Ai-based building defect response system with emotion analysis and construction information integration
Patent Information
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2025-06-13
- Publication Date
- 2026-08-12
Smart Images
Figure 112025066317419-PAT00001_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to the field of building defect response technology, and more specifically, to an AI-based appraisal analysis and construction information-linked building defect response system that collects and analyzes voice, text, and image-based complaint data using artificial intelligence, and enables a rapid and accurate response to building defect complaints through appraisal analysis, defect type classification, construction history linkage, schedule optimization, and blockchain-based data integrity assurance. Background Technology
[0003] Traditionally, systems for receiving and processing complaints regarding building defects, such as in apartments, have operated primarily through manual processes. When a defect complaint was received from a resident, the manager would verify the details, manually classify them, establish a processing schedule, or forward the matter to the construction company. This structure followed a uniform procedure regardless of the urgency of the complaint or the resident's emotional state, leading to delays even in situations requiring an immediate response. In particular, the system lacked the ability to analyze emotions and reflect situations where the complaint involved heightened emotion or repetitive protests.
[0004] Furthermore, voice-based complaint submission capabilities were limited or non-existent, requiring users to input text directly or agents to manually record the details. Consequently, the digitization and documentation of defect complaints were insufficient, and there were limitations in data standardization and follow-up tracking. The image-based defect complaint submission function was also limited to simple uploads, preventing automatic classification or severity analysis, and making it difficult to consistently identify defect types.
[0005] Furthermore, the analysis of defect locations and the linkage with construction history were conducted in a fragmented manner, and there was a lack of functionality to automatically retrieve detailed information such as construction materials, contractors, and construction years in real time through integration with BIM (Building Information Modeling) data. As a result, it was time-consuming or errors could occur in assessing the precision of defect causes and determining the contractor's liability.
[0006] In the case of defect resolution schedules, assignments and adjustments were mostly made manually based on the manager's judgment, which presented a limitation in that readjustments due to technician schedule conflicts or resource unavailability could not be made immediately. Even when urgent complaints arose, adjustments had to be made individually based on the existing schedule, leading to operational inefficiencies such as conflicts in work order and duplication of technician assignments.
[0007] Furthermore, defect complaints and processing records are stored only in a central database, making them vulnerable to falsification, alteration, or omission. While a storage method capable of flawlessly recording the entire process of complaint reception and response is required to ensure the reliability of complaint processing and secure objective evidence in the event of a dispute, conventional systems lacked this capability.
[0008] Therefore, conventional technology contains technical limitations in various aspects, such as the absence of analysis of civil complaint assessment status, lack of automatic defect type classification functions, insufficient real-time linkage of BIM information, non-automation of defect processing schedules, and inadequate assurance of integrity based on blockchain. Consequently, there is an increasing need for an automated, intelligent defect response system that ensures efficiency and reliability. Prior art literature
[0010] Published Patent 10-2023-0043419 (AI-based building defect inspection system) Published Patent 10-2004-0072429 (Building defect management system and method using a portable terminal) Published Patent 10-2021-0096353 (Building defect management method, server and system) Published Patent 10-2025-0019895 (Artificial intelligence-based apartment facility maintenance management system and method) The problem to be solved
[0011] The present invention was developed to improve upon the aforementioned problems, and aims to provide a system that efficiently and reliably automates the entire process, from the receipt of building defect complaints to analysis, the establishment of processing schedules, and the management of response results, by utilizing artificial intelligence and blockchain technology. Through this, the invention seeks to fundamentally resolve the inefficiencies and schedule conflicts caused by subjective judgments, delays in complaint response, and data integrity issues that appear in the existing defect management process, and to ensure transparency, speed, and objectivity throughout the entire defect response process.
[0012] Specifically, the present invention aims to provide a technology that automatically collects and standardizes resident complaint information in the form of voice, text, and images, accurately classifies the type and severity of the complaint based on this data, and automatically determines whether the complaint is urgent by quantitatively evaluating the resident's emotional state. Furthermore, by utilizing Building Information Modeling (BIM), it supports a more rapid and accurate response by automatically linking construction history, material information, and responsible contractor information regarding the location of the complaint in real time.
[0013] In particular, the present invention aims to provide a technical means to dynamically optimize response schedules and technician assignments immediately when urgent complaints arise or conflicts with existing work schedules are anticipated, by automating the establishment and readjustment of defect complaint processing schedules using an artificial intelligence-based reinforcement learning algorithm. Through this, the purpose is to maximize the efficient utilization of technicians and equipment resources and minimize delays in complaint response times.
[0014] Furthermore, the present invention aims to provide an environment where data can be utilized as reliable, objective evidence in the event of future disputes by recording all data from the defect complaint reception and processing process in the form of hashes on a blockchain network to prevent data tampering and continuously guarantee integrity.
[0015] Overall, the main objective of the present invention is to provide a comprehensive and efficient defect response platform that includes not only response after a defect occurs but also prediction of the possibility of future defects and preemptive preventive management, by establishing an intelligent defect management system that combines artificial intelligence and blockchain. means of solving the problem
[0017] To achieve the above objectives, the AI-based sentiment analysis and construction information-linked building defect response system according to the present invention is an AI-based sentiment analysis and construction information-linked building defect response system for receiving and promptly responding to building defect complaints from residents, comprising: a complaint reception unit that collects input data from residents to generate complaint data so that residents can easily report defect complaints using voice, text, and images; a sentiment analysis unit that analyzes voice or text-based complaint data transmitted from the complaint reception unit using a natural language processing-based artificial intelligence model to quantitatively determine the resident's sentiment state and determine a sentiment score and whether the complaint is urgent; a defect classification unit that analyzes image data transmitted from the complaint reception unit using a deep learning-based defect classification model to generate defect types and severity scores; a BIM data linkage unit that automatically searches for construction history, related material information, and responsible contractor information for a corresponding location in a Building Information Modeling (BIM) database based on building location information linked to the complaint data and links it to the complaint data; and the complaint data received by the complaint reception unit, the analysis result of the sentiment analysis unit, the analysis result of the defect classification unit, and BIM data It is configured to include a blockchain storage unit that converts all processed data, including the results of the integration unit, into a hash form and stores it in a blockchain network to ensure data integrity.
[0018] The aforementioned civil complaint reception department,
[0019] It can be configured to include a STT (Speech-to-Text) unit that converts voice into text in real time when a resident inputs a voice complaint, a multi-image processing unit capable of simultaneously uploading multiple defect images, and a resident notification unit that provides the processing status and results of received complaints to the resident in real time.
