Thermal fault early warning method of machine room inspection robot with multi-mode perception

Abstract

The invention relates to the field of thermal fault early warning of machine room inspection robots, in particular to a multi-modal sensing thermal fault early warning method of a machine room inspection robot, which comprises the following steps: acquiring infrared thermal image, acoustics, vibration and gas concentration data through a multi-modal sensor array, and generating voxelization temperature distribution; according to the method, temperature features are extracted through wavelet transform and a fractional Brownian motion algorithm and are fused into an equipment global temperature feature set; constructing noise, vibration and gas concentration tower extraction auxiliary features, and combining non-extensive entropy and a graph attention model to be fused into a high-dimensional fault feature pool; building a fiber bundle model by taking the machine room topology as a base space, and locating a thermal fault after optimizing parameters by a simulated annealing algorithm; and finally, a dynamic thermodynamic diagram and the like are generated through digital twinning, so that accurate early warning is realized.

Description

TECHNICAL FIELD

[0001] The application relates to the field of thermal fault early warning of a machine room inspection robot, in particular to a thermal fault early warning method of a multi-modal perception machine room inspection robot. BACKGROUND

[0002] With the acceleration of digital transformation, data center machine room equipment presents a high-density and high-power trend. The centralized deployment of servers, switches and storage devices makes the thermal environment of the machine room increasingly complex. Local overheating and failure of the heat dissipation system have become the main cause of equipment downtime or data loss, which seriously threatens the stable operation of the machine room. The traditional machine room thermal fault early warning method has the following limitations: It mainly relies on single-point temperature sensors or infrared thermal imagers for local temperature measurement, which can only capture surface temperature changes and cannot associate multi-source fault precursors such as vibration abnormalities, noise mutations and gas leaks during equipment operation, resulting in early weak faults being easily ignored, high false positive and false negative rates, and difficulty in extracting deep features reflecting the nature of thermal faults such as transient temperature fluctuations of chip-level micro-hot spots and complex thermal gradients formed by airflow turbulence inside the cabinet, making it difficult to achieve early warning. The influence of the machine room equipment topology on heat propagation has not been considered, and the diffusion path of the thermal fault between equipment cannot be modeled, resulting in ambiguous fault location and difficulty in accurately locating the root component. Therefore, a thermal fault early warning method of a multi-modal perception machine room inspection robot is needed to solve the above problems. SUMMARY

[0003] In order to solve the technical problems in the background art, the application provides a scene self-adaptive adjustment method based on an intelligent AI terminal.

[0004] The purpose of the application can be achieved by the following technical solutions: The application provides a thermal fault early warning method of a multi-modal perception machine room inspection robot, and the specific steps are as follows: Step 1, machine room data acquisition: the multi-modal sensor array carried by the robot collects infrared thermal images, acoustic, vibration and gas concentration data of the machine room, and generates voxelized temperature distribution using a three-dimensional temperature field reconstruction algorithm; The inspection robot moves according to the preset inspection path, and the multi-modal sensor array starts collecting data at each monitoring point. The multi-modal sensor array includes an infrared thermal imager, an acoustic sensor, a vibration sensor and a gas sensor, and real-time equipment data sets are obtained, including infrared images, equipment operating noise, equipment vibration signals and gas concentration around the equipment. The collected multiple infrared images are arranged to obtain an infrared image sequence; The solution adopts a binocular structured light camera, which consists of two infrared cameras and one infrared projector. The infrared projector emits infrared light with preset patterns, such as stripes and dot patterns, to enhance the surface texture features of the equipment in the computer room, making it easier for the camera to recognize and match. The camera and projector are installed at the front end of the robot. An infrared projector projects a pre-encoded structured light pattern onto the surface of the cabinet. The structured light pattern resembles a sequence of binary code stripes. The light is distorted by the shape or distance of the equipment surface. Two infrared cameras synchronously acquire images with the distorted pattern. Each acquisition generates an image from each camera, forming a binocular image pair. The acquired binocular images undergo epipolar correction processing. By using camera calibration parameters, the image coordinates are transformed to the epipolar constraint plane, ensuring that the search for corresponding points is performed only in the same row, improving matching efficiency and accuracy. A block-based matching algorithm is used to search for matching points in the epipolar-corrected binocular images. Centered on a pixel in the left image, a certain size image block is selected. The image block with the highest similarity is searched in the corresponding row of the right image to determine the matching point pair and generate a disparity image. Based on the principle of triangulation, combined with the baseline distance obtained from camera calibration, the camera focal length, and the disparity image, the depth value of each point on the computer cabinet surface is calculated pixel by pixel, finally generating a structured light depth image. Each pixel value in the image represents the actual distance of the corresponding point from the camera, providing accurate spatial location information for the reconstruction of the three-dimensional temperature field. The structured light depth image and infrared image sequence are input into the U-Net-LSTM network. The encoder extracts multi-scale temperature features through convolution and outputs a temperature feature map. The LSTM layer captures the temperature change line in the time dimension through bidirectional LSTM. The decoder outputs a voxelized temperature distribution. ,in The dimensions are horizontal, vertical, and depth, with t representing a time point, covering the three-dimensional space inside the device. It should be noted that the voxelized temperature distribution is achieved by fusing two-dimensional temperature information from infrared thermal imaging with three-dimensional spatial information from structured light depth maps via a network, and then reconstructing the data through an encoder-temporal modeling-decoder process. The final output is the temperature value of each tiny region in the three-dimensional space inside the device.

