12274 results about "Classification result" patented technology

The invention provides a mine ecological risk prediction method and system based on artificial intelligence, and the method comprises the steps: collecting multi-source ecological monitoring data through a distributed sensor network disposed in a mine region, carrying out the time-space alignment processing to generate a time-space correlation feature set, calling a pre-trained ecological risk prediction model to predict the features, and carrying out the prediction of the features. Generating a risk conduction mode set including a risk propagation path, an initial node, a propagation direction and strength, constructing an ecological risk evolution network based on the risk conduction mode set, and displaying a topological structure and a time sequence dependency relationship of risk nodes, and according to the ecological risk evolution network, generating a risk early warning instruction set containing a risk grade division result and an ecological restoration strategy, and sending the risk early warning instruction set to an ecological management platform, thereby comprehensively and accurately predicting the ecological risk of the mine, providing an effective risk early warning and ecological restoration strategy, and ensuring the ecological safety of the mine.

The invention relates to the field of defect detection, in particular to a PCB defect detection method and system based on image recognition. The method comprises the following steps: collecting a multidirectional PCB detection image, carrying out pixel-level registration correction and adaptive pixel stability compensation, and constructing a space-time stability compensation image sequence; performing reverse pyramid structure division on the space-time stability compensation image sequence, performing normalized similarity probability calculation, and constructing an initial region classification result; based on an initial region classification result, depth image visual analysis is carried out, pseudo defect comprehensive elimination optimization is carried out, and a pseudo defect purification high-confidence image is constructed; pCB connection defect identification is carried out on the pseudo defect purification high-confidence image, global defect point distribution marking is carried out, and a defect point space distribution diagram is constructed. According to the invention, high-credibility, high-precision and high-closed-loop PCB defect detection is realized.

The invention provides an equipment supervision data real-time analysis method and system, and the method comprises the steps: calculating an energy value of a frequency band corresponding to a sub-band signal, determining the energy characteristics of the surface roughness and dust deposition distribution uniformity of a blade in each sub-band in a square sum form, and generating energy distribution vectors which are arranged according to a frequency band sequence; performing time-frequency analysis on the vibration signal by using the acquired real-time rotating speed information of the gas turbine, identifying a frequency component corresponding to the eccentric distance value of the rotor and the vibration frequency offset, and determining the amplitude of the frequency component; according to the classification result and the real-time rotating speed information, the comprehensive influence index of the material density change rate and the ash deposition layer thickness difference is calculated, the fault early warning level is determined through a weighted average method, and an early warning signal is generated; and according to the change trend of the early warning signal and the blade edge wear degree, a random forest algorithm is adopted to predict a development curve of the vibration frequency offset in the future target time, and real-time equipment supervision analysis data are obtained.

Disclosed in the present invention is an SDN-architecture-based routing method for guaranteeing network QoS. The method comprises: on the basis of network topology information, performing abstraction to obtain a directed graph; then, on the basis of an improved K-means algorithm, dividing data into mouse flows, elephant flows and mammoth flows; defining a traffic transmission constraint condition; constructing a DQN neural network; training the DQN neural network by using the maximum Q value, which corresponds to an action a having the maximum reward function R, as an objective, so as to obtain a QoS guarantee routing model; obtaining an optimal path weight value; determining a data packet sending path on the basis a Dijkstra algorithm; and finally issuing path information to a switch. The present invention avoids the situations in which data processing performed by using a conventional K-means algorithm is not stable enough and a classification result is not accurate enough. During the training of a DQN neural network, data is extracted from an experience pool according to the priority of data, such that the neural network pays more attention to important data, thereby improving the learning efficiency of the neural network and the performance of the algorithm.

The invention provides a crop whole growth cycle identification method and system based on deep learning, and the method comprises the steps: obtaining a multispectral image sequence of a target crop in a continuous time period through an image collection device, and carrying out the standardized illumination adjustment processing of each image frame in the multispectral image sequence, obtaining a standard illumination image set corresponding to the multispectral image sequence; carrying out crop region segmentation processing on each image, extracting a local feature region related to a target crop, generating a standardized crop image set containing the local feature region, inputting the standardized crop image set into a pre-trained multi-task deep learning model, extracting combined features through parallel convolution branches, and carrying out image segmentation processing on the combined features; and executing cross-stage correlation analysis in the full connection layer, outputting a multi-task classification result corresponding to the target crop growth cycle, and generating a stage identification report corresponding to the target crop full growth cycle. According to the invention, the crop whole growth cycle identification precision and the agricultural management efficiency can be improved.

