Model migration method and device for edge autonomous diagnosis and medium

Abstract

The invention discloses a model migration method and device for edge autonomous diagnosis and a medium, and relates to the technical field of autonomous navigation. The method comprises the following steps: scanning a navigation area through a laser radar to construct an environment map, and determining the position of navigation equipment; acquiring a target point, selecting an optimal initial point according to the position of the navigation equipment, and calculating a first path according to the optimal initial point and the target point; detecting an obstacle in real time, and performing local path planning on the first path to obtain a second path; acquiring noise data of the navigation equipment in the second path, the optimal initial point and the target point through a high-speed acquisition tool; performing enhancement, Fourier transform and splicing on the noise data to obtain a color picture, and diagnosing the color picture to construct a first noise model; and migrating the first noise model to obtain a second noise model. According to the method, the unmanned vehicle-mounted equipment can autonomously monitor the equipment fault, and meanwhile, the convenience of fault diagnosis of the mechanical equipment is improved.

Description

Technical Field

[0001] This application relates to the field of autonomous navigation technology, and particularly to a model migration method, device, and medium for edge autonomous diagnosis. Background Art

[0002] An autonomous navigation vehicle can integrate sensor devices such as sensors and lidar, as well as positioning systems such as GPS or inertial navigation systems, to perceive the surrounding environment in real time, avoid collisions, and drive autonomously according to a pre-set route.

[0003] However, in a complex environment, such as when there are obstacles blocking or in a dynamic environment, the autonomous navigation vehicle may be affected in terms of collisions with other vehicles or pedestrians, and the navigation accuracy and stability of the vehicle. Even if a certain amount of data is collected, existing methods may lack effective ways to extract key features in the fault signals when processing and analyzing this data, or may not be able to accurately judge the difference between fault signals and normal signals. Moreover, existing fault diagnosis methods often only focus on local or single fault signals, lacking comprehensive consideration of global fault signals, which affects the accuracy and comprehensiveness of the diagnosis results.

[0004] Through the above analysis, the problems and defects of the existing technology are as follows:

[0005] In the existing technology, the autonomous navigation device in a complex environment collects fault data one-sidedly, cannot extract the key features in the fault signals, and only focuses on local or single fault signals, lacking comprehensive consideration of global fault signals. Summary of the Invention

[0006] The embodiments of this application provide a model migration method, device, and medium for edge autonomous diagnosis, which can solve the problems that the autonomous navigation device in the existing technology collects fault data one-sidedly in a complex environment, cannot extract the key features in the fault signals, and only focuses on local or single fault signals, lacking comprehensive consideration of global fault signals.

[0007] In a first aspect, an embodiment of the present application provides a model migration method for edge autonomous diagnosis. The method includes: scanning a navigation area with a lidar to construct an environmental map and determine the position of a navigation device, where the lidar and camera are located on the navigation device; obtaining a target point, selecting an optimal initial point based on the position of the navigation device, and calculating a first path based on the optimal initial point and the target point; detecting obstacles in real time and performing local path planning on the first path to obtain a second path; obtaining noise data of the navigation device at the second path, the optimal initial point, and the target point through a high-speed acquisition tool; strengthening, performing Fourier transform, and splicing the noise data to obtain a color picture, and diagnosing the color picture to construct a first noise model; migrating the first noise model to obtain a second noise model to reduce the number of model parameters.

[0008] In an implementation manner of the present application, scanning the navigation area with a lidar to construct an environmental map and determine the position of the navigation device specifically includes: during the movement of the navigation device, scanning with a lidar to collect point cloud data of the navigation area, and shooting a video with a camera to collect the video of the navigation area; performing feature clustering division on the point cloud data and combining it with the video to construct an environmental map; sending and receiving point cloud data of multiple navigation devices through ROS communication; matching the point cloud data with the environmental map to obtain the pose and position of the navigation device.

