158 results about "Feature model" patented technology

The invention relates to the technical field of robots, and discloses a robot nondestructive testing method and system based on artificial intelligence, and the method comprises the steps: obtaining multi-modal sensing data and motion state parameters; constructing a three-dimensional material feature model and analyzing to generate an initial defect feature evaluation model; loading material attributes and detection precision constraints to generate a comprehensive model, and combining motion parameter simulation to obtain path regulation and control data; training the dynamic defect prediction model to obtain a detection path correction model and generating initial detection parameters; obtaining dynamic correction simulation information based on the initial parameters and the like; constructing a multi-objective optimization model, and optimizing to obtain an optimal detection path parameter; and generating a safety detection probability by combining real-time model reasoning, and adjusting a detection strategy to realize dynamic path matching. The system comprises a data acquisition module, a feature modeling and analysis module, a comprehensive evaluation and simulation module and the like. The method improves the detection precision, efficiency and adaptability, and is suitable for nondestructive detection of complex targets.

The invention discloses a PLC controller fault detection system, and relates to the technical field of industrial control equipment fault detection.The system is characterized in that PLC operation environment data is acquired through a multi-physical-quantity holographic acquisition module, and after the PLC operation environment data is cleaned and subjected to feature fusion through a data preprocessing module, a fault model is constructed through a multi-physical-quantity fusion model module; the self-adaptive threshold value judgment module dynamically calculates and judges a threshold value and evaluates a state; the fault traceability analysis module constructs a propagation path diagram based on a model and historical cases, and realizes accurate traceability of a fault source and a propagation process; according to the invention, multi-physical-quantity holographic acquisition and feature fusion algorithms are integrated, multi-dimensional parameters are monitored synchronously, a comprehensive feature model is constructed, the fault identification precision is improved, and misjudgment is avoided; dynamic optimization is achieved through self-adaptive threshold judgment, the early warning accuracy is improved, meanwhile, accurate traceability is achieved through an element fault propagation algorithm, a path is optimized in combination with historical cases, the downtime is shortened through full-process intelligent support, and the maintenance cost is reduced.

The invention discloses a workflow-based multi-modal water conservancy large model decision support method, which comprises the following steps of: S1, preprocessing multi-modal input data based on a local model, and converting a data format; s2, constructing a modular workflow supporting parallel processing through a specified DSL (Digital Subscriber Line) framework; s3, constructing a feature model suitable for the water conservancy field through the multi-modal data subjected to synchronous acquisition and processing, and performing field adaptation and fine adjustment on the feature model; s4, on the basis of the feature model subjected to domain adaptation and fine adjustment, a specified workflow is constructed, and multi-modal task resources are dynamically scheduled; and S5, combining a local knowledge base with networking data by using a dynamic fusion algorithm to realize dual-channel knowledge fusion for executing the multi-modal task in the S4 in parallel. According to the invention, real-time acquisition and parallel processing of multi-modal data can be realized, and the stability of a local knowledge base and the real-time performance of networking data are dynamically balanced through a dual-channel knowledge fusion mechanism.

The invention discloses an H-bridge key equipment service life and system reliability evaluation method and system for a cascade network construction type energy storage system, and belongs to the technical field of power system automation. The method comprises the following steps: firstly, extracting task profile parameters under multiple time scales, and constructing a time sequence feature model; secondly, estimating a hot spot temperature sequence of the IGBT device and the capacitor based on a multilayer feedforward neural network; then, in combination with a continuous extreme point paired temperature cycle extraction method and a Miner linear cumulative damage criterion, the damage factor and the residual life of the device are evaluated; then, task profile samples are expanded based on a generative adversarial network with gradient penalty, and life distribution and reliability indexes of key devices under different profiles are calculated; and finally, based on H-bridge series structure mapping device level information, constructing a system level reliability model, obtaining system failure rate, average fault-free operation time and a reliability function, and realizing health state perception and reliability quantitative evaluation of the energy storage system.

