Gear transmission design method based on large model dialogue questions and answers

Abstract

The invention discloses a gear transmission design method based on large model dialogue questions and answers, and relates to the technical field of mechanical design, and the method comprises the steps: obtaining first information which comprises gear transmission terms, gear transmission design logic, gear transmission parameters and physical relations among the parameters; training a preset natural language processing model according to the first information; a domain knowledge base is constructed, a fine tuning training data set is constructed according to the domain knowledge base, and the domain knowledge base comprises a material type selection rule, a precision grade rule and a characteristic classification rule; performing fine tuning on the natural language processing model according to the fine tuning training data set; user input is obtained and input into the fine-tuned natural language processing model, a gear transmission design parameter combination is output, and the user input is a natural language or a semi-structured statement. According to the invention, intelligent, automatic and one-stop design functions of gear transmission are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical design, and in particular to a gear transmission design method based on large-model dialogue question and answer. Background Art

[0002] Traditional gear transmission design requires designers to have extensive gear design knowledge, experience, and mechanical design skills. The design work involves a large number of tedious and repetitive tasks, which is time-consuming and labor-intensive.

[0003] There are two problems with the existing gear automation design method: 1. Users typically interact with design software by manually selecting design solutions and initial parameters. This step still requires designers to have extensive knowledge and experience. 2. The design of the gear only focuses on the design of the basic parameters of the gear, but does not include the specific structural form of the gear. This requires designers to select and model the gear structure themselves later.

[0004] Therefore, a gear transmission design method based on large model dialogue question and answer was developed to solve the above problems. Summary of the Invention

[0005] The present invention proposes a gear transmission design method based on large-model dialogue question-answering to solve the existing problem of time-consuming and labor-intensive manual selection of design parameters.

[0006] The present invention achieves the above-mentioned purpose through the following technical solutions: The present invention provides a gear transmission design method based on large-model dialogue question-answering, comprising: Acquiring first information, the first information including gear transmission terminology, gear transmission design logic, gear transmission parameters, and physical relationships between the parameters; Training a preset natural language processing model based on the first information; Building a domain knowledge base and building a fine-tuning training data set based on the domain knowledge base, wherein the domain knowledge base includes material selection rules, precision grade rules, and property classification rules; Fine-tuning the natural language processing model according to the fine-tuning training dataset; Obtain user input and input it into the fine-tuned natural language processing model to output gear transmission design parameters, wherein the user input is natural language or semi-structured sentences.

[0007] Furthermore, obtaining the first information includes: Acquire original data, including text content on gear transmission in mechanical design manuals, text content on gear strength calculation in ISO 6336, text content on gear precision standards in ISO 1328, academic papers on gear transmission, and case reports on industrial gear design; performing structured extraction on the raw data; performing term alignment processing on the original data; The original data is segmented into sections.

[0008] Furthermore, training a preset natural language processing model according to the first information includes: Build the DeBERTa-v3 model; According to the first information, the DeBERTa-v3 model is trained based on a training task, and the training adopts a Transformer architecture to pre-train the text in an autoregressive manner. The training task includes masked language modeling and causal language modeling.

[0009] Furthermore, a domain knowledge base is constructed, including: Constructing material selection rules, wherein the material selection rules are to mark the matching relationship between load type, rotation speed, environment and material; Constructing an accuracy grade rule, wherein the accuracy grade rule is to construct a "working machine type-accuracy grade" mapping table based on ISO 1328; A characteristic classification rule is constructed, wherein the characteristic classification rule is a prime mover and working machine characteristic classification table.

[0010] Furthermore, a fine-tuning training dataset is constructed based on the domain knowledge base, including: Decomposing the rules in the domain knowledge base into corresponding independent variables; Permuting and combining the independent variables to generate all possible parameter combinations, each parameter combination corresponding to a gear transmission design scenario; Perform natural language templates for each parameter combination; Each parameter combination after natural language template is diversified and enhanced, and the diversified enhancement includes synonym replacement, sentence structure change, parameter range expansion and context enrichment.

