Manufacturing industry production process optimization method and device based on large model

Abstract

The invention provides a manufacturing industry production process optimization method and device based on a large model. Multi-modal data, including text logs, equipment parameters, image data and sensor signals, in the production process are collected in real time through a sensor network, a production management system and quality detection equipment; performing cleaning, labeling and format conversion on the collected data to enable the collected data to adapt to the input requirements of the large language model; analyzing and reasoning the preprocessed data through a large language model subjected to industry data fine adjustment, and generating an optimization suggestion and a decision scheme; an output result of the large language model is converted into a production instruction, and production parameters and a technological process are adjusted in real time through an interaction interface with production equipment; the optimization effect is evaluated by monitoring key indexes in the production process, and the result is fed back to the large language model, so that closed-loop optimization is realized. According to the scheme, the production process of the manufacturing industry can be systematically optimized, and the intelligent level of the manufacturing industry is improved.

Description

Technical Field

[0001] This invention relates to the field of large-scale model technology, and in particular to a method and apparatus for optimizing production processes in the manufacturing industry based on large-scale models. Background Technology

[0002] Traditional manufacturing processes rely primarily on human experience, historical data analysis, and simple automation tools for optimization. However, with the expansion of production scale and the increase in process complexity, the amount of production data is exploding, making it difficult for traditional methods to efficiently process this massive amount of data.

[0003] In recent years, Large Language Models (LLMs) have made significant progress in natural language processing and multimodal data fusion. Their powerful language understanding and generation capabilities enable them to handle complex text, image, and sensor data. However, applying LLMs to production process optimization in the manufacturing industry still faces challenges. On the one hand, manufacturing data is highly specialized and complex, requiring targeted training and fine-tuning of the model; on the other hand, production process optimization demands real-time performance and accuracy, while the response speed and decision-making accuracy of LLMs need further improvement.

[0004] Currently, while some research attempts to introduce artificial intelligence technology into the manufacturing industry, most focus on optimizing single aspects, such as quality inspection or equipment failure prediction, lacking a systematic optimization of the entire production process. Therefore, developing a manufacturing process optimization method and system based on large language models is of great significance for improving the intelligence level of the manufacturing industry. Summary of the Invention

[0005] This invention provides a method and apparatus for optimizing manufacturing processes based on a large model, which can systematically optimize manufacturing processes and improve the level of intelligence in the manufacturing industry.

[0006] According to one aspect of the present invention, a method for optimizing manufacturing processes based on a large model is provided, comprising: Through sensor networks, production management systems, and quality inspection equipment, multimodal data during the production process is collected in real time, including text logs, equipment parameters, image data, and sensor signals; the collected data is cleaned, labeled, and formatted to adapt it to the input requirements of large language models. By using a large language model fine-tuned with industry data, the preprocessed data is analyzed and reasoned to generate optimization suggestions and decision-making solutions. The output of the large language model is converted into production instructions, and production parameters and processes are adjusted in real time through the interaction interface with the production equipment. By monitoring key indicators in the production process, the optimization effect is evaluated, and the results are fed back to the large language model to achieve closed-loop optimization.

[0007] Optionally, the process of cleaning, labeling, and format conversion of the collected data includes: Noise is removed by statistical filtering, missing values ​​are filled by linear interpolation, and outliers are identified and marked based on the 3σ principle. The YOLO object detection algorithm was used to assist manual annotation of image and video data. The annotation labels included equipment operating status labels and product defect labels. At the same time, texture features, morphological features and spatial location features were extracted. The text logs were segmented using a word segmentation tool combined with a professional dictionary for the manufacturing industry, and a semantic graph of the production scenario was constructed through part-of-speech tagging and semantic association analysis. Text data, image feature data, and sensor time-series data are converted into tensor formats compatible with large language models, while preserving the metadata of data relationships.

[0008] Optionally, the step of analyzing and reasoning about the preprocessed data using a large language model fine-tuned with industry data to generate optimization suggestions and decision-making solutions includes: Perform semantic deep analysis on production logs and operation records to identify potential problems such as equipment failure precursors, process bottlenecks, and non-standard operations, and extract optimization points in the production process; Frame-level analysis of images and video data from the production site is performed to detect the wear status of equipment parts, product assembly accuracy, and production environment compliance, and output quantitative test results. By using a cross-modal attention mechanism, text semantic features, image visual features, and sensor temporal features are dynamically weighted and fused to generate an optimization scheme that includes equipment parameter adjustment values, process step change suggestions, and production scheduling priorities.