[0020] The aforementioned emotion analysis department,
[0021] It may be configured to include an input data preprocessing unit that receives and preprocesses voice or text complaint data from residents and refines noise and unstructured expressions; an emotion quantitative evaluation unit that uses a pre-trained deep learning-based natural language processing model to deeply analyze the context of the input data and subdivide the types of negative emotions into emotion categories to quantitatively evaluate the emotional state; and an urgency determination unit that automatically classifies the complaint as an urgent processing target when the emotion score evaluated by the emotion quantitative evaluation unit exceeds a preset threshold or when specific keywords or emotion patterns are repeatedly detected.
[0022] The above defect classification unit is,
[0023] It may be configured to include an image preprocessing unit that performs noise removal preprocessing on defect images received from a complaint reception unit, an object detection and classification unit that automatically detects defect objects within defect images and classifies the types of defects using a deep learning-based defect classification model, and a severity evaluation unit that analyzes the size, shape, and degree of damage of defect objects detected by the object detection and classification unit to score the severity of each defect and determine whether an emergency response is necessary.
[0024] The above BIM data linkage unit is,
[0025] It may be configured to include a location information extraction unit that extracts defect occurrence location information of complaints received from tenants; a construction history inquiry unit that automatically retrieves and provides detailed construction history, including structural member information, materials used, construction company information, and construction year, from the building's BIM data based on the defect occurrence location information provided by the location information extraction unit; and a report provision unit that searches for similar defect repair records at similar locations within previously received complaint data and provides the existing similar defect repair records to the manager to provide a report so that the manager can respond to newly received defects.
[0026] The above blockchain storage unit is,
[0027] It may be configured to include a data structuring unit that collects complaint reception data and structures it into JSON form, a data hash conversion unit that converts the structured data into an encrypted hash value using the SHA-256 hash algorithm, a blockchain network record unit that stores the hash value in real-time on a blockchain network to continuously maintain data integrity, and a data integrity verification and notification unit that periodically verifies the hash value recorded on the blockchain and the original data, detects whether the data has been tampered with, and immediately notifies the administrator.
[0029] The present invention may further include an AI-based defect processing schedule optimization unit that inputs the appraisal score determined by the appraisal analysis unit, whether it is an urgent complaint, the defect type and severity score generated by the defect classification unit, whether there is a need for urgent response, and the technicians available for defect repair, their schedules, and the current availability of defect repair equipment into a response schedule optimization algorithm to automatically assign a defect processing schedule and a technician in charge of work corresponding to a new complaint, and performs real-time schedule readjustment while minimizing conflicts with the existing defect processing schedule when a new urgent defect processing complaint occurs.
[0030] The above AI-based defect processing schedule optimization unit is,
[0031] It may be configured to include a reinforcement learning decision unit configured to train a policy network by defining an appraisal score, whether there is an urgent complaint, defect type, severity score, technician, and equipment availability information as a state vector, determining the defect processing priority, technician and equipment assignment, and work start time as actions using a deep learning-based reinforcement learning algorithm, and setting the reward to be proportional to the reduction of the urgent complaint delay time and the increase in resource utilization rate; a defect processing schedule generation and update unit configured to automatically generate a preliminary schedule including the start time, work order, assigned technician, and necessary equipment for each defect by applying the actions calculated by the reinforcement learning decision unit, and dynamically update the defect processing schedule through partial recalculation when an urgent complaint or resource change occurs; and a defect processing schedule notification and verification unit configured to synchronize the finalized defect processing schedule in real time with the administrator dashboard and mobile app, send automatic notifications to technicians and residents, and record the schedule version identification hash value on the blockchain to verify integrity. Effects of the invention
[0033] The AI-based sentiment analysis and construction information-linked building defect response system according to the present invention provides the following effects.
[0034] First, by introducing artificial intelligence technology throughout the entire defect complaint handling process, errors and inefficiencies caused by manual work by managers can be minimized by automatically classifying complaint types and urgency. In particular, by quantitatively analyzing residents' emotional states to quickly determine the urgency of complaints, it is possible to prevent emotional conflicts from escalating into complaints in advance.
[0035] Second, by using deep learning-based image analysis technologies such as YOLOv5 and Vision Transformer (ViT), the type and severity of defects can be automatically determined and objectively evaluated, thereby improving the accuracy and consistency of complaint response and significantly reducing the workload of managers.
[0036] Third, by linking with a BIM (Building Information Modeling) database, construction history, material information, and information on the responsible contractor can be automatically retrieved in real time, thereby improving the accuracy of analyzing the causes of complaints and responding to them. Furthermore, by enabling the clear identification of liability, confusion arising during the complaint handling process can be prevented and disputes with the construction company can be minimized.
[0037] Fourth, through reinforcement learning-based schedule optimization technology, automatic generation and real-time dynamic readjustment of complaint processing schedules are possible. This enables immediate and efficient resource allocation and schedule adjustment even when urgent complaints arise or resource availability fluctuates. As a result, the speed of response to defects can be significantly increased, and the efficient utilization of technicians and equipment can be maximized.
[0038] Fifth, through a blockchain-based data integrity guarantee function, all records regarding the receipt, analysis, response, and scheduling of complaints are transparently stored in an immutable form, thereby providing objective and reliable records throughout the entire complaint response process. This establishes trust among residents, managers, and construction companies and creates an environment where records can be utilized as evidence in the event of a dispute, thereby minimizing legal disputes and costs.
[0039] Overall, the present invention has the effect of increasing the efficiency, accuracy, and transparency of responding to defect complaints, reducing the workload of managers, and simultaneously increasing resident satisfaction and system reliability. Brief explanation of the drawing
[0041] FIG. 1 is an example diagram of the configuration of the AI-based sentiment analysis and construction information-linked building defect response system of the present invention. FIG. 2 is an overall functional block diagram of the AI-based emotion analysis and construction information-linked building defect response system of the present invention. Figure 3 is a functional block diagram of the complaint reception department and the appraisal analysis department, Figure 4 is a functional block diagram of the defect classification unit and the BIM data linkage unit, Fig. 5 is a functional block diagram of a blockchain storage unit and an AI-based defect processing schedule optimization unit, Figure 6 is a flowchart illustrating the operation process of the AI-based emotion analysis and construction information-linked building defect response system of the present invention. Specific details for implementing the invention
[0042] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings.
[0043] However, the present invention is not limited to the embodiments disclosed below but will be implemented in various different forms.
[0044] The embodiments described in this specification are provided to ensure that the disclosure of the invention is complete and to fully inform those skilled in the art of the scope of the invention.
[0045] And the present invention is defined only by the scope of the claims.
[0046] Accordingly, in some embodiments, well-known components, well-known operations, and well-known techniques are not specifically described to avoid the invention being interpreted ambiguously.
[0047] Additionally, throughout the specification, the same reference numerals refer to the same components, and the terms used (mentioned) in this specification are for describing embodiments and are not intended to limit the invention.
[0048] In this specification, the singular form includes the plural form unless specifically stated otherwise in the text, and components and operations referred to as 'comprising (or comprising)' do not exclude the presence or addition of one or more other components and operations.
