Search

CN-121979918-A - Construction management and BIM component intelligent association method based on semantic space analysis

CN121979918ACN 121979918 ACN121979918 ACN 121979918ACN-121979918-A

Abstract

The invention discloses a construction management and BIM component intelligent association method based on semantic space analysis, which comprises the steps of collecting an original voice signal for processing, converting the voice signal into text information, carrying out entity identification on the text information, extracting an entity, extracting space coordinates and attribute information of components, constructing a three-dimensional space index, searching out all components located in a space position range by using the three-dimensional space index, filtering to obtain candidate components, sorting the candidate components based on the space coordinates and the component types of the candidate components, generating a mapping relation between relative sequence numbers and component identifications, mapping the relative sequence numbers of the entity into the component identifications by using the mapping relation, calculating the matching degree and the total matching degree of each candidate component in multiple dimensions to obtain a target BIM component, associating and storing detection results with the target BIM component, and updating the attribute or state of the corresponding component in a BIM model. The invention can improve the information association efficiency and the matching accuracy.

Inventors

  • HUANG CHENGUANG
  • CHEN KAI
  • WANG WENYUAN
  • LUO JIE
  • LI BING
  • Tan Yidao
  • ZHU JIAJUN
  • ZHANG KANG

Assignees

  • 中国建筑第四工程局有限公司

Dates

Publication Date
20260505
Application Date
20251231

Claims (10)

