference to Figures 2 to 10.

[0047] Figure 2 illustrates a schematic diagram of an example yield screening task according to an embodiment of the present disclosure.

[0048] Yield rate refers to the ratio between the number of qualified products and the total number of products produced during the manufacturing process, reflecting the stability of the manufacturing process and the level of product quality. The yield rate is affected by factors such as equipment status, process parameters, raw material characteristics, and environmental conditions. For example, the yield rate may decrease if there are fluctuations in equipment temperature or unstable raw material quality in a particular batch of products. In the embodiments of this disclosure, by analyzing historical data, real-time data, and traceability data, yield anomalies can be investigated and located.

[0049] Yield testing can be widely applied in various scenarios, such as electronic product manufacturing, automotive parts production, and semiconductor chip manufacturing.

[0050] Taking Figure 2 as an example, yield rate screening tasks can be applied to the circuit board production stage in electronic product manufacturing. In this scenario, by analyzing production and quality inspection data, it can be found that the defect rate in certain specific areas of the circuit board is abnormally high. Using yield rate screening tasks, the commonalities of these defective products can be further analyzed, such as production time, equipment parameters, and operators, thereby identifying potential influencing factors. For example, if it is found that defective products are mainly concentrated within the production time of a certain machine, it can be preliminarily determined that the machine may have a malfunction or improper parameter settings. By inspecting and adjusting the equipment, the defect rate can be effectively reduced, and product quality improved.

[0051] Yield assessment tasks can also be applied to engine component manufacturing in automotive parts production. Due to the complexity of the processes involved in engine component production, numerous factors influence yield. Yield assessment tasks allow for in-depth analysis of production data, identifying factors highly correlated with defective products, such as material batches and processing parameters. For example, if a particular batch of materials is found to significantly reduce the yield of components, the material supplier can be changed promptly or the material handling process adjusted, thereby improving the product qualification rate.

[0052] Yield monitoring tasks can also be applied to the photolithography process in semiconductor chip manufacturing. Photolithography is a critical step in chip manufacturing and has a significant impact on yield. Yield monitoring tasks allow for the monitoring and analysis of various parameters during the photolithography process, such as exposure time and photoresist thickness. If a decrease in chip photolithography yield is detected within a certain period, the causes of the yield decline can be identified by analyzing the changes in relevant parameters. For example, excessively long or short exposure times will affect the photolithography effect, thus reducing chip yield. By adjusting parameters such as exposure time, photolithography yield can be effectively improved, and production costs reduced.

[0053] Figure 3 shows a flowchart of a method 30 for processing yield information according to an embodiment of the present disclosure.

[0054] Method 30 can be executed at a terminal device or a server (such as terminal 120 or server 110 as shown in FIG1). Method 30 includes the following operations S301 to S305. Of course, method 30 may also include more or fewer operations, and this disclosure is not limited thereto.

[0055] In operation S301, based on the first database, the analysis module determines abnormal yield information of industrial products.

[0056] Optionally, the first database refers to a database used to store quality inspection data related to product quality checks; therefore, the first database is also called the quality inspection database. For example, the first database stores quality inspection data for products at each stage of the process. The analysis module can detect abnormalities in the yield rate of specific batches of products through statistical analysis of the quality inspection data. Of course, this disclosure is not limited to this.

[0057] Optionally, the analysis module refers to a software or hardware module used to analyze data related to yield, also known as an analysis device for processing yield information. Optionally, the analysis module is used to provide big data analysis services based on machine learning. Machine learning refers to the technology of extracting patterns from large amounts of data stored in multiple databases (including but not limited to a first database) using algorithms and statistical models, and making predictions and classifications based on data patterns. The machine learning method used in this embodiment is trained on various types of data generated during the production process to identify implicit relationships between data, thereby providing support for the analysis of the causes of yield anomalies.

[0058] Optionally, the analysis module can also employ big data analytics. Big data refers to massive amounts of data generated from multiple data sources, including structured and unstructured data, which may involve information such as production records, equipment testing data, and supply chain data. By processing big data, the analysis module can extract hidden patterns from the massive amounts of data. For example, in a production scenario, statistical analysis of millions of quality inspection data points can identify implicit relationships between product yield and specific production processes, thus providing data support for production adjustments. Furthermore, the analysis module can also use big data analytics to compare data from different batches and time periods to identify potential systemic problems.

[0059] Optionally, yield anomaly information is information indicating yield anomaly events. Specifically, yield anomaly information indicates deviations from normal yield rates identified based on statistical information related to quality inspection data. Yield anomaly information includes, but is not limited to, exceeding defect rate limits, a continuous increase in defect rate, and clustering of defects in location, time of occurrence, or personnel. For example, when quality inspection data in the first database indicates that the product defect rate exceeds the expected range within a certain period, the analysis module can determine that a yield anomaly event exists; similarly, if the quality inspection data shows a trend of defective products appearing in a concentrated manner in a certain process, this is also considered a yield anomaly event. The analysis module provides a basis for investigating the causes of anomalies by analyzing the changing trends of defect rates and the clustering of quality inspection data. Of course, this disclosure is not limited to this.

[0060] In operation S302, based on the yield anomaly information, the analysis module generates a query request and sends the query request to the question-and-answer module.

[0061] Optionally, the query request is a signaling transmission between the analysis module and the question-answering module, and the query request includes semantic information indicating that the question-answering module should determine the cause of the yield anomaly. The question-answering module is also referred to as a question-answering device for processing yield information.

[0062] The query request fields include at least one of the following: defect type, defect phenomenon, defect occurrence conditions, and query target information. Defect type clarifies the specific category of the defect, such as product appearance defects or performance non-compliance in product quality inspection. Defect phenomenon includes exceeding standards and continuous escalation; for example, exceeding a preset standard range for a product indicator is considered exceeding standards, and a continuously rising product defect rate is considered continuous escalation. It also includes clustering phenomena. Defect occurrence conditions require special explanation; for example, the <qualifier> in the following text only takes effect when the problem phenomenon is "clustering," used to limit the scope or characteristics of the cluster. For example, when statistically analyzing product defects in a factory, clustering within a specific workshop, production line, or time period can be specified. Query target information is the identifier for the analysis module to initiate a query for qualitative analysis of the cause of the defect, such as "What is the cause?" or "Why is it happening?".

[0063] Specifically, the query request can be represented as: " + exceeding the standard or continuously increasing or (<qualifier> + clustering) + what is the reason?" For example, for defects detected in the welding process, a query request can be generated as "What is the reason for the welding defect exceeding the standard?"; if significant data fluctuations occur during the placement process, an instruction can be generated as "What is the reason for the continuous increase in placement defects?"; and if a clustering trend is observed in the packaging process, a request can be generated as "What is the reason for the clustering during the packaging process?". Of course, this disclosure is not limited to this.