[0049] Unless otherwise defined, all terms used in this specification (including technical and scientific terms) may be used in a meaning that is commonly understood by those skilled in the art to which the present invention belongs.
[0050] Also, terms defined in commonly used dictionaries are not interpreted ideally or excessively unless otherwise defined.
[0051] Hereinafter, preferred embodiments of the present invention will be described with reference to the attached drawings.
[0052] The present invention will be described in detail with reference to FIGS. 1 to 6.
[0053] The AI-based emotional analysis and construction information-linked building defect response system (900) according to the present invention is a system designed to effectively manage defect complaints from building occupants and to respond quickly and accurately according to the occupant's emotional state and defect type.
[0054] The present system (900) is configured to include a complaint reception unit (100), an appraisal analysis unit (200), a defect classification unit (300), a BIM data linkage unit (400), and a blockchain storage unit (500). It may also include an AI-based defect processing schedule optimization unit (600). These components of the present invention operate on a building defect response server (700), and the AI-based appraisal analysis and construction information-linked building defect response system (900) includes a building defect response server (700) and a blockchain network (800).
[0055] The complaint reception department (100) generates complaint data by collecting input data from residents so that residents can easily report defect complaints using voice, text, and images.
[0056] The complaint reception department (100) supports residents in easily reporting complaints in the form of voice, text, or images via a smartphone app or web.
[0057] Specifically, voice is received from the resident via a microphone in PCM or WAV format, text in UTF-8 based JSON, and images in JPEG or PNG format. The input data is validated at the client end and then transmitted to the server (700) to generate complaint data. This complaint data is stored in the internal server (700) in a structured JSON format that includes a complaint ID, time of report, exact location information within the building, and reporter ID.
[0058] The emotion analysis unit (200) uses a natural language processing (NLP)-based artificial intelligence model to quantitatively determine the emotional state of the resident using voice or text data received from the complaint reception unit (100) and determines the emotion score and whether it is an urgent complaint.
[0059] The emotion analysis unit (200) probabilistically predicts multidimensional emotions such as anger, anxiety, dissatisfaction, neutrality, and positivity using a KoBERT or KLUE-based Transformer model, and quantifies the emotion score in the range of 0 to 100 based on this. In addition, if the emotion score exceeds a preset threshold (e.g., 70 points) or if specific keywords indicating urgency (e.g., "collapse," "leakage," "power outage") are detected, it is automatically classified as an urgent complaint and responded to as a priority.
[0060] The defect classification unit (300) analyzes image data transmitted from the complaint reception unit (100) using a deep learning-based defect classification model to generate the type and severity score of the defect.
[0061] The defect classification unit (300) analyzes image data received from the complaint reception unit (100) using a hybrid model combining deep learning-based YOLOv5 and Vision Transformer (ViT), and automatically classifies it into more than 30 predefined defect types, such as cracks, leaks, mold, condensation, paint peeling, tile breakage, and pipe leaks.
[0062] Specifically, the defect classification unit (300) quickly detects the location and area of a defect object in a complaint image using the YOLOv5 model, extracts the detected area, and performs a more precise defect type classification through the Vision Transformer (ViT). This hybrid structure, composed of two stages, simultaneously improves object detection speed and classification accuracy, and is designed to ensure high classification accuracy and reliability, especially given the characteristics of building defects that appear in various types and forms.
[0063] The defect classification unit (300) generates a final severity score (0 to 1.0) by applying a predefined severity weight (e.g., leakage 1.5, crack 1.2, etc.) to each type and area of each classified defect, and uses this to determine the priority of defect response.
[0064] The BIM data linkage unit (400) automatically searches for the construction history, material information, and responsible contractor information of the corresponding location in the BIM database (440) based on the building location information linked to the complaint data and links it with the complaint data.
[0065] The BIM data linkage unit (400) automatically searches for relevant information in a building information model (BIM) database (440) built in the IFC (Industry Foundation Classes) format based on the location information of the building included in the complaint data.
[0066] Specifically, the coordinates of the complaint location are used to search for construction components at that location using a 3D R-Tree spatial index, and information on construction history (construction year, maintenance records), related materials (material name, manufacturer, specifications, etc.), and the responsible contractor (company name, contact information) is extracted. This information is automatically linked with complaint data to support immediate utilization in the complaint response process.
[0067] The blockchain storage unit (500) converts the complaint data received by the complaint reception unit (100), the analysis results of the appraisal analysis unit (200), the analysis results of the defect classification unit (300), and the results of the BIM data linkage unit (400) into a hash form and stores them in the blockchain network (800) to ensure data integrity.
[0068] The blockchain storage unit (500) integrates all processing results, including complaint data, appraisal analysis results, defect classification results, and BIM linkage data, into JSON format and converts them into encrypted hash values using the SHA-256 algorithm. The generated hash values are immediately stored in a private blockchain network (800) to prevent data alteration and falsification. The original data is periodically verified using the hash values recorded in the blockchain to continuously maintain data integrity and to enable its use as evidence when necessary.
[0070] The building defect response system (900) of the present invention operates as shown in the following example.
[0071] A resident uploads two photos of a bathroom ceiling leak using a smartphone app and reports it via voice, saying, "Water keeps dripping, please come quickly!"
[0072] The complaint reception department (100) immediately converts the voice into text, generates complaint data, and transmits it to the building defect response server (700).
[0073] The emotion analysis unit (200) analyzes the converted text data to detect high anger and dissatisfaction, calculates the emotion score as 85 points, and classifies it as an urgent complaint.
[0074] The defect classification unit (300) analyzes the uploaded image and identifies it as a "leakage" type, and calculates the severity score as 0.9 because the leakage area is wide.
[0075] The BIM data linkage unit (400) extracts information on the construction member closest to the reported location (e.g., "bathroom plumbing system, PVC 20A, 2022 construction by Company A") and links it with the complaint data.
[0076] The blockchain storage unit (500) integrates all of the above results to generate a hash value and records it in the blockchain network (800) to ensure the integrity of the data.
[0078] The complaint reception unit (100) of the present invention supports various input forms (voice, text, image) so that a resident can easily and quickly report defects, and in particular, improves the efficiency of complaint processing by including real-time text conversion (STT) for voice input, processing of simultaneous upload of multiple images, and a real-time notification function for the status of complaint processing.
[0079] The complaint reception unit (100) is configured to include an STT (Speech-to-Text) unit (110), a multi-image processing unit (120), and a resident notification unit (130).
[0080] The STT unit (110) converts complaints reported by residents via voice through a smartphone or web app into text in real time. The STT unit (110) can use cloud-based speech recognition engines such as Google Cloud Speech-to-Text, AWS Transcribe, or Naver CLOVA Speech, and can utilize a user-customized speech recognition model pre-tuned with specialized domain terms related to apartment defect complaints to increase speech recognition accuracy.