A layout optimization method of water quality monitoring points based on a RF-C-SOM clustering algorithm includes: preprocessing collected water quality data to obtain preprocessed water quality data that are used as data, and using water quality categories as labels to train a random forest model to determine feature importance of water quality indicators; selecting important features based on the feature importance and model training accuracy, performing dimensionality reduction on the preprocessed water quality data to obtain dimension-reduced data; performing a fuzzy clustering on the dimension-reduced data to obtain a water quality section classification result, and based on it, determining initial weight values of a self-organizing mapping algorithm; initializing neurons and training a self-organizing mapping network model with the initial weight values; obtaining a point clustering result through the self-organizing mapping network model; and conducting a water quality index evaluation for the point clustering result before and after screening.

In some embodiments, at a first time, a computing system receives a classification request specifying query content. The system executes a first pre-trained AI-generated content classifier in the set of AI-generated content classifiers on the query content to compute a set of confidence scores proportional to predicted probabilities that the query content includes content data produced by a generative AI model. The set of confidence scores is generated based on parameters learned by the first pre-trained AI-generated content classifier during a training stage. The classification results are output based on the set of confidence scores at approximately the first time.

The invention relates to the technical field of data processing, and discloses a real-time data processing analysis method and system of an industrial PLC. The method comprises the following steps: transmitting temperature, pressure, current, vibration and acoustic parameters acquired by multiple sensors to an industrial PLC (Programmable Logic Controller) in real time to obtain multi-source heterogeneous original data; preprocessing the multi-source heterogeneous original data to obtain standardized fusion data; correlation calculation and anomaly recognition are carried out through the multivariate analysis model, and an abnormal state classification result is obtained; dynamically adjusting data interaction frequency and sampling rate between the edge nodes and the central PLC, and generating a real-time control decision instruction; and matching the real-time control decision instruction with the current motor load fluctuation state, and outputting the optimal frequency conversion control parameter. According to the invention, the response delay of the system is reduced, the control precision and reliability are improved, the dynamic balance between the safety and the energy efficiency is realized, and the system can intelligently adjust the control strategy according to the real-time safety situation.

The invention relates to the technical field of image recognition, in particular to an image classification system and method based on the image recognition technology, and the system comprises an image collection module which is used for obtaining original image data to be classified; the preprocessing module is used for carrying out denoising, normalization and size standardization processing on the image; the feature extraction module is used for extracting multi-level features of the image by adopting a deep convolutional neural network; the classification decision module is used for weighting fusion features based on an attention mechanism and outputting a classification result; the output module is used for displaying the classification labels and confidence scores; according to the method, the input quality is optimized by dynamically selecting a preprocessing strategy, the multi-scale representation capability is enhanced by adopting a parallel convolution path and a feature pyramid structure, the robustness of the system is improved by integrating an adversarial sample detection and defense mechanism, and the dynamic scheduling and mixing precision acceleration of computing resources are realized by introducing an edge computing optimization technology. And the operation efficiency is obviously improved on the premise of ensuring the classification precision.

A defect prediction method based on a multi-feature parallel multi-stage neural network (MF-PMSNN), includes: obtaining a trajectory dataset, and preprocessing data of a defect of a workpiece in additive manufacturing (AM); building an MF-PMSNN, and evaluating an output classification result based on evaluation indicators; and performing real-time defect prediction, and deploying a trained MF-PMSNN model to a production environment. The present disclosure combines and effectively matches thermal imaging-based in-situ monitoring data and X-ray computed tomography (XCT)-based in-situ monitoring data to ensure temporal and spatial consistency between the thermal imaging-based in-situ monitoring data and the XCT-based in-situ monitoring data. In this way, a molten pool status and a pore of the workpiece can be captured more comprehensively. The MF-PMSNN is proposed to obtain a molten pool status and the porosity distribution in the data and perform defect prediction.