[0064] Optionally, the question-answering module provides question-answering services based on a large model using Retrieval Augmented Generation (RAG). RAG technology is a technical solution that integrates information retrieval mechanisms with a large-scale generative model. Its operation involves first retrieving information related to the question from a second database, then injecting the retrieved information along with preliminary answers into the large model's prompts to generate the final answer. For example, when dealing with issues related to product quality, production processes, or equipment status, the question-answering module can generate answers that reflect the actual situation by querying relevant troubleshooting manuals and white papers. Compared to a general-purpose large model, the question-answering module, by incorporating private domain expertise, can effectively avoid the illusion problem that may occur in generative models and protect the security of private domain data during information processing, thus meeting the high accuracy and high security requirements of practical applications.

[0065] Furthermore, the question-and-answer module retrieves knowledge related to yield processing and organically integrates the acquired information into the generation process, thereby constructing a logical chain for fault analysis. For example, in one aspect of this disclosure, when the analysis module sends a query request asking "What are the causes of component defect clustering?", the question-and-answer module can quickly retrieve data from the second database and generate a qualitative analysis result that fits the actual production situation, providing a reference for locating the cause of the anomaly. Of course, this disclosure is not limited to this; the question-and-answer module can play a similar role in many other scenarios, thereby providing data support and decision-making basis for production and quality management.

[0066] Optionally, although both the question-answering module and the analysis module are integrated into the yield processing business process, they are decoupled, exchanging information through signaling interaction. Decoupling means that each module operates independently, transmitting query requests and feedback information through a predefined signaling interface, allowing each module to be upgraded or replaced independently without affecting the overall system operation. For example, after the analysis module generates a query request, the question-answering module receives the request and returns the search results through signaling interaction, thereby reducing system complexity and facilitating maintenance and expansion. Of course, this disclosure is not limited to this.

[0067] In operation S303, based on the query request, the question-answering module performs semantic retrieval in the second database and generates multiple yield analysis dimensions and sorted data for the multiple yield analysis dimensions based on the semantic retrieval results.

[0068] Optionally, the second database is a knowledge base that is adaptively iterated by the question-and-answer module. It automatically stores the conclusions of influencing factor analysis (including data analysis conclusions and offline experimental analysis conclusions), solutions to defect problems, and quality confirmation results, and periodically summarizes this data into text to form and update the knowledge base, thus constituting a closed loop in the business system process. Here, "adaptive" refers to the question-and-answer module's ability to automatically adjust its behavior or structure based on changes in the external environment or internal state to better cope with changes and optimize its performance. Specifically, in the embodiments of this disclosure, "adaptive" is reflected in the question-and-answer module's ability to automatically iterate and update the second database, adjusting its content based on influencing factor analysis conclusions, experimental analysis, and other data. "Adaptive update" refers to the periodic adjustments and updates performed by the question-and-answer module based on an adaptive mechanism. Optionally, in the embodiments of this disclosure, "adaptive update" is manifested in the question-and-answer module periodically summarizing influencing factor analysis, yield problem solutions, and other data into text and updating it to the second database, thereby continuously improving and refining the yield investigation process and content.

[0069] Optionally, the second database (i.e., the knowledge base) is not built based on offline data. Instead, the business system serves as the engine for data collection, continuously updating the knowledge base with conclusions generated during yield checks to improve timeliness and reduce the cost of manual knowledge organization. For example, in a production line, equipment inspection data, operation records, and experimental analysis results are automatically integrated into the second database, providing rich data support for subsequent semantic retrieval. Of course, this disclosure is not limited to this.

[0070] Optionally, semantic retrieval refers to parsing the semantics of a query request to match information corresponding to the semantics of the query request from a second database. In practical applications, semantic retrieval can identify historical influencing factors related to production anomalies in the second database and sort them according to their relevance to the root causes of defect resolution. For example, regarding the phenomenon of a continuously rising defect rate in the surface mount technology (SMT) process, semantic retrieval can extract text data from the second database that are similar to the query "What are the reasons for the continuous rise in surface mount defects?" and perform correlation analysis on various factors such as temperature, pressure, and raw materials. Of course, this disclosure is not limited to this.

[0071] Optionally, the sorted data for yield analysis dimensions refers to the data formed after sorting each yield analysis dimension according to the semantic search results. This sorted data provides a reference for subsequent analysis in the analysis module, enabling the selection of one or more of the most valuable yield analysis dimensions for in-depth analysis. For example, if the semantic search results show that dimensions such as temperature fluctuation, operating time, and equipment aging are all related to yield anomalies, the sorted data will arrange these dimensions according to their correlation, so as to select the appropriate yield analysis dimension for troubleshooting in operation S304. Of course, this disclosure is not limited to this.

[0072] Optionally, the first and second databases are also decoupled. That is, the first and second databases do not directly interact during data storage and processing; instead, they rely on signaling mechanisms to transmit information. The first database stores quality inspection data from the production process, while the second database serves as an adaptive iterative knowledge base for the question-and-answer module, storing data such as influencing factor analysis conclusions, solutions to defective problems, and quality confirmation results. In one aspect of this disclosure, the data in the first database is processed by the analysis module, while the question-and-answer module accesses information in the second database only through a predefined signaling interface during semantic retrieval, thus ensuring that there is no direct data interaction between the two. Of course, this disclosure is not limited to this.

[0073] In operation S304, based on the sorted data for yield analysis dimensions, the analysis module selects one or more yield analysis dimensions from multiple yield analysis dimensions.

[0074] Optionally, as mentioned above, the analysis module may employ machine learning and big data-related analysis schemes to analyze factors influencing yield through different yield analysis dimensions. Machine learning requires significant computation time when processing massive amounts of data; therefore, reducing the number of dimensions involved in yield analysis can effectively reduce the computational load. The selection process prunes the yield analysis dimensions, filtering out dimensions with less influence based on their impact on the analysis results. For example, if temperature and humidity have a significant impact on yield, while pressure has a relatively low impact, the pressure dimension can be removed, making subsequent data analysis more efficient.

[0075] In operation S305, based on one or more selected yield analysis dimensions, the analysis module performs data analysis based at least on data in the first database to generate yield influencing factors.

[0076] Optionally, the data analysis implemented by the analysis module refers to using machine learning algorithms to train historical data, automatically extracting features from the data, and building a predictive model to identify anomalies. For example, decision trees, neural networks, or support vector machines can be used to model the relationship between various yield-influencing factors and yield anomalies, thereby identifying yield-influencing factors. In contrast, the vector retrieval based on the RAG model implemented by the question-answering module focuses on matching text information similar to the input query in a large-scale semantic vector space. For example, when it is necessary to retrieve information related to a certain anomaly from a large number of documents, the RAG model can quickly find relevant qualitative descriptions based on vector similarity, while machine learning methods predict quantitative yield-influencing factors through data modeling. Of course, this disclosure is not limited to this.