[0081] Specifically, voice data in PCM or WAV format is received, converted to a sampling rate of 16 kHz, and then transmitted to a cloud server via real-time streaming. The server returns the converted text result to the client within one second, allowing the resident to immediately check the converted result on the screen as soon as they report.
[0082] The multi-image processing unit (120) supports the resident in uploading multiple defect images simultaneously when reporting defects. When multiple images are selected from the resident's smartphone app or web client and transmitted to the server simultaneously, the server immediately compresses the uploaded images and optimizes their size (e.g., reducing the size of each image file to 2MB or less and applying a JPEG compression rate of 80% or more), then generates a unique ID (UUID) for each image and stores it in a server storage (S3, Azure Blob Storage, etc.) along with the complaint ID. The multi-image processing unit (120) reduces the image processing load through asynchronous processing and provides the resident with a pleasant reporting experience.
[0083] The resident notification unit (130) notifies the resident that the complaint has been received immediately upon receipt, and subsequently delivers the processing status (received, processing in progress, completed, etc.) and the final processing result in real time. This notification function is implemented using messaging platforms such as Firebase Cloud Messaging (FCM), Apple Push Notification (APNs), and KakaoTalk, and sends a notification message to the resident's smartphone app or SMS whenever the processing status changes.
[0084] The resident notification unit (130) automatically sends a notification when there is a delay in processing a complaint, thereby minimizing requests from residents to check the status of their complaint processing and improving resident satisfaction and trust.
[0086] The complaint reception department (100) operates as shown in the following example.
[0087] A resident discovers a leak in the living room ceiling, opens a smartphone app, says "There is a leak in the living room ceiling" in voice, and uploads three related photos.
[0088] The STT unit (110) immediately converts the voice into text and displays it on the resident's screen, and the resident checks the text conversion content in real time.
[0089] The multi-image processing unit (120) automatically compresses three images uploaded simultaneously and quickly saves them to the server.
[0090] The resident notification unit (130) immediately sends a "Complaint has been successfully received" message to the resident upon receipt of the report, and subsequently sends real-time notifications whenever the complaint enters the processing stage or is completed, thereby providing a transparent processing status.
[0092] The emotion analysis unit (200) of the present invention is a key component for accurately identifying the emotional state of a resident based on the resident's voice or text input during the complaint reception process, and for objectively and quickly determining the urgency of a defect complaint.
[0093] The emotion analysis unit (200) is configured to include an input data preprocessing unit (210), an emotion quantitative evaluation unit (220), and an urgency determination unit (230).
[0094] The input data preprocessing unit (210) preprocesses the resident's voice or text complaint data received from the complaint reception unit (100) into a form suitable for analysis. Voice data is converted into text through noise removal, deletion of silent segments, and STT (Speech-to-Text), and text data is refined into a form suitable for a natural language processing model by performing typographical correction, removal of repetitive expressions, deletion of special characters, and standardization of unstructured expressions (colloquial, abbreviated expressions, etc.) through a regular expression-based preprocessing algorithm.
[0095] Specifically, background noise is removed using open-source speech preprocessing libraries such as WebRTC or Noise Suppression, and text data is tokenized using HuggingFace's Tokenizers.
[0096] The emotion quantitative evaluation unit (220) uses a pre-trained deep learning-based natural language processing (NLP) model to analyze the emotional state of a resident based on pre-processed text data, and quantitatively evaluates various emotional states that appear in context. This evaluation unit utilizes Transformer-based models such as KoBERT and KLUE-RoBERTa that have been pre-trained using a Korean civil complaint dataset (approximately 100,000 cases or more), and evaluates the negative emotions of the resident that appear in context by subdividing them into detailed categories such as anger, anxiety, dissatisfaction, and irritation.
[0097] The final emotional state is obtained by extracting probability scores for each emotion type from a Softmax layer, and summing them to produce a final normalized emotion score ranging from 0 to 100 points.
[0098] For example, in a complaint, if the emotion quantitative evaluation unit (220) predicts a probability of 60% anger, 20% anxiety, 10% dissatisfaction, and 10% neutrality, the weights of each emotion are applied to convert this into a score (e.g., anger 1.0, anxiety 0.8, dissatisfaction 0.5) and the final score (e.g., 74 points) is quantitatively derived.
[0099] The urgency determination unit (230) automatically classifies a complaint as an emergency processing target based on the appraisal score evaluated by the appraisal quantitative evaluation unit (220), if it exceeds a preset emergency processing threshold (e.g., 70 points) or if specific emergency keywords (e.g., "leakage," "power outage," "danger," "collapse") are found in the complaint text.
[0100] These urgent complaints are immediately registered in the priority processing list within the complaint response system and are promptly conveyed to the relevant management personnel via notifications to support rapid action.
[0102] The emotion analysis unit (200) operates as shown in the following example.
[0103] A resident discovers a serious leak in the bathroom ceiling and reports in an urgent voice, "Water is dripping heavily from the bathroom ceiling, please take action immediately!"
[0104] The input data preprocessing unit (210) immediately converts voice data into text and refines noise and unstructured expressions to convert them into standard text form.
[0105] The emotion quantitative evaluation unit (220) analyzes the preprocessed text using a KoBERT model to detect high levels of anger (80%) and anxiety (15%), and calculates an emotion score of 85 points.
[0106] The urgency determination unit (230) recognizes that the calculated appraisal score has exceeded the urgency threshold (70 points) and automatically registers the complaint in the urgency processing list to induce a quick on-site response by the manager.
[0108] The defect classification unit (300) of the present invention is a key component for automatically classifying the type of defect by analyzing defect image data collected from residents and quantitatively evaluating the severity to determine the necessity of an emergency response.
[0109] The defect classification unit (300) receives image data transmitted from the complaint reception unit as input and processes it in three stages: preprocessing -> defect detection and classification -> severity evaluation.
[0110] The defect classification unit (300) is configured to include an image preprocessing unit (310), an object detection and classification unit (320), and a severity evaluation unit (330).
[0111] The image preprocessing unit (310) removes information unnecessary for analysis from the defect image transmitted through the complaint reception unit (100) and performs a preprocessing process to increase the accuracy of defect detection. Specifically, it follows the following procedure.
[0112] Resolution normalization: Resize the input image to 640x640 or 512x512 pixels to fit the model input format.
[0113] Noise Removal: Removes visual noise caused by illumination, shadows, blurring, etc. by applying an OpenCV-based Gaussian blur or anisotropic diffusion filter.
[0114] Brightness and Saturation Correction: Adaptive Histogram Equalization (AHE) is applied to normalize brightness and enhance contrast so that defective areas are clearly distinguished.