A method for early diagnosis of Parkinson's disease based on multimodal deep learning is provided. Audio-visual data of a to-be-diagnosed subject while performing a speech task is acquired. The audio-visual data are preprocessed to extract a plurality of audio segments and a plurality of video segments. A face image sequence is extracted from each of the plurality of video segments. A Mel-spectrogram of each of the plurality of audio segments is calculated. The face image sequence and the Mel-spectrogram are input into a multimodal deep learning model to output a classification result for Parkinson's disease early diagnosis of the to-be-diagnosed subject. A system for early diagnosis of Parkinson's disease based on multimodal deep learning is also provided.

The invention relates to a multimedia equipment control method and system based on adaptive protocol matching. The method comprises the following steps: firstly, collecting communication protocol data and extracting features to form a protocol feature data set; secondly, performing confidence coefficient verification on a data set type identification result, and if the confidence coefficient is lower than a threshold value, extracting a depth feature through a convolutional neural network; and calling a national standard protocol adaptation conversion engine based on a classification result, converting the heterogeneous protocol into data in a unified format, and generating a secure communication data frame through national secret algorithm encryption and bidirectional identity verification. And finally, performing hash check and structured analysis on the data frame, constructing an interaction information packet in combination with user behavior characteristics, and dynamically optimizing a control strategy through real-time feedback data to generate a self-adaptive interaction control scheme. According to the method, through a technical closed loop of intelligent analysis-standard conversion-security reinforcement-closed loop optimization, the high efficiency, the intelligent level and the security performance of multimedia equipment management are remarkably improved.

The invention provides an emergency event visualization cooperation method and terminal equipment, and the method comprises the steps: obtaining a historical monitoring data set of an emergency event in a target region, carrying out the preprocessing, and generating a standardized event feature set; training a time sequence prediction model according to the dynamic environment parameters to obtain an event evolution prediction model; and constructing a geographic region matching model based on the static geographic parameters, and generating a visual layer template associated with the target region. And collecting a current environment data flow of the target area in real time, performing classification through the event evolution prediction model to obtain an event classification result, and calling a corresponding visual layer template from the geographic area matching model according to the event classification result. And fusing the diffusion path, the intensity level and the called visual layer template in the event classification result to generate a dynamic visual instruction set, and rendering and outputting an emergency event evolution situation map through an interactive interface. According to the invention, the timeliness of emergency response and decision reliability can be improved.

The invention provides a precise interference method and system for pilot signals of a remote controller of an unmanned aerial vehicle, and the method comprises the steps: receiving and processing an image transmission signal and a remote control signal of the unmanned aerial vehicle, and analyzing a starting time, a transmission period and different types of image transmission protocol classification results of the image transmission signal; extracting the time period of the remote control signal and the occurrence time of a pilot symbol in the remote control signal; generating optimized interference signal parameter configuration by utilizing a reinforcement learning algorithm, analyzing historical data by utilizing a long short-term memory network to obtain a historical analysis result, and predicting an optimal interference opportunity; monitoring a wireless communication environment around the unmanned aerial vehicle by using a real-time spectrum sensing technology so as to identify a potential conflict frequency band, and selecting a target interference frequency band by using a frequency agility technology; determining a precise interference method based on the historical analysis result, the optimal interference opportunity and the target interference frequency band; according to the invention, the interference precision and effect on the pilot signal of the remote controller of the unmanned aerial vehicle are improved.

The invention belongs to the field of intelligent question-answering systems, and relates to a financial intelligent question-answering method and system based on mixed retrieval and dynamic query, and the method comprises the steps: constructing an FAQ library and a financial office database; receiving a user question, calling a historical chat record of the user, and calling the large language model to perform intention recognition; calling a large language model to perform semantic rewriting in combination with the real intention of the user to generate a standardized query statement; obtaining a candidate list from the FAQ library by adopting a mixed retrieval mode; if no candidate questions and answers exist in the candidate list, replying the standardized answers; otherwise, screening candidate questions and answers of Top-N, judging whether the candidate questions and answers of Top-1 are personalized questions or not, if yes, calling a financial office database to generate a query result, and replying according to the query result and the candidate questions and answers of Top-N; according to the method, semantic retrieval and keyword retrieval are mixed through the user intention classification result and the conflict detection rule, the candidate list is obtained according to the mixed retrieval mode, and the retrieval recall rate is increased.