[0005] Step 2, Cooperative Temperature Features: The voxelized temperature distribution is decomposed into multiple scales by wavelet transform, and the Hurst exponent features at each scale are extracted by fractional Brownian motion algorithm. Then, the wavelet energy features are fused to output a set of global temperature features of the device that includes the surface and internal heat distribution of the device. The voxelized temperature distribution is decomposed into multiple layers using a wavelet algorithm with a low-pass filter (HY) and a high-pass filter (HT). The decomposition scales for each filter are set to micro, meso, and macro scales. The micro scale captures localized, minute temperature abrupt changes, such as chip-level hotspots and localized heat dissipation anomalies. The meso scale reflects regional temperature gradient changes, such as heat transfer between server modules. The macro scale extracts global temperature trends, such as the overall heat dissipation status of the server rack and the influence of ambient temperature. The decomposition calculation logic using the recursive rules of the low-pass filter is as follows: ,in Let k be the low-frequency quantity at the k-th decomposition scale, where k is the decomposition scale number. For convolution operations, Represents the tensor product. , and For low-pass filters in different directions; the decomposition calculation logic using the recursive rules of high-pass filters is as follows: Where T is an abbreviation for voxelization temperature distribution. , and These are high-pass filters in different directions. , and These are high-frequency components in different directions, which, when superimposed, constitute a high-frequency quantity. ; The surface and internal regions of the cabinet are labeled as region GS and region FI, respectively. Hurst exponents are calculated at k scales of wavelet decomposition to quantify the self-similarity of temperature distributions at different scales. The voxelized temperature distribution is then distributed across a spatial grid at scale k, and for any spatial location... , or Extract the time series of this spatial location , where N is the total number of time points; the mean temperature at this spatial location over time is calculated accordingly. Then, the cumulative deviation of the temperature series is obtained based on the mean. Its calculation logic is as follows: , The cumulative deviation over the first n time steps is calculated. These n time steps are then combined into a subsequence length, where n < N. The difference between the maximum and minimum subsequence lengths is taken to obtain the range. The standard deviation of the subsequence length is then calculated. The rescaled range is obtained by dividing the range by the standard deviation. Fitting by least squares method The slope is the Hurst exponent H, and its calculation logic is as follows: Where C is the intercept of the linear fit, is the natural logarithm function, argmin is the value of the parameter H that minimizes the expression in the parentheses; repeat the above steps for each scale to obtain the Hurst index of each scale ; extract the instantaneous thermal mutation features of each scale through high-frequency quantity, and calculate the energy of the high-frequency quantity of each scale The calculation logic is: , locate the modulus maximum value point in the high-frequency quantity, extract the hot spot coordinates (r, t) and the corresponding temperature peak value; cross-scale fusion of voxelized temperature distribution, fractional Brownian motion features and wavelet features to construct the global temperature feature set of the device.