[0009] Optionally, the step of converting the output of the large language model into production instructions, and adjusting production parameters and processes in real time through an interface with production equipment, includes: Based on the reasoning results of the large language model, and combined with the weight coefficients of the production target, a differentiated optimization scheme is generated. The implementation effects of each optimization scheme in the production scenario are simulated by digital twin models. The efficiency improvement rate, cost reduction rate and quality pass rate improvement of each scheme are quantitatively evaluated to determine the optimal scheme. The optimal solution is transformed into specific production instructions that conform to industrial communication protocols. The instructions include equipment parameter adjustment thresholds, changes in the execution sequence of process steps, and production resource scheduling and allocation schemes. At the same time, instruction execution documentation is generated. When the production environment changes, the solution reselection mechanism is automatically triggered to generate a temporary optimized solution and execute it.

[0010] Optionally, the step of evaluating the optimization effect by monitoring key indicators in the production process and feeding the results back to the large language model to achieve closed-loop optimization includes: Real-time monitoring includes key indicators such as production efficiency, equipment utilization rate, product qualification rate, raw material loss rate, and energy consumption rate, with the monitoring frequency consistent with the data acquisition frequency; The optimization effect is quantitatively evaluated using a weighted scoring method. The scoring weights correspond to the production target weight coefficients, and the evaluation results are divided into excellent, good, qualified, and unqualified. When the evaluation result is qualified or below, the detailed evaluation data is fed back to the large language model, the model parameters are updated using incremental learning, the adjusted optimization strategy is generated and issued for execution; when the evaluation result is excellent or good, the current optimization scheme and corresponding production scenario data are recorded for subsequent model reinforcement training.

[0011] Optionally, the method further includes: providing a natural language interactive interface, enabling production managers to input questions via voice or text to obtain system optimization suggestions and decision support; wherein, the natural language interactive interface supports three input methods: voice, text, and handwriting; the output includes optimization suggestions and decision support in natural language text, voice broadcast, and visual charts.

[0012] Optionally, the fine-tuning of the large language model includes: Acquire training data including production logs, equipment manuals, process standards, failure cases, and quality inspection specifications from various sub-scenarios within the manufacturing industry; Fine-tuning was performed using mini-batch gradient descent, with batch size set between 32 and 64, initial learning rate of 1e-5, decaying by 50% every 20 rounds, and at least 100 iterations. After fine-tuning, the model achieved an accuracy of at least 92% in identifying production problems, a feasibility verification pass rate of at least 88% for optimization suggestions, and an inference response time of no more than 500ms.

[0013] According to another aspect of the present invention, a manufacturing process optimization device based on a large model is provided, comprising: The data acquisition unit is used to collect multimodal data in real time during the production process through sensor networks, production management systems, and quality inspection equipment, including text logs, equipment parameters, image data, and sensor signals; and to clean, label, and convert the collected data to adapt it to the input requirements of large language models. The large language model analysis unit is used to analyze and reason about preprocessed data using a large language model that has been fine-tuned with industry data, and to generate optimization suggestions and decision-making solutions. The intelligent decision-making unit is used to convert the output of the large language model into production instructions and adjust production parameters and process flow in real time through the interaction interface with the production equipment. The real-time feedback unit is used to evaluate the optimization effect by monitoring key indicators in the production process and feed the results back to the large language model to achieve closed-loop optimization.

[0014] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute the large-model-based manufacturing process optimization method according to any embodiment of the present invention.

[0015] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the manufacturing process optimization method based on a large model as described in any embodiment of the present invention.

[0016] The solution presented in this invention leverages the powerful language understanding and generation capabilities of a large language model, combined with production data from the manufacturing industry, to achieve dynamic optimization of the production process through multimodal data fusion, intelligent decision-making algorithms, and real-time feedback mechanisms. The system can automatically identify bottlenecks in the production process, generate optimization suggestions, and adjust production parameters in real time, thereby improving production efficiency, reducing production costs, and enhancing product quality. This invention is applicable to various manufacturing scenarios and has broad industrial application prospects.

[0017] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a manufacturing process optimization method based on a large model, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of a manufacturing process optimization device based on a large model, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device that implements the manufacturing process optimization method based on a large model according to embodiments of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] This invention provides a method for optimizing manufacturing processes based on a large model, which may include the following steps: S110: Through sensor networks, production management systems, and quality inspection equipment, collect multimodal data in real time during the production process, including text logs, equipment parameters, image data, and sensor signals; clean, label, and convert the collected data to adapt it to the input requirements of large language models.

[0023] Specifically, sensor networks are responsible for capturing real-time physical signals on the production floor, the Manufacturing Execution System (MES) records business data throughout the entire production process, and quality inspection equipment focuses on product quality-related data. Real-time acquisition ensures that the dynamic changes in the production process are reflected synchronously, avoiding distorted optimization decisions due to data lag. Among the multimodal data, text logs contain unstructured text information such as production instructions and fault records; equipment parameters cover quantifiable equipment operating indicators such as temperature, pressure, and operating speed; image data is a visual representation of the production floor, product appearance, and equipment status; and sensor signals are the raw physical quantities captured by various sensors.