The invention discloses a commodity evaluation monitoring method and system based on multi-dimensional data, and relates to the technical field of evaluation monitoring, and the method comprises the steps: collecting the multi-dimensional data, and carrying out the preprocessing of the multi-dimensional data; performing feature extraction on the preprocessed multi-dimensional data, and fusing the extracted features into a comprehensive feature vector; based on the comprehensive feature vector, through a space-time convolutional neural network, identifying and evaluating the space-time distribution of the hot spots; based on the spatial and temporal distribution of the evaluation hotspots, identifying areas and time periods of the evaluation hotspots, and evaluating the trend of the hotspots according to an identification result; and performing dynamic early warning according to the trend of predicting and evaluating the hot spots, and generating an evaluation report. According to the method, spatial-temporal feature modeling is carried out on data through STCNN, and the distribution conditions of evaluation hotspots in different time and spaces are identified. Based on the spatio-temporal characteristics, the area and time period of the hot spot can be accurately positioned and evaluated, and the diffusion trend of the hot spot can be predicted and evaluated.

The invention relates to the technical field of deep learning and computer vision, and discloses a histogram interactive attention-based pavement crack segmentation network processing method and system, so as to enhance the edge detail fidelity and improve the crack segmentation precision. The method comprises the steps of image preprocessing, up-sampling, down-sampling, feature fusion and image reconstruction processing. Wherein global feature modeling in the intensity sub-boxes and among the sub-boxes is realized by constructing a histogram interactive attention module (HIA); a double-branch detail enhancement feedforward module (DDEF) is introduced to enhance spatial detail and high-frequency edge information expression; meanwhile, a Fourier jump enhancement module (FFSM) is adopted to jointly refine jump connection features in a spatial domain and a frequency domain. Through the synergistic effect of the modules, the network can realize continuous recovery and structural consistency modeling of a crack boundary in a complex pavement environment, so that the accuracy and the stability of a segmentation result are remarkably improved.

The invention discloses a modularized extensible Internet of Things multi-scene dynamic regulation and control system and a regulation and control method, and belongs to the field of intelligent regulation and control. The invention discloses a modularized extensible Internet of Things multi-scene dynamic regulation and control system and method. The system comprises a data acquisition unit, a cloud analysis unit and a regulation and control execution unit. According to the invention, the problem of single scene in the prior art is solved, the efficiency and reliability of data transmission are ensured by collecting multi-scene data of the Internet of Things in real time and automatically selecting the optimal communication protocol and link according to the network condition, cloud analysis is combined, and according to the preset scene feature model, the data is subjected to multi-scene data transmission. According to the invention, the scene type can be automatically identified and the corresponding regulation and control rule can be called, the accurate and real-time regulation and control of the Internet of Things equipment can be realized according to the regulation and control rule, the operation efficiency of the equipment and the overall performance of the system can be effectively improved, the regulation and control process can be evaluated and analyzed through feedback information, the regulation and control rule can be automatically adjusted, and a continuously optimized closed-loop control system can be formed.

The invention provides an AGC hydropower station intelligent control method based on multi-source data fusion. Constructing a control feature vector of the multi-dimensional feature; performing spatial-temporal feature modeling on the control feature vector, and extracting a time sequence dependency relationship between power grid load change and hydraulic dynamic response and a spatial coupling effect between units; establishing a multi-objective optimization function, and dynamically adjusting the weight coefficient of each objective through fuzzy logic according to the current working condition; a self-adaptive differential evolution algorithm is adopted to carry out on-line optimization on an active power distribution coefficient of a unit and PID parameters of a speed regulator, and the requirements of guide vane opening change rate constraint and water hammer effect avoidance are met. According to the method, multi-source heterogeneous data such as power grid, hydraulic engineering and equipment states can be effectively fused, multi-target dynamic optimization control is realized through the space-time attention model and the adaptive differential evolution algorithm, and the control precision, the response speed and the equipment operation safety of the hydropower station AGC system are remarkably improved.

An AGV robot vision acquisition and avoidance control method based on deep thinking relates to the field of non-electrical variable control or adjustment systems, and comprises the following steps: acquiring real-time motion parameters of an automatic guided vehicle, calculating a key path point set and constructing feature vectors; collecting a visual image and depth data, extracting obstacle features through grid division, and generating a three-dimensional feature model; mapping the path points to a three-dimensional model to generate a dynamic scene sequence, and calculating a distance value with a dynamic target to determine a risk level; analyzing the scene by using a large language model, generating an obstacle avoidance scheme containing an obstacle avoidance track, speed and steering, and storing the obstacle avoidance scheme in a scheme library; current scene features are obtained, candidate schemes are selected through similarity matching, and the optimal scheme is selected for execution through safety scoring. By implementing the method, the obstacle avoidance success rate of the automated guided vehicle can be improved.