[0077] Optionally, the generated yield-influencing factors involve various parameters and conditions affecting product yield, including but not limited to equipment status, process parameters, raw material characteristics, and environmental conditions. For example, slight fluctuations in temperature or pressure during a certain production stage may be related to low yield; similarly, batch-to-batch differences in raw materials may also affect product yield. During the data analysis process implemented in the analysis module, correlation analysis is performed on quality inspection data and yield anomaly events to determine the yield-influencing factors.

[0078] In response to yield anomalies, the analysis module performs data analysis to obtain quantitative results, including one or more of the following: quality inspection data analysis, product history analysis, and production parameter correlation analysis. For example, when analyzing data from each process stage of a batch of products layer by layer, inspection data from different production stages can be compared to identify data anomalies; when analyzing production parameters, correlation tests can be performed between equipment operating data and process conditions to identify potential anomalies. The identified data analysis results or yield-influencing factors are then presented to the personnel responsible for yield management (e.g., yield specialists) to conduct further offline experimental analysis.

[0079] Offline experimental analysis includes design drawing review, equipment review, material analysis, process changeover verification, and operational comparison experiments. For example, when yield-influencing factors indicate an abnormal operating status of a certain piece of equipment, yield personnel can conduct equipment review; if yield-influencing factors indicate deviations in raw material characteristics, material analysis can be performed to identify the root cause of the low yield. Optionally, each offline experiment can be specifically validated based on the data analysis results.

[0080] Optionally, the analysis module can perform data analysis not only based on data from the first database, but also by combining data from the third and fourth databases for comprehensive analysis. The third database can be a resume database, and the fourth database can be a production database. When conducting joint data analysis, aggregating different types of data from each database helps to form a more comprehensive analytical perspective.

[0081] Optionally, a history database refers to a database used to store history data. It is a database used to record detailed information about a product throughout its manufacturing, testing, and distribution processes. History data, in turn, indicates the product's production path. For example, production time, operation records, and test results for each process can be stored in the history database. A production database, on the other hand, is a database used to store production data. It typically stores various process parameters, equipment status, and operating data during the production process. Production data refers to parameters recorded during product production, such as equipment temperature, pressure, and output statistics. This disclosure is not limited to these specific examples.

[0082] Optionally, the first, third, and fourth databases are all decoupled from the second database associated with the question-and-answer model. The fourth database is the production database, where real-time production parameters are accessed in multiple application scenarios. This decoupling helps prevent the question-and-answer module from potentially impacting production data. For example, when real-time monitoring and production adjustments are performed, data delays or errors caused by the question-and-answer module's operations can be avoided, thus ensuring the stable operation of the production process.

[0083] Optionally, more comprehensive data support can be obtained by jointly searching data from multiple databases. By integrating pre-processed data from the first database, detailed product history records from the third database, and data reflecting real-time process conditions from the third database, more perspectives can be covered during analysis, thereby improving the accuracy of identifying yield anomalies. For example, if the information in a single database is insufficient, joint data analysis can fill the information gaps; alternatively, depending on actual needs, searches and data analysis can be performed only in one database to more quickly pinpoint the cause of yield anomalies.

[0084] Therefore, in one aspect of this disclosure, efficient location and analysis of yield anomalies are achieved through the interaction between the question-and-answer module and the analysis module. In a big data environment, embodiments of this disclosure effectively reduce the computation time required for machine learning by reducing invalid yield analysis dimensions, and quickly locate the causes of yield anomalies (i.e., yield-influencing factors). For example, when searching different batches of products, comparative analysis using key parameters such as temperature and humidity can accurately locate anomalies, thereby providing a reference for process adjustments and equipment maintenance.

[0085] Next, some optional details of embodiments of the present disclosure will be described with reference to FIG4. FIG4 illustrates a schematic diagram of the interaction process between the yield troubleshooting auxiliary tool and the business process according to embodiments of the present disclosure.

[0086] For example, in the yield statistics phase, the analysis module extracts quality inspection data from the first database in the first cycle and performs statistics based on dimensions such as product type, defect type, and time period. For example, for a certain batch of products, after the system collects data from each process, it statistically analyzes the product yield, defect rate, and occurrence of various defects in each process. The resulting data is processed and stored in the fifth database. The fifth database can also be called the analysis results database, and its function is to store the result data processed by the analysis module.

[0087] The analysis results data refers to any data processed by the analysis module. The analysis results data includes statistical data, such as product yield, defect rate, frequency of defects, and abnormal fluctuations in each process, presented in the form of numbers, charts, reports, etc., and serving as information to assist in judging the current production status. For example, within a certain statistical period, the yield percentage, the number of various defects, and trend charts generated by the analysis module all constitute the analysis results data, providing a reference for subsequent quality control. Optionally, the analysis results data also includes the yield anomaly information and factors affecting yield described above. This disclosure is not limited to these.

[0088] In one optional embodiment, for a batch of products, the analysis module can extract process-related data from the first database within a first cycle and perform statistical analysis based on multiple dimensions such as product type, equipment model, and operator to calculate the yield, defect rate, and frequency of defects in each process. The generated statistical data is stored in a fifth database. Subsequently, within a second cycle, the analysis module retrieves the stored data from the fifth database and compares it with historical data to generate a data comparison report and trend chart. For example, if the defect rate of a certain process shows an upward trend in consecutive cycles, the analysis module will automatically generate a chart and push the comparison results to the yield manager in the form of a data report. The first cycle and the second cycle can be the same or different. This disclosure is not limited thereto.

[0089] Furthermore, in another example, for yield fluctuations caused by equipment malfunctions on a production line, the analysis module can retrieve relevant historical records from the fifth database, correlate them with current data to generate specific statistical data and early warning reports, and deliver them to relevant personnel through the user interface. Additionally, at the end of each cycle, the analysis module can automatically generate and update reports, storing detailed data and charts in the fifth database for subsequent tracking and analysis.

[0090] During the yield anomaly investigation phase, the analysis module executes operation S301 to determine whether there are any yield anomaly events based on statistical information related to the quality inspection data. When an abnormal defect rate is detected, such as exceeding the defect rate, a continuous increase in the defect rate, or a clustering of defects in terms of location, occurrence time, and operator, the analysis module will determine that an anomaly event exists. For non-emergency anomalies, the analysis module stores the relevant information in a second database and pushes the anomaly information periodically according to a preset cycle. For example, if the defect rate of a certain process slightly increases compared to the previous cycle, the analysis module generates an anomaly report and pushes it cyclically within a predetermined time. Conversely, if an anomaly event is determined to be an emergency, the analysis module immediately stores the anomaly information in the analysis results database and immediately pushes a warning to the user via SMS, email, or analysis module notification, while marking it as emergency in the database for long-term storage and subsequent verification. Subsequently, the analysis module executes operation S302 to generate a query request based on the above yield anomaly information and sends the request to the question-and-answer module to reduce the dimensions of yield analysis.