[0024] Since the collected data comes from multiple sources, it may contain issues such as noise, inconsistent formatting, and lack of annotation. Directly inputting it into a large language model can lead to biased or even invalid analysis results. Therefore, it is necessary to optimize data quality through cleaning, annotation, and format conversion. Cleaning mainly involves removing noisy data, filling in missing values, and correcting outliers to ensure the accuracy and reliability of the data. Annotation adds semantic tags to data without clear meaning to help the large language model understand the data's connotation. Format conversion unifies data of different types and formats into a standard format that the large language model can recognize and process, eliminating data format barriers.

[0025] S120. Using a large language model fine-tuned with industry data, the preprocessed data is analyzed and reasoned to generate optimization suggestions and decision-making solutions.

[0026] Specifically, general-purpose large language models lack manufacturing industry expertise, such as specific process standards, equipment failure patterns, and production optimization logic. By fine-tuning the model through input of specialized data such as historical production data, process standards, and failure cases from the manufacturing industry, the model can accurately understand the specific needs of the manufacturing scenario, avoiding ineffective suggestions caused by generalized analysis. Natural language processing technology is used to parse the semantics of text logs and identify potential problems; image recognition technology is combined with visual data analysis, while quantitative data such as sensor signals and equipment parameters are integrated for multi-dimensional correlation analysis to find core problems, such as a decrease in product qualification rate due to abnormal equipment parameters. Generating optimization suggestions and decision-making solutions is the output of the model analysis. These suggestions and solutions must be practical, covering specific content such as equipment parameter adjustment values, process step optimization directions, and production scheduling priorities.

[0027] S130: The output of the large language model is converted into production instructions, and production parameters and process flow are adjusted in real time through the interaction interface with the production equipment.

[0028] Specifically, the optimization suggestions and decision-making schemes output by the large language model are essentially strategic content and cannot be directly recognized and executed by production equipment. Therefore, they need to be transformed into production instructions, breaking down the abstract optimization suggestions into specific operational instructions that the equipment can understand, such as "adjust the operating speed of equipment A to 300 r / min" or "advance the execution sequence of process B before process C." The instruction format must conform to the communication protocol standards of industrial equipment, such as OPC UA and Modbus. The interaction interface with the production equipment is to achieve compatible integration between the model output and equipment control, ensuring that instructions can be accurately and stably transmitted to the target equipment. Real-time adjustment of production parameters and process flow is the execution action, emphasizing the timeliness of optimization. When bottlenecks or anomalies occur during production, problems can be solved immediately by adjusting parameters and optimizing processes, preventing the expansion of production losses.

[0029] S140. By monitoring key indicators in the production process, the optimization effect is evaluated, and the results are fed back to the large language model to achieve closed-loop optimization.

[0030] Specifically, monitoring key indicators (KPIs) during the production process clarifies the basis for performance evaluation. These KPIs must be set around production goals and include quantifiable metrics such as production efficiency, product qualification rate, raw material loss rate, equipment utilization rate, and energy consumption rate. Real-time tracking of these KPIs provides a clear picture of the actual effectiveness of optimization measures. Evaluating the optimization results requires reaching clear conclusions, not only judging the effectiveness but also analyzing the reasons for any positive or negative outcomes. The results are fed back to the large-scale language model, with feedback data including KPI changes and problem attribution analysis, allowing the model to understand the actual effectiveness of previous optimization schemes. Based on the feedback data, the model adjusts its analytical logic and optimization strategies. If the optimization effect is unsatisfactory, the model will reanalyze the data and revise parameter suggestions to ensure that the optimization scheme continuously adapts to dynamic changes in production, rather than being a one-time static optimization.

[0031] In this embodiment of the invention, the collected data is cleaned, labeled, and converted in format, including: Noise is removed by statistical filtering, missing values ​​are filled by linear interpolation, and outliers are identified and marked based on the 3σ principle. The YOLO object detection algorithm was used to assist manual annotation of image and video data. The annotation labels included equipment operating status labels and product defect labels. At the same time, texture features, morphological features and spatial location features were extracted. The text logs were segmented using a word segmentation tool combined with a professional dictionary for the manufacturing industry, and a semantic graph of the production scenario was constructed through part-of-speech tagging and semantic association analysis. Text data, image feature data, and sensor time-series data are converted into tensor formats compatible with large language models, while preserving the metadata of data relationships.