[0009] In an implementation manner of the present application, strengthening, performing Fourier transform, and splicing the noise data to obtain a color picture and diagnosing the color picture to construct a noise diagnosis model specifically includes: performing filtering processing and time-domain segmentation on the noise data to obtain a basic data segment; performing a fast Fourier transform on the basic data segment to convert the basic data segment from the time domain to the frequency domain to obtain a frequency-domain data segment; dividing the frequency-domain data segment into frequency-domain signals in three frequency bands, low, medium, and high, according to a preset frequency-domain interval; extracting component information of the frequency-domain signals and performing zero-padding in the frequency domain; performing an inverse Fourier transform on the frequency-domain signals after zero-padding in the frequency domain to convert the frequency-domain signals from the frequency domain to the time domain to obtain a first time-domain data segment.

[0010] In an implementation manner of the present application, after performing an inverse Fourier transform on the frequency-domain signals after zero-padding in the frequency domain to convert the frequency-domain signals from the frequency domain to the time domain to obtain a first time-domain data segment, the method further includes: segmenting the first time-domain data segment into segments one by one in a one-dimensional time series according to a preset length to obtain a second time-domain data segment for rearranging the spatial dimension of the time-domain data segment; splicing the second time-domain data segment into a two-dimensional matrix, normalizing the numerical values of the two-dimensional matrix of the second time-domain data segment, and splicing it into a color picture in the spatial dimension; inputting the color picture into a network model and collecting the historical fault types of the navigation device; matching the color picture with the historical fault types to obtain a noise diagnosis model.

[0011] In one implementation of the present application, the noise model is migrated to reduce the number of model parameters, which specifically includes: obtaining training data, using the first noise model for prediction to obtain a first prediction result, where the training data includes noise data and fault types; reducing a preset number of layers, convolutional kernels, and neurons in the first noise model to obtain a second noise diagnosis model; using the second noise diagnosis model to predict the training data to obtain a second prediction result.

[0012] In one implementation of the present application, after using the second noise diagnosis model to predict the training data to obtain a second prediction result, the method further includes: calculating the cross-entropy loss between the second prediction result and the fault type, and calculating the KL divergence between the second prediction result and the first prediction result to obtain a loss function; adjusting the loss function through the training data.

[0013] In one implementation of the present application, obstacles are detected in real time, and local path planning is performed on the first path to obtain a second path, which specifically includes: using sensors and cameras to identify obstacles in real time and collecting obstacle data, where the obstacle data includes the speed, position, and shape of the obstacles; extracting the edges, textures, and motion trajectories of the obstacle data to identify the obstacle types, where the obstacle types include dynamic obstacles, static obstacles, large obstacles, and small obstacles; diagnosing whether to detour or wait according to the obstacle types; in the case of detouring, generating a second path through a local path planning algorithm.

[0014] In one implementation of the present application, the method further includes: performing visual display during the process of enhancing, Fourier-transforming, and splicing the noise data to obtain a color picture.

[0015] In a second aspect, an embodiment of the present application further provides a model migration device for edge autonomous diagnosis. The device includes at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is capable of: scanning a navigation area through a lidar to construct an environmental map and determine the position of the navigation device, where the lidar and the camera are located on the navigation device; obtaining a target point, selecting an optimal initial point according to the position of the navigation device, and calculating a first path according to the optimal initial point and the target point; detecting obstacles in real time, performing local path planning on the first path to obtain a second path; obtaining noise data of the navigation device at the second path, the optimal initial point, and the target point through a high-speed acquisition tool; enhancing, Fourier-transforming, and splicing the noise data to obtain a color picture, and diagnosing the color picture to construct a first noise model; migrating the first noise model to obtain a second noise model to reduce the number of model parameters.