[0091] In the influencing factor analysis phase, the yield assessment tool automatically performs multiple data processing tasks. The question-and-answer module executes operation S303, performing a semantic search in the second database to generate multiple yield analysis dimensions and ranking data for those dimensions based on the search results. For example, the question-and-answer module can rank multiple dimensions such as equipment status, production process, raw material characteristics, and operator performance according to semantic relevance, facilitating the identification of factors that may affect yield. In operation S304, the question-and-answer module sends the ranking data to the analysis module, which then selects one or more dimensions from multiple dimensions for detailed analysis in operation S305. Based on the selected yield analysis dimensions, the analysis module performs data analysis in operation S306, utilizing at least the data from the first database, to generate yield influencing factors for specific production stages. For example, when an abnormal equipment temperature or improper process operation is detected, the data analysis will provide targeted evidence for subsequent offline experimental verification.

[0092] The yield influencing factors generated during the influencing factor analysis phase are automatically pushed to the user responsible for yield management. Simultaneously, the yield management user can also proactively trigger data acquisition of yield influencing factors through the user interface. Upon receiving the yield influencing factors, the yield management user will conduct offline experimental analysis based on the actual situation. This process includes, but is not limited to, design drawing review, equipment review, material analysis, process changeover verification, and operational comparison experiments. For example, if data analysis shows abnormal equipment operation, a dedicated investigation can be conducted on the equipment; if raw material data fluctuates, in-depth testing of the raw materials can be performed. Each experiment can verify the data analysis conclusions from different perspectives, providing multi-level data support for determining the specific causes of low yield.

[0093] For the adverse influencing factors identified in the influencing factor analysis phase, those responsible for yield improvement develop solutions based on offline experimental results during the yield defect resolution phase, thereby improving product yield. For example, upon discovering an abnormality in temperature control during a production process, an equipment adjustment or maintenance plan is developed; or, upon detecting fluctuations in raw material quality, measures such as replacing raw materials or optimizing the procurement process are taken. Solutions at this stage often combine multiple data analysis results and experimental verification to gradually eliminate potential hazards affecting yield and formulate practical improvement measures.

[0094] Furthermore, as shown in Figure 4, yield-influencing factors are directly stored in the fifth database, serving as related knowledge to the second database. For example, in a specific production batch, yield-influencing factors such as temperature, humidity, equipment operating status, and process operation data are directly saved in the fifth database after collection. Subsequently, through related query requests, the second database can retrieve this data to support the output of relevant query results for the RAG large model.

[0095] In one aspect of this disclosure, a decoupled design is employed between the fifth and second databases. Both databases have fields indicating defect type identifiers, allowing for data interaction between them via associated query requests. For example, on a production line, after process data and yield-influencing factors are stored in the fifth database by the analysis module, the question-and-answer module can first find possible defect type identifiers in the second database, then initiate an associated query request containing that defect type identifier. This query request then retrieves statistical information related to the defect type identifier stored in the fifth database. Finally, based on this statistical information, qualitative information related to yield-influencing factors (including, but not limited to, the previously described sorted data for yield analysis dimensions) is generated without directly interfering with the original data storage. This ensures that both the analysis module and the question-and-answer module can operate independently, facilitating subsequent maintenance and upgrades.

[0096] During the quality verification phase, the user responsible for quality confirms the product yield based on the solutions implemented. When the product yield returns to normal, the business process is immediately closed; if the product yield has not yet recovered, the yield troubleshooting tool will notify the person responsible for yield to restart the influencing factor analysis and defect resolution steps. For example, in a yield check, if the equipment condition improves but the raw material issue remains unresolved, further analysis and adjustments to the raw materials are needed until the quality representative confirms that product quality has returned to normal. This cyclical feedback mechanism ensures that anomalies in each stage of the production process can be continuously investigated and improved.

[0097] During the quality verification phase, the Q&A module automatically stores the analysis conclusions confirmed by the quality manager, the conclusions of offline experimental analysis, the solutions to defects, and the quality verification results. It also periodically summarizes and organizes this data into text, forming and updating a second database. For example, during a troubleshooting process for equipment malfunctions, the confirmed data analysis results and experimental verification records are stored in the knowledge base for future reference when troubleshooting similar issues, thereby continuously improving the yield troubleshooting tool. The closed-loop storage of data at each stage and the continuous updating of the knowledge base provide data support and experience accumulation for quality management and anomaly handling in actual production processes.

[0098] The following describes some optional details of embodiments of the present disclosure with reference to FIG5. FIG5 shows a schematic diagram of a question-and-answer module according to an embodiment of the present disclosure, which can be used to implement some operations according to the embodiment of the present disclosure, such as operations S303 and S304. The question-and-answer module can also implement more or fewer operations. Of course, the present disclosure is not limited thereto.

[0099] As shown in Figure 5, both the question-and-answer module and the analysis module can interact with the user interface.

[0100] The user interface shown in Figure 5 serves as the interface for interaction between the yield rate auxiliary troubleshooting tool and the user, serving both the yield rate supervisor and the quality supervisor. This user interface supports interaction through various methods such as text, images, and voice. For example, the yield rate supervisor can use this user interface to query quality inspection data to understand the various indicators of product quality testing; query history data to trace the product's past production and testing records; and query production data to grasp various parameter information during the production process. The quality supervisor can use this user interface to confirm whether the yield rate is acceptable, such as determining whether the current product's yield rate meets the established standards. Simultaneously, the quality supervisor can also determine whether the current business process needs to be redirected back to the yield rate supervisor for reconfirmation of yield-related issues; for example, when an abnormal trend in the yield rate is detected, the process can be redirected for further analysis.

[0101] As mentioned earlier, this user interface supports proactively pushed data from the analysis module. The pushed data covers yield statistics across different dimensions, such as yield rates for different product models by product type; yield rates corresponding to various defect types by defect type; and yield rates for different time periods, such as daily, weekly, and monthly. In addition, it also pushes out events indicating abnormal yield rates, such as defect rates exceeding acceptable limits; continuously rising defect rates over a period of time; clustering of defect locations, such as frequent occurrences of defective products in a specific area of ​​the production line; clustering of defect occurrence times, such as higher defect rates during specific time periods each day; and clustering of operators, with some operators handling products with a higher defect rate. Furthermore, yield-influencing factors pushed by the analysis module are also displayed to the user through this user interface.