[0032] Specifically, statistical filtering filters out meaningless noise data generated by environmental interference and electromagnetic radiation from the sensor, such as instantaneously fluctuating temperature values ​​and messy sensor signals, by selecting reasonable fluctuation ranges in the data. Linear interpolation is suitable for scenarios where a small amount of data is missing due to data transmission delays or brief sensor offline times during the production process. It supplements missing values ​​by linear fitting of adjacent valid data, ensuring the continuity of the data sequence and meeting the requirements of large language models for the integrity of time-series data. The 3σ principle (deviation from the data mean by 3 times the standard deviation) can accurately mark extreme data caused by equipment failure, operational errors, etc., such as the overload pressure when the equipment is overloaded or the serious deviation data of product size, and only mark it instead of deleting it directly.

[0033] The YOLO object detection algorithm boasts advantages such as strong real-time performance and accurate positioning. It can quickly identify core targets such as key equipment components and product bodies from production site images / videos, reducing repetitive manual annotation work. Then, manual review is used to correct algorithm annotation deviations. Equipment operation status labels and product defect labels are the core semantic information of visual data in manufacturing scenarios, providing clear guidance for large language models to understand the connotation of images. Extracting texture features, morphological features, and spatial location features is the key to transforming visual data into data that can be quantified and analyzed by the model, allowing the large language model to interpret the production status information behind the images.

[0034] Text logs contain a large number of manufacturing industry terms, and general word segmentation tools are prone to errors. Combining them with a manufacturing industry-specific dictionary can ensure the accuracy of term segmentation. Part-of-speech tagging helps large language models clarify the grammatical functions of words in the production scenario, while semantic association analysis focuses on the logical relationships between words, such as the causal relationship between "equipment A" and "excessive temperature", and the association between "process B" and "product qualification rate". Constructing a semantic graph of the production scenario is the core objective, which connects scattered text information into a structured network, allowing large language models to quickly locate key information based on the graph, thereby improving the efficiency and accuracy of analysis and reasoning.

[0035] Text data, image feature data, and sensor time-series data originally belonged to different types with significant format differences, making them difficult for large language models to process uniformly. However, tensor format is the standard input format for large language models, which can transform various types of data into numerical matrices that the model can compute, achieving normalization of multi-source data. Preserving data relationships and metadata is a core requirement for production process optimization in the manufacturing industry. Metadata includes key information such as data acquisition timestamps, equipment numbers, production batches, and process names, ensuring that the correspondence between different types of data is not lost. This allows large language models to trace the production context of the data during analysis, avoiding biases in optimization suggestions caused by isolated analysis.

[0036] In this embodiment of the invention, a large language model fine-tuned with industry data is used to analyze and reason about the preprocessed data to generate optimization suggestions and decision-making schemes, including: Perform semantic deep analysis on production logs and operation records to identify potential problems such as equipment failure precursors, process bottlenecks, and non-standard operations, and extract optimization points in the production process; Frame-level analysis of images and video data from the production site is performed to detect the wear status of equipment parts, product assembly accuracy, and production environment compliance, and output quantitative test results. By using a cross-modal attention mechanism, text semantic features, image visual features, and sensor temporal features are dynamically weighted and fused to generate an optimization scheme that includes equipment parameter adjustment values, process step change suggestions, and production scheduling priorities.

[0037] Specifically, production logs and operation records contain textual information about the entire production process. Semantic depth analysis, based on models fine-tuned with manufacturing industry data, combines industry expertise to analyze the logical connections and hidden information behind the text. Identifying early signs of equipment failure involves analyzing the semantic relationships between vague descriptions such as "abnormal equipment noise" and "temperature fluctuations" in the logs and historical failure cases to predict potential equipment malfunctions in advance. For example, identifying early signs of bearing wear from a "spindle rotation jamming" log. Linking process bottlenecks involves mining the semantic connections between logs from different processes to pinpoint the core nodes causing decreased production efficiency. For example, analyzing the relationship between "extended waiting time for process C" and "longer processing cycle for process B" identifies process B as a bottleneck. Non-standard operations are identified by referring to a manufacturing industry operation standard dictionary to identify deviations from standard practices in the operation records. Extracting optimization points from the production process involves transforming identified potential problems into actionable optimization directions, such as optimizing production scheduling for process bottlenecks and developing standardized guidelines for non-standard operations.

[0038] The video consists of consecutive frames, and frame-by-frame analysis avoids missing momentary anomalies, such as brief misalignments during product assembly or minor wear on equipment parts, ensuring comprehensiveness and accuracy of the inspection. The wear status of equipment parts is determined by analyzing texture changes and morphological defects in the surface images of the parts; product assembly accuracy is assessed by comparing the geometric positional deviations between the product assembly image and the standard model; and production environment compliance is determined by analyzing temperature and humidity indicators, the status of safety protection facilities, and material placement specifications in the on-site images to determine whether production environment requirements are met. Quantitative inspection results are output, providing specific numerical values ​​such as a wear area percentage of 3%, an assembly deviation of 0.2 mm, and an ambient temperature exceeding the standard range by 2°C, transforming visual data into quantitative data consistent with equipment parameters and sensor signals.