[0115] This preprocessing provides a foundation for subsequent deep learning models to detect defective objects more accurately.
[0116] The object detection and classification unit (320) receives a preprocessed image as input, automatically detects defective objects within the image, and classifies them into predefined defect types. The following deep learning structure is utilized.
[0117] Model structure: Using a YOLOv5 or YOLOv8-based backbone, the location of defective objects is quickly detected, and then the features of the detected area are analyzed using a Vision Transformer (ViT) or Swin Transformer, and finally the type of defect is determined.
[0118] Defect type classification: It is classified into one of more than 30 predefined defect types, such as cracks, leaks, mold, paint peeling, and concrete spalling.
[0119] Multi-object recognition: When there are multiple defective objects in an image, each is individually detected and classified, and an ID and location coordinates (x, y, w, h) are assigned to each.
[0120] The severity evaluation unit (330) calculates the severity of each defect based on the size, shape, and degree of damage of the detected defect object. At this time, the following criteria and methods are used.
[0121] Area-based analysis: Calculates the size relative to the entire image based on the area ratio of the object detection box.
[0122] Shape-based analysis: Assigns additional weight if the object has irregular boundaries (e.g., waveforms or diffusion patterns).
[0123] Apply defect type weights: For example, a severity weight of 1.5 is assigned to each defect type, such as for leaks, cracks, and mold, and 1.0 is assigned.
[0124] Calculation of final severity score: Severity = (Area ratio X Shape weight) X Type weight, and if this value exceeds a certain standard (e.g., 0.7 or higher), it is determined that an emergency response is required.
[0126] The defect classification unit (300) operates as shown in the following example.
[0127] The resident submits a photo of mold and cracks mixed on the living room wall through the complaint reception department (100).
[0128] The image preprocessing unit (310) resizes the input image to 640×640 size and removes noise to obtain a clear outline.
[0129] The object detection and classification unit (320) detects two objects, mold and crack, using YOLOv5 and accurately classifies them into mold (object 1) and crack (object 2) using ViT.
[0130] The severity evaluation unit (330) calculates scores of 0.25 and 0.12 × 1.2 = 0.144, respectively, considering that the area of the mold object is 25% of the total image and has a weight of 1.0, and the area of the crack object is 10% and has a weight of 1.2, and determines the severity to be moderate or higher.
[0131] This information is linked to the emotion analysis unit (200) and the AI-based defect processing schedule optimization unit (600) and is set as a priority processing target.
[0133] The BIM data linkage unit (400) of the present invention is a core component that links the location of a defect reported as a complaint with the building's design and construction information, thereby supporting rapid and precise defect response based on construction history and similar case data. This component is connected to a BIM (Building Information Modeling) database and significantly improves the accuracy and efficiency of defect response through location-based linkage and past case-based response functions.
[0134] The BIM data linkage unit (400) is configured to include a location information extraction unit (410), a construction history inquiry unit (420), and a report provision unit (430).
[0135] The location information extraction unit (410) receives location information (e.g., text selection, dropdown, map click, GPS coordinates, etc.) entered by the resident when registering a defect complaint and normalizes it into a form that can be used in the BIM system.
[0136] Text-based location information: For example, the phrase "Room 302 bathroom ceiling" is analyzed using the building-unit-space-part system and mapped as "D3-F302-BATHROOM-CEILING".
[0137] Coordinate-based information: Location information collected from mobile devices via GPS is converted into the coordinate system of the BIM model (the building's local reference coordinate system). At this stage, accurate location mapping is achieved using the Helmert 7-parameter transformation method or GIS map matching techniques.
[0138] BIM Element Mapping: The converted location information is linked to individual objects (e.g., IfcWall, IfcSlab, IfcPipeSegment, etc.) within the IFC or Revit model to extract the element IDs of the corresponding defect locations.
[0139] The construction history lookup unit (420) automatically retrieves detailed construction history from the BIM database (440) based on the element ID derived from the location information extraction unit (410). The items to be retrieved are as follows:
[0140] Structural member information: Structural type (slab, wall, piping, etc.), dimensions, strength, etc. at the location
[0141] Material Information: Type of material (PVC, RC, wood, etc.), Specifications (Standards, Manufacturer, Durability)
[0142] Construction Company Information: Names of prime and subcontractors responsible for the construction, licenses, contact information, etc.
[0143] Construction Year and Records: Initial construction date of the relevant area, maintenance history (if any), remodeling status, etc.
[0144] This data is extracted via an automatic query method from a PostgreSQL + PostGIS or NoSQL-based BIM integrated database and stored in an integrated data structure linked to complaint IDs.
[0145] The report providing unit (430) searches for cases corresponding to similar locations and types of defects in the BIM DB as well as the past complaint record repository and provides a comparable case report to the manager.
[0146] Similar Case Search: Extracts cases satisfying similar conditions among past defect complaints through multidimensional filtering based on defect type (e.g., cracks, leaks, etc.), location (space ID), severity score, etc.
[0147] Automatic Report Generation: Automatically generates a report for extracted historical cases that includes summary information such as the date of occurrence, handling method (maintenance / replacement), processing time, and recurrence status.
[0148] Format and delivery method: It can be viewed in PDF / HTML format on the administrator dashboard, and when selected, related images and response results are also provided.
[0150] The BIM data linkage unit (400) operates as shown in the following example.
[0151] A resident registers a complaint and a photo showing mold on the "ceiling of the master bedroom in Unit 303."
[0152] The location information extraction unit (410) combines the information of 'Room 303', 'master bedroom', and 'ceiling' to extract the element ID 'FL03-RM03-CEIL-002' within the BIM model.
[0153] The construction history lookup unit (420) automatically extracts from the BIM database (440) that the element is a gypsum board ceiling constructed by A Construction in 2020 and has a pipe leak repair history in February 2023.
[0154] The report provider (430) finds three past complaint records corresponding to the ‘mold’ defect and the ‘three-story master bedroom ceiling’ location within the same complex and provides information such as the repair method (anti-mold application, pipe leak inspection), and average processing time (2 days) to the manager in the form of a report.
[0156] The blockchain storage unit (500) of the present invention is a component for protecting core data generated during the complaint processing process from tampering and for technically guaranteeing the transparency and integrity of the processing history. This configuration stores the receipt of defect complaints, the analysis results and response history therefrom, etc., in a blockchain network (800), thereby enabling anyone to verify whether the data has been altered and securing the reliability and auditability of the complaint processing.
[0157] The blockchain storage unit (500) is configured to include a data standardization unit (510), a data hash conversion unit (520), a blockchain network record unit (530), and a data integrity verification and notification unit (540).
[0158] The data standardization unit (510) collects not only complaint data generated from the complaint reception unit (100) but also analysis result data such as appraisal analysis results, defect classification results, and BIM linkage information, and organizes them into a structured JSON format.