[0102] Optionally, in Figure 5, the question-and-answer module may also obtain yield confirmation information from the user interface, the yield confirmation information indicating information related to yield resolution; based on the yield confirmation information, the question-and-answer module updates the second database; based on the question data from the user interface, the question-and-answer module performs semantic retrieval in the updated second database and generates answer data related to yield influencing factors based on the semantic retrieval results; and the answer data related to yield influencing factors is presented through the user interface. Of course, this disclosure is not limited thereto.

[0103] Optionally, yield confirmation information can be data entered by the quality control personnel through the user interface during the yield troubleshooting process to confirm the resolution status of defective issues. This information can be presented in various forms such as text, images, and voice, and its content includes a description of the defect, cause analysis, improvement measures, and relevant data indicators. As shown in Figure 5, yield confirmation information obtained from the user interface through the question-and-answer module indicates that the defective issue has been resolved. The information is recorded and stored in the second database as a basis for subsequent knowledge updates.

[0104] Furthermore, the user interface supports question-and-answer interaction, allowing direct interaction between the user interface and the question-and-answer module, enabling users to obtain answers directly through the interface. The interface supports various interaction methods for users to acquire knowledge, such as text, images, and voice. Text interaction utilizes natural language understanding, with the question-and-answer module summarizing and generalizing using a large RAG model to output results as required. Image interaction is more intelligent; users input relevant images, and the question-and-answer module uses a large RAG model to recognize the image content and output the answer. Voice interaction requires initial text recognition, after which the question-and-answer module uses a large RAG model to understand the meaning and output the answer. Regardless of the interaction method, the expected output answer may include textual descriptions, relevant case images, data analysis charts, and other indicators.

[0105] To achieve this interactive approach, the RAG large model needs to learn from data in a continuously updated second database and output the content the user expects. Specifically, as shown in Figure 5, new data generated from each yield analysis conclusion confirmed by the quality control officer can be saved in text form and may be associated with charts. This data is first stored in the second database accessible to the question-and-answer module, i.e., the knowledge base.

[0106] During the generation of quality inspection data, history data, or production data, the data is stored in various formats, including text, images, charts, SQL datasets, relational datasets, and time-series datasets, in the first database (quality inspection database), the third database (history database), and the fourth database (production database), respectively, to adaptively update these three databases. For example, information such as defect descriptions, equipment status charts, and process parameters generated during the production process of a product are recorded in their respective databases for subsequent querying and analysis.

[0107] Furthermore, after data analysis by the analysis module, the data confirmed by the quality control team will be transformed into new knowledge. This summarized and organized knowledge information, including text descriptions, case images, data analysis charts, and core indicators, will be transferred to the second database (knowledge base) for systematic organization. The resulting knowledge content is dynamically updated, continuously supplemented and iterated as new defect data is generated. For example, the cause analysis and improvement measures for a certain product model, after confirmation by the quality control team, will form a new knowledge entry containing detailed descriptions and data indicators, and will be stored in the second database in real time.

[0108] In another aspect of this disclosure, the RAG big model in the question-answering module acquires yield-related knowledge according to a pre-defined calling method. The question-answering module combines existing text-based knowledge with adaptive fine-tuning to generate new knowledge, and stores this knowledge in the second database (knowledge base) used by RAG in the form of text, images, and data metrics. For example, when the system receives confirmation of a product yield anomaly, the RAG big model generates new knowledge containing root cause analysis and improvement measures based on the updated second database, and pushes it to the user through various interactive methods on the user interface.

[0109] Throughout the process, after the question-and-answer module obtains the yield confirmation information input by the user, it updates the knowledge base accordingly. By calling the RAG model, the module provides information related to resolving defects based on data in the knowledge base and user questions, achieving seamless integration of the large model with the business process. This disclosed embodiment enables the knowledge of yield investigation, root cause analysis, and improvement measures to be continuously updated and improved, providing data support and reference for subsequent product quality improvement.

[0110] The following describes some optional details of embodiments of the present disclosure with reference to FIG6. FIG6 illustrates a schematic diagram of the process of data preparation and application by the question-and-answer module according to an embodiment of the present disclosure. Of course, the present disclosure is not limited thereto.

[0111] During the stage where the question-answering module prepares data for the second database, it receives yield confirmation information from the user interface. This yield confirmation information indicates information related to yield resolution. The information transmitted from the user interface, after processing by the question-answering module, is used to update the data in the second database. This processing includes text segmentation and vectorization. Text segmentation involves breaking long texts into smaller units, such as sentences or paragraphs, for subsequent processing. For example, a technical document containing multiple paragraphs is segmented into individual sentences or paragraphs for processing one by one. Vectorization is the process of converting text into numerical vectors, making it understandable and processable by a computer. Common methods include bag-of-words models, TF-IDF, and word embeddings. For example, word embedding models are used to convert each word into a high-dimensional vector, thereby capturing the semantic relationships between words. The vectorized data is stored in the second database (i.e., the knowledge base used by the RAG large model).

[0112] In addition to yield confirmation information, content such as white papers can be processed through text segmentation and vectorization and stored in a second database. For example, a white paper on production process optimization can be divided into multiple paragraphs, and each paragraph can be converted into a vector and stored in the database for subsequent retrieval and analysis. Of course, this disclosure is not limited to this.

[0113] In the application phase of the question-and-answer module, users input question data, and the question-and-answer data is also processed by text segmentation and vectorization. Then, semantic retrieval is performed in the updated second database to retrieve professional knowledge related to the problem, thereby generating answer data related to factors affecting yield. This answer data is finally presented through the user interface.

[0114] For example, a user asks, "How can I improve the yield rate of my production line?" The question-and-answer module first performs text segmentation and vectorization on the question. Then, it retrieves professional knowledge related to keywords such as "production line," "improve," and "yield rate" from a second database. Through semantic retrieval, the module may find the latest yield rate solution confirmed by the quality control officer. The module extracts the relevant content from this article and generates answer data, such as "Through lean manufacturing methods, the yield rate of the production line can be effectively improved." Finally, the module presents this answer to the user through the user interface.

[0115] Optionally, in one embodiment of this disclosure, the analysis module stores yield influencing factors in a fifth database decoupled from the second database, and both the fifth and second databases have fields indicating defect type identifiers. To utilize data in the fifth database as much as possible, the question-and-answer module can perform semantic retrieval in the updated second database to determine defect type identifiers associated with the problem data. The question-and-answer module then generates and sends an association query request containing the defect type identifier to the fifth database. Subsequently, the question-and-answer module receives a response to the association query request from the fifth database, and based on the response to the association query request, generates answer data related to the yield influencing factors.