[0006] Step three, multi-feature extraction: the inspection robot constructs a noise tower, a device vibration tower and a gas concentration tower, outputs the corresponding features and integrates them into an auxiliary decision feature set, and fuses the auxiliary decision features with the global temperature feature set of the device through a non-extensive entropy algorithm and a graph attention model to construct a high-dimensional temperature fault feature pool; The noise tower identifies the features through power spectrum and empirical mode decomposition (EMD), frames the device operating noise and adds a Hanning window, performs fast Fourier transform on each frame signal through the Welch method, calculates the corresponding power spectrum, and calculates the power spectrum distribution curve by averaging multiple frames of power spectrum. The device operating noise is decomposed by EMD to obtain a plurality of intrinsic mode functions (CL), and the noise energy ratio and instantaneous frequency of each intrinsic mode function are calculated. The noise energy calculation logic of a single intrinsic mode function is: , wherein and are the start time and end time of the calculation process, and dt is the time increment; the total noise energy of all intrinsic mode functions is obtained, and the noise energy ratio is obtained by dividing the single noise energy by the total noise energy; the noise instantaneous frequency The calculation logic is: , wherein is the sampling interval, is the instantaneous phase, and unwrap is the phase unwrapping function; The device vibration tower identifies the vibration features through differential geometry manifold algorithm and quantum resonance analysis, and the specific identification steps are as follows: S001: Obtain the device vibration signal xc, select the embedding dimension mc through the false neighbor point method, set the time delay rk, and construct the phase space matrix through the embedding dimension and the time delay, the construction logic is: , wherein K is the total sampling time point of the vibration device, and each row of the matrix corresponds to a phase space point; S002: Obtain the neighborhood points corresponding to each phase space point through the k-nearest neighbor algorithm, and perform PCA dimension reduction on the neighborhood points to obtain the local tangent space; S003: The quantum resonance analysis algorithm performs Fourier transform on the device vibration signal to obtain the vibration frequency spectrum ,Will The embedded potential function is calculated as follows: ,in It is a small change; S004: Substitute the potential energy function into the Schrödinger equation to obtain the vibrational Hamiltonian. Its calculation logic is as follows: Where hr is the reduced Planck constant, and mc is the equivalent mass, which is obtained by extracting the mass of equipment components from the database. The Laplace operator is used; the vibration characteristic equation is constructed and solved based on the Hamiltonian. The equation formula is as follows: The vibration energy EC is obtained. The vibration energy is divided by the reduced Planck constant to obtain the eigenfrequency. The normal eigenfrequency is obtained from historical data. The eigenfrequency and the normal eigenfrequency are combined and their absolute values ​​are taken to obtain the eigenfrequency deviation. Gas concentration towers extract gas characteristics through fractional-order diffusion field reconstruction, specifically: G001: Acquire data on the gas concentration in the computer room (UE), construct a fractional-order diffusion equation based on the gas concentration data, and the calculation logic is as follows: Where is the air diffusion coefficient of the computer room. If the source term is... A value of 0 indicates no leakage, while a value greater than 0 indicates the presence of leakage. For time t fractional derivative; G002: Based on the fractional diffusion equation, the fractional diffusion field under normal conditions is obtained. It can be trained using historical fault-free data. The deviation between the real-time field and the normal field is marked as the diffusion anomaly index. The power spectrum distribution curve, noise energy ratio, instantaneous frequency, local tangent space, intrinsic frequency deviation, and diffusion anomaly index are integrated into an auxiliary judgment feature set; Calculate the joint non-extensive entropy of the feature xm of the device's global temperature feature set and the feature xv of the auxiliary decision feature set. Its calculation logic is as follows: , of which Summing the square of the joint probability distribution, where q is the non-extensive parameter; embedding the joint non-extensive entropy into the mutual information formula to obtain the correlation coefficient between any feature in the auxiliary judgment feature set and the features in the device's global temperature feature set. The logic for calculating its correlation coefficient is as follows: , The larger the value, the stronger the nonlinear correlation between the two features; the auxiliary features are screened by arranging the features of the auxiliary judgment feature set according to the magnitude of the correlation coefficient, and retaining the first m auxiliary features to obtain the screened auxiliary feature set; The features of the screening auxiliary feature set and the device global temperature feature set are taken as nodes, and the correlation coefficient is taken as the initial edge weight to construct a feature correlation graph. The feature correlation graph is input into a graph attention model and an attention coefficient is output. The graph attention model specifically calculates the attention coefficient of each temperature node with all connected auxiliary nodes by the following logic: wherein and are the original feature vectors of the auxiliary nodes and the temperature nodes, m2 and m3 are the numbers of the auxiliary nodes and the temperature nodes, WT is a shared weight matrix, which is obtained by mapping the original features to a high-dimensional space, is an attention parameter vector, which is used to learn the importance of feature correlation, is an activation function; is larger; all attention coefficients are obtained by stacking two layers of GAT, and global correlation features are obtained. The global correlation features are maximally pooled to obtain a pooling result. The high-dimensional fault feature pool is obtained by concatenating the pooling result and the feature nodes.