The invention discloses a lithium battery health state assessment method and system, and particularly relates to the technical field of battery health state assessment. The method comprises the following steps: performing time sequence alignment and structured preprocessing on multi-source operation data of a target lithium battery in a plurality of historical work cycles to construct a structured data set; constructing a spatial-temporal characteristic residual error map based on residual error mapping analysis, and extracting a spatial heterogeneity index; in combination with a spatial heterogeneity index, generating regional degradation feature mapping; through high-dimensional feature embedding and evolution path clustering, a heterogeneous aging mode is identified, and a classification result is generated; evaluating the health state grade of the target lithium battery according to the regional degradation characteristic mapping and heterogeneous aging mode classification result; whether the battery has a local potential thermal runaway risk or not is judged based on the evaluation result, and a corresponding risk early warning signal and a safety disposal suggestion are generated, so that the nonlinear influence of the lithium battery aging heterogeneity can be accurately identified, and the health state evaluation precision and the safety risk early warning capability are effectively improved.

The invention discloses a long-tail image recognition method based on multi-modal semantic generation and image-text fusion. The method comprises the following steps: extracting structured semantic description from a tail image; carrying out semantic rewriting and enhancement based on a multi-modal visual language model, and generating an image semantic extension description; the image semantic extension description is optimized based on semantic duplicate judgment and a style alignment mechanism, and an optimized text description set is obtained; inputting the optimized text description set into a text graph model, generating a tail class image sample, performing semantic and visual quality screening, and constructing to obtain an enhanced image set for training; constructing a training data set based on the original long-tail data set and the enhanced image set, and training an image-text fusion classification model; and inputting a to-be-identified image into the trained image-text fusion classification model, and outputting classification results of all categories. According to the method, the discrimination capability in a long-tail distribution scene is enhanced, and the method has a stronger generalization characteristic.

The invention belongs to the field of data processing, and discloses an archive arrangement multi-mode intelligent AI classification system and method. Comprising the following steps: introducing a cross-modal consistency loss function and structured constraint loss into a deployed dual-channel generator, and carrying out facies evaluation on original multi-modal data and evolutionary multi-modal data based on a modal quality evaluator; through conflict identification and digestion mechanism loop optimization, features are extracted based on the enhanced multi-modal data to obtain a quality scoring triple; a PPO algorithm is used for performing dynamic routing decision, quality evaluation closed-loop optimization is introduced, path rules are executed according to decision actions, a multi-modal feature matrix is constructed based on comprehensive features, a multi-modal classifier is designed based on Attention-Transformer, a modal weight adaptive adjustment module is introduced, a classification result verification mechanism is added, and closed-loop optimization is formed; and intelligent AI classification of the multi-modal archives is realized.

The invention relates to the technical field of machine tool casting detection, and discloses an automatic detection system for surface defects of machine tool precision castings. The system comprises a surface information acquisition core module, a first defect identification core module, a second defect identification core module and a defect type fusion core module. The surface information acquisition module is used for synchronously acquiring real-time optical images and process parameter data in production aiming at the surface of the casting part, and constructing a defect diagnosis characteristic spectrum and an auxiliary text according to the real-time optical images and the process parameter data; the first defect recognition module inputs the atlas and the auxiliary text into a pre-training double-flow convolutional neural network to generate a first classification result of defect types; a second defect identification module extracts defect mechanism characteristic values from the atlas and matches the defect mechanism characteristic values with a pre-stored defect mechanism knowledge base to obtain a second classification result; and the defect type fusion module fuses the two types of results to determine a target defect type. The system solves the problems of single detection information and identification deviation in the prior art, improves the detection accuracy and real-time performance, and meets the requirements of different production scenes.

The invention relates to an intelligent vacuum coating quality monitoring method based on artificial intelligence, which comprises the following steps: collecting process parameters in real time through a coating equipment sensor, dynamically monitoring vacuum degree change, deposition rate deviation and temperature gradient, and standardizing data flow to obtain a quantitative trend curve of process parameter fluctuation; combining the defect microcosmic feature description set with gas flow fluctuation and power supply voltage jitter to obtain a correlation analysis result of process parameters and defect forms, and calibrating defect types according to a defect feature library to obtain an updated defect classification basis; and inputting real-time new surface image data through the updated defect classification basis, performing defect detection for illumination reflection differences and stripe deflection angles, obtaining a preliminary defect classification result, and adjusting classification weight parameters.