[0011] Furthermore, fine-tuning the natural language processing model according to the fine-tuning training dataset includes: A supervised fine-tuning strategy is adopted, with the fine-tuning training dataset as input and the corresponding gear transmission design parameter combination as the target output, and the cross entropy loss is calculated and gradient backpropagation is performed; A contrastive learning mechanism is introduced to add noise to the positive samples of each parameter combination to generate negative samples. A triplet is constructed, which includes the input, the positive sample, and the negative sample. The InfoNCE loss function is used to shorten the semantic distance between the positive sample and the input. Phased training strategy: first freeze the underlying parameters of the natural language processing model and only fine-tune the top-level classifier. After the loss converges, unfreeze the full model parameters of the natural language processing model for end-to-end training; Set a dynamic learning rate decay strategy, with an initial learning rate of 1e-5, which decays to 0.8 times the original value every 1000 steps; Use mixed precision training technology, set the global gradient clipping threshold to 1.0, and the batch size to 32; Monitor the perplexity metric on the validation set and terminate training early if it does not decrease for three consecutive epochs.

[0012] Furthermore, it also includes parametric modeling of the gear transmission design parameter combination according to the SolidWorks API called by the C# code.

[0013] Furthermore, the gear transmission design parameter combination is parametrically modeled according to the SolidWorks API called by the C# code, including: Add the SolidWorks COM library to the project; Calling a gear structure template in SolidWorks based on an OpenDoc6 method, wherein the gear structure template includes a forged web gear template, a forged integral hole-free gear template, a forged integral gear template, a cast single-spoke gear template, a cast web gear template, and a cast double-spoke gear template; The equation of the gear structure template is obtained according to the GetEquationMgr method, and the corresponding gear transmission design parameter combination is written into the equation to complete the modeling.

[0014] Verify the output gear transmission design parameter combination according to the domain knowledge base, including: Rule matching verification: Match the output gear transmission design parameter combination with the load-material correspondence table in the material selection rule, and generate an early warning when the tooth surface contact stress exceeds the allowable stress of the selected material; Numerical range check: Perform boundary check on module and tooth width coefficient according to ISO 6336 standard, and mark as abnormal value when module m < 1mm or tooth width b > 20m; Logical consistency check: Verify whether the tooth number combination satisfies the transmission ratio error <2%, and check whether the hardness difference of the paired gears meets the hardening criterion of HB1-HB2 ≥ 30; Conflict detection and verification: When there is a conflict between the accuracy level and the surface roughness Ra value, the conflict resolution module based on the decision tree is activated; Verification result grading: categorizes exceptions into three levels: fatal error, warning, and prompt.

[0015] The beneficial effect of the gear transmission design method based on large-model dialogue question and answer of the present invention is that: the present invention provides a new gear transmission design mode of dialogue and review. The user inputs the design requirements in natural language, and the software automatically completes the gear design. During the design process, the design parameters and plans are given to the user for inspection and selection. After confirmation, a three-dimensional model and engineering drawings are generated, realizing the intelligent, automated, and one-stop design function of the gear transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This application provides a flow chart of a gear transmission design method based on large-model dialogue question and answer. DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the embodiments of the present invention. It should be understood that the described embodiments are only a portion of the embodiments of the present invention, not all of them. Generally, the components of the embodiments of the present invention described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations.

[0018] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0019] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0020] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0021] like Figure 1 As shown in FIG, a gear transmission design method based on large-scale model dialogue question answering includes: S1: Acquire first information, where the first information includes gear transmission terminology, gear transmission design logic, gear transmission parameters, and physical relationships between the parameters; S2: Training a preset natural language processing model based on the first information; S3: Build a domain knowledge base and construct a fine-tuning training data set based on the domain knowledge base, wherein the domain knowledge base includes material selection rules, precision grade rules, and property classification rules; S4: Fine-tune the natural language processing model according to the fine-tuning training dataset; S5: Obtain user input and input it into the fine-tuned natural language processing model to output a gear transmission design parameter combination, wherein the user input is a natural language or a semi-structured statement.

[0022] In one embodiment, obtaining the first information includes: Acquire original data, including text content on gear transmission in mechanical design manuals, text content on gear strength calculation in ISO 6336, text content on gear precision standards in ISO 1328, academic papers on gear transmission, and case reports on industrial gear design; performing structured extraction on the raw data; performing term alignment processing on the original data; The original data is segmented into sections.