[0007] Step four, fiber bundle model prediction: a fiber bundle model is constructed with the machine room device topology as the base space and the high-dimensional fault feature pool as the fiber vector. The model parameters are optimized using a simulated annealing algorithm, and the F1 score is taken as the target. The parameters are iteratively updated according to the Metropolis criterion. The characteristic propagation anomaly degree is calculated according to the optimized model to determine and locate the thermal fault; The physical topology of the machine room device is taken as the base space B, such as the location distribution of servers, switches and cooling systems. It is discretized into grid nodes b, each grid node corresponds to a device component or region, and the topological relationship of the base space is represented by an adjacency matrix. If two grid points are physically connected, the corresponding adjacency matrix is 1, and vice versa. The fiber vector is the high-dimensional fault feature pool of any grid node b. The fiber vector is obtained by mapping the high-dimensional fault feature pool to the base space; The connection coefficient of the base space is obtained by the following logic: wherein and are the numbers of any two nodes, is the connection coefficient, is a topological constraint coefficient that controls the correlation between propagation strength and physical distance, and dm is the physical distance. The fiber bundle model is constructed in this way; ​The simulated annealing algorithm optimizes the fiber cluster parameters, marks the initial topological constraint coefficient as a to-be-optimized parameter, the initial topological constraint coefficient is provided by historical data, marks the thermal fault prediction performance of the fiber cluster model as a thermal energy value HM, randomly disturbs the to-be-optimized parameter to generate a new to-be-optimized parameter, and obtains a new thermal energy value of the new to-be-optimized parameter, and the thermal energy value and the current thermal energy value are subtracted to obtain a thermal energy difference, if the thermal energy difference is less than 0, the new to-be-optimized parameter is directly accepted, and the current parameter is updated to the new parameter, if the thermal energy difference is greater than 0, the random disturbance is continued, when the iteration number reaches a preset value, the iteration is stopped, and the optimal topological constraint coefficient is output and delivered to the fiber cluster model; Extract the contact coefficient of all adjacent nodes of each node in the base space to form a plurality of local contact coefficient sets, calculate the average contact coefficient of each node, obtain the maximum deviation value of the contact coefficient and the average contact coefficient of the node, mark the maximum deviation value as the thermal fault abnormality degree, and obtain the thermal fault abnormality degree of each node in the base space; extract the preset abnormality threshold in the database, and determine the nodes with the thermal fault abnormality degree exceeding the preset abnormality threshold as potential fault points; the potential fault points are clustered in space, and adjacent fault points with a physical distance less than a preset distance are aggregated into a predicted fault area; Step five, digital twin fault scene mapping: the base space physical coordinates are aligned with the digital twin three-dimensional coordinates, and the optimized fiber cluster feature data is imported; the future thermal diffusion trend is simulated to generate a dynamic thermal map and a vibration cloud chart; The base space physical coordinates are accurately aligned with the three-dimensional coordinates of the digital twin model, and the optimized fiber cluster feature data such as abnormality degree and contact coefficient are imported; the thermal diffusion trend of the future period is simulated through the digital twin to generate a visual dynamic thermal map and a vibration cloud chart; the inspection robot sends the predicted fault area, the dynamic thermal map and the vibration cloud chart to the client.

[0008] Compared with the prior art, the beneficial effects of the present application are: By fusing multi-modal features with non-extended entropy and graph attention model, a high-dimensional fault feature pool is constructed, which effectively deals with interference in the complex environment of the machine room, improves the robustness of the model, and adapts to the general early warning needs of different equipment; By combining wavelet transform, fractional Brownian motion algorithm and other methods, nonlinear features and micro-mutation signals of the thermal distribution of the equipment are extracted, early weak faults that are difficult to detect by traditional methods can be identified, and advanced early warning is realized; Based on the fiber cluster model, a device topological base space and a feature propagation path are constructed, combined with parameter optimization, the diffusion law of thermal faults between devices is accurately described, and accurate mapping from feature anomalies to physical locations is realized. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced as follows, and the following drawings are not drawn in scale, and the emphasis is on showing the main idea of the present application.

[0010] Figure 1 Method steps block diagram of the present application; Figure 2 Algorithm framework of the present application Figure One ; Figure 3 Algorithm framework of the present application Figure Two . DETAILED DESCRIPTION

[0011] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings, and obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor also belong to the scope of protection of the present application.