  1. 1. A construction management and BIM component intelligent association method based on semantic space analysis is characterized by comprising the following steps: Step 1, acquiring an original voice signal of a field person in a construction environment, and performing noise reduction treatment on the original voice signal to obtain a noise-reduced voice signal; step 2, converting the noise-reduced voice signal into text information, carrying out entity identification on the text information, and extracting an entity containing a space position, a component type, a relative sequence number, a detection parameter and a detection result; Step3, analyzing a project IFC file of the target building, and extracting the space coordinates and attribute information of the component; step 4, based on the spatial position of the entity, searching out all components positioned in the spatial position range by utilizing the three-dimensional spatial index, and filtering according to the component type of the entity to obtain candidate components; Step 5, sorting the candidate components according to a preset space coordinate sorting rule based on the space coordinates and the component types of the candidate components, generating a mapping relation between the relative sequence numbers and the component identifications based on the sorting result, and mapping the relative sequence numbers of the entities into the component identifications by utilizing the mapping relation; Step 6, calculating the matching degree of each candidate component in multiple dimensions, and carrying out weighted summation on the matching degree of each dimension according to preset weights to obtain the total matching degree of each candidate component; And 7, positioning the corresponding component in the BIM model according to the component identification of the target BIM component, storing the detection result and the component in a correlated way, and updating the attribute or state of the component.
  2. 2. The method for intelligently associating construction management with BIM components based on semantic space analysis according to claim 1, wherein the step 1 specifically includes: Step 11, acquiring original voice signals of field personnel in a construction environment by adopting a directional acoustic acquisition device, wherein the directional acoustic acquisition device is integrated with a micro vibration mechanism or an air flow dust removing mechanism with a self-cleaning function, and dust in the construction environment is removed by the micro vibration mechanism or the air flow dust removing mechanism; step 12, cutting the collected original voice signals according to preset frame length and frame movement to obtain a series of voice frames, and applying a Hamming window to each voice frame to carry out windowing treatment; Step 13, carrying out real-time filtering treatment on each windowed voice frame by adopting a self-adaptive filter based on a least mean square algorithm, wherein the self-adaptive filter tracks noise characteristics in real time through the least mean square algorithm and dynamically adjusts step factors thereof so as to filter power frequency interference and background mechanical noise and obtain a denoised voice frame; Step 14, carrying out high-frequency pre-emphasis treatment on the denoised voice frame by adopting a first-order high-pass filter to obtain a denoised voice signal, wherein the transfer function H (z) of the first-order high-pass filter is H (z) =1- μz -1 , wherein μ is a pre-emphasis coefficient, the value range is 0.9 to 0.98 so as to enhance the voice details of a 3kHz to 8kHz frequency band, and z represents a complex variable; And 15, calculating the signal-to-noise ratio of the noise-reduced voice signal, and if the signal-to-noise ratio is lower than a preset threshold, automatically adjusting the gain of the directional acoustic acquisition device and then re-acquiring the voice signal.
  3. 3. The intelligent association method of construction management and BIM components based on semantic space analysis according to claim 1, wherein the step 2 specifically includes: Step 21, constructing a voice recognition model and training, and inputting the noise-reduced voice signal into the trained voice recognition model for processing to obtain text information; and step 22, preprocessing the text information, identifying an entity containing a predefined category from the preprocessed text information by utilizing a pre-trained entity identification model, and performing confidence verification and output on an entity identification result.
  4. 4. The intelligent association method of construction management and BIM components based on semantic space analysis according to claim 3, wherein the step 21 of constructing and training a speech recognition model specifically includes: step 211, obtaining a plurality of voice samples from a building detection corpus, wherein each voice sample is marked with a corresponding normalized text to form a voice-text pair, and dividing the voice-text pair into a training set, a verification set and a test set according to a preset proportion, wherein the training set, the verification set and the test set are respectively used for model training, parameter tuning and performance evaluation; Step 212, a pre-trained voice recognition model based on an encoder-decoder converter architecture is adopted as a basic model, wherein the encoder is composed of a plurality of layers of converter blocks and is used for processing input voice characteristics, and the decoder is composed of the plurality of layers of converter blocks and is used for generating a corresponding text sequence; Step 213, freezing parameters of the bottom layer transducer block of the encoder, so that the parameters do not participate in updating in the back propagation process, and the extraction capacity of the speech recognition model on the general speech features is reserved; Step 214, training the speech recognition model by using the training set, and in the training process, performing iterative training on parameters of a high-level transducer block of the encoder and parameters of all transducer blocks of the decoder, optimizing by using a AdamW optimizer, and performing training speed by using a mixed precision training strategy; Step 215, performing parameter tuning on the voice recognition model through the verification set, and performing performance evaluation on the voice recognition model through the test set to finally obtain a trained voice recognition model; the step 22 specifically includes: step 221, defining the category of the entity, including spatial position, component type, relative sequence number, detection parameter and detection result; Step 222, preprocessing the text information, including text cleaning and normalization, word segmentation, removal of stop words and punctuation and index and vector preparation, wherein the text cleaning and normalization refers to removal of irrelevant symbols and redundant words and unification of term formats, the word segmentation refers to segmentation of the text according to words to obtain a series of words, the stop words and punctuation removal refers to removal of stop words and punctuation symbols without practical significance and preservation of effective words, and the index and vector preparation refers to conversion of each word into an index and preparation of vector representation corresponding to a vocabulary; Step 223, the entity recognition model is composed of a text vectorization representation layer, a bidirectional context feature fusion layer and a conditional random field labeling and optimal decoding layer; step 224, inputting each word after pretreatment into a text vectorization representation layer, and converting each word into a high-dimensional semantic vector through a pre-trained word vector model; Step 225, inputting the high-dimensional semantic vector into a bidirectional context feature fusion layer, processing the high-dimensional semantic vector from left to right and from right to left through a bidirectional long-short-term memory network, capturing context information of each vocabulary, and outputting a context feature vector; 226, inputting the context feature vector into a conditional random field labeling and optimal decoding layer, labeling each vocabulary by adopting a predefined BIO sequence labeling system, and decoding by utilizing an optimal label Viterbi algorithm to obtain an optimal entity label sequence; Step 227, calculating the confidence coefficient of each entity label output by the conditional random field label and the optimal decoding layer, triggering a manual confirmation process if the confidence coefficient is lower than a set threshold value, outputting an entity identification result after manual review, and outputting the entity identification result if the confidence coefficient is not lower than the set threshold value, wherein the entity identification result comprises an entity category, an entity value and the corresponding confidence coefficient.