[0016] In a third aspect, an embodiment of the present application further provides a non-volatile computer storage medium for model migration of edge autonomous diagnosis, storing computer-executable instructions, which are set as follows: scanning a navigation area through a lidar to construct an environmental map and determine the position of a navigation device, where the lidar and camera are located on the navigation device; obtaining a target point, selecting an optimal initial point according to the position of the navigation device, and calculating a first path according to the optimal initial point and the target point; detecting obstacles in real time, performing local path planning on the first path to obtain a second path; obtaining noise data of the navigation device at the second path, the optimal initial point, and the target point through a high-speed acquisition tool; strengthening, performing Fourier transform, and splicing the noise data to obtain a color picture, and diagnosing the color picture to construct a first noise model; migrating the first noise model to obtain a second noise model to reduce the number of model parameters.

[0017] The model migration method, device, and medium for edge autonomous diagnosis provided by the embodiments of the present application comprehensively capture the operation data of the navigation device in a complex environment through high-speed acquisition of sensors such as lidar and cameras; obtain the noise data of the navigation device on the path through a high-speed acquisition tool, strengthen, perform Fourier transform, and splice it, and finally convert it into a color picture for diagnosis, innovatively converting complex noise data into an intuitive and easy-to-analyze image form, and comprehensively considering the global fault signal, which provides convenience for constructing a noise model; migrating the first noise model to obtain a second noise model can significantly reduce the number of model parameters, improve the operation efficiency and deployability of the model. It realizes unmanned non-contact autonomous acquisition, and through pose adjustment, data enhancement, and model migration algorithms, it realizes lightweight, autonomous, and accurate monitoring of equipment failures of unmanned vehicle-mounted devices, while improving the convenience and practicality of mechanical equipment fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments and descriptions thereof of the present application are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0019] Figure 1 It is a flowchart of a model migration method for edge autonomous diagnosis provided by an embodiment of the present application;

[0020] Figure 2 It is a schematic diagram of the composition of a navigation device in a model migration method for edge autonomous diagnosis provided by an embodiment of the present application;

[0021] Figure 3 It is a navigation flowchart of a model migration method for edge autonomous diagnosis provided by an embodiment of the present application;

[0022] Figure 4Flowchart of noise data collection for a model migration method of edge autonomous diagnosis provided by an embodiment of the present application;

[0023] Figure 5 Schematic diagram of the basic data segment for a model migration method of edge autonomous diagnosis provided by an embodiment of the present application;

[0024] Figure 6 Schematic diagram of enhancing, Fourier transforming, and splicing the noise data for a model migration method of edge autonomous diagnosis provided by an embodiment of the present application;

[0025] Figure 7 First noise model migration schematic diagram for a model migration method of edge autonomous diagnosis provided by an embodiment of the present application;

[0026] Figure 8 Internal structure schematic diagram of a model migration device for edge autonomous diagnosis provided by an embodiment of the present application. Detailed implementation manners

[0027] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be clearly and completely described below in conjunction with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0028] The embodiments of the present application provide a model migration method, device, and medium for edge autonomous diagnosis, which solve the problems in the prior art that the autonomous navigation device collects partial fault data in a complex environment, cannot extract the key features in the fault signal, only focuses on local or single fault signals, and lacks comprehensive consideration of global fault signals.

[0029] The technical solutions proposed in the embodiments of the present application will be described in detail below with reference to the drawings.

[0030] Figure 1 Flowchart of a model migration method for edge autonomous diagnosis provided by an embodiment of the present application. As Figure 1 shown, a model migration method for edge autonomous diagnosis provided by an embodiment of the present application specifically includes the following steps:

[0031] Step 10: Scan the navigation area through a lidar to construct an environmental map and determine the position of the navigation device. The lidar and camera are located on the navigation device;

[0032] In the embodiments of the present application, a lidar can be installed on the autonomous navigation trolley to achieve automatic obstacle avoidance, as Figure 2As shown, the vehicle-mounted industrial computer is used to achieve path planning and tracking, the high-speed acquisition tool is used to collect the ambient noise of the surrounding scene and form a network, and finally the real-time collected data is analyzed through the model migration algorithm.