[0039] Different modalities of data contribute differently to production optimization. Cross-modal attention mechanisms allow the model to automatically focus on features more critical to the current problem, avoiding information redundancy or weakening of key information caused by averaging fusion. Dynamic weights are dynamically calculated by the model based on the real-time production scenario, ensuring the fusion results align with real-time production needs. The three types of fused features correspond to the text analysis, visual analysis, and sensor data processing results mentioned earlier. Equipment parameter adjustment values ​​provide specific and actionable standards for parameter modification, process step change suggestions clarify the direction of process optimization, and production scheduling priorities address resource allocation issues.

[0040] In this embodiment of the invention, the output of the large language model is converted into production instructions, and production parameters and processes are adjusted in real time through an interactive interface with production equipment, including: Based on the reasoning results of the large language model, combined with the weight coefficients of the production target, a differentiated optimization scheme is generated. By simulating the implementation effects of various optimization schemes in production scenarios using digital twin models, the efficiency improvement rate, cost reduction rate, and quality pass rate improvement of each scheme are quantitatively evaluated to determine the optimal scheme. The optimal solution is transformed into specific production instructions that conform to industrial communication protocols. The instructions include equipment parameter adjustment thresholds, changes in the execution sequence of process steps, and production resource scheduling and allocation schemes. At the same time, instruction execution documentation is generated. When the production environment changes, the solution reselection mechanism is automatically triggered to generate a temporary optimized solution and execute it.

[0041] Specifically, the inference results of the large language model form the basis for solution generation, namely the problems identified by the model and the initial optimization directions. Production goals typically cover three core dimensions: efficiency, cost, and quality. Enterprises can customize weights according to their actual scenarios and also adaptively generate solutions based on industry benchmarks and their own historical data. Generating differentiated optimization solutions involves generating at least three solutions with different focuses based on different weight combinations. For example, Solution 1 focuses on efficiency improvement, Solution 2 on cost control, and Solution 3 on quality stability. Each solution proposes a different approach to address the core problem.

[0042] Digital twin models recreate the equipment layout, process flow, and material flow logic of the production site, constructing a virtual scenario completely synchronized with the physical production system to ensure the authenticity and reliability of the simulation results. Simulating the implementation effects of each optimization scheme involves inputting the differentiated schemes generated earlier into the virtual scenario one by one. Quantitative evaluation of the key indicators for each scheme serves as the basis for selection. These indicators include: efficiency improvement rate (corresponding to the expected increase in output per unit time), cost reduction rate (corresponding to the expected savings in raw material loss and energy consumption), and quality pass rate improvement (corresponding to the expected increase in the percentage of qualified products). The optimal scheme is determined by calculating the comprehensive score of each scheme using a weighted scoring method, selecting the scheme with the highest comprehensive score to ensure that the scheme achieves optimal benefits under the guidance of the company's set goals.

[0043] Industrial production equipment requires adherence to unified communication standards to recognize commands. Therefore, the solution needs to be converted into a format compatible with mainstream industrial protocols such as OPC UA, Modbus, and Profinet. Specific production commands are the outcome of this conversion and must be practical: equipment parameter adjustment thresholds should clearly specify directly operable values, rather than vague statements like "reduce temperature" or "adjust speed"; changes in the execution sequence of process steps should clearly define the logic before and after each step; and production resource scheduling and allocation schemes should clearly define the allocation rules for equipment, personnel, and materials. A command execution instruction document should be generated, containing detailed operating steps and key precautions to help operators quickly grasp the execution points and avoid production failures due to improper operation.

[0044] Changes in the production environment encompass various unforeseen circumstances, including sudden equipment failures, fluctuations in raw material performance, exceeding quality inspection standards, and urgent order changes, all of which directly impact the effectiveness of the original optimal solution. By monitoring key signals in the production environment in real time, when changes exceed preset thresholds, a reselection process can be automatically initiated without manual intervention. Temporary solutions must be characterized by rapid generation and implementation. Based on the changed production environment, the inference results of the large language model are re-invoked, the target weight coefficients are adjusted, the optimal temporary solution adapted to the unforeseen scenario is selected, and it is converted into production instructions for execution, minimizing losses caused by unforeseen circumstances.