[0159] The data hash conversion unit (520) converts data structured in JSON into an encrypted fixed-length hash value using the SHA-256 hash algorithm.
[0160] Processing method: Serialize JSON into a string -> Apply SHA-256 -> Generate a 64-character hash value.
[0161] Ensuring security: Since identical data generates identical hash values, it is easy to verify whether data has been altered by comparing hash values.
[0162] The blockchain network record book (530) stores the generated hash value in the blockchain network (800) in real time. Through this process, the complaint processing data is safely preserved as a blockchain record that cannot be tampered with.
[0163] The data integrity verification and notification unit (540) periodically compares the hash value recorded on the blockchain with the original data currently stored to verify integrity, and if an anomaly is detected, it immediately sends a warning to the administrator. This module performs the role of detecting and responding to security threats such as forgery or data loss in advance.
[0165] The blockchain storage unit (500) operates as shown in the following example.
[0166] When a resident registers a complaint regarding a leak in the living room ceiling, the complaint information and analysis results are organized in JSON format.
[0167] The data hash conversion unit (520) converts this JSON into a hash value using the SHA-256 algorithm.
[0168] The blockchain network record book (530) stores the hash value along with the complaint ID in the blockchain network (800).
[0169] After 10 minutes, the data integrity verification and notification unit (540) regenerates the hash value for the complaint and compares it with the blockchain value.
[0170] If they do not match, a notification of suspected tampering is immediately sent to the administrator via email and push notifications.
[0172] The AI-based defect processing schedule optimization unit (600) of the present invention is a module designed to automatically generate a defect processing schedule by comprehensively considering the results of appraisal analysis, defect classification results, and information on technicians and equipment resources during the complaint processing process, assign appropriate technicians, and adjust conflicts between schedules in real time. In particular, even when a new urgent complaint arises, the schedule can be readjusted quickly and efficiently while minimizing conflicts with the existing schedule.
[0173] The AI-based defect processing schedule optimization unit (600) operates based on the following procedure:
[0174] 1. Collect various analysis results and resource status related to defect complaints as input
[0175] 2. Automatically generates work schedules through a schedule optimization algorithm
[0176] 3. When a new urgent complaint arises, determine whether it conflicts with the existing schedule and adjust the schedule in real time.
[0178] The main data input to the AI-based defect processing schedule optimization unit (600) is as follows.
[0179] An emotional score (e.g., 0 to 100) and whether it is an urgent complaint (e.g., Boolean) transmitted from the emotional analysis unit (200)
[0180] The defect type (e.g., leakage, crack, etc.) and severity score (e.g., 0 to 1.0) provided by the defect classification unit (300)
[0181] Emergency Response Status: Automatically determined based on emotional score or severity score
[0182] Technician Information: Technician ID, Skill Type, Proficiency, Available Schedule Slots
[0183] Equipment Information: List of required equipment by defect type, current availability and location information for each piece of equipment
[0185] The AI-based defect processing schedule optimization unit (600) inputs the above key data into a corresponding schedule optimization algorithm to automatically assign a defect processing schedule and a technician in charge of work corresponding to a new complaint, and performs real-time schedule readjustment while minimizing conflicts with the existing defect processing schedule when a new urgent defect processing complaint occurs.
[0186] The specific configuration of the AI-based defect processing schedule optimization unit (600) is examined in detail below.
[0188] The AI-based defect processing schedule optimization unit (600) according to the present invention is a scheduling system that autonomously learns and optimizes a defect processing schedule based on civil complaint data by introducing a reinforcement learning technique. In particular, it integrates the assessment state, defect characteristics, resource information, etc., into a single state vector to construct a reinforcement learning-based decision policy, and through this, automatically derives the priority of civil complaints, the assignment of technicians and equipment, and the start time of work.
[0189] The AI-based defect processing schedule optimization unit (600) is configured to include a reinforcement learning decision unit (610), a defect processing schedule generation and update unit (620), and a defect processing schedule notification verification unit (630).
[0190] The reinforcement learning decision unit (610) defines the emotional score, whether it is an urgent complaint, the type of defect, the severity score, the technician, and the equipment availability information as state vectors, and uses a deep learning-based reinforcement learning algorithm to determine the priority of defect processing, the assignment of technicians and equipment, and the time of work commencement as actions, and is configured to train a policy network by setting the reward to be proportional to the reduction of the urgent complaint delay time and the increase in resource utilization rate.
[0191] The reinforcement learning decision unit (610) operates with the following elements.
[0192] ● State Definition
[0193] To quantitatively represent the defect complaint situation, the following items are defined as state vectors.
[0194] yes:
[0195] o Appraisal Score (0~100, level of resident dissatisfaction)
[0196] o Urgent complaint status (1: Urgent, 0: General)
[0197] o Defect Types (One-hot encoding: 30 classifications including cracks, leaks, mold, etc.)
[0198] o Severity score (0.0~1.0)
[0199] o Technician Availability (Number of technicians capable of responding to the defect, skill level, available working hours, etc.)
[0200] o Equipment availability status (availability of essential equipment, location, occupancy rate, etc.)
[0202] ● Definition of Action
[0203] The policy network outputs the following behavior.
[0204] o Determination of priority processing for the relevant complaint
[0205] o Determination of commencement time
[0206] o Decision on Technician and Equipment Assignment
[0208] ● Reward Function
[0209] The reward for learning is structured by reflecting the following elements.
[0210] o The shorter the response time to urgent complaints, the higher the compensation.
[0211] o The higher the utilization rate of technicians or equipment, the higher the reward.
[0212] o Penalties are imposed in case of schedule conflicts or processing delays
[0214] ● Deep learning model
[0215] Policy functions are typically designed using Proximal Policy Optimization (PPO) or Deep Q-Network (DQN) structures and are iteratively trained to output the optimal action for a given state.
[0217] The defect processing schedule generation and update unit (620) is configured to automatically generate a preliminary schedule including the start time, work sequence, responsible technician, and necessary equipment for each defect by applying the actions calculated by the reinforcement learning decision unit (610), and to dynamically update the defect processing schedule by partial recalculation when an urgent complaint or resource change occurs.
[0218] The defect processing schedule generation and update unit (620) generates and updates the following schedule based on the output result of the reinforcement learning decision unit (610).
[0219] ● Automatic generation of preliminary schedules
[0220] For the entered complaints, a preliminary schedule is prepared by synthesizing information on the start time, priority, assigned technician, and required equipment.
[0221] For example, a specific schedule is created such as "May 21, 10:00 AM, Technician Kim assigned for leak complaint, Wet Equipment No. 1 reserved".
[0223] ● Dynamic recalculation function
[0224] In cases where new urgent complaints are added, or resource changes occur such as schedule changes for existing technicians or equipment failures, the entire schedule is not recalculated; instead, only the parts of the schedule that conflict with the relevant complaint are selectively adjusted.
