CN-122022432-A - Intelligent body auxiliary method for detecting strength of metal material by field instrumented indentation method

CN122022432ACN 122022432 ACN122022432 ACN 122022432ACN-122022432-A

Abstract

The invention discloses an agent auxiliary method for detecting strength of a metal material by an on-site instrumented indentation method, which comprises the following steps of S1, receiving a component to be detected and on-site information, inputting information of the component to be detected by a user, S2, extracting key characteristic fields from the input information, carrying out structuring and vectorization processing, S3, executing vector similarity matching in a tool knowledge base and an analysis mode knowledge base to generate an initial evidence packet, carrying out iterative updating to obtain a scene evidence packet, S4, generating an arbitration rule based on the scene evidence packet, calculating combination credibility, and outputting a recommended result, and S5, outputting a detection auxiliary instruction packet comprising recommended tools and analysis modes, recommended reasons and installation steps to an operator through a result output module. According to the invention, an operator can automatically recommend the tool and the analysis mode according to the form, the size, the type and the load-displacement curve of the detection component input by the operator under the condition of insufficient experience, and the tool installation step and the recommendation reason are provided.

Inventors

PENG YANG
ZHENG HAOYANG
Cao kuang
DONG JUN

Assignees

南京工业大学
南京工大建设工程技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (6)