[0116] For example, when investigating a problem, the question-and-answer module can statistically analyze historical data to obtain the frequency of occurrence, types of abnormal manifestations, and probabilities of each type of abnormality. This information is then integrated with semantic search results to generate qualitative answer data. For instance, regarding the problem of "production line downtime," the question-and-answer module might leverage the association between the second and fifth databases. By generating a query request with a corresponding defect type identifier for "production line downtime," it can retrieve a response in the fifth database—that the "production line downtime" occurred 50 times, 30 of which were due to equipment malfunction and 20 due to operational errors. Combined with the semantic search results, the question-and-answer module can generate the answer: "Production line downtime problems are mainly caused by equipment malfunction (60%) and operational errors (40%)." This answer data is presented through the user interface to help users understand the root cause of the problem.

[0117] Next, some optional details of embodiments of the present disclosure will be described with reference to FIG7. FIG7 shows a schematic diagram of the data structure of data stored in the second database according to an embodiment of the present disclosure. Of course, the present disclosure is not limited thereto.

[0118] In yield rate investigation scenarios, the data format in the second database is not limited, but the common data format is text QA pairs. The question-and-answer module receives input from the user interface, where the question data consists of two parts: the first part is a description of the defect type, such as "<defect type> exceeds the standard" or "<defect type> continues to rise", or it can be expressed as "<qualifier> occurs in clusters", etc.; the second part is usually a question about the cause, such as "What is the cause?" This format facilitates unified recognition and processing by the system.

[0119] For past adverse issues, the answer data typically contains multiple layers of information to provide comprehensive support for subsequent investigation and analysis. The answer data includes at least one of the following: descriptive information about the adverse issue in the problem data, the type of manifestation corresponding to the adverse issue, the adverse influencing factors corresponding to the type of manifestation, experimental conclusions regarding the adverse influencing factors, and solutions corresponding to the experimental conclusions. This disclosure is not limited to these.

[0120] Figure 7 presents the hierarchical structure of this answer data in the form of a tree diagram, where each node contains probability information indicating the probability of the content specified in that node occurring. However, it's important to note that the presentation format of the answer data is not limited to a tree diagram. Depending on the actual needs, the answer data can also be presented in the form of articles or paragraphs; the specific display method will be adjusted according to user requirements or interface design requirements.

[0121] In a tree diagram format, the answer data typically starts from the most basic level and gradually expands to more detailed analysis. For example, the first level might simply be a brief description of the problem (i.e., descriptive information about the problem), such as "a certain type of defect has exceeded the standard" or "a certain type of defect continues to rise." The second level further provides statistical data related to the problem, such as the frequency of occurrence in history, the time period of occurrence, and the manifestation of the problem under different circumstances. The third level could be a detailed analysis of the abnormal manifestations of defective problems, listing different clustering patterns (such as point clustering, corner clustering, etc.) and their corresponding probabilities and influencing factors.

[0122] As an example, firstly, the answer data will be displayed at the root node (problem problem) in Figure 7, showing the number or frequency of the problem in history, helping to understand the prevalence of the problem. For example, if a certain type of problem has occurred 50 times in the past three months, the second database will record this information for subsequent analysis.

[0123] Secondly, the answer data will describe the abnormal behavior of the problem at the anomaly nodes. Specifically, different types of clustering problems may exhibit different anomaly behaviors (i.e., the types of manifestations of the problem). For example, in map clustering problems, different behaviors such as corner clustering, point clustering, line clustering, or ring clustering may be observed. The probability of occurrence of each anomaly type will also be recorded and analyzed to help investigators identify the trends of different anomalies.

[0124] Furthermore, the answer data will also provide the influencing factors for each abnormal manifestation (i.e., the adverse influencing factors corresponding to the type of defect) at the data analysis conclusion node. This includes the name of the influencing factor, the analysis method, the importance ranking of multiple influencing factors, and the probability of association between each influencing factor and the defect. For example, in the case of corner clustering issues, temperature fluctuations might be found to be a significant influencing factor with a 70% probability of affecting the problem, while airflow issues might have a 30% probability of affecting the problem. This information will provide guidance for subsequent investigation and improvement.

[0125] Furthermore, for each factor affecting yield, the answer data will include conclusions drawn from offline experimental analysis (i.e., experimental conclusions for the adverse influencing factor) at the experimental conclusion node. This includes the experimental analysis method, the specific root cause of the defect, and the probability of it causing the problem. For example, if experiments show that the probability of corner aggregation significantly increases under specific temperature conditions, then this temperature fluctuation can be identified as the root cause and specifically pointed out in the analysis conclusion. For instance, temperature fluctuations lead to a 50% probability of aggregation. Optionally, the user interface will present relevant experimental analysis data to help yield personnel gain a deeper understanding of the root cause of the problem.

[0126] Finally, the answer data will also include, at the "improvement methods" node, the improvement methods targeting the root causes of the problems and their implementation effects (i.e., the solutions corresponding to the experimental conclusions of the adverse influencing factors). Improvement methods may include adjusting production parameters, optimizing equipment configuration, or improving operating procedures, and the implementation effects will be verified through subsequent data statistics and tracking. For example, by adjusting the temperature control system, the frequency of related adverse problems may be significantly reduced, and the improvement effect will be reflected in the answer data.

[0127] For example, when the yield specialist queries "What are the reasons for the clustering of a certain type of defect?", the question-and-answer module first identifies the defect type and then queries the fifth database (which is associated with the second database and is also the database where the analysis module outputs analysis results) to retrieve the most recent experimental data and historical statistical data based on that defect type. The retrieved data, after text segmentation and vectorization, is stored in the second database.

[0128] The question-answering module retrieves various data related to poor clustering through semantic retrieval. By integrating with the RAG large model, these retrieval results are transformed into answer data, which is then displayed on the user interface, showing specific anomalies, influencing factors, and experimental analysis results.

[0129] If, during subsequent investigations, the quality control representative provides yield confirmation information for a specific question and its corresponding answer, the Q&A module will update this information and store the updated data in a second database as QA pairs. In this way, both question and answer data are vectorized and stored in a paired data format. Each question-and-answer pair will function as a complete data unit, supporting further analysis and querying.

[0130] For example, when encountering a similar problem again, the question-and-answer module can quickly retrieve relevant answers based on stored QA pairs, thereby reducing repetitive query processes and improving troubleshooting efficiency. Secondly, the vectorized data format allows for more efficient data processing and analysis. Through vectorization, the system can capture semantic information in text data, avoiding simple keyword matching and thus improving the accuracy and reliability of retrieval. Finally, with the addition of more yield confirmation information, the question-and-answer module can continuously optimize the data content in the second database, gradually accumulating knowledge and forming a self-updating and optimizing knowledge base.