[0023] In one embodiment, training a preset natural language processing model based on the first information includes: Build the DeBERTa-v3 model; According to the first information, the DeBERTa-v3 model is trained based on a training task, and the training adopts a Transformer architecture to pre-train the text in an autoregressive manner. The training task includes masked language modeling and causal language modeling.

[0024] In one embodiment, building a domain knowledge base includes: Constructing material selection rules, wherein the material selection rules are to mark the matching relationship between load type, rotation speed, environment and material; Constructing an accuracy grade rule, wherein the accuracy grade rule is to construct a "working machine type-accuracy grade" mapping table based on ISO 1328; A characteristic classification rule is constructed, wherein the characteristic classification rule is a prime mover and working machine characteristic classification table.

[0025] In one embodiment, constructing a fine-tuning training dataset based on the domain knowledge base includes: Decomposing the rules in the domain knowledge base into corresponding independent variables; Permuting and combining the independent variables to generate all possible parameter combinations, each parameter combination corresponding to a gear transmission design scenario; Perform natural language templates for each parameter combination; Each parameter combination after natural language template is diversified and enhanced, and the diversified enhancement includes synonym replacement, sentence structure change, parameter range expansion and context enrichment.

[0026] In one embodiment, fine-tuning the natural language processing model according to the fine-tuning training dataset includes: A supervised fine-tuning strategy is adopted, with the fine-tuning training dataset as input and the corresponding gear transmission design parameter combination as the target output, and the cross entropy loss is calculated and gradient backpropagation is performed; A contrastive learning mechanism is introduced to add noise to the positive samples of each parameter combination to generate negative samples. A triplet is constructed, which includes the input, the positive sample, and the negative sample. The InfoNCE loss function is used to shorten the semantic distance between the positive sample and the input. Phased training strategy: first freeze the underlying parameters of the natural language processing model and only fine-tune the top-level classifier. After the loss converges, unfreeze the full model parameters of the natural language processing model for end-to-end training; Set a dynamic learning rate decay strategy, with an initial learning rate of 1e-5, which decays to 0.8 times the original value every 1000 steps; Use mixed precision training technology, set the global gradient clipping threshold to 1.0, and the batch size to 32; Monitor the perplexity metric on the validation set and terminate training early if it does not decrease for three consecutive epochs.

[0027] In one embodiment, the method further includes performing parametric modeling on the gear transmission design parameter combination according to a SolidWorks API called by a C# code.

[0028] In one embodiment, parametric modeling of the gear transmission design parameter combination is performed according to the SolidWorks API called by the C# code, including: Add the SolidWorks COM library to the project; Calling a gear structure template in SolidWorks based on an OpenDoc6 method, wherein the gear structure template includes a forged web gear template, a forged integral hole-free gear template, a forged integral gear template, a cast single-spoke gear template, a cast web gear template, and a cast double-spoke gear template; The equation of the gear structure template is obtained according to the GetEquationMgr method, and the corresponding gear transmission design parameter combination is written into the equation to complete the modeling.

[0029] The C# code calls the SolidWorks API and the corresponding gear structure template to implement parametric modeling of the large and small gears.

[0030] Specifically, C# code calls the SolidWorks API by referencing the SolidWorks API library. Before using the SolidWorks API in C#, add the SolidWorks COM library to the project, including: SolidWorks.Interop.sldworks.dll and SolidWorks.Interop.swconst.dll.

[0031] The gear structure template is called using the OpenDoc6 method, and parametric modeling is implemented by using the GetEquationMgr method to obtain the template equation, write the corresponding parameters, and then rebuild the model.

[0032] The gear structure templates include: Forged Web Gear.PRTDOT, Forged Integral Gear (No Hole).PRTDOT, Forged Integral Gear.PRTDOT, Cast Single Spoke Gear.PRTDOT, Cast Web Gear.PRTDOT, Cast Double Spoke Gear.PRTDOT.

[0033] The method for making the gear structure template is to input the parameter size equation in each template, constrain the size in the modeling process with the corresponding equation, and complete the modeling according to the gear structure.

[0034] Step 5: Create a part drawing.

[0035] The way to use C# code to call the SolidWorks API to create a drawing is to use the CreateDrawing method to directly generate a drawing through a part document.