The invention provides a retrieval enhancement generation method and system based on a large language model. The method comprises the steps that input query information of a user is obtained; performing rewriting and key information extraction on the input query information based on a large language model to obtain a first query instruction and a second query instruction; according to the input query information, identifying a user intention based on a large language model and performing intention classification to obtain an intention classification result; according to the first query instruction and the second query instruction, performing multi-path mixed retrieval through a preset multi-source knowledge base to obtain an initial retrieval result; according to the intention classification result, performing result rearrangement and screening on the initial retrieval result to obtain an enhanced cue word; and inputting the input query information and the enhanced cue word into the large language model to obtain a retrieval enhanced generation result. According to the method, the query demand and intention of the user can be accurately recognized, related information is retrieved in an open domain in a high-precision manner, and the instantaneity, the accuracy, the specialty and the reply quality of information retrieval are improved.

The invention relates to the technical field of text classification, and provides a text classification method based on a large model and a rule engine, and the method comprises the steps: S1, storing multi-level rule classification labels, and constructing a classification rule template library; s2, receiving text data from various data sources, and preprocessing the text data; s3, performing rule matching on the text data based on the classification rule through a rule engine, and outputting a rule classification result; and S4, when any one of the following conditions is met, large language model classification is triggered: a, a classification rule is not matched; b, matching a classification rule, wherein the rule confidence is smaller than a rule confidence threshold; c, the text data length exceeds the preset text data length; d, matching a specific business scene label; outputting a model classification result; and S5, when the rule engine classification in the S3 and the large language model classification in the S4 are parallel, executing the strategy. The output reliability and the service continuity are guaranteed, and the method is suitable for scenes with high accuracy requirements such as financial compliance examination and the like.

The invention discloses a transformer substation scene three-dimensional semantic segmentation method based on point cloud enhancement and attention guidance, and the method specifically comprises the steps: carrying out the region division and category labeling of three-dimensional point cloud data in a transformer substation scene, setting a semantic tag based on an equipment type, and constructing a semantic segmentation point cloud data set adaptive to a power scene; a three-dimensional semantic segmentation model is constructed, and collaborative modeling of local details and global context is realized through a point cloud enhancement module and an attention guidance coding module; a weighted loss function and a parameter optimizer training model are adopted to solve the problem of unbalanced equipment category distribution in a substation scene; and performing semantic segmentation on the substation scene point cloud data based on the trained model, and outputting a high-precision equipment classification result. According to the method provided by the invention, through collaborative design of point cloud enhancement and attention guidance, the accuracy and robustness of three-dimensional semantic segmentation in a complex scene are effectively improved, and the method is particularly suitable for a scene with dense substation equipment and a fine structure.

The invention discloses an unmanned car washer stain panorama identification system. The system operation process specifically comprises the following steps: acquiring panorama image data of a target car; preprocessing the panoramic image data to obtain a standardized panoramic image set; performing stain area identification on the standardized panoramic image set based on a deep learning model to generate an initial stain distribution diagram; performing stain type classification on the initial stain distribution diagram according to a stain feature database to generate a stain classification result set; generating a dynamic cleaning path instruction set based on the stain classification result set and a cleaning strategy library; real-time images in the cleaning process are collected in real time, real-time stain residue analysis is conducted, and finally a cleaning effect feedback report is generated. The method has the following advantages and effects that the system of multi-dimensional stain feature recognition, classification and dynamic decision can be fused, so that the core contradiction that the cleaning strategy is not matched with the stain features in the prior art is solved.

The invention provides an interactive data query and display method and system based on a large language model. The method comprises the steps that a user query request and a service database used for query are obtained; based on the user query request, obtaining a query intention classification result of the user query request and a key entity in the user query request; generating a structured query parameter based on the obtained query intention classification result and the key entity; based on the generated structured query parameters, through interactive processing, rewriting a user query request to conform to a standard query grammar; constructing a query cue word based on the rewritten user query, and generating an SQL statement through a pre-trained large language model based on the constructed query cue word; obtaining a data query result from a business database for query based on the generated SQL statement; based on the query result, a chart is automatically generated to display the query result, and the chart type is determined according to the chart display type of the historical query result of the user.

