Search

CN-122019713-A - Machine tool after-sale operation and maintenance intelligent question-answering method based on intention recognition

CN122019713ACN 122019713 ACN122019713 ACN 122019713ACN-122019713-A

Abstract

The invention provides an after-sale operation and maintenance intelligent question-answering method of a machine tool based on intention recognition, which comprises the steps of firstly carrying out standardized processing on multi-source operation and maintenance data and constructing a multi-dimensional intention recognition data set, then utilizing a thousand-question model to extract semantic and intention characteristics, utilizing a DeepSeek model to extract field professional characteristics and training a high-precision intention recognition model based on an improved LORA algorithm, then carrying out characteristic alignment through a knowledge graph of the operation and maintenance field of the machine tool and adopting a dual-model collaborative attention mechanism to realize characteristic fusion, constructing a dynamic search enhancement generation mechanism based on the recognized intention, outputting a self-adaptive structured answer, and finally optimizing model parameters and knowledge graph through a feedback iteration mechanism.

Inventors

  • ZHANG ZHENGFEI

Assignees

  • 安徽省通驰数控机床有限公司

Dates

Publication Date
20260512
Application Date
20260122

Claims (10)

  1. 1. The machine tool after-sale operation and maintenance intelligent question-answering method based on intention recognition is characterized by comprising the following steps of: Step S1, preprocessing multi-source operation and maintenance data, namely carrying out standardized and structured conversion on multi-source data such as natural language problems, machine tool models/parameters, fault logs, historical maintenance records and the like in a machine tool after-sales scene, and synchronously constructing a multi-dimensional intention recognition marking data set; S2, extracting dual-model features, namely extracting the semantics and the intention features of natural language problems through a first pre-training language model, extracting the professional features of related field data through a second pre-training language model, training an intention recognition model based on an improved LORA algorithm, and outputting intention labels and confidence; s3, knowledge graph alignment and attention fusion, namely constructing a knowledge graph in the machine tool operation and maintenance field, performing semantic alignment on the features extracted in the S2 and the knowledge graph nodes, and generating a fusion feature vector through a dual-model collaborative attention mechanism; Step S4, intent-driven dynamic retrieval enhancement generation, namely dynamically retrieving related knowledge based on the intent label output in the step S2, combining the fusion feature vector generated in the step S3, constructing a prompt word and generating a structured operation and maintenance solution; And S5, feedback iterative optimization, namely, based on feedback data of the solution by a user, optimizing and updating the model parameters in the step S2, the knowledge graph in the step S3 and the generation strategy in the step S4.
  2. 2. The after-sales operation and maintenance intelligent question-answering method of the machine tool based on the intention recognition of claim 1, wherein in the step S1, the constructed multi-dimensional intention recognition marking dataset covers five core intention types of equipment maintenance, fault diagnosis, parameter debugging, accessory consultation and operation guidance.
  3. 3. The machine tool after-sale operation and maintenance intelligent question-answering method based on intention recognition according to claim 1, wherein in the step S2, the improved LORA algorithm performs parameter updating by inserting a trainable low-rank matrix pair in a network layer of the first pre-training language model encoder and trains by adopting a mixed loss function fused with cross entropy loss and contrast learning loss, and the first pre-training language model is a meaning thousand-question model, and the second pre-training language model is a DeepSeek model.
  4. 4. The after-sales operation and maintenance intelligent question-answering method of the machine tool based on intention recognition according to claim 1, wherein the dual-model collaborative attention mechanism in the step S3 specifically takes the intention feature extracted by the first pre-training language model as a query vector, takes the domain feature extracted by the second pre-training language model and the feature enhanced by the knowledge graph as a key vector and a value vector respectively, and generates the fusion feature vector after calculating attention weights.
  5. 5. The after-sales operation and maintenance intelligent question-answering method of the machine tool based on intention recognition according to claim 1, wherein the dynamic searching in the step S4 comprises the steps of selecting corresponding index fragments in a vector database according to the intention labels for preliminary screening, and adopting a comprehensive similarity algorithm for secondary sorting of the preliminary screening results, wherein the comprehensive similarity algorithm is as follows: , wherein, Is a user problem vector; knowledge segments in a vector database; balance coefficient; Representing cosine similarity calculation; Is an intention weight factor; And calculating for semantic similarity.
  6. 6. The after-sales operation and maintenance intelligent question-answering method of the machine tool based on intention recognition according to claim 1, wherein in the step S4, the constructed prompt word is a four-in-one template containing task description, constraint conditions, related knowledge and context information.
  7. 7. The after-sales operation and maintenance intelligent question and answer method for machine tools based on intention recognition according to claim 1 is characterized in that in the step S4, when the structured operation and maintenance solution is generated, an output strategy is dynamically adjusted according to the confidence coefficient output by the step S2, a structured maintenance flow is output when the confidence coefficient is higher than Gao Zhixin degrees threshold, a candidate scheme and a query are output when the confidence coefficient is between a high confidence coefficient threshold and a medium confidence coefficient threshold, an troubleshooting guide is output when the confidence coefficient is lower than the medium confidence coefficient threshold, and when the generated solution relates to accessory replacement, the adaptive accessory information obtained from the knowledge graph query is output in an associated mode.
  8. 8. The after-sales operation and maintenance intelligent question-answering method of the machine tool based on intention recognition according to claim 1, wherein the optimization updating in the step S5 comprises the step of performing reinforcement learning optimization on the generation strategy in the step S4 through a PPO algorithm by taking user satisfaction, maintenance success rate and intention recognition accuracy as reward signals.
  9. 9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 8 when the computer program is executed.
  10. 10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 8.