[0036] In one embodiment, verifying the output gear transmission design parameter combination according to the domain knowledge base includes: Rule matching verification: Match the output gear transmission design parameter combination with the load-material correspondence table in the material selection rule, and generate an early warning when the tooth surface contact stress exceeds the allowable stress of the selected material; Numerical range check: Perform boundary check on module and tooth width coefficient according to ISO 6336 standard, and mark as abnormal value when module m < 1mm or tooth width b > 20m; Logical consistency check: Verify whether the tooth number combination satisfies the transmission ratio error <2%, and check whether the hardness difference of the paired gears meets the hardening criterion of HB1-HB2 ≥ 30; Conflict detection and verification: When there is a conflict between the accuracy level and the surface roughness Ra value, the conflict resolution module based on the decision tree is activated; Verification result grading: categorizes exceptions into three levels: fatal error, warning, and prompt.

[0037] The present invention is further described below by taking a specific application of the present invention as an example: Design a high-speed gear drive for this belt conveyor reducer (parallel shaft structure, with two supports arranged asymmetrically relative to the pinion). Given an input power of P = 11 kW, a pinion speed of n1 = 970 r / min, a gear ratio of u = 3.2, and an electric motor drive, the belt conveyor should operate smoothly and maintain a constant direction of rotation, with a 15-year service life (assuming 250 operating days per year) and a two-shift system.

[0038] Model output: Based on your design requirements, the initial parameters are as follows: Input power (kW): 11; Input speed (r / min): 970; Gear ratio: 3.2; Working life (h): 60000; Configuration of supporting gears: asymmetric configuration; Pinion material type: alloy forged steel; Pinion material grade: 40Cr; Pinion material tooth surface hardness (HBW): 280; Pinion material tooth center hardness (HBW): 280; Big gear material type: carbon forged steel; Big gear material grade: 45; Tooth surface hardness of large gear material (HBW): 240; Large gear material tooth center hardness (HBW): 240; Accuracy level: 7; Prime mover characteristics: uniform and stable; Working machine characteristics: uniform and stable; There are many gear materials that meet your design requirements. Please confirm your selection based on the actual working conditions! Logical check: Material matching: 40Cr (alloy forged steel) is suitable for medium speed and medium load, and 45 steel (carbon forged steel) meets the hardness difference requirements; Accuracy level: Belt conveyor → Level 7 (knowledge base mapping is correct); Characteristic classification: Electric motor → 0, conveyor → 0 (according to the classification table). 1. This method combines a large model with a thinking chain + knowledge base to enable the large model to accurately understand the user's design requirements and accurately select the preliminary design parameters.

[0039] 2. This method realizes the complete design of gear tooth profile and structure, and users can independently select the structural scheme to be adopted according to their needs and automatically complete all modeling work.

[0040] 3. This method uses the big model by using the online general big model API and standardizing the output of the big model using the thinking chain + knowledge base method, which can achieve accurate output effects, eliminate the complex technical steps of big model deployment and fine-tuning, and improve the ease of use of the technology.

[0041] 4. This method as a whole achieves the effect of inputting user natural language design requirements and outputting design plans, three-dimensional models and engineering drawings. Through a new design mode of dialogue and review, the design process is made interactive and customized, realizing the intelligent, automated, and one-stop design function of gear transmission.

[0042] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A gear transmission design method based on large model dialogue question and answer, characterized in that: include: Acquiring first information, the first information including gear transmission terminology, gear transmission design logic, gear transmission parameters, and physical relationships between the parameters; Training a preset natural language processing model based on the first information; Building a domain knowledge base and building a fine-tuning training data set based on the domain knowledge base, wherein the domain knowledge base includes material selection rules, precision grade rules, and property classification rules; Fine-tuning the natural language processing model according to the fine-tuning training dataset; Obtain user input and input it into the fine-tuned natural language processing model to output a gear transmission design parameter combination, wherein the user input is a natural language or a semi-structured sentence.

2. A gear transmission design method based on large model dialogue question and answer according to claim 1, characterized in that: Obtaining first information, including: Acquiring original data, including text content on gear transmission in mechanical design manuals, text content on gear strength calculation in ISO 6336, text content on gear precision standards in ISO 1328, academic papers on gear transmission, and case reports on industrial gear design; Performing structured extraction on the original data; performing term alignment processing on the original data; The original data is segmented into sections.