[0033] As an alternative embodiment, the navigation area is scanned by a lidar to construct an environmental map and determine the position of the navigation device, which may specifically include: Step 101: During the movement of the navigation device, scan through the lidar to collect the point cloud data of the navigation area, and take videos through the camera to collect the videos of the navigation area;

[0034] In this step, the video data captured by the camera provides additional information about the environment, such as color, texture, and dynamic changes. These information are very useful for constructing a more accurate environmental map and making more complex navigation decisions.

[0035] Step 102: Perform feature clustering and division on the point cloud data, and combine with the video to construct an environmental map;

[0036] In this step, the points in the point cloud data are usually very dense. In order to construct a useful environmental map, it is necessary to cluster and divide these points, group the points belonging to the same object into one category, and then use the information such as color and texture in the video data to further enrich the environmental map.

[0037] Step 103: Through ROS communication, send and receive the point cloud data of multiple navigation devices;

[0038] In this step, the initial points are selected according to the positions of the point clouds of multiple devices, the existing map is divided into grids, and ROS topic communication is used to realize the sending and receiving of coordinate points;

[0039] Step 104: Match the point cloud data with the environmental map to obtain the pose and position of the navigation device.

[0040] In this step, the map can be rasterized to achieve multi-point navigation and adjust the pose after reaching the target point.

[0041] In addition, it is also possible to obtain the data representation of the lidar, the pose ξ=(ξ x ,ξ y ,ξ θ ) of the vehicle and its translation amount in the two-dimensional plane, where ξ x and ξ y represent the translation amounts in the x and y directions, and ξ θRepresents the rotation amount of a two-dimensional plane; the lidar data frames continuously scanned within a period of time can generate a sub-map. The sub-map adopts a probabilistic grid map representation model. Each grid has two states: hit and miss. P represents the probability of grid hit. The scanning frame can be mapped to the pose change T of the sub-map through a functional relationship ξ , and the specific formula is as follows:

[0042]

[0043] The constructed map is divided into grids at a certain resolution to obtain the grid map for vehicle navigation; according to the result of feature matching, calculate the relative transformation (including rotation and translation) between the navigation device and the global map to obtain the current pose of the navigation device.

[0044] Step 20: Obtain the target point, select the optimal initial point according to the position of the navigation device, and calculate the first path according to the optimal initial point and the target point;

[0045] In this step, with multiple navigation devices and multiple initial points, overall path planning is achieved according to all given target points and the global map. The Dijkstra and A* algorithms can be used for global path planning to calculate the optimal route as the global optimal route.

[0046] In addition, after completing single-point navigation, the vehicle takes the completed target point as the starting pose, and at the same time takes the uncompleted target point as the next navigation target point, and repeats the single-point navigation process to achieve multi-point navigation. As Figure 3 shown, given the poses of n target points, determine the position information of all target points that the robot needs to visit, and clarify the starting position (pose) of the robot. For example, set the variable i = 1 to track the index of the currently going target point, obtain the current real-time pose of the robot, subscribe to the message of the movebase server, set the waiting time for the movebase service, and the robot checks whether its current position matches the position of the target point to determine whether it has reached the target point. In the case of successfully connecting to the movebase server, go to the i(1)th target point: according to the current target point pose information, the robot starts to navigate to this point; send a control command to the robot to make it start to move; the robot subscribes to the information of the movebase server to obtain its position information in real time. If the robot has not reached the target point, continue to execute the subsequent navigation steps. If the robot has reached the current target point, set this point as the starting position for the next navigation, i = i + 1: increment the target point index by 1 and prepare to go to the next target point. Check whether i is equal to n to determine whether all n target points have been visited, indicating that all target points have been visited and the robot navigation process ends.