[0225] The recalculation algorithm used at this time derives the optimal adjustment plan by re-calling a Conflict-Aware Rescheduler based on subgraph search or a reinforcement learning agent.
[0227] The defect processing schedule notification verification unit (630) is configured to synchronize the confirmed defect processing schedule in real time with the administrator dashboard and mobile app, send automatic notifications to technicians and residents, and record the schedule version identification hash value in the blockchain network (800) to verify integrity.
[0228] The defect handling schedule notification and verification unit (630) transmits the confirmed schedule information to the relevant parties and performs a blockchain storage function to prevent falsification or alteration of the schedule history.
[0229] ● Schedule synchronization and notification features
[0230] o Real-time synchronization with the administrator dashboard and mobile app
[0231] o Send work assignments and changes to technicians via push notifications
[0232] o Send complaint status updates to residents (e.g., "Technician ○○ scheduled to visit at 14:00")
[0234] ● Schedule integrity verification
[0235] o Structure schedule information (task ID, start time, person in charge, etc.) in JSON format and generate a SHA-256 hash
[0236] o Store the generated hash on the blockchain network
[0237] Whether the schedule has changed since then can be verified by comparing hash values.
[0239] The AI-based defect processing schedule optimization unit (600) operates as shown in the following example.
[0240] This was organized as an actual operational example by reflecting the existing work details of technicians A, B, and C, whether there were any schedule changes, and the process of rescheduling in the event of urgent complaints.
[0241] Step 1: Initial complaint reception and preliminary schedule creation
[0242] At 9:00 AM, resident A filed a complaint regarding "bathroom ceiling leak."
[0243] The appraisal analysis department (200) determined the appraisal score to be 8.7 and the urgent complaint status to be 'yes'.
[0244] ㆍThe defect classification unit (300) classifies the defect type as ‘leakage’ and the severity score as 9.5.
[0245] ㆍThe BIM data linkage unit (400) checks the pipe aging history at the corresponding location.
[0247] The reinforcement learning decision unit (610) makes a decision with the following state vector:
[0248] ㆍScore: 8.7
[0249] ㆍUrgency: Yes
[0250] ㆍDefect Type: Leakage
[0251] Severity: 9.5
[0252] ㆍTechnician Status:
[0253] o A: No morning work
[0254] o B: Detailed inspection of exterior wall cracks scheduled for 1 PM
[0255] o C: Non-urgent wallpaper repair work in progress since morning
[0257] Reward function settings:
[0258] ㆍ Minimizing delays in urgent complaints: High compensation
[0259] ㆍ Utilizing Technician Idle Time: Intermediate Rewards
[0260] ㆍ Existing schedule delay: Penalty compensation granted (within acceptable range)
[0262] The defect handling schedule generation and update unit (620) generates the following preliminary schedule:
[0263] ㆍ A: Prioritize deployment to leak complaint sites starting at 11:00 AM
[0264] ㆍ B: Maintain scheduled afternoon work
[0265] ㆍ C: Keep existing work
[0267] The schedule is synchronized with the relevant schedule manager and technician through the defect processing schedule notification and verification unit (630), and the hash value is recorded on the blockchain.
[0269] Step 2: Occurrence of New Urgent Complaints and Rescheduling
[0270] ㆍ At 10:40 AM, Resident B filed an urgent complaint regarding "underground parking lot drainage backflow."
[0271] ㆍ Appraisal Analysis Score: 9.3, Defect Type: 'Drainage Backflow', Severity: 9.8
[0273] The reinforcement learning decision unit (610) updates the state vector:
[0274] ㆍ Technician A: Scheduled to respond to leak complaints starting at 11:00 AM
[0275] ㆍ Technician B: Work scheduled for 1 PM
[0276] ㆍ Technician C: Non-urgent work in progress
[0278] The reinforcement learning policy network makes dynamic decisions as follows:
[0279] ㆍ Technician A remains unchanged due to existing complaint handling.
[0280] Technician B postpones afternoon work by one hour and is prioritized for handling urgent civil complaints starting at 11:30.
[0281] ㆍ Technician C maintains non-urgent work
[0283] The defect processing schedule generation and update unit (620) dynamically recalculates and updates the schedule as follows:
[0284] ㆍ B's detailed diagnosis schedule has been delayed by one hour and moved to 14:00.
[0285] ㆍ C has no task changes
[0286] ㆍ A responds to the leak site as before
[0288] The following items are reflected in real-time through the defect handling schedule notification and verification unit (630):
[0289] ㆍ B receives notification for urgent civil complaint response and prepares to move
[0290] ㆍ Synchronize changes to B's existing task schedule to the administrator dashboard
[0291] ㆍ The changed schedule hash value is generated and recorded on the blockchain
[0293] The AI-based defect processing schedule optimization unit (600) utilizes a policy network of a reinforcement learning algorithm to dynamically determine whether to delay or maintain the existing work schedule based on the relative importance of urgent complaints. At this time, the key judgment criteria are reflected in the priority adjustment logic defined within the compensation function, and comprehensively consider the urgency of the complaint, the severity of the defect, the appraisal score, and the current status of the technician.
[0294] The system analyzes complaint data in real time and can automatically generate or update defect resolution schedules. It also incorporates risk avoidance logic to prevent schedule conflicts, such as ensuring minimum intervals between tasks, preventing equipment duplication, and minimizing technician movement paths. In particular, even if a specific technician already has fixed tasks, if an urgent complaint arises, the system compares the importance of the existing schedule; if it is determined that the complaint requires higher priority, the existing schedule is postponed and the urgent complaint is assigned first.
[0295] Conversely, the system may be designed so that schedule changes are not made when existing tasks are more important than urgent complaints, or when the technician's expertise and grounds for priority assignment are sufficient. These criteria for scheduling decisions are designed through reward engineering within the reinforcement learning model and are adjusted to simultaneously minimize task conflicts and optimize resources.
[0296] The changed schedule is immediately synchronized in real-time with the technician's mobile app and the administrator dashboard, and technicians automatically receive notifications regarding the newly assigned work order, time, and location. This enables on-site response without confusion caused by schedule changes and ensures the system's feasibility and responsiveness in the field.
[0298] According to the present invention having the above-described configuration, the speed and accuracy of complaint processing can be dramatically improved by automating the entire process from the receipt of building defect complaints to analysis, schedule optimization, and recording of response results based on AI. When a complaint is received, various forms of data such as text, voice, and images can be processed in an integrated manner, thereby increasing user accessibility, and the urgency of the complaint and the severity of the defect can be quantitatively determined through sentiment analysis and defect classification algorithms.