[0045] In this embodiment of the invention, the optimization effect is evaluated by monitoring key indicators in the production process, and the results are fed back to the large language model to achieve closed-loop optimization, including: Real-time monitoring includes key indicators such as production efficiency, equipment utilization rate, product qualification rate, raw material loss rate, and energy consumption rate, with the monitoring frequency consistent with the data acquisition frequency; The optimization effect is quantitatively evaluated using a weighted scoring method. The scoring weights correspond to the production target weight coefficients, and the evaluation results are divided into excellent, good, qualified, and unqualified. When the evaluation result is qualified or below, the detailed evaluation data will be fed back to the large language model, the model parameters will be updated using incremental learning, the adjusted optimization strategy will be generated and issued for execution; when the evaluation result is excellent or good, the current optimization scheme and the corresponding production scenario data will be recorded for subsequent model reinforcement training.

[0046] Specifically, production efficiency focuses on output per unit time, reflecting the speed of process operation; equipment utilization rate refers to the ratio of actual equipment operating time to planned operating time, reflecting resource utilization efficiency; product qualification rate is directly related to quality targets, measuring the effect of optimization solutions on quality improvement; raw material loss rate and energy consumption rate specifically correspond to cost control targets, quantifying the effectiveness of optimization measures in cost reduction. Maintaining consistency between monitoring frequency and data collection frequency ensures that each set of collected data corresponds to real-time indicator status, avoiding data misalignment due to frequency differences.

[0047] The weighted scoring method assigns different weights to each indicator based on its importance, and then calculates the overall score by combining the actual improvement / decrease of each indicator. When generating optimization solutions, production target weights are already set according to the company's needs. These weights are used during evaluation to ensure that the optimization direction is consistent with the evaluation criteria, avoiding the contradiction of a solution meeting the goals but receiving a low evaluation score. Dividing the evaluation results into four levels clarifies the gradient boundaries of the optimization effect. Different levels correspond to different processing strategies. Compared to vague descriptions, the tiered results are more practically instructive, allowing managers to quickly determine whether adjustments to the optimization solution are needed.

[0048] For evaluation results of "qualified" or below, detailed evaluation data includes not only specific deviation values ​​for the five major indicators but also problem attribution analysis, allowing the large language model to accurately pinpoint the shortcomings of the optimization solution. The incremental learning approach eliminates the need for a full model retraining, fine-tuning relevant parameters only based on feedback deviation data, ensuring the model quickly adapts to the problem. Generating and implementing the adjusted optimization strategy achieves an evaluation-feedback-correction closed loop, reducing production losses. For excellent or good evaluation results, the current optimization solution and corresponding production scenario data are recorded. Through reinforcement training, the model learns the optimal optimization logic for similar scenarios, improving its ability to generate accurate solutions in similar production scenarios in the future.

[0049] In this embodiment of the invention, the method may further include the following steps: It provides a natural language interactive interface, enabling production managers to obtain system optimization suggestions and decision support by inputting questions through voice or text. The natural language interactive interface supports three input methods: voice, text, and handwriting. The output includes optimization suggestions and decision support in natural language text, voice broadcast, and visual charts.

[0050] The natural language interface is designed to align with the daily communication habits of managers. In the production environment, managers may be near equipment and unable to operate a keyboard; in such cases, they can ask questions directly via voice. For more precise descriptions of problems or detailed data queries, text input is available. The system quickly analyzes the questions entered by managers, combining them with preceding production data and model analysis results to provide targeted practical suggestions and decision-making support. This allows managers to quickly obtain effective information and shorten the decision-making cycle.

[0051] On the input side, voice input is suitable for mobile office scenarios in production sites, allowing users to initiate queries without manual operation; text input is suitable for questions that require precise expression, contain professional terminology or specific parameters, ensuring accurate transmission of information; handwriting input is suitable for scenarios without keyboards or where users are not proficient in text input, allowing them to initiate requests by writing text, numbers, or even simple diagrams, lowering the operational threshold. On the output side, multiple formats cater to different receiving habits and usage scenarios. Natural language text facilitates managers' retention, review, or forwarding, with clear and easy-to-understand expressions and no professional technical barriers; voice broadcasting is suitable for busy managers, allowing them to receive core information without looking at the screen; visual charts transform abstract data and optimization effects into intuitive graphics, helping managers quickly capture key information, more accurately judge the feasibility of optimization solutions, and improve decision-making efficiency.

[0052] In this embodiment of the invention, fine-tuning of the large language model includes: Acquire training data including production logs, equipment manuals, process standards, failure cases, and quality inspection specifications from various sub-scenarios within the manufacturing industry; Fine-tuning was performed using mini-batch gradient descent, with batch size set between 32 and 64, initial learning rate of 1e-5, decaying by 50% every 20 rounds, and at least 100 iterations. After fine-tuning, the model achieved an accuracy of at least 92% in identifying production problems, a feasibility verification pass rate of at least 88% for optimization suggestions, and an inference response time of no more than 500ms.