[0012] Please refer to Figures 1-3 The present application provides a multi-modal perception machine room inspection robot thermal fault early warning method, and the specific steps are as follows: Step one, machine room data acquisition: through the multi-modal sensor array carried by the robot, the infrared thermal image, acoustic, vibration and gas concentration data of the machine room are collected, and the voxelized temperature distribution is generated by using the three-dimensional temperature field reconstruction algorithm; The inspection robot moves according to the preset inspection path, and the multi-modal sensor array starts to collect every time it reaches a monitoring point. The modal sensor array includes an infrared thermal imager, an acoustic sensor, a vibration sensor and a gas sensor, and the machine room equipment data set is obtained in real time, including infrared images, equipment running noise, equipment vibration signals and equipment surrounding gas concentration. The collected multi-frame infrared images are arranged to obtain an infrared image sequence; A binocular structured light camera scheme is adopted, which is composed of two infrared cameras and an infrared projector. The infrared projector emits a preset pattern of infrared light, such as a stripe and a dot matrix pattern, to enhance the texture features of the machine room equipment surface, facilitate camera recognition and matching, and the camera and the projector are installed at the front end of the robot; The infrared projector projects a pre-encoded structured light pattern to the cabinet surface, such as a binary code stripe sequence, and the light is deformed due to the shape or distance of the device surface. Two infrared cameras synchronously capture images with the deformed pattern. Each capture generates an image for the left and right cameras, forming a binocular image pair. The collected binocular images are subjected to epipolar rectification processing, which converts the image coordinates to the epipolar constraint plane through camera calibration parameters, ensuring that the corresponding point search is only in the same row, thereby improving the matching efficiency and accuracy. A block matching algorithm is used to search for matching points in the epipolar rectified binocular images. Taking a certain pixel point in the left image as the center, a certain size of image block is taken, and the image block with the highest similarity in the corresponding row of the right image is searched to determine the matching point pair and generate a disparity image DT. The disparity is the horizontal pixel coordinate difference of the corresponding points. According to the principle of triangulation, combined with the baseline distance BC, camera focal length FG, and disparity image obtained by camera calibration, the depth value of each point on the cabinet surface is calculated pixel by pixel through the formula , and the structured light depth image ZR is finally generated. Each pixel value in the image represents the actual distance of the corresponding point from the camera, providing accurate spatial position information for three-dimensional temperature field reconstruction. The structured light depth image and infrared image sequence are input into the U-Net-LSTM network. The encoder extracts multi-scale temperature features through convolution and outputs temperature feature maps. The LSTM layer captures the temperature change line in the time dimension through bidirectional LSTM. The decoder outputs the voxelized temperature distribution , where is the horizontal, vertical, and depth directions, and t is the time point, covering the three-dimensional space inside the device. It should be noted that the voxelized temperature distribution is reconstructed through the network by fusing the two-dimensional temperature information of the infrared thermal image and the three-dimensional spatial information of the structured light depth image, and finally outputs the temperature value of each small region in the three-dimensional space inside the device.

[0013] Step two, cooperative temperature features: through wavelet transform, the voxelized temperature distribution is multi-scale decomposed, and the Hurst index features of each scale are extracted combined with the fractional Brownian motion algorithm, and then the wavelet energy features are fused to output the device global temperature feature set containing the device surface and internal heat distribution; The voxelized temperature distribution is multi-layer decomposed through the low-pass filter HY and the high-pass filter HT by the wavelet algorithm. The decomposition scale of each filter is set to the microscopic scale, the mesoscopic scale, and the macroscopic scale. The microscopic scale captures local small temperature mutations, such as chip-level hot spots and local heat dissipation abnormalities. The mesoscopic scale reflects the regional temperature gradient changes, such as heat transfer between server modules. The macroscopic scale extracts the global temperature trend, such as the overall heat dissipation state of the cabinet and the environmental temperature influence. The decomposition calculation logic of the low-pass filter recursive rule is: , where Let k be the low-frequency quantity at the k-th decomposition scale, where k is the decomposition scale number. For convolution operations, Represents the tensor product. , and For low-pass filters in different directions; the decomposition calculation logic using the recursive rules of high-pass filters is as follows: Where T is an abbreviation for voxelization temperature distribution. , and These are high-pass filters in different directions. , and These are high-frequency components in different directions, which, when superimposed, constitute a high-frequency quantity. ; The surface and internal regions of the cabinet are labeled as region GS and region FI, respectively. Hurst exponents are calculated at k scales of wavelet decomposition to quantify the self-similarity of temperature distributions at different scales. The voxelized temperature distribution is then distributed across a spatial grid at scale k, and for any spatial location... , or Extract the time series of this spatial location , where N is the total number of time points; the mean temperature at this spatial location over time is calculated accordingly. Then, the cumulative deviation of the temperature series is obtained based on the mean. Its calculation logic is as follows: , The cumulative deviation over the first n time steps is used to construct a subsequence length, where n < N. It should be noted that the cumulative deviation reflects the temperature's deviation from the mean. The range is obtained by subtracting the maximum and minimum values ​​of the subsequence length; the range reflects the fluctuation range of the cumulative deviation sequence. The standard deviation of the subsequence length is then calculated, and the rescaled range is obtained by dividing the range by the standard deviation. Fitting by least squares method The slope is the Hurst exponent H, and its calculation logic is as follows: Where C is the intercept of the linear fit, Let H be the natural logarithm function, and argmin be the value of the parameter H that minimizes the expression within the parentheses. It should be noted that H = 1 indicates strong self-similarity, and H = 0 indicates weak self-similarity. Repeating the above steps for each scale yields the Hurst exponent for each scale. ; The transient thermal mutation characteristics at each scale are extracted by high-frequency quantities, and the energy of high-frequency quantities at each scale is calculated. This reflects the intensity of temperature fluctuations at this scale, and its calculation logic is as follows: , positioning the modulus maximum point in the high-frequency quantity, extracting the hotspot coordinates (r, t) and the corresponding temperature peak value; the voxelized temperature distribution, fractional Brownian motion characteristics and wavelet characteristics are fused across scales to construct a device global temperature feature set; the device global temperature feature set covers the temperature of the cabinet equipment surface and the inside, which not only reflects the coverage range from the outside to the inside in space, but also highlights the integrated description of the overall temperature state of the cabinet equipment, accurately reflecting the comprehensive situation of the temperature of each part of the cabinet equipment.