The invention discloses a text classification method, device and system based on a large language model, a storage medium and a product, and relates to the technical field of text classification, the method comprises the following steps: receiving a to-be-classified text and a classification label input or selected by a user through a user interaction interface of the text classification system; and according to the to-be-classified text and the classification label, constructing prompt words, and calling a pre-trained large language model to perform different-level classification on the to-be-classified text to generate a classification result. According to the technical scheme, the low-cost and low-use-threshold text classification method is provided based on the large language model, classification of different levels can be achieved, in addition, due to the fact that the text classification result is in a light-weight and convenient-to-analyze JSON format, the system can visually display the classification result to a user and can also provide an interface for a third-party application to call, and the user experience is improved. And the application scene of text classification is expanded.

The invention relates to the technical field of target detection, in particular to a photovoltaic panel detection method and system based on machine vision, and the method comprises the following steps: obtaining the displacement data of a photovoltaic panel cell, calculating the displacement gradient of a connection point, dividing a rigid region and a non-rigid region, generating a deformation classification result, and constructing a global deformation field. Generating a time sequence deformation trend matrix; calculating a displacement mean value of a bracket fixing point; adjusting a deformation compensation parameter; generating an adaptive compensation parameter, adjusting a welding point position, calculating an abnormal region Euclidean distance, and generating a defect matching identification result. According to the method, rigid and non-rigid areas are accurately divided by calculating the displacement gradient of a connection point, a global deformation field is constructed in combination with the displacement change of a feature point, trend continuity is ensured, a compensation parameter is adjusted by utilizing a fixed point displacement mean value, a deformation increment factor is calculated, the weight and stability are corrected, the position of a welding point is adjusted, and inverse deformation correction is optimized. The Euclidean distance is combined to match defects, so that the detection precision and stability are improved.

The invention relates to a remote sensing image ground feature recognition method based on deep learning, and the method comprises the steps: carrying out the data collection and preprocessing of a ground surface target region, eliminating the position deviation through geometric correction, processing the illumination difference through combination with radiation equalization, and generating a ground feature registration image; performing multi-dimensional feature fusion processing on the image, performing tensor fusion on vegetation spectral features, earth surface texture features, point cloud data features and linear ground feature features, and constructing a ground feature fusion matrix; a bilateral convolutional neural network is adopted to extract spectral response characteristics and spatial correlation characteristics, and characteristic interaction is realized through an attention mechanism to generate a ground feature probability distribution diagram; and finally, carrying out noise filtering, boundary refining and vectorization conversion processing on a classification result, and outputting a ground feature classification vector diagram. According to the method, three technical bottlenecks of insufficient cooperative utilization of multi-source heterogeneous data, insufficient spectrum-space feature fusion and poor GIS compatibility are solved, and the operation efficiency of territorial investigation, disaster monitoring and other scenes can be remarkably improved.

The invention provides an intelligent identification method and system for an asymmetric plate shape defect, and the method comprises the steps: collecting the multi-modal image data of a to-be-detected plate shape surface, and carrying out the preprocessing of the multi-modal image data, and obtaining a standardized image; performing feature extraction on the standardized image based on an asymmetric feature enhancement algorithm to obtain an asymmetric feature vector; inputting the asymmetric feature vector into a pre-trained asymmetric defect identification model to generate a preliminary defect classification result and a defect area thermodynamic diagram; according to the thermodynamic diagram of the defect region, segmenting a defect boundary in combination with a geometric constraint optimization algorithm, and determining morphological parameters and spatial positions of asymmetric defects; and based on the morphological parameters and the spatial positions, correcting the preliminary defect classification result through a dynamic threshold adjustment algorithm to obtain an identification result, thereby alleviating the technical problem of low accuracy of asymmetric plate shape defect identification in the prior art.

The invention relates to an industrial equipment maintenance intelligent question-answering system based on multi-agent collaboration. Wherein the input unit is used for receiving text, voice, image or equipment scanning and other multi-mode user input information and analyzing the information into structured problem information; the scheduling unit performs semantic understanding and problem classification on the structured problem information based on the fine-tuned cross-language pre-training language model and an incremental training mechanism; the processing unit calls a corresponding domain agent according to the classification result, and generates an intelligent question and answer processing result including predictive maintenance suggestions, structured reply content and semantic annotation information; and the fusion unit fuses the local knowledge base, the graph database and the networking retrieval information, performs multi-hop semantic reasoning on the intelligent question and answer processing result, and generates multi-modal reply information including text description, image screenshots, prediction curves and recommendation links. The system can support multi-language and multi-mode intelligent question answering and predictive maintenance in a complex industrial maintenance scene.