[0053] The manufacturing industry segmentation covers specific sub-sectors such as machining, electronic assembly, chemical production, and automotive parts manufacturing, ensuring that the model can adapt to the specific needs of different manufacturing scenarios. Production logs provide dynamic data from real production processes, helping the model understand actual production logic; equipment manuals contain the operating principles, operating procedures, and parameter thresholds of various production equipment, enabling the model to master relevant professional knowledge; process standards clearly define the technical requirements of each production link, ensuring that the optimization suggestions generated by the model comply with industry standards; failure cases record the manifestations, causes, and solutions of equipment failures, improving the model's ability to identify fault precursors and generate repair suggestions; quality inspection specifications clearly define product qualification standards and testing methods, providing a basis for the model to predict quality risks and optimize quality control.

[0054] Mini-batch gradient descent reduces computational resource consumption per training iteration, accelerating training iteration speed. Compared to stochastic gradient descent, it avoids excessive parameter update fluctuations, ensuring the model converges to a better state. Specific parameter settings are optimized for the characteristics of manufacturing industry data: a batch size of 32-64 ensures the model learns sufficient sample information in each training iteration without causing computational delays due to excessively large batches; an initial learning rate of 1e-5 is a low setting, preventing the model from compromising its general capabilities due to excessive initial updates; a 50% decay every 20 epochs gradually reduces parameter adjustment amplitude in the later stages of training, allowing the model to accurately converge to the optimal state suitable for the manufacturing scenario; and at least 100 iterations ensure the model fully learns the industry knowledge and optimization logic from the training data. The accuracy rate of production problem identification is no less than 92%, ensuring that the model can accurately capture core issues such as equipment failures and process bottlenecks, reducing missed and false judgments; the feasibility verification pass rate of optimization suggestions is no less than 88%, ensuring that the suggestions output by the model meet the actual production conditions; the inference response time is no more than 500ms to meet the needs of real-time optimization of the production process, ensuring that the model can quickly output analysis results and adapt to the timeliness requirements of dynamic adjustments on the production site.

[0055] like Figure 2 As shown, this embodiment of the invention provides a manufacturing process optimization device based on a large model, the device comprising: The data acquisition unit 210 is used to collect multimodal data in real time during the production process through sensor networks, production management systems and quality inspection equipment, including text logs, equipment parameters, image data and sensor signals; and to clean, label and convert the collected data to adapt it to the input requirements of large language models. The large language model analysis unit 220 is used to analyze and reason about the preprocessed data through a large language model that has been fine-tuned with industry data, and to generate optimization suggestions and decision-making solutions. The intelligent decision-making unit 230 is used to convert the output of the large language model into production instructions and adjust production parameters and process flow in real time through the interaction interface with the production equipment. The real-time feedback unit 240 is used to evaluate the optimization effect by monitoring key indicators in the production process and feed the results back to the large language model to achieve closed-loop optimization.

[0056] It is understood that the structures illustrated in the embodiments of the present invention do not constitute a specific limitation on the manufacturing process optimization device based on a large model. In other embodiments of the present invention, the manufacturing process optimization device based on a large model may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0057] The information interaction and execution process between the various units in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description of the method embodiment of the present invention, and will not be repeated here.

[0058] Figure 3 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0059] like Figure 3 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0060] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0061] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as manufacturing process optimization methods based on large models.

[0062] In some embodiments, the large-model-based manufacturing process optimization method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the large-model-based manufacturing process optimization method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute the large-model-based manufacturing process optimization method by any other suitable means (e.g., by means of firmware).

[0063] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0064] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0065] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0066] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0067] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0068] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0069] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0070] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A manufacturing process optimization method based on a large model, characterized in that, include: Through sensor networks, production management systems, and quality inspection equipment, multimodal data during the production process is collected in real time, including text logs, equipment parameters, image data, and sensor signals; the collected data is cleaned, labeled, and formatted to adapt it to the input requirements of large language models. By using a large language model fine-tuned with industry data, the preprocessed data is analyzed and reasoned to generate optimization suggestions and decision-making solutions. The output of the large language model is converted into production instructions, and production parameters and processes are adjusted in real time through the interaction interface with the production equipment. By monitoring key indicators in the production process, the optimization effect is evaluated, and the results are fed back to the large language model to achieve closed-loop optimization.