[0014] Step three, multi-feature extraction: the inspection robot constructs a noise tower, a device vibration tower and a gas concentration tower, outputs corresponding features and integrates them into an auxiliary judgment feature set, and fuses the auxiliary judgment features with the device global temperature feature set through a non-extensive entropy algorithm and a graph attention model to construct a high-dimensional temperature fault feature pool; The noise tower identifies the features through the power spectrum diagram and empirical mode decomposition (EMD), frames the device running noise and adds a Hanning window, performs fast Fourier transform on each frame signal through the Welch method, calculates the corresponding power spectrum, and obtains the power spectrum distribution curve by averaging multiple frame power spectrums; the device running noise is decomposed by EMD to obtain a plurality of intrinsic mode functions (CL), each intrinsic mode function corresponds to a vibration mode of different frequency scales, and is suitable for capturing transient noise during device start-stop and sudden failure; the noise energy proportion and instantaneous frequency of each intrinsic mode function are calculated, specifically, the noise energy calculation logic of a single intrinsic mode function is as follows: , wherein and is the starting time and the ending time of the calculation process, and dt is the time increment; the total noise energy of all intrinsic mode functions is obtained by dividing the single noise energy by the total noise energy; it should be noted that the role of empirical mode decomposition (EMD) is to adaptively decompose the complex noise signal into a plurality of intrinsic mode functions; these intrinsic mode functions correspond to vibration modes of different frequency scales, and can accurately capture transient features in noise, such as abnormal noise during device start-stop and sudden failure; by EMD decomposition, the noise energy proportion and instantaneous frequency of each intrinsic mode function can be further extracted, which can effectively reflect the subtle changes of the device running state, such as the abnormal energy of specific frequency noise generated by the fault component, providing key auxiliary judgment basis for thermal fault early warning and enhancing the sensitivity to early weak faults; noise energy proportion; noise instantaneous frequency The calculation logic is as follows: , wherein is the sampling interval, is the instantaneous phase, and unwrap is a phase unwrapping function used to eliminate phase mutations; The device vibration tower identifies vibration features through differential geometric manifold algorithm and quantum resonance analysis, and the specific identification steps are as follows: S001: Obtain the device vibration signal xc, select the embedding dimension mc by the false nearest neighbor method, set the time delay rk, and construct the phase space matrix by the embedding dimension and the time delay, and the construction logic is: , where K is the total sampling time point of the vibration device, and each row of the matrix corresponds to a phase space point; S002: Obtain the neighborhood points corresponding to each phase space point by the k-nearest neighbor algorithm, and perform PCA dimension reduction on the neighborhood points to obtain the local tangent space; It should be noted that the core significance of the local tangent space is to capture the low-dimensional intrinsic structure of high-dimensional vibration data, and to provide key geometric feature support for identifying device minor faults and nonlinear vibration patterns; PCA (Principal Component Analysis) is a commonly used dimension reduction algorithm, which maps high-dimensional data to low-dimensional space through linear transformation, while reducing data dimension while preserving main features of data; The role of PCA is to reduce the dimension of the phase space neighborhood points obtained by the k-nearest neighbor algorithm to obtain the local tangent space, which can effectively extract key information from high-dimensional vibration data, eliminate redundant noise, and accurately capture the low-dimensional intrinsic structure of data, thereby identifying minor faults of the device, such as early bearing wear and component loosening; S003: Quantum resonance analysis algorithm performs Fourier transform on the device vibration signal to obtain the vibration frequency spectrum , the vibration frequency spectrum is embedded into the potential function , and the calculation logic is: , where is a small change, used to represent the integral range, the spatial distribution of vibration energy, and abnormal vibration will cause local peak value of the potential function; S004: Substitute the potential function into the Schrödinger equation to obtain the vibration Hamiltonian , and the calculation logic is: , where hr is the reduced Planck constant, and mc is the equivalent mass, which is obtained from the database by extracting the mass of the device components, is the Laplace operator; Based on the Hamiltonian, the vibration characteristic equation is constructed and solved, and the equation formula is: , the vibration energy EC is obtained, and the intrinsic frequency is obtained by dividing the reduced Planck constant by the normal intrinsic frequency obtained from the historical data, and the absolute value of the intrinsic frequency and the normal intrinsic frequency is obtained. The intrinsic frequency deviation is obtained; The intrinsic frequency deviation is a key indicator reflecting the device structure state and fault characteristics, which often occurs earlier than the changes of vibration amplitude and temperature and other macroscopic characteristics, and can be used as an auxiliary signal for temperature fault to realize early warning of temperature fault; The gas concentration tower extracts gas characteristics by fractional order diffusion field reconstruction, specifically: G001: UE that acquires gas concentration data in the computer room. Its sampling points cover key areas of the computer room, such as gaps between server racks, air conditioning vents, and the bottom of equipment. A fractional-order diffusion equation is constructed based on the gas concentration data. The calculation logic is as follows: Where is the air diffusion coefficient of the computer room. If the source term is... A value of 0 indicates no leakage, while a value greater than 0 indicates the presence of leakage. For time t fractional derivative; G002: Based on the fractional diffusion equation, the fractional diffusion field under normal conditions is obtained. It can be trained using historical fault-free data. The deviation between the real-time field and the normal field is marked as the diffusion anomaly index. The power spectrum distribution curve, noise energy ratio, instantaneous frequency, local tangent space, intrinsic frequency deviation, and diffusion anomaly index are integrated into an auxiliary judgment feature set; Calculate the joint non-extensive entropy of the feature xm of the device's global temperature feature set and the feature xv of the auxiliary decision feature set. Its calculation logic is as follows: , of which Summing the square of the joint probability distribution, where q is a non-extensive parameter used to describe non-equilibrium states and relationships with long-range correlations or nonlinearities, such as the coupling relationship between temperature and vibration in a computer room; embedding the joint non-extensive entropy into the mutual information formula to obtain the correlation coefficient between any feature in the auxiliary judgment feature set and the features of the equipment's global temperature feature set. The logic for calculating its correlation coefficient is as follows: , The larger the value, the stronger the nonlinear correlation between the two features; the auxiliary features are screened by arranging the features of the auxiliary judgment feature set according to the magnitude of the correlation coefficient, and retaining the first m auxiliary features to obtain the screened auxiliary feature set; Using features from the auxiliary feature set and the device's global temperature feature set as nodes, a feature association graph is constructed with the association coefficient as the initial edge weight. This feature association graph is then input into a graph attention model, which outputs attention coefficients. Specifically, the graph attention model works by applying attention coefficients to each temperature node. Calculate its connection with all connected auxiliary nodes. Attention coefficient Its calculation logic is as follows: ,in and Here, m1 represents the original feature vectors of the auxiliary nodes and temperature nodes, m2 and m3 are the indices of the auxiliary nodes and temperature nodes, and WT is the shared weight matrix, which is obtained by mapping the original features to a high-dimensional space. This is the attention parameter vector, used to learn the importance of feature associations. For activation functions; The greater, the stronger the auxiliary feature explains the temperature feature; all attention coefficients are obtained and two-layer GAT stacks are performed to obtain global correlation features, the global correlation features are maximum-pooled to obtain a pooling result, and the pooling result and feature nodes are spliced to obtain a high-dimensional fault feature pool.