The invention provides a writing track optimization system based on machine learning, and relates to the technical field of machine learning and artificial intelligence, and the system comprises a data collection module which synchronously collects inertial measurement and visual track data, and generates standardized writing data through dynamic alignment and coordinate unification; the feature extraction and fusion module is used for extracting spatio-temporal features and fusing multi-source features through a cross-modal attention mechanism; the model building module is used for building a double-branch multi-task collaborative model, classifying branches to predict character categories, regression branches to generate a space-time coordinate sequence, and designing a multi-objective optimization function fusing cross entropy loss, trajectory distance error and shape constraint loss; and the trajectory generation and recognition module is based on a two-stage training strategy of dynamic division, optimizes the overall shape similarity of the trajectory in the first training stage, enhances the distance precision of trajectory points in the second training stage, and realizes high-precision trajectory reconstruction and character recognition. According to the method, the problem of writing jitter is effectively solved, and the track smoothness and the recognition robustness are improved.

The invention relates to the field of mechanical design and manufacturing, and provides a three-dimensional model processing method, a training and application method of a system, equipment and a medium, and a target boundary representation three-dimensional model is obtained based on a preset complex manufacturing feature model and a boundary representation three-dimensional model. The complex manufacturing feature model is added to the boundary representation three-dimensional model containing the simple structure manufacturing feature model, so that the boundary representation three-dimensional model containing the simple structure manufacturing feature model and the complex structure manufacturing feature model is synthesized. According to the three-dimensional model processing scheme provided by the invention, the complex manufacturing feature model is added in the boundary representation three-dimensional model, so that when the boundary representation three-dimensional model obtained by the processing scheme is used for training an image classification system, the image classification efficiency is improved; the generalization ability of the image classification system and the classification accuracy of the manufacturing feature model in the boundary representation three-dimensional model can be improved, so that the production efficiency and quality of process products are improved.

The invention relates to the technical field of digital employees, and discloses a digital employee creation method and system based on artificial intelligence, and the method comprises the steps: obtaining and analyzing post data, generating a digital employee portrait and a dynamic knowledge base for creating digital employees based on the post data, and creating the digital employees based on the digital employee portrait and the dynamic knowledge base. The system corresponds to the method. According to the method, the digital employee portrait including the skill map and the character feature model is generated through the post data, and the digital employee can dynamically adjust the skill weight based on the parameters in the character feature model by combining real-time updating of the dynamic knowledge base, so that deep cooperation of the skill and the character is realized; when a digital employee processes an instruction, the digital employee accurately calls a core skill and an associated skill, adapts to a personalized scene through character features, and meanwhile, depends on dynamic knowledge updating, the autonomous response and sustainable evolution ability of the digital employee to a complex business scene is enhanced, and the intelligent level and the actual business adaptation degree of the digital employee are effectively improved.

The invention provides an itinerary generation method and system based on a large language model. The itinerary generation method comprises the steps of obtaining a user social portrait, wherein the user social portrait is fused with at least one of historical activity information, geographic position information and social information of a user; performing multi-dimensional feature modeling on the interest points in the area where the user is located to obtain interest point embedding; the social portraits and the interest points of the users are embedded and converted into natural language descriptions, and the natural language descriptions and real-time context information are fused to generate context prompts; reasoning an activity intention of the user based on context prompt through a large language model, generating a natural language travel description, and then generating a structured activity plan by using the natural language travel description; according to the structured activity plan, candidate interest points are matched in a preset range in an area where the user is located, and recommended interest points are determined in the candidate interest points; and performing path optimization on the recommended interest points to generate a trip.

The invention discloses a method for improving code writing efficiency. The method comprises the following steps: S1, acquiring and preprocessing a user programming instruction sequence; s2, constructing a behavior feature model, and generating a semantic map sequence in combination with big data analysis; s3, analyzing a user code and a project structure, and extracting context features of the code; s4, based on the semantic map sequence and code context features, using Transform to predict a current position code sequence under grammar constraints, and generating candidate segments; s5, grading and sorting the candidate segments according to the context structure matching degree, the semantic similarity and the confidence coefficient of the behavior feature model; s6, presenting the candidate code snippet with the highest score to the user and recording the response behavior of the user; and S7, storing the candidate code snippets and user feedback data, and updating model parameters and generation logic in real time. According to the method, through behavior-driven modeling, context fusion prediction and a multi-factor dynamic scoring mechanism, the code writing efficiency and quality are remarkably improved.