1. The intelligent agent auxiliary method for detecting the strength of the metal material by the field instrumented indentation method is characterized by comprising the following steps of: S1, receiving the information of a member to be detected and the site, and inputting the material type name, the shape and the size of the member, the magnetic attraction of the surface of the member, the coating information and a load-displacement curve chart by a user; s2, extracting key feature fields from the input information, carrying out structuring and vectorization processing, outputting structured prompt words through a large model, and forming a high-dimensional embedded vector; S3, performing vector similarity matching in the tool knowledge base and the analysis mode knowledge base to generate an initial evidence packet, calculating a divergence value through divergence judgment and a notch positioning network, inquiring supplementary information as required, and performing iterative updating to obtain a scene evidence packet; s4, generating an arbitration rule based on a scene evidence packet, performing three-level check on the detection surface accessibility and tool installation, the support rigidity and test piece stability, the material property and the cycle curve characteristic, calculating the combination credibility, and outputting a recommended result; S5, outputting a detection auxiliary instruction packet containing recommended tools, recommended analysis modes, recommended reasons and installation steps to an operator through a result output module.
2. The method for assisting an agent for detecting the strength of a metal material by an on-site instrumented indentation method according to claim 1, wherein the step S1 is specifically: The user inputs the information of the component to be detected and the field detection information to the intelligent body through a man-machine interaction interface, the input content supports text input, meanwhile, the JPG, PNG pictures and PDF format files are compatible for uploading, the input information comprises a material type name, a component form size, component surface magnetic attraction property, coating information and a load-displacement curve graph, the component form is circular, rectangular and special-shaped, and the size parameters are the diameter of the circular component, the length of the rectangular component is multiplied by the width of the rectangular component, and the thickness of the component.
3. The method for assisting the intelligent agent for detecting the strength of the metal material by the field instrumented indentation method according to claim 2, wherein the step S2 is specifically: Based on a general large model, carrying out semantic understanding and disambiguation on the input information, removing redundant information and unifying expression specifications; Converting the unstructured text subjected to standardized processing into structured data in the form of key value pairs through a rule engine and a feature engineering algorithm, and extracting core feature fields which are strongly related to a target task, wherein the core feature fields comprise structural morphology, size parameters, material type names, surface magnetic attraction, surface coating information and the like; Encoding the core feature field and the curve form label to form 67-dimensional structured feature vectors serving as a structured feature set; Based on the automatic generation function of the large model prompting word, generating a structured prompting word according to the input information, and converting the fuzzy expression of the user into the indentation detection technical term; and calling a preset embedding model, mapping the structured prompt word information into 768-dimensional high-dimensional vectors, forming vectorization input information which can be used for retrieving a vector knowledge base, wherein the dimensions of the vectors are consistent with the dimensions of the field data vectors stored in the vector knowledge base, and ensuring the effectiveness of similarity calculation.
4. The method for assisting an agent for detecting the strength of a metal material by an on-site instrumented indentation method according to claim 3, wherein the step S3 is specifically: The general large model executes cosine similarity retrieval, bifurcation judging and notch positioning network algorithm, the cosine similarity retrieval is used for retrieving and generating an initial evidence packet set, and the bifurcation judging and notch positioning network is used for calculating bifurcation values for the initial evidence packet and feeding back prompt words needing to be supplemented and optimized; based on a general large model, taking 768-dimensional high-dimensional embedded vectors as search probes, adopting a cosine similarity algorithm, respectively executing search in an analysis mode vector knowledge base and a tool vector knowledge base to obtain a tool evidence set and an analysis mode evidence set, and merging to form an initial evidence packet set; Each piece of evidence in the initial evidence packet set is encoded into 40-dimensional evidence feature vectors, wherein the 40-dimensional evidence feature vectors comprise tool candidate conclusion category encoding pointed by the evidence, analysis mode candidate conclusion category encoding and establishment condition coverage encoding; The method comprises the steps of calling a bifurcation judgment and notch positioning network, generating a 3-dimensional constraint sequence vector of a detection surface reachable and tooling installation limit item, a support rigidity and test piece stability limit item, a material property and curve characteristic limit item according to an instrumented indentation method industry rule, taking a 40-dimensional evidence feature vector and a 3-dimensional constraint sequence vector as scene input vectors, outputting a 64-dimensional evidence matching vector through bifurcation judgment operation, and carrying out weighted summarization on the 64-dimensional evidence matching vector to obtain an evidence summarization vector; When the 1-dimensional divergence value is larger than or equal to the divergence threshold, the candidate conclusion of the initial evidence packet is judged to be scattered, key information gaps exist, core input information which needs to be supplemented is accurately positioned based on the 12-dimensional missing field score vector, a targeted information supplementing prompt is generated, a structured prompt word is updated after the supplementing information is received, a 768-dimensional high-dimensional embedded vector is generated through remapping, and a similarity retrieval and divergence judgment flow is returned to be executed again until the scene evidence packet set meeting the threshold requirement is generated; the construction of the tool knowledge base and the analysis mode knowledge base comprises data acquisition, structural conversion and vectorization storage, and specifically comprises the following steps: Performing questionnaires and depth interviews on industry experts to obtain experience knowledge about analysis modes and tool adaptation; collecting analysis mode names used by different metal materials and standard load-displacement curve graphs of different metal materials during historical detection, collecting tool instruction manuals matched with an indentation instrument and technical guidelines of newly developed tools, and sorting the experience knowledge, the instruction manuals and the historical detection data into unstructured data of documents and pictures; Carrying out cleaning treatment on document data, carrying out sentence segmentation operation on text data according to semantic logic to obtain cleaned data, converting the cleaned data according to a preset fixed format to obtain structured data, wherein the structured data of a tool knowledge base and an analysis mode knowledge base adopt a key value pair and adapting label format, the structured data core field of the tool knowledge base comprises tool types, adapting member forms, size ranges, supporting rigidity threshold values, material magnetism requirements, space limited adapting grades, applicable detection postures and environment adapting conditions, the structured data core field of the analysis mode knowledge base comprises analysis mode names, adapting material types, a strength ratio range, curve characteristics and inversion precision indexes, carrying out cleaning treatment on load-displacement curve data, extracting numerical value characteristics and morphological characteristics from load curves, unloading points and unloading curves of load-displacement curves, and carrying out artificial marking on the load-displacement curves of different metal materials to obtain structured data; And converting the processed structured data into 768-dimensional vectors through the same embedded model to obtain generated vector data, and storing the vector data into a vector database to obtain an analysis mode vector knowledge base and a tool vector knowledge base.
5. The method for assisting an agent for detecting the strength of a metal material by an on-site instrumented indentation method according to claim 4, wherein the step S4 is specifically: Dividing a scene evidence package set into a tool knowledge base evidence and an analysis mode knowledge base evidence according to evidence sources, extracting tool candidate conclusions, analysis mode candidate conclusions and establishment condition fields from each evidence to obtain a candidate conclusion node set; Initializing a establishment condition gap count and a feasible marker for a candidate combination set, determining establishment condition check references of each combination, and carrying out hierarchical constraint check according to three-dimensional constraint sequence vector priority; The first stage performs detection surface reachability and tooling installation checking on the candidate combination set, adopts detection gesture, installation space and surface magnetic attraction fields in the structural feature set, checks the detection surface reachability limit items in the rule judgment and tooling installation conditions, and eliminates candidate combinations which do not meet the limit items; the second stage executes the stable check of the support rigidity and the test piece on the reserved candidate combination, adopts the input component size parameter and the input test piece surface state field to check the support rigidity threshold value and the adapting surface state condition in the rule judgment, and reserves the candidate combination with sufficient support rigidity and attached to the test piece surface; Thirdly, checking the reserved candidate combination for material attribute and cycle curve characteristic, checking the adaptive material type and the adaptive cycle curve characteristic condition in rule arbitration by adopting the input material type and curve form label field, and updating the establishment condition gap count; calculating the credibility of each candidate combination based on the evidence matching vector weight carried in the scene evidence packet and the establishment condition gap count; When the combined reliability is larger than the reliability threshold, outputting a candidate combination with highest combined reliability as a recommended combination, combining a candidate combination with next highest combined reliability, marking that the conditions of all combinations are satisfied, and when the combined reliability is smaller than the reliability threshold, optimizing and adjusting a scene evidence packet set.
6. The method for assisting an agent for detecting the strength of a metal material by an on-site instrumented indentation method as set forth in claim 5, wherein the step S5 is specifically: The method comprises the steps of integrating output information to generate a standardized detection auxiliary instruction packet, wherein the instruction packet is a recommended tool, a recommended reason and an installation step, the installation step at least comprises flatness adjustment, stability inspection and abnormality investigation prompt, and the recommended analysis mode and the recommended reason at least comprise material attribute and cycle curve characteristic check basis or similar case basis.