[0047] Step 30: Detect obstacles in real time, perform local path planning on the first path, and obtain the second path;

[0048] As an alternative embodiment, detecting obstacles in real time, performing local path planning on the first path, and obtaining the second path may specifically include: Step 301: Use sensors and cameras to identify obstacles in real time and collect obstacle data, where the obstacle data includes the speed, position, and shape of the obstacles;

[0049] Step 302: Extract the edges, textures, and motion trajectories of the obstacle data to identify the obstacle types, where the obstacle types include dynamic obstacles, static obstacles, large obstacles, and small obstacles;

[0050] In this step, analyze the texture features of the obstacles to distinguish different types of obstacles, such as the ground, walls, vehicles, etc.; if the obstacle is dynamic, extract the motion trajectory by analyzing its continuous position changes; static and large obstacles such as trees, large obstacles

[0051] Step 303: Diagnose whether to detour or wait according to the obstacle type;

[0052] For example, dynamic and large obstacles: pedestrians, you can wait or detour.

[0053] Step 304: In the case of detouring, generate the second path through the local path planning algorithm.

[0054] In this step, during the actual navigation process of the vehicle, if an obstacle appears, the vehicle cannot move along the originally planned globally optimal route. The DWA algorithm is used for local path planning to avoid obstacles and select the current optimal path to best conform to the globally optimal path.

[0055] Step 40: Obtain the noise data of the navigation device at the second path, the optimal initial point, and the target point through a high-speed acquisition tool;

[0056] In this step, obtain the serial number of the device, identify the IP address of the device through the constructed local area network, and test whether the connection is intact. Refer to Figure 4 , and then collect the sound signals at different positions of the navigation device.

[0057] Step 50: Strengthen, perform Fourier transform, and splice the noise data to obtain a color picture, and diagnose the color picture to construct the first noise model; as Figure 6 shown.

[0058] As an alternative embodiment, strengthening, performing Fourier transform, and splicing the noise data to obtain a color picture, and diagnosing the color picture to construct a noise diagnosis model may specifically include:

[0059] Step 501: Filter and perform time-domain segmentation on the noise data to obtain basic data segments;

[0060] In this step, first, filter the sound signal to remove irrelevant components, obtain a signal with better representation ability, and perform data augmentation on this basis to expand the data and alleviate the problem of insufficient samples. The data augmentation specifically performs time-domain segmentation on the device vibration sound data collected by the sensor, and each segmented signal segment serves as the basic data source for the TSFT transform, as Figure 5 shown.

[0061] Step 502: Perform a fast Fourier transform on the basic data segments to convert the basic data segments from the time domain to the frequency domain, obtaining frequency-domain data segments;

[0062] Step 503: Divide the frequency-domain data segments into frequency-domain signals in three frequency bands: low, medium, and high, according to a preset frequency-domain interval;

[0063] Step 504: Extract the component information of the frequency-domain signals and perform zero-padding in the frequency domain;

[0064] In this step, as Figure 7 in the backbone network, in order to make the length of the transformed signal sub-band the same as that of the original signal, it is necessary to perform zero-padding in the frequency domain on the divided frequency signals.

[0065] Step 505: Perform an inverse Fourier transform on the frequency-domain signals after zero-padding in the frequency domain to convert the frequency-domain signals from the frequency domain to the time domain, obtaining the first time-domain data segments.

[0066] As an optional embodiment, after performing an inverse Fourier transform on the frequency-domain signals after zero-padding in the frequency domain to convert the frequency-domain signals from the frequency domain to the time domain and obtaining the first time-domain data segments, the method may further include: Step 506: Segment the first time-domain data segments in one-dimensional time series according to a preset length to obtain second time-domain data segments, so as to rearrange the time-domain data segments in the spatial dimension;

[0067] For example, the lengths of the 3 sub-band signals after separation are L(i), i = 1,..., M 2 , where M 2 represents the signal length, and perform spatial dimension rearrangement on the transformed sub-band signals, that is, segment the signals in one-dimensional time series according to the length M,

[0068] Step 507: Concatenate the second time-domain data segments into a two-dimensional matrix, normalize the numerical values of the two-dimensional matrix of the second time-domain data segments, and splice them into a color picture in the spatial dimension;

[0069] In this step, a two-dimensional M×M matrix is spatially spliced, and the specific formula is as follows. Finally, the numerical values of the two-dimensional matrices of the three segments of signals are normalized to the range of 0-255, and then they are spliced into a color image in the spatial dimension.