[0131] This approach not only effectively solves the large model illusion problem and ensures the accuracy of the answers, but also protects the security of the fifth database. For example, experimental data and historical statistical data are stored in the fifth database, indexed by product batch information or time information during storage. Content in the answer data, such as "the type of anomaly manifestation of the problem," "the conclusions of influencing factors for the problem under each anomaly manifestation type," and "improvement methods and implementation effects for the root causes of problems," will be accompanied by a list of relevant batch or time information.

[0132] If a user wants to query further details of a particular answer, the RAG large model can extract batch or time information from the answer data and link it to detailed result data through indexing. This allows users to obtain more precise analytical data. For example, a user might want to view the improvement effect over a specific time period; the system can link the experimental data and results for that time period through indexing and display them to the user, enabling them to make more accurate decisions.

[0133] According to another aspect of this disclosure, an electronic device is also provided for implementing the method according to an embodiment of this disclosure. FIG8 shows a schematic diagram of an electronic device 2000 according to an embodiment of this disclosure.

[0134] As shown in Figure 8, the electronic device 2000 may include one or more processors 2010 and one or more memories 2020. The memories 2020 store computer-readable code, which, when executed by the one or more processors 2010, can perform the methods described above.

[0135] The processor in this disclosure embodiment can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, operations, and logic block diagrams disclosed in this disclosure embodiment. The general-purpose processor can be a microprocessor or any conventional processor, and can be based on an x86 architecture or an ARM architecture.

[0136] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0137] For example, the methods or apparatus according to embodiments of this disclosure can also be implemented using the architecture of the computing device 3000 shown in FIG. 9. As shown in FIG. 9, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. Storage devices in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the methods provided in this disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, the architecture shown in FIG. 9 is merely exemplary, and one or more components in the computing device shown in FIG. 9 may be omitted as needed when implementing different devices.

[0138] According to another aspect of this disclosure, a computer-readable storage medium is also provided. Figure 10 shows a schematic diagram of the storage medium 4000 according to this disclosure.

[0139] As shown in FIG10, computer-readable instructions 4010 are stored on the computer storage medium 4020. When the computer-readable instructions 4010 are executed by a processor, the methods according to embodiments of the present disclosure described with reference to the above figures can be performed. The computer-readable storage medium in the embodiments of the present disclosure may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memory used in the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0140] This disclosure also provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform a method according to an embodiment of this disclosure.

[0141] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0142] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0143] The exemplary embodiments of this disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will understand that various modifications and combinations of these embodiments or their features can be made without departing from the principles and spirit of this disclosure, and such modifications should fall within the scope of this disclosure. For example, the following provides an overview of some aspects of this disclosure, which can be combined with any other aspects.

[0144] Aspect 1: A method for processing yield information is provided, comprising: determining yield anomaly information of industrial products by an analysis module based on a first database; generating a query request and sending the query request to a question-and-answer module based on the yield anomaly information; performing semantic retrieval in a second database based on the query request and generating multiple yield analysis dimensions and ranking data for the multiple yield analysis dimensions based on the semantic retrieval results; sending the ranking data for the yield analysis dimensions to the analysis module by the question-and-answer module; selecting one or more yield analysis dimensions from the multiple yield analysis dimensions based on the ranking data for the yield analysis dimensions; and performing data analysis by the analysis module based at least on data in the first database to generate yield influencing factors based on the selected one or more yield analysis dimensions.

[0145] Aspect 2: In combination with any other aspect, the analysis module is used to provide big data analysis services based on machine learning, the question answering module is used to provide question answering services based on a large model generated by retrieval enhancement, and the analysis module is decoupled from the question answering module.

[0146] Aspect 3: In conjunction with any other aspect, the query request is a signaling transmitted between the analysis module and the question-answering module, and the query request includes semantic information for instructing the question-answering module to determine the cause of the yield anomaly.

[0147] Aspect 4: In combination with any other aspect, the fields of the query request include at least one of the following: defect type, defect phenomenon, defect occurrence conditions, and query target information.

[0148] Aspect 5: In combination with any other aspect, the first database is used to store quality inspection data related to product quality inspection, and the second database is used to store knowledge data related to yield anomalies that are adaptively updated by the question-and-answer module, and the first database and the second database are decoupled.

[0149] Aspect 6: In combination with any other aspect, the data analysis performed by the analysis module based at least on data in the first database to generate yield influencing factors includes: the data analysis performed by the analysis module based on data in the first database, the third database, and the fourth database to generate yield influencing factors, wherein the third database is used to store historical data indicating the production path of the product, and the fourth database is used to store production data indicating the production parameters of the product.

[0150] Aspect 7: In combination with any other aspect, the method further includes: obtaining yield confirmation information received from the user interface by the question-and-answer module, the yield confirmation information indicating information related to yield resolution; updating the second database by the question-and-answer module based on the yield confirmation information; performing semantic retrieval in the updated second database by the question-and-answer module based on question data from the user interface and generating answer data related to yield influencing factors based on the results of the semantic retrieval; and presenting the answer data related to yield influencing factors through the user interface.

[0151] Aspect 8: In combination with any other aspect, the answer data includes at least one of the following: descriptive information of the problem data, the manifestation type corresponding to the problem, the adverse influencing factors corresponding to the manifestation type, the experimental conclusions of the adverse influencing factors, and the solutions corresponding to the experimental conclusions.

[0152] Aspect 9: In conjunction with any other aspect, the answer data is presented on the user interface in the form of a tree diagram, and each node of the tree diagram has probability information indicating the probability of the content indicated in the node occurring.

[0153] Aspect 10: In conjunction with any other aspect, the analysis module stores yield-influencing factors in a fifth database, which is decoupled from the second database, and both the fifth and second databases have fields associated with defect type identifiers.

[0154] Aspect 11: A device for processing yield information is provided, the device comprising a decoupled analysis module and a question-answering module, and a decoupled first database and a second database, wherein the analysis module is configured to: determine yield anomaly information of industrial products based on the first database; and generate a query request based on the yield anomaly information and send the query request to the question-answering module; the question-answering module is configured to: perform semantic retrieval in the second database based on the query request and generate multiple yield analysis dimensions and ranking data for the multiple yield analysis dimensions based on the results of the semantic retrieval; and send the ranking data for the yield analysis dimensions to the analysis module; wherein the analysis module is further configured to: select one or more yield analysis dimensions from the multiple yield analysis dimensions based on the ranking data for the yield analysis dimensions; and perform data analysis based on the selected one or more yield analysis dimensions, at least based on data in the first database, to generate yield influencing factors.