Description

Intelligent body auxiliary method for detecting strength of metal material by field instrumented indentation method Technical Field The invention relates to the field of material mechanical property detection and intelligent auxiliary decision making, in particular to an intelligent agent method for tool selection, analysis mode selection and result closed loop error correction in instrumented indentation method field detection. Background The instrumented indentation method is used as a micro-damage mechanical property testing means, and load-displacement response is loaded and recorded on the surface of a material through a pressure head, so that key mechanical parameters such as yield strength, tensile strength and the like are obtained through inversion. However, software for instrumented indentation testing typically provides multiple analysis modes to accommodate the elastoplastic behavior and hardening characteristics of different metallic materials. For on-site operators (e.g., those skilled in the civil engineering arts), it is often difficult to accurately determine the analysis mode based on information such as load-displacement curves or yield ratios due to lack of knowledge of the metallic materials. Under the condition of wrong analysis mode selection before inversion of the load-displacement curve, systematic deviation can occur in predicted results such as yield strength, tensile strength and the like obtained by subsequent inversion, and serious data distortion is caused. The reliability of the indentation test result is reduced, and the rework rate and the misjudgment risk of field detection are greatly increased. Secondly, the on-site indentation detection needs to select a proper clamping and supporting tool according to the shape and working condition of the detected component, and common types comprise a V-shaped cushion block, a U-shaped bracket, a magnetic clamp, a vacuum chuck and the like. In different industry applications, the shape and size, the surface state, the installation space, the detection attitude and the field environment of the measured object are greatly different, and the applicability of each tool is not only dependent on geometric size and shape matching, but also limited by various factors such as magnetism of materials, supporting rigidity, limited space and the like. The indentation process requires that the pressure head is in stable contact with the surface of the test piece and keeps vertical pressure, and once the tool is improperly selected or installed, the support rigidity is insufficient, the load-displacement curve can be fluctuated, non-uniform in circulation or distorted, so that the mechanical curve obtained by the test is discrete or even distorted. Therefore, an intelligent auxiliary scheme facing the field is needed, and the analysis mode selection and the tool selection are formed into a standardized process, so as to reduce the detection threshold and improve the detection consistency and reliability. Disclosure of Invention The invention aims to overcome the defects of the prior art, provides an intelligent agent auxiliary method for detecting the strength of a metal material on-site instrumented indentation method, and can automatically recommend a tool and an analysis mode according to the form, the size, the type and the load-displacement curve of a detection component input by an operator under the condition of insufficient experience of the operator, and provide a tool installation step and a recommendation reason. The technical scheme adopted by the invention is that the intelligent agent auxiliary method for detecting the strength of the metal material by an on-site instrumented indentation method comprises the following steps: S1, receiving the information of a member to be detected and the site, and inputting the material type name, the shape and the size of the member, the magnetic attraction of the surface of the member, the coating information and a load-displacement curve chart by a user; s2, extracting key feature fields from the input information, carrying out structuring and vectorization processing, outputting structured prompt words through a large model, and forming a high-dimensional embedded vector; S3, performing vector similarity matching in the tool knowledge base and the analysis mode knowledge base to generate an initial evidence packet, calculating a divergence value through divergence judgment and a notch positioning network, inquiring supplementary information as required, and performing iterative updating to obtain a scene evidence packet; s4, generating an arbitration rule based on a scene evidence packet, performing three-level check on the detection surface accessibility and tool installation, the support rigidity and test piece stability, the material property and the cycle curve characteristic, calculating the combination credibility, and outputting a recommended result; S5, outputting a detection auxiliary i