The invention discloses an AR 6DoF tracking and large space identification method based on a WeChat applet, and the method comprises the following steps: S1, obtaining the position and attitude data of equipment, and constructing a data set; s2, constructing a virtual-real space coordinate system alignment standard, and preliminarily aligning a virtual object and a real scene; s3, point cloud data are collected and processed, PointNet + + is used for denoising and simplification, and a three-dimensional space feature model is constructed; s4, in combination with ICP and a global optimization alignment algorithm, realizing accurate docking of a virtual object and a real scene; s5, updating the data of the dynamic sensor, and keeping the real-time stability of the virtual object and the real scene; and S6, based on the updated alignment relationship, virtual object space repositioning is carried out, and synchronization is continued. According to the method, the ICP algorithm and the least square method global optimization alignment algorithm are combined, accurate alignment of virtual and real space coordinate systems is achieved, and the purpose of large space recognition is achieved.

The invention discloses a shore power load prediction method and system based on a self-attention mechanism, and the method comprises the steps: building a high-precision load prediction model through combining a historical load time sequence, an environment feature and a ship feature as input features, and employing a multi-head self-attention mechanism of a Transform model; multi-modal feature representation is generated through feature embedding and position coding, a global dependency relationship between a time sequence and multi-modal features is captured by using a multi-head self-attention mechanism, deep features are further extracted in combination with a feedforward neural network, and an accurate load prediction result can be generated after the model is subjected to multiple rounds of training and verification. According to the method, the load prediction precision, the prediction efficiency and the robustness and adaptability of the model are improved, and energy optimization of a green port is supported.

The invention relates to the technical field of computers, and discloses a test case generation method, device and equipment and a storage medium, and the method comprises the steps: constructing a scene feature model based on the multi-source sensing data of a vehicle collected in a to-be-tested target scene, and matching a target source scene from a pre-constructed test template library based on the scene feature model; mapping test configuration information corresponding to the target source scene to the target scene to generate an initial test template adaptive to the target scene; and generating a target test case in combination with the scene feature model, the initial test template, the multi-modal interaction rule base and the target test case evaluation rule. The scene coverage rate, the scene adaptability and the test efficiency of the intelligent cabin test can be improved.

The invention discloses a frequency converter fault prediction method based on artificial intelligence. The method comprises the following steps: firstly, acquiring sensing data sets corresponding to a plurality of parameters of the frequency converter respectively, selecting a target parameter from the plurality of parameters, and further acquiring a plurality of target data sets corresponding to the target parameter in historical data; secondly, constructing and training an integrated learning model, calculating the correlation degree between any two data sets in each target data set of the target parameters through the model, and combining the correlation degree into a correlation array of the target data sets; then, analyzing and processing the correlation array corresponding to each target data set, calculating to obtain a plurality of representative vectors of each target data set, and constructing a spatial feature model based on the representative vectors; and finally, collecting a real-time data set of the target parameter in the current time period, calculating to obtain a real-time representative vector, and inputting the real-time representative vector into the spatial feature model to obtain a fault prediction result. The method can effectively improve the accuracy and efficiency of frequency converter fault prediction.

The invention discloses an evaluation method for a multi-modal large model, and the method comprises the following steps: S1, data collection: collecting multi-modal data from a multi-modal data source, including images, voices and / or texts, the data of each modal including features related to the performance of the model; s2, model training: training a multi-modal large model by using the collected multi-modal data; s3, performance evaluation: performing performance evaluation on the trained multi-modal large model; s4, model optimization: optimizing the multi-modal large model according to a performance evaluation result; and S5, result analysis and suggestion: carrying out result analysis on the optimized multi-modal large model, and providing a corresponding optimization suggestion. The scheme has the advantages of less manual participation, high efficiency and high accuracy, and is suitable for the evaluation requirements of various multi-modal large models.