Description

Machine tool after-sale operation and maintenance intelligent question-answering method based on intention recognition Technical Field The invention relates to the field of intersection of artificial intelligence, natural language processing and industrial operation and maintenance, in particular to an after-sale operation and maintenance intelligent question-answering method of a machine tool based on intention recognition. Background Along with the intelligent upgrade of the manufacturing industry, the numerical control machine tool is used as core equipment of the modern manufacturing industry, the stable and efficient operation of the numerical control machine tool is crucial, and the response speed, the diagnosis accuracy and the specialty of solutions of after-sales operation and maintenance service directly influence the continuity of a production line and the economic benefit of enterprises. However, the machine tool after-market operation and maintenance field mainly depends on the following modes, and all have significant limitations: The traditional manual expert mode is characterized in that the remote or on-site diagnosis is carried out by severely relying on engineers with abundant experience, the mode is slow in response, high in cost and scarce in expert resources, and massive and concurrent customer consultation demands are difficult to deal with in a large scale. The question-answering system based on rules or traditional FAQ can only process predefined simple and fixed problems, has poor flexibility, can not understand complex and changeable fault scenes described by natural language of users, and can not generate dynamic solutions. The intelligent question and answer based on a single general large model is that general language models such as GPT, religion and the like are directly applied, and the models have strong natural language understanding and generating capability, but lack of deep expertise in the field of machine tools, so that deviation exists in understanding of professional terms, fault mechanisms and maintenance flows, and the generated answer is always reasonable in look and cannot be operated, and even potential safety hazards exist. Therefore, an intelligent operation and maintenance question-answering technology which can deeply integrate domain knowledge, accurately understand user intention, effectively utilize multi-source data and continuously evolve is needed so as to improve the automation and intelligent level of machine tool after-sale service. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides an after-sale operation and maintenance intelligent question-answering method based on intention recognition, which aims to solve the defects of the prior art in aspects of professional suitability, question-answering accuracy, scene coverage, intention recognition accuracy and the like and improve the quality and efficiency of after-sale operation and maintenance service of a machine tool through dual-model advantage complementation, field knowledge graph alignment, dynamic feedback mechanism, multi-dimensional intention recognition system, improved LORA algorithm training and intention driving retrieval enhancement generation. The invention adopts the following technical scheme: An intelligent after-sale operation and maintenance question-answering method of a machine tool based on intention recognition comprises the following steps: Step S1, preprocessing multi-source operation and maintenance data, namely carrying out standardized and structured conversion on multi-source data such as natural language problems, machine tool models/parameters, fault logs, historical maintenance records and the like in a machine tool after-sales scene, and synchronously constructing a multi-dimensional intention recognition marking data set; S2, extracting dual-model features, namely extracting the semantics and the intention features of natural language problems through a first pre-training language model, extracting the professional features of related field data through a second pre-training language model, training an intention recognition model based on an improved LORA algorithm, and outputting intention labels and confidence; s3, knowledge graph alignment and attention fusion, namely constructing a knowledge graph in the machine tool operation and maintenance field, performing semantic alignment on the features extracted in the S2 and the knowledge graph nodes, and generating a fusion feature vector through a dual-model collaborative attention mechanism; Step S4, intent-driven dynamic retrieval enhancement generation, namely dynamically retrieving related knowledge based on the intent label output in the step S2, combining the fusion feature vector generated in the step S3, constructing a prompt word and generating a structured operation and maintenance solution; And S5, feedback iterative optimization, namely, based on feedback da