[0070]

[0071] In the formula, P(j,k), where j = 1,......, M and k = 1,......, M, represents the value of a certain element in the converted matrix. j and k are the specific positions of the element values, round() is the rounding function, and Max() and Min() represent taking the maximum value and the minimum value respectively.

[0072] Step 508: Input the color image into the network model and collect the historical fault types of the navigation device;

[0073] Step 509: Match the color image with the historical fault types to obtain a noise diagnosis model.

[0074] In this step, the image features of the color image are correlated and analyzed with the fault features extracted from the historical data to find the fault patterns or trends.

[0075] Step 60: Migrate the first noise model to obtain a second noise model to reduce the number of model parameters. As Figure 7 shown.

[0076] As an optional embodiment, migrating the first noise model to reduce the number of model parameters may specifically include: Step 601: Obtain training data, use the first noise model for prediction to obtain a first prediction result, and the training data includes noise data and fault types; Step 602: Reduce a preset number of layers, convolutional kernels, and neurons in the first noise model to obtain a second noise diagnosis model; Step 603: Use the second noise diagnosis model to predict the training data to obtain a second prediction result.

[0077] As an optional embodiment, after using the second noise diagnosis model to predict the training data to obtain a second prediction result, the method may further include: Step 604: Calculate the cross-entropy loss between the second prediction result and the fault types, and calculate the KL divergence between the second prediction result and the first prediction result to obtain a loss function; Step 605: Adjust the loss function through the training data.

[0078] In this step, the first noise model is trained well, and then the output result q of the first noise model with a larger number of parameters is used as the target value of a second noise diagnosis model with a smaller number of parameters, prompting the second noise diagnosis model to learn the output result of the first noise model. The training target loss function is as follows:

[0079] L = CE(y, p) + αCE(q, p)

[0080] Wherein, CE() is the cross-entropy function, y is the one-hot encoding of the true label, q is the output result of the first noise model, and p is the output result of the second noise diagnosis model; in order to transfer the similar information of the data learned by the original network to the small network, an improved activation function, the sofmax-T function, is selected. The formula is as follows:

[0081]

[0082] Meanwhile, the kernel method is used to improve the feature extraction ability of the model, and the transfer of the model feature extraction ability is mined from the mapping relationship between the two in the high-dimensional space. Therefore, the final loss function is as follows:

[0083]

[0084] Wherein, KLdiv is the KL divergence, Q S , Q T is the probability distribution obtained by softmax of the output of the fully connected layer, is the probability distribution obtained by the above formula from the outputs of the fully connected layers of the pre-trained network and the lightweight network, α, β, and T are all hyperparameters, and y true is the label of the training sample.

[0085] As an optional embodiment, the method may further include: performing visual display during the process of strengthening, Fourier transform, and splicing of the noise data to obtain a color picture.

[0086] In this step, the processing process of the noise data is visually displayed.

[0087] The above is the method embodiment proposed in this application. Based on the same inventive concept, the embodiment of this application also provides a model migration device for edge autonomous diagnosis, and its structure is as Figure 8 shown.