An AI-powered system for automated, genre-specific, and context-dependent video summarization, consisting of: an input module configured to receive and store healthcare video data comprising a plurality of frames, the input module comprising a storage module configured to store the input video; a centralized processing unit comprising a plurality of modules implemented by an AI-assisted processor, a memory, and a graphics processing unit, wherein the memory stores instructions executed by the processor and the graphics processing unit, the centralized processing module comprising: a data preprocessing module configured to receive an input video having a plurality of frames and normalize the plurality of frames to generate preprocessed frames; a genre-specific complexity calculation module configured to quantify the perceptual complexity of each preprocessed frame using genre-specific metrics to generate frame-by-frame complexity values; a subshot-wise analysis module configured to aggregate the frame-wise complexity values using a sliding window approach to generate subshot complexity vectors; a time complexity graph module configured to model temporal relationships between sub-shots by creating a graph in which sub-shots are represented as nodes and temporal similarities between sub-shots are represented as edges; an adaptive thresholding module configured to dynamically adjust complexity thresholds based on local and global trends in the subshot complexity vectors; a complexity-based model selection module configured to classify video frames as complex or non-complex based on dynamically adjusted complexity thresholds and pass each frame to an appropriate neural network architecture; a spatial feature extraction module connected to the complexity-based model selection module and configured to extract spatial features from each frame using the corresponding neural network architecture and apply spatial pyramid pooling to the extracted features; a temporal feature modeling module configured to process the spatial features to capture temporal dependencies in the video; and a video summary generation module configured to predict frame importance values, select frames using diversity-aware optimization, and generate a summarized video while maintaining procedural accuracy and relevance; and an output module connected to the central processing unit and configured to display the aggregated video via a user interface, wherein the user interface also facilitates uploading of the input video.

The present disclosure extracts contextual product feature model from requirement specification documents where the conventional methods fail to perform. Initially, the system receives a plurality of requirement specification documents pertaining to a product, a domain dictionary, a plurality of configuration parameters, and a plurality of extraction patterns. A product feature model is generated using a NLP based feature extraction technique. The product feature model includes a plurality of product feature elements comprising a feature area, a major feature and a plurality of features arranged hierarchically and classified into feature types. Further, a plurality of ContextType associations like core, client, geography and market are extracted for each of the plurality of features using a ContextType extraction technique. Finally, the plurality of ContextType associations is updated in the product feature model to obtain a contextual product feature model. Various types of feature exports can be generated using a natural language interface.

The invention discloses an automatic optimization system for an administrative approval process, which relates to the field of administrative approval and comprises a twin modeling module, a simulation prediction module, a scheme generation module, a feature modeling module, a dynamic routing module, a semantic analysis module, a change analysis module, a behavior intervention module, a fusion inference module and a decision support module. According to the method, a real process structure and an operation rule can be accurately copied, bottleneck nodes can be identified in different pressure scenes, intervention measures are ensured to consider efficiency, throughput and resource utilization rate, task allocation is efficient and fair by dynamically generating a routing instruction, the response speed and accuracy of approval to regulation updating are improved, and the method is suitable for popularization and application. Quantitative policy implementation effect prediction is provided for decision makers, the scheme is ensured to have transparency and auditing performance, and manual review and risk control are facilitated.

The invention relates to the technical field of big data analysis and user behavior modeling, and provides a user portrait generation method and system based on big data. The method comprises the following steps: extracting a time sequence and interaction frequency characteristics from user behavior data, constructing a user access mode characteristic set and analyzing data missing distribution; boundary features are extracted based on missing distribution, and if the correlation between the boundary features and complete data is significant, surrounding data are detected to identify behavior rules; extracting a potential association path of the missing area, and deducing a user behavior pattern prototype; spreading the behavior rule to the missing region, and generating a data inference value; fusing the path and the boundary features, determining the internal structure of the missing region, and generating a user activity sequence; and verifying the consistency of the internal structure and the inferred value to form a complete user feature model. The missing data is inferred through the context, and the accuracy and integrity of the user portrait are improved.

The invention belongs to the technical field of intelligent education and personalized learning, and provides a personalized learning path recommendation method based on a course knowledge point sequence. The method aims at solving the problems that existing learning path recommendation logic association is insufficient, the personalized matching degree is low, and the optimization efficiency is not high. The method comprises the following steps: constructing a knowledge point sequence model by adopting a topological sorting algorithm on the basis of a curriculum knowledge point pre-modification and post-modification relationship, a difficulty level and dependence intensity, and generating a structured knowledge relationship graph; a learner feature model and a knowledge resource model are established, learner features are described from the dimensions of learning targets, cognitive levels, learning styles and the like, and the matching degree with knowledge points is calculated; and constructing an optimization model taking learning cost minimization, mastering degree maximization and path continuity optimization as targets, performing iterative optimization by using an improved intelligent optimization algorithm, and generating a learning path with logic rationality and personalized adaptability. According to the method, the scientificity and learning efficiency of learning path recommendation can be effectively improved.