[0015] Step four, fiber bundle model prediction: a fiber bundle model is constructed with the machine room equipment topology as the base space and the high-dimensional fault feature pool as the fiber vector, and a simulated annealing algorithm is used to optimize the model parameters, taking the F1 score as the target, iteratively updating the parameters according to the Metropolis criterion, and then calculating the feature propagation anomaly degree according to the optimized model to determine and locate the hot fault; The physical topology of the machine room equipment is taken as the base space B, such as the location distribution of servers, switches and cooling systems, which are discretized into grid nodes b, each grid node corresponds to a device component or region, and the topological relationship of the base space is represented by an adjacency matrix. If two grid points are physically connected, the corresponding adjacency matrix is 1, and vice versa. The fiber vector is the high-dimensional fault feature pool of any grid node b, and the fiber vector is obtained by mapping the high-dimensional fault feature pool to the base space; The liaison coefficient is a key component of the fiber bundle feature propagation algorithm, which quantifies the propagation rule of the high-dimensional fault feature pool in the device topology space. The core idea is that the propagation strength of the feature between adjacent nodes is positively correlated with the feature difference and negatively correlated with the physical distance. The calculation logic of the liaison coefficient is: wherein and are the numbers of any two nodes, is the liaison coefficient, is the topological constraint coefficient, which controls the correlation between propagation strength and physical distance, and dm is the physical distance, thereby completing the construction of the fiber bundle model; Simulated annealing algorithm is used to optimize the fiber bundle parameters. The initial topological constraint coefficient is marked as the parameter to be optimized, which is provided by historical data. The thermal fault prediction performance of the fiber bundle model is marked as the thermal energy value HM, and the lower the thermal energy value, the better the parameter. Random perturbation is performed on the parameter to be optimized to generate a new parameter to be optimized. The new thermal energy value of the new parameter to be optimized is obtained. The thermal energy difference is obtained by subtracting the current thermal energy value from the new thermal energy value. If the thermal energy difference is less than 0, the new parameter to be optimized is directly accepted, and the current parameter is updated to the new parameter. If the thermal energy difference is greater than 0, the random perturbation is continued. When the number of iterations reaches the preset value, the optimization of the topological constraint coefficient is stopped, and the optimal topological constraint coefficient is output to the fiber bundle model. Extract the liaison coefficient of all adjacent nodes of each node in the base space to form several local liaison coefficient sets, calculate the liaison coefficient mean value of each node, obtain the maximum deviation value of the liaison coefficient and the liaison coefficient mean value of the node, mark the maximum deviation value as the thermal fault abnormality degree, and obtain the thermal fault abnormality degree of each node in the base space; extract a preset abnormality threshold in the database, and determine a node with a thermal fault abnormality degree exceeding the preset abnormality threshold as a potential fault point; spatial clustering is performed on the potential fault point, and adjacent fault points with a physical distance less than a preset distance are aggregated as a predicted fault area; Step five, digital twin fault scenario mapping: aligning the base space physical coordinates with the digital twin three-dimensional coordinates, importing the optimized fiber cluster feature data; simulating the future thermal diffusion trend to generate dynamic thermal maps and vibration cloud maps; The base space physical coordinates are accurately aligned with the three-dimensional coordinates of the digital twin model, and the optimized fiber cluster feature data such as abnormality degree and liaison coefficient are imported; the future thermal diffusion trend is simulated through the digital twin to generate visual dynamic thermal maps and vibration cloud maps; the inspection robot sends the predicted fault area, dynamic thermal maps and vibration cloud maps to the client.