[0088] Figure 8 is a schematic diagram of the internal structure of a model migration device for edge autonomous diagnosis provided by the embodiment of this application. As shown in Figure 8 shown, the device includes:

[0089] At least one processor 201;

[0090] And a memory 202 communicatively connected to at least one processor;

[0091] ​Among them, the memory 202 stores instructions executable by at least one processor. The instructions are executed by at least one processor 201 to enable at least one processor 201 to: scan a navigation area through a lidar to construct an environmental map and determine the position of a navigation device, where the lidar and camera are located on the navigation device; obtain a target point, select an optimal initial point according to the position of the navigation device, and calculate a first path according to the optimal initial point and the target point; detect obstacles in real time, perform local path planning on the first path to obtain a second path; obtain noise data of the navigation device at the second path, the optimal initial point, and the target point through a high-speed acquisition tool; strengthen, perform Fourier transform, and splice the noise data to obtain a color picture, and diagnose the color picture to construct a first noise model; migrate the first noise model to obtain a second noise model to reduce the number of model parameters.

[0092] Some embodiments of the present application provide a Figure 1 non-volatile computer storage medium for model migration of edge autonomous diagnosis corresponding to, storing computer-executable instructions, and the computer-executable instructions are set to: scan a navigation area through a lidar to construct an environmental map and determine the position of a navigation device, where the lidar and camera are located on the navigation device; obtain a target point, select an optimal initial point according to the position of the navigation device, and calculate a first path according to the optimal initial point and the target point; detect obstacles in real time, perform local path planning on the first path to obtain a second path; obtain noise data of the navigation device at the second path, the optimal initial point, and the target point through a high-speed acquisition tool; strengthen, perform Fourier transform, and splice the noise data to obtain a color picture, and diagnose the color picture to construct a first noise model; migrate the first noise model to obtain a second noise model to reduce the number of model parameters.

[0093] The embodiments in the present application are all described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the embodiments of the Internet of Things devices and media, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0094] The systems and media provided by the embodiments of the present application correspond one-to-one with the methods. Therefore, the systems and media also have beneficial technical effects similar to the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0095] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0096] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0097] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0099] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0100] The memory may include non-permanent memory in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

[0101] A computer-readable medium includes permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transitory media that can be used to store information that can be accessed by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0102] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0103] The above description is only for the embodiments of the present application and is not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A model migration method for edge autonomous diagnosis, characterized in that: The method comprises: Scanning the navigation area by laser radar to construct an environment map and determine the position of the navigation device, where the laser radar and the camera are located; Acquire a target point, select an optimal initial point according to the position of the navigation device, and calculate a first path according to the optimal initial point and the target point; Detect obstacles in real time, perform local path planning on the first path, and obtain a second path; Acquire noise data of the navigation device on the second path, the optimal initial point and the target point by a high-speed acquisition tool; Performing enhancement, Fourier transformation and splicing on the noise data to obtain a color image, and diagnosing the color image to construct a first noise model; The first noise model is migrated to obtain a second noise model to reduce the number of model parameters.

2. The model migration method for edge autonomous diagnosis according to claim 1, characterized in that: Scanning the navigation area through a laser radar to construct an environmental map and determine the location of the navigation device, specifically including: During the movement of the navigation device, the laser radar is used to scan and collect point cloud data of the navigation area, and the camera is used to shoot and collect video of the navigation area; Performing feature clustering on the point cloud data, and combining it with the video to construct an environment map; Send and receive point cloud data from multiple navigation devices through ROS communication; The point cloud data is matched with the environment map to obtain the posture and position of the navigation device.

3. The model migration method for edge autonomous diagnosis according to claim 1, characterized in that: The noise data is enhanced, Fourier transformed and spliced ​​to obtain a color image, and the color image is diagnosed to build a noise diagnosis model, specifically including: The noise data is filtered and time-domain segmented to obtain basic data segments; Performing a fast Fourier transform on the basic data segment to convert the basic data segment from the time domain to the frequency domain to obtain a frequency domain data segment; Dividing the frequency domain data segment into frequency domain signals of three frequency bands: low, medium and high according to a preset frequency domain interval; Extracting component information of the frequency domain signal and performing frequency domain zero padding; An inverse Fourier transform is performed on the frequency domain signal after the frequency domain zero padding to convert the frequency domain signal from the frequency domain to the time domain to obtain a first time domain data segment.