The application discloses a design method, device and equipment for rotary body part turning processing and a storage medium. The processing feature model is established based on the processing features of an aviation rotary body part and in units of a part family, and then a part processing database is formed, so that the features of various aviation rotary body parts are effectively summarized and integrated, a standardized and parameterized turning processing program can be quickly formed, the programming quality is improved, the automation of aviation rotary body part programming is realized, and then the production cycle is shortened and the production efficiency is improved. Geometric feature analysis is performed on a part to be designed, part family classification, part size and part material are obtained, a corresponding processing feature model is selected in the part processing database according to the part family classification, and the feature elements of the part to be designed are input, so that a parameterized primary turning processing program can be directly obtained, and the programming efficiency is effectively improved. Furthermore, simulation verification is performed, and the accuracy of the turning processing program is ensured.

The application provides a data acquisition method and system based on flexible design of structured table, comprising: obtaining table data; establishing a structured table feature model according to the table data; constructing a table knowledge database, and calling table related attributes from the table knowledge database to perform corresponding table flexible design and obtain a corresponding feature matrix; aggregating the corresponding feature matrix according to the structured table feature model to generate a multifunctional structured table; and establishing an associated mapping relationship between a production process and the structured table, integrating the multifunctional structured table with a data terminal for application, and realizing acquisition of structured table data. The application establishes a multifunctional structured table model and performs flexible design, realizes software and hardware integrated application with a diversified data acquisition terminal, and effectively improves data structured, integrated acquisition and management capability in a satellite development process.

The invention provides a multi-modal feature splicing method and a multi-modal data processing method based on a rotation position coding technology, and relates to the technical field of artificial intelligence such as rotation position coding, multi-modal feature splicing and large language model training. The method comprises the following steps: acquiring to-be-coded data of different modes from different data acquisition channels; performing feature modeling on the features of the to-be-coded data of each mode in the time dimension in a low-frequency band by using a preset rotation position coding technology to obtain coded low-frequency features; performing feature modeling on the features of the to-be-coded data of each mode in the spatial dimension in a middle-high frequency band in a mutual crossing manner by using a rotation position coding technology to obtain coded middle-high frequency features; and splicing the coded low-frequency features and the coded medium-high-frequency features into multi-modal combined features. According to the scheme, long-period component modeling time is used, the long sequence dependence capturing capability is enhanced, meanwhile, the remote attenuation problem can be relieved, the long dimension and the wide dimension share the same frequency range, the problem of spatial asymmetry caused by the fact that the long dimension and the wide dimension are concentrated in different frequency bands for modeling can be avoided, and symmetry is guaranteed.

The application discloses an automatic auxiliary programming system for die electrode processing, and belongs to the technical field of computer-aided manufacturing; the automatic auxiliary programming system comprises an electrode feature recognition module, a processing scheme prediction module, a scheme optimization module, a simulation verification module and a post-processing module; the electrode feature recognition module is used for acquiring a three-dimensional model of a die, and obtaining electrode feature parameters and a feature model of a plurality of electrodes to be processed according to the three-dimensional model; the processing scheme prediction module is used for obtaining an initial processing scheme according to the electrode feature parameters; wherein the processing scheme comprises a processing cutting point, a processing track and processing parameters; the scheme optimization module is used for optimizing the initial processing scheme according to a preset historical case database to obtain an optimized processing scheme; the simulation verification module is used for simulating and verifying the optimized processing scheme, and obtaining a target processing scheme according to a verification result; and the post-processing module is used for converting the target processing scheme into a code executable by a numerical control machine tool.

The application discloses a kind of train vehicle fault image intelligent analysis methods based on lightweight deep learning technology, comprising: obtaining train image data, then the train fault component to be detected in image is labeled, then the overall data is divided into training set, test set two parts;Data enhancement operation is carried out, the hyperparameter required by algorithm is configured, and the training sample is input into the MobileDetectNet neural network model formed after improvement to carry out feature learning;The learned feature model is input to the new vehicle image to identify faults, and finally the area of the fault occurs is framed and the corresponding alarm information is output.The MobileDetectNet network intelligent identification model of the application has few parameters, occupies less memory and has high recognition rate, can improve the recognition rate of EMU fault detection under the original equipment CPU environment, reduce the work intensity of artificial, shorten the maintenance operation time, reduce the missed detection probability, so as to ensure the safe operation of EMU.