2. The method according to claim 1, characterized in that, The process of cleaning, labeling, and format conversion of the collected data includes: Noise is removed by statistical filtering, missing values ​​are filled by linear interpolation, and outliers are identified and marked based on the 3σ principle. The YOLO object detection algorithm was used to assist manual annotation of image and video data. The annotation labels included equipment operating status labels and product defect labels. At the same time, texture features, morphological features and spatial location features were extracted. The text logs were segmented using a word segmentation tool combined with a professional dictionary for the manufacturing industry, and a semantic graph of the production scenario was constructed through part-of-speech tagging and semantic association analysis. Text data, image feature data, and sensor time-series data are converted into tensor formats compatible with large language models, while preserving the metadata of data relationships.

3. The method according to claim 1, characterized in that, The process involves analyzing and reasoning about preprocessed data using a large language model fine-tuned with industry data to generate optimization suggestions and decision-making solutions, including: Perform semantic deep analysis on production logs and operation records to identify potential problems such as equipment failure precursors, process bottlenecks, and non-standard operations, and extract optimization points in the production process; Frame-level analysis of images and video data from the production site is performed to detect the wear status of equipment parts, product assembly accuracy, and production environment compliance, and output quantitative test results. By using a cross-modal attention mechanism, text semantic features, image visual features, and sensor temporal features are dynamically weighted and fused to generate an optimization scheme that includes equipment parameter adjustment values, process step change suggestions, and production scheduling priorities.

4. The method according to claim 1, characterized in that, The process of converting the output of the large language model into production instructions and adjusting production parameters and processes in real time through an interface with production equipment includes: Based on the reasoning results of the large language model, and combined with the weight coefficients of the production target, a differentiated optimization scheme is generated. The implementation effects of each optimization scheme in the production scenario are simulated by digital twin models. The efficiency improvement rate, cost reduction rate and quality pass rate improvement of each scheme are quantitatively evaluated to determine the optimal scheme. The optimal solution is transformed into specific production instructions that conform to industrial communication protocols. The instructions include equipment parameter adjustment thresholds, changes in the execution sequence of process steps, and production resource scheduling and allocation schemes. At the same time, instruction execution documentation is generated. When the production environment changes, the solution reselection mechanism is automatically triggered to generate a temporary optimized solution and execute it.

5. The method according to claim 1, characterized in that, The process of monitoring key indicators during the production process, evaluating the optimization effect, and feeding the results back to the large language model to achieve closed-loop optimization includes: Real-time monitoring includes key indicators such as production efficiency, equipment utilization rate, product qualification rate, raw material loss rate, and energy consumption rate, with the monitoring frequency consistent with the data acquisition frequency; The optimization effect is quantitatively evaluated using a weighted scoring method. The scoring weights correspond to the production target weight coefficients, and the evaluation results are divided into excellent, good, qualified, and unqualified. When the evaluation result is qualified or below, the detailed evaluation data is fed back to the large language model, the model parameters are updated using incremental learning, the adjusted optimization strategy is generated and issued for execution; when the evaluation result is excellent or good, the current optimization scheme and corresponding production scenario data are recorded for subsequent model reinforcement training.

6. The method according to claim 1, characterized in that, The method further includes: providing a natural language interactive interface, enabling production managers to input questions via voice or text to obtain system optimization suggestions and decision support; wherein, the natural language interactive interface supports three input methods: voice, text, and handwriting; the output includes optimization suggestions and decision support in natural language text, voice broadcast, and visual charts.

7. The method according to claim 1, characterized in that, The fine-tuning of the large language model includes: Acquire training data including production logs, equipment manuals, process standards, failure cases, and quality inspection specifications from various sub-scenarios within the manufacturing industry; Fine-tuning was performed using mini-batch gradient descent, with batch size set between 32 and 64, initial learning rate of 1e-5, decaying by 50% every 20 rounds, and at least 100 iterations. After fine-tuning, the model achieved an accuracy of at least 92% in identifying production problems, a feasibility verification pass rate of at least 88% for optimization suggestions, and an inference response time of no more than 500ms.

8. A manufacturing process optimization device based on a large model, characterized in that, include: The data acquisition unit is used to collect multimodal data in real time during the production process through sensor networks, production management systems, and quality inspection equipment, including text logs, equipment parameters, image data, and sensor signals; and to clean, label, and convert the collected data to adapt it to the input requirements of large language models. The large language model analysis unit is used to analyze and reason about preprocessed data using a large language model that has been fine-tuned with industry data, and to generate optimization suggestions and decision-making solutions. The intelligent decision-making unit is used to convert the output of the large language model into production instructions and adjust production parameters and process flow in real time through the interaction interface with the production equipment. The real-time feedback unit is used to evaluate the optimization effect by monitoring key indicators in the production process and feed the results back to the large language model to achieve closed-loop optimization.

9. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the manufacturing process optimization method based on a large model according to any one of claims 1-7.

10. A computer-readable medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the manufacturing process optimization method based on a large model as described in any one of claims 1-7.

Images