[0299] Furthermore, the BIM data integration function enables rapid and accurate cause diagnosis by automatically linking existing construction history and related information, and utilizes a reinforcement learning-based schedule optimization algorithm to intelligently determine the assignment of technicians and equipment, work priorities, and start times. In particular, it enhances on-site responsiveness by providing a real-time dynamic scheduling function that partially recalculates existing schedules to minimize conflicts while increasing resource utilization in the event of new urgent complaints.
[0300] Furthermore, by recording all data regarding the complaint reception and processing process using a hash-based system via a blockchain storage, an immutable and flawless record is secured, which can be utilized as reliable evidence in the event of future disputes. This establishes trust between residents and managers, ensures operational transparency between construction companies and technicians, enhances the quality of the overall defect response service, and maximizes operational efficiency.
[0301] Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art to which the present invention pertains should understand that the present invention can be implemented in other specific forms without changing its technical concept or essential features, and therefore the embodiments described above should be understood as illustrative in all respects and not restrictive.
[0302] Furthermore, the scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention. Explanation of the symbols
[0304] 100...Complaint Reception Desk 110...STT Department 120...Multiple Image Processing Unit 130...Tenant Notification Board 200...Emotion Analysis Department 210...Input Data Preprocessing Unit 220...Quantitative Appraisal Department 230... Urgency Decision Department 300...Defect Classification Department 310...Image Preprocessing Section 320...Object Detection and Classification Unit 330...Severity Assessment Department 400...BIM Data Integration Section 410...Location information extraction unit 420...Construction History Inquiry 430...Report Provider 440...BIM Database 500...Blockchain Storage 510...Data structuring section 520...Data hash conversion unit 530...Blockchain Network Register 540...Data integrity verification and notification section 600...AI-based Defect Processing Schedule Optimization Unit 610...Reinforcement learning decision unit 620...Defect handling schedule creation and update section 630...Defect Resolution Schedule Notification Verification Department 700... Building defect response server 800...Blockchain Network 900...AI-based sentiment analysis and construction information-linked building defect response system
Claims
Claim 1 An AI-based appraisal analysis and construction information-linked building defect response system for receiving and promptly responding to building defect complaints from tenants, comprising: a complaint reception unit that collects input data from tenants to generate complaint data so that tenants can easily report defect complaints using voice, text, and images; an appraisal analysis unit that analyzes voice or text-based complaint data transmitted from the complaint reception unit using a natural language processing-based artificial intelligence model to quantitatively determine the tenant's emotional state and decide the appraisal score and whether it is an urgent complaint; a defect classification unit that analyzes image data transmitted from the complaint reception unit using a deep learning-based defect classification model to generate defect types and severity scores; and a BIM data linkage unit that automatically searches for construction history, related material information, and responsible contractor information for a corresponding location in a Building Information Modeling (BIM) database based on building location information linked to the complaint data, and links it to the complaint data. A blockchain storage unit that guarantees data integrity by converting all processing data, including complaint data received by the complaint reception unit, analysis results from the appraisal analysis unit, analysis results from the defect classification unit, and results from the BIM data linkage unit, into a hash form and storing it in a blockchain network; and an AI-based defect processing schedule optimization unit that inputs the appraisal score determined by the appraisal analysis unit, whether it is an urgent complaint, the defect type and severity score generated by the defect classification unit, whether an urgent response is necessary, and the technicians available for defect repair, their schedules, and the current availability of defect repair equipment into a response schedule optimization algorithm to automatically assign a defect processing schedule and a technician in charge of work corresponding to a new complaint, and performs real-time schedule readjustment while minimizing conflicts with existing defect processing schedules when a new urgent defect processing complaint occurs.The AI-based defect processing schedule optimization unit includes: a reinforcement learning decision unit configured to train a policy network by defining appraisal scores, urgent complaint status, defect type, severity score, technicians, and equipment availability information as state vectors, determining defect processing priority, technician and equipment assignment, and work start time as actions using a deep learning-based reinforcement learning algorithm, and setting rewards proportional to the reduction of urgent complaint delay time and the increase in resource utilization rate; a defect processing schedule generation and update unit configured to automatically generate a preliminary schedule including start time, work sequence, assigned technician, and required equipment for each defect by applying the actions calculated by the reinforcement learning decision unit, and to dynamically update the defect processing schedule through partial recalculation when urgent complaints or resource changes occur; and a defect processing schedule notification and verification unit configured to synchronize the confirmed defect processing schedule in real-time with an administrator dashboard and a mobile app, send automatic notifications to technicians and residents, and record a schedule version identification hash value on a blockchain to enable integrity verification.An AI-based sentiment analysis and construction information-linked building defect response system comprising: the responsible contractor and related material information retrieved by the BIM data linkage unit is used as a key parameter by the reinforcement learning decision unit to assign the most suitable technician and necessary equipment for a specific defect; the reinforcement learning decision unit defines a cumulative reward function including the processing delay time of an urgent complaint, the idle time of a technician or equipment, and whether a delay in an existing defect processing schedule has occurred as input variables, respectively; and a policy network is configured to output an action corresponding to a state vector according to the cumulative reward function; and the defect processing schedule generation / updating unit is characterized by selectively recalculating by searching only the subgraph corresponding to the conflicting part among the schedule graphs composed of the precedence relationship between defects and resource allocation relationship, using a partial graph search-based Conflict-Aware rescheduling means when a new urgent complaint or resource change occurs. Claim 2 delete Claim 3 delete Claim 4 The AI-based sentiment analysis and construction information-linked building defect response system according to claim 1, wherein the defect classification unit comprises: an image preprocessing unit that performs noise removal preprocessing on a defect image received from a complaint reception unit; an object detection and classification unit that automatically detects defect objects within a defect image and classifies the type of defect using a deep learning-based defect classification model; and a severity evaluation unit that analyzes the size, shape, and degree of damage of defect objects detected by the object detection and classification unit to score the severity of each defect and determine whether an emergency response is necessary. Claim 5 The AI-based appraisal analysis and construction information-linked building defect response system according to claim 1, wherein the BIM data linkage unit comprises: a location information extraction unit that extracts defect occurrence location information of a complaint received from a resident; a construction history inquiry unit that automatically retrieves and provides detailed construction history including structural member information, materials used, construction company information, and construction year from the building's BIM data based on the defect occurrence location information provided by the location information extraction unit; and a report provision unit that searches for similar defect repair history at similar locations in previously received complaint data and provides the existing similar defect repair history to the manager to provide a report so that the manager can respond to newly received defects. Claim 6 delete Claim 7 delete Claim 8 delete
Citation Information
Patent Citations
Movement management system for maintenance person
JP2007119107A
COLLECTIVE BUILDING VoC(Voice of Customer) PROCESSING SYSTEM
KR102460292B1
Consolidated system of analyzing civil complaint
KR102658555B1
Blockchain-based tenant management system for defect repair history management and dispute prevention
KR102774646B1