  5. 5. The intelligent association method between construction management and BIM components based on semantic space parsing according to claim 3, wherein the step 22 further includes: Step 23, acquiring a construction detection text, extracting action-target-result triples from the construction detection text by using a pre-trained semantic role annotation model and outputting the action-target-result triples, wherein the method specifically comprises the following steps of: Step 231, generating text samples based on voice-text pairs, wherein each text sample is marked with an action-target-result triplet, and dividing the marked text samples into a training set, a verification set and a test set according to a preset proportion, wherein the action represents a specific action executed in a detection scene, the target represents a specific component of action, and the result represents a state or conclusion obtained after the action is executed; step 233, monitoring the performance of the semantic role annotation model by using the verification set in the training process; step 234, after training is completed, evaluating the performance of the semantic role annotation model by using the test set; And 235, inputting the building construction detection text to be analyzed into a trained semantic role marking model, wherein the semantic role marking model analyzes the input building construction detection text and outputs identified action entities, target entities and result entities, and the action entities, the target entities and the result entities are automatically combined into a structured action-target-result triplet.
  6. 6. The intelligent association method of construction management and BIM components based on semantic space analysis according to claim 1, wherein the step3 specifically includes: Step 31, analyzing the project IFC file through an IFC analyzer, and extracting the space coordinates and attribute information of each component in the project IFC file, wherein the attribute information comprises the type of the component, the axis position and the floor to which the component belongs; Step 32, taking a three-dimensional boundary cube of the whole building as a root node; Step 33, recursively dividing the root node into eight sub-nodes along X, Y, Z coordinate axis directions to form an octree node structure, wherein each sub-node represents a cube region in the building space; Step 34, judging the geometric center coordinates of each component or the space range of the outer packet and the child node according to the space coordinates and attribute information of the components, and registering the geometric center coordinates or the space range of the outer packet and the child node into the child node containing the component; And 35, storing the octree node structure and a member list contained in each child node to form a three-dimensional space index.
  7. 7. The intelligent association method of construction management and BIM components based on semantic space analysis according to claim 1, wherein the step 4 specifically includes: Step 41, mapping the spatial position of the entity into a specific spatial position range, wherein the floor description is mapped into a corresponding Z coordinate range, and the plane area description is mapped into X and Y coordinate ranges through spatial object definition or axis network coordinate derivation in the BIM model; Step 42, taking the space position range as a query range, executing range query in the three-dimensional space index, namely recursively judging whether the coordinate range of a child node is intersected with the query range from a root node, if so, entering the child node to continue to search, and collecting all components in the child node intersected with the query range; And 43, performing type filtering on the components according to the types of the components of the entity to obtain a candidate component list containing space coordinates and axis positions.
  8. 8. The intelligent association method of construction management and BIM components based on semantic space analysis according to claim 1, wherein the step 5 specifically includes: Step 51, extracting the space coordinates of the candidate components; step 52, dynamically determining a space coordinate ordering rule according to the component types, and ordering all candidate components according to the space coordinate ordering rule: If the component type is a column, firstly sorting according to the ascending order of the X-axis coordinates of the candidate components in a building coordinate system, and if the difference value of the X-axis coordinates of two adjacent candidate components is smaller than a preset threshold value, sorting according to the ascending order of the Y-axis coordinates; If the component type is a beam or a wall, firstly sequencing the Y-axis coordinates of the candidate components in an ascending order according to the space coordinates of the candidate components in a building coordinate system, and if the difference value of the Y-axis coordinates of two adjacent candidate components is smaller than a preset threshold value, sequencing the Y-axis coordinates of the two adjacent candidate components in an ascending order according to the X-axis coordinates; when the space coordinates of different candidate components overlap, the axis positions of the candidate components are read and compared for sorting; Step 53, compiling a corresponding relative sequence number for each candidate component according to the sequencing result of the current component type, generating a unique code for each relative sequence number, acquiring a floor and an area where the space coordinates of the candidate component are located according to the space coordinates of the candidate component, generating a unique component identifier for each candidate component according to the floor, the area, the component type and the code, and generating a mapping table containing the mapping relation between the relative sequence number and the component identifier under the current component type; Step 54, determining the target component type according to the component type of the entity, finding a corresponding mapping table according to the target component type, and matching the corresponding component identifier from the mapping table according to the relative sequence number of the entity.
  9. 9. The intelligent association method of construction management and BIM components based on semantic space analysis according to claim 1, wherein the step 6 calculates the matching degree of each candidate component in a plurality of dimensions, and performs weighted summation on the matching degrees of the dimensions according to preset weights to obtain the total matching degree of each candidate component, and the method specifically comprises the following steps: step 61, calculating the matching degree of each candidate component in a plurality of dimensions of space, type, sequence number, history record and detection parameters respectively; For the space matching degree S1, the space matching degree S1 is calculated based on Euclidean distance d between the geometric center of the candidate component and a reference coordinate point analyzed from detection parameters, wherein the equation is S1=1- (d/Dmax), dmax is a preset maximum span of a region, and when d > Dmax, S1=0; for the type matching degree S2, the type matching degree S2 is calculated based on the cosine similarity of the word vector between the type name of the candidate component and the component type text of the entity, and the formula is S2=cos (theta), wherein theta is the word vector included angle between the type name of the candidate component and the component type text of the entity; For the sequence number matching degree S3, if the candidate components are successfully matched according to the mapping relationship, s3=1.0, if the candidate components are partially matched, s3=0.8, and if the candidate components are not matched, s3=0; For the history matching degree S4, querying the frequency of hits of the candidate component described by the same relative sequence number in the history, if hit, s4=0.9+0.1× (hit number/total detection number of the candidate component), otherwise s4=0.5; Judging whether the attribute parameters of the candidate components are consistent with the detection parameters of the entity or not according to the detection parameter matching degree S5, if so, s5=1.0, if so, s5=0.6, otherwise, s5=0; step 62, calculating the total matching degree S of each candidate component by a weighted summation formula: S = w1·S1+ w2·S2+ w3·S3+ w4·S4+ w5·S5; wherein w1 represents a weight coefficient of a space dimension, w2 represents a weight coefficient of a type dimension, w3 represents a weight coefficient of a sequence number dimension, w4 represents a weight coefficient of a history dimension, w5 represents a weight coefficient of a detection parameter dimension, and specific numerical values of w1+w2+w3+w4+w5=1, w1, w2, w3, w4 and w5 are set by themselves according to the member densities of different items.
  10. 10. The intelligent association method of construction management and BIM components based on semantic space analysis according to claim 1, wherein the step 7 specifically includes: Step 71, packaging the spatial position of the entity, the component type of the entity, the relative sequence number of the entity, the semantic role labeling result, the original voice signal, the voice signal after noise reduction, the picture shot on site and the component identifier of the target BIM component into a detection record in a JSON format; Step 72, storing the structured data in the detection record into a relational database, storing the unstructured data into the non-relational database, and establishing an association index; Step 73, through an API interface of the BIM platform, locating a corresponding component in the BIM model according to the component identifier of the target BIM component, and writing the detection result into a custom attribute field of the component; And step 74, triggering a visual update instruction of the BIM according to the state of the detection result, and updating the material color or the display state of the target BIM component in the BIM.