3. The gear transmission design method based on large model dialogue question and answer according to claim 1 is characterized in that: Training a preset natural language processing model according to the first information includes: Build the DeBERTa-v3 model; According to the first information, the DeBERTa-v3 model is trained based on a training task, and the training adopts a Transformer architecture to pre-train the text in an autoregressive manner. The training task includes masked language modeling and causal language modeling.

4. The gear transmission design method based on large model dialogue question and answer according to claim 1 is characterized in that: Build a domain knowledge base, including: Constructing material selection rules, wherein the material selection rules are to mark the matching relationship between load type, rotation speed, environment and material; Establishing precision grade rules, wherein the precision grade rules are to establish a "working machine type-precision grade" mapping table based on ISO 1328; A characteristic classification rule is constructed, wherein the characteristic classification rule is a prime mover and working machine characteristic classification table.

5. A gear transmission design method based on large model dialogue question and answer according to claim 1 or 4, characterized in that: Construct a fine-tuning training dataset based on the domain knowledge base, including: Decomposing the rules in the domain knowledge base into corresponding independent variables; Permuting and combining the independent variables to generate all possible parameter combinations, each parameter combination corresponding to a gear transmission design scenario; Perform natural language templates for each parameter combination; Each parameter combination after natural language template is diversified and enhanced, and the diversified enhancement includes synonym replacement, sentence structure change, parameter range expansion and context enrichment.

6. The gear transmission design method based on large model dialogue question and answer according to claim 5 is characterized in that: Fine-tuning the natural language processing model according to the fine-tuning training dataset includes: A supervised fine-tuning strategy is adopted, with the fine-tuning training dataset as input and the corresponding gear transmission design parameter combination as the target output, and the cross entropy loss is calculated and gradient backpropagation is performed; A contrastive learning mechanism is introduced to add noise to the positive samples of each parameter combination to generate negative samples. A triplet is constructed, which includes the input, the positive sample, and the negative sample. The InfoNCE loss function is used to shorten the semantic distance between the positive sample and the input. Phased training strategy: first freeze the underlying parameters of the natural language processing model and only fine-tune the top-level classifier. After the loss converges, unfreeze the full model parameters of the natural language processing model for end-to-end training; Set a dynamic learning rate decay strategy, with an initial learning rate of 1e-5, which decays to 0.8 times the original value every 1000 steps; Use mixed precision training technology, set the global gradient clipping threshold to 1.0, and the batch size to 32; Monitor the perplexity metric on the validation set and terminate training early if it does not decrease for three consecutive epochs.

7. The gear transmission design method based on large model dialogue question and answer according to claim 1 is characterized in that: The method also includes performing parametric modeling on the gear transmission design parameter combination according to a SolidWorks API called by a C# code.

8. The gear transmission design method based on large model dialogue question and answer according to claim 7 is characterized in that: The gear transmission design parameter combination is parametrically modeled according to the SolidWorks API called by the C# code, including: Add the SolidWorks COM library to the project; Calling a gear structure template in SolidWorks based on an OpenDoc6 method, wherein the gear structure template includes a forged web gear template, a forged integral hole-free gear template, a forged integral gear template, a cast single-spoke gear template, a cast web gear template, and a cast double-spoke gear template; The equation of the gear structure template is obtained according to the GetEquationMgr method, and the corresponding gear transmission design parameter combination is written into the equation to complete the modeling.

9. The gear transmission design method based on large model dialogue question and answer according to claim 1 is characterized in that: The method further includes verifying the output gear transmission design parameter combination according to the domain knowledge base, wherein the verification step includes: Rule matching verification: Match the output gear transmission design parameter combination with the load-material correspondence table in the material selection rule, and generate an early warning when the tooth surface contact stress exceeds the allowable stress of the selected material; Numerical range check: Perform boundary check on module and tooth width coefficient according to ISO 6336 standard, and mark as abnormal value when module m < 1mm or tooth width b > 20m; Logical consistency check: Verify whether the tooth number combination satisfies the transmission ratio error <2%, and check whether the hardness difference of the paired gears meets the hardening criterion of HB1-HB2 ≥ 30; Conflict detection and verification: When there is a conflict between the accuracy level and the surface roughness Ra value, the conflict resolution module based on the decision tree is started.

Images