4. The model migration method for edge autonomous diagnosis according to claim 3 is characterized in that: After performing inverse Fourier transform on the frequency domain signal after zero padding in the frequency domain to convert the frequency domain signal from the frequency domain to the time domain to obtain the first time domain data segment, the method further includes: Dividing the first time domain data segments segment by segment according to a preset length in a one-dimensional time sequence to obtain second time domain data segments, so as to rearrange the time domain data segments in a spatial dimension; splicing the second time domain data segments into a two-dimensional matrix, normalizing the values ​​of the two-dimensional matrix of the second time domain data segments, and splicing them into a color picture in the spatial dimension; inputting the color image into a network model and collecting historical fault types of the navigation device; The color image is matched with the historical fault types to obtain a noise diagnosis model.

5. The model migration method for edge autonomous diagnosis according to claim 1, characterized in that: Migrating the noise model to reduce the number of model parameters includes: Acquire training data, and use the first noise model to perform prediction to obtain a first prediction result, wherein the training data includes noise data and a fault type; Reducing the number of layers, convolution kernels, and neurons of the first noise model by a preset number to obtain a second noise diagnosis model; The second noise diagnosis model is used to predict the training data to obtain a second prediction result.

6. The model migration method for edge autonomous diagnosis according to claim 5, characterized in that: After predicting the training data using the second noise diagnosis model to obtain a second prediction result, the method further includes: Calculating the cross entropy loss between the second prediction result and the fault type, and calculating the KL divergence between the second prediction result and the first prediction result to obtain a loss function; The loss function is adjusted by the training data.

7. The model migration method for edge autonomous diagnosis according to claim 1, characterized in that: Detecting obstacles in real time and performing local path planning on the first path to obtain a second path specifically includes: Using sensors and cameras to identify obstacles in real time and collect obstacle data, including the speed, position, and shape of the obstacles; Extracting the edge, texture and motion trajectory of the obstacle data, and identifying the obstacle type, wherein the obstacle type includes dynamic obstacles, static obstacles, large obstacles and small obstacles; Diagnose detour or wait according to the type of obstacle; In the case of a detour, a second path is generated by a local path planning algorithm.

8. The model migration method for edge autonomous diagnosis according to claim 4, characterized in that: The method further comprises: In the process of enhancing, Fourier transforming and splicing the noise data to obtain a color picture, a visual display is performed.

9. A model migration device for edge autonomous diagnosis, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Scanning the navigation area by laser radar to construct an environment map and determine the position of the navigation device, wherein the laser radar and the camera are located in the navigation device; Acquire a target point, select an optimal initial point according to the position of the navigation device, and calculate a first path according to the optimal initial point and the target point; Detect obstacles in real time, perform local path planning on the first path, and obtain a second path; Acquire noise data of the navigation device on the second path, the optimal initial point and the target point by a high-speed acquisition tool; Performing enhancement, Fourier transformation and splicing on the noise data to obtain a color image, and diagnosing the color image to construct a first noise model; The first noise model is migrated to obtain a second noise model to reduce the number of model parameters.

10. A non-volatile computer storage medium for model migration of edge autonomous diagnosis, storing computer executable instructions, characterized in that: The computer executable instructions are configured to: Scanning the navigation area by laser radar to construct an environment map and determine the position of the navigation device, wherein the laser radar and the camera are located in the navigation device; Acquire a target point, select an optimal initial point according to the position of the navigation device, and calculate a first path according to the optimal initial point and the target point; Detect obstacles in real time, perform local path planning on the first path, and obtain a second path; Acquire noise data of the navigation device on the second path, the optimal initial point and the target point by a high-speed acquisition tool; Performing enhancement, Fourier transformation and splicing on the noise data to obtain a color image, and diagnosing the color image to construct a first noise model; The first noise model is migrated to obtain a second noise model to reduce the number of model parameters.

Images