Description

Construction management and BIM component intelligent association method based on semantic space analysis Technical Field The invention relates to the technical field of building engineering information, in particular to a construction management and BIM component intelligent association method based on semantic space analysis. Background In the construction quality detection of the building engineering, the operation habit of field personnel and the attribute specification of the BIM model are obviously different, and the specific problems and technical details are as follows: 1. The information association efficiency is low, A member in the existing BIM model uses A management mode with 'floor + region + type + code' (such as '3F-A-COL-005') as A unique identifier, and detection personnel are required to finish multi-step operation association results of floor screening, region positioning, member type filtering and code searching in A BIM software environment. The multi-stage retrieval method is long in time consumption in complex projects, and cannot meet the instantaneity requirement of on-site high-frequency detection work. The reason for this is that the BIM component identification system is optimized towards the design and modeling stage, but lacks the capability of fast retrieval mechanisms and natural language mapping for the job scenario here. 2. The manual matching error rate is high, A plurality of components with the appearance similar to the specification are present in the same floor and the same areA (like 10 columns and beams of the same type of floor), and absolute codes (like '3F-A-COL-005' and '3F-A-COL-006') are easy to be confused when specific components are distinguished in A visual or text list manually, so that the binding error of detection datA is caused. The generation reason is that the existing coding identification scheme is biased to the structured chemical unique identification on the information expression, cannot directly support on-site semantic description (such as a second root column on the left side), and lacks a visualization or semantic verification mechanism. 3. The stability of relying on external hardware is poor, namely the two-dimensional code, RFID and other labeling positioning schemes need to be pre-labeled on the components in the construction stage, but dust covering labels in the construction environment, the labels fall off when the components are hoisted, and the condition that a plurality of labels need to be attached to a large-volume component (such as a shear wall) can influence the stability and the integrity of information reading. The reason for this is that the physical carriers of these hardware identification methods are susceptible to construction environments and are costly to maintain, and cannot be reliably associated with BIM model properties when tag information is missing or damaged. 4. The BIM model only contains absolute coding attributes (such as component ID and axis coordinate), and lacks relative sequence number attributes (such as X-th root and left-hand root) based on spatial arrangement, so that natural language description of field personnel cannot be directly matched with model identification, and secondary conversion of positions and numbers must be carried out manually, thereby increasing operation time and error probability. The generation reason is that the BIM data structure at present does not establish a relative space positioning field aiming at a construction detection scene, and a combination mechanism of semanteme and space analysis is lacked. Disclosure of Invention Therefore, the invention aims to provide a construction management and BIM component intelligent association method based on semantic space analysis, which does not need to modify a BIM model bottom data structure, and realizes intelligent association of precise matching of a construction site fuzzy voice instruction and the BIM component by fusing acoustic enhancement, field self-adaptive semantic analysis and a three-dimensional space index technology. In order to achieve the technical purpose, the invention adopts the following technical scheme: The invention provides a construction management and BIM component intelligent association method based on semantic space analysis, which comprises the following steps: Step 1, acquiring an original voice signal of a field person in a construction environment, and performing noise reduction treatment on the original voice signal to obtain a noise-reduced voice signal; step 2, converting the noise-reduced voice signal into text information, carrying out entity identification on the text information, and extracting an entity containing a space position, a component type, a relative sequence number, a detection parameter and a detection result; Step3, analyzing a project IFC file of the target building, and extracting the space coordinates and attribute information of the component; step 4, based on the spat