[0155] Aspect 12: An analysis apparatus for processing yield information is provided, comprising: one or more memories; and one or more processors coupled to the one or more memories, the one or more processors being configured to: determine yield anomaly information of industrial products based on a first database; generate a query request based on the yield anomaly information and send the query request to a question-and-answer apparatus for processing yield information; receive sorted data for yield analysis dimensions from the question-and-answer apparatus for processing yield information; select one or more yield analysis dimensions from a plurality of yield analysis dimensions based on the sorted data for yield analysis dimensions; and perform data analysis based on the selected one or more yield analysis dimensions, at least based on data in the first database, to generate yield influencing factors.

[0156] Aspect 13: A question-and-answer apparatus for processing yield information is provided, comprising: one or more memories; and one or more processors coupled to the one or more memories, the one or more processors being configured to: receive a query request based on yield anomaly information from an analysis apparatus for processing yield information; perform a semantic search in a second database based on the query request and generate multiple yield analysis dimensions and sorted data for the multiple yield analysis dimensions based on the results of the semantic search; and send the sorted data for the yield analysis dimensions to the analysis apparatus for processing yield information.

[0157] Aspect 14: An electronic device is provided, comprising: one or more processors; and one or more memories, wherein the memories store a computer-executable program that, when executed by the processor, performs the method described in any aspect.

[0158] Aspect 15: A computer-readable storage medium is provided having computer instructions stored thereon, which, when executed by a processor, implement the method described in any aspect.

Claims

1. A method for processing yield information, comprising: Based on the first database, the analysis module determines abnormal yield information of industrial products. Based on the yield anomaly information, the analysis module generates a query request and sends the query request to the question-and-answer module; Based on the query request, the question-answering module performs semantic retrieval in the second database and generates multiple yield analysis dimensions and sorted data for the multiple yield analysis dimensions based on the semantic retrieval results; The question-and-answer module sends the sorted data for the yield analysis dimension to the analysis module; Based on the sorted data for yield analysis dimensions, the analysis module selects one or more yield analysis dimensions from the plurality of yield analysis dimensions; as well as Based on one or more selected yield analysis dimensions, the analysis module performs data analysis based at least on data in the first database to generate yield influencing factors.

2. The method as described in claim 1, wherein, The analysis module is used to provide big data analysis services based on machine learning, and the question answering module is used to provide question answering services based on a large model generated by retrieval enhancement. The analysis module and the question answering module are decoupled.

3. The method as described in claim 2, wherein, The query request is a signaling transmitted between the analysis module and the question-answering module, and the query request includes semantic information for instructing the question-answering module to determine the cause of the yield abnormality.

4. The method of claim 2, wherein, The fields of the query request include at least one of the following: defect type, defect phenomenon, defect occurrence conditions, and query target information.

5. The method of claim 1, wherein, The first database is used to store quality inspection data related to product quality inspection, and the second database is used to store knowledge data related to yield anomalies that are adaptively updated by the question-and-answer module. The first database and the second database are decoupled.

6. The method of claim 1, wherein, The process of generating yield influencing factors by the analysis module based at least on data in the first database includes: The analysis module performs data analysis based on data from the first, third, and fourth databases to generate factors affecting yield. The third database is used to store historical data indicating the production path of a product, and the fourth database is used to store production data indicating the production parameters of a product.

7. The method of claim 1, wherein, The method further includes: The question-and-answer module obtains yield confirmation information received from the user interface, which indicates information related to resolving yield issues. Based on the yield confirmation information, the question-and-answer module updates the second database; Based on question data from the user interface, the question-answering module performs semantic retrieval in the updated second database and generates answer data related to yield influencing factors based on the semantic retrieval results; and The user interface presents the answer data related to the factors affecting yield.

8. The method of claim 7, wherein, The answer data includes at least one of the following: The problematic data contains inappropriate descriptive information. The types of manifestations corresponding to the defects, The adverse influencing factors corresponding to the aforementioned manifestation type The experimental conclusions of the adverse influencing factors, and The solutions corresponding to the experimental conclusions.

9. The method of claim 8, wherein, The answer data is presented on the user interface in the form of a tree diagram, and each node of the tree diagram has probability information, which indicates the probability of the content indicated in the node occurring.

10. The method of claim 7, wherein, The analysis module stores yield-influencing factors in a fifth database, which is decoupled from the second database. Both the fifth and second databases have fields indicating defect type identifiers. The step of the question-and-answer module performing semantic retrieval in the updated second database and generating answer data related to yield-influencing factors based on the semantic retrieval results includes: The question-and-answer module performs a semantic search in the updated second database to determine the bad type identifier associated with the question data; The question-and-answer module generates and sends an associated query request with the bad type identifier to the fifth database; The question-answering module receives a response to the related query request from the fifth database, and The question-and-answer module generates answer data related to factors affecting yield based on the response to the associated query request.

11. An apparatus for processing yield information, the apparatus comprising a decoupled analysis module and a question-and-answer module, and a decoupled first database and a second database, wherein, The analysis module is configured as follows: Based on the first database, abnormal yield information of industrial products is determined; and Based on the yield anomaly information, a query request is generated and sent to the question-and-answer module; The question-and-answer module is configured as follows: Based on the query request, semantic retrieval is performed in the second database, and multiple yield analysis dimensions and sorted data for the multiple yield analysis dimensions are generated based on the results of the semantic retrieval. Send the sorted data for the yield analysis dimension to the analysis module; The analysis module is further configured as follows: Based on the sorted data for yield analysis dimensions, select one or more yield analysis dimensions from multiple yield analysis dimensions; and Based on one or more selected yield analysis dimensions, data analysis is performed using at least the data from the first database to generate yield influencing factors.

12. An analysis device for processing yield information, comprising: One or more memory units; and One or more processors coupled to the one or more memories, the one or more processors being configured to: Based on the first database, abnormal yield information of industrial products is identified; Based on the yield anomaly information, a query request is generated and sent to the question-and-answer device that processes yield information. Receive sorted data for yield analysis dimensions from the question-and-answer device that processes yield information; Based on the sorted data for yield analysis dimensions, select one or more yield analysis dimensions from multiple yield analysis dimensions; as well as Based on one or more selected yield analysis dimensions, data analysis is performed using at least the data from the first database to generate yield influencing factors.

13. A question-and-answer device for processing yield information, comprising: One or more memory units; and One or more processors coupled to the one or more memories, the one or more processors being configured to: Receive query requests based on yield anomaly information from the analysis device that processes yield information; Based on the query request, semantic retrieval is performed in the second database, and multiple yield analysis dimensions and sorted data for the multiple yield analysis dimensions are generated based on the results of the semantic retrieval. as well as The sorted data for the yield analysis dimension is sent to the analysis device that processes yield information.

14. An electronic device comprising: One or more processors; and One or more memories, wherein the memories store a computer-executable program that, when executed by the processor, performs the method of any one of claims 1-10.

15. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the method as claimed in any one of claims 1-10.