[0016] The above formulas are all dimensionless numerical calculations, and specific dimensionless can be standardized, etc., which will not be described here. The formulas are obtained by collecting a large amount of data of equipment room equipment (such as infrared thermal image, vibration, gas concentration, etc.) through software simulation, which fits the real situation of thermal fault early warning scene, wherein the preset parameters (such as wavelet decomposition scale, abnormality threshold, etc.) are set by the person skilled in the art according to the actual situation of the equipment room equipment type, layout, etc.

[0017] The above embodiments can be realized all or partially by software, hardware, firmware or other any combination. If realized by software, it can be presented in the form of computer program product, which contains one or more computer instructions or programs. Loading or executing these instructions or programs on a computer can generate all or part of the processes or functions described in the embodiments of the method (such as voxelized temperature distribution generation, feature correlation graph construction, fiber cluster model parameter optimization, etc.). The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. Computer instructions can be stored in a computer-readable storage medium, or transmitted from one medium to another, for example, from one website, computer, server or data center to another through wired (such as cable) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium accessible by the computer, or a data storage device such as a server, data center, etc. containing a set of one or more available media, and the available media includes magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as DVD), semiconductor media (such as solid state disk) and the like.

[0018] It should be understood that the sequence numbers of the steps in each embodiment of the method do not represent the execution sequence, and the execution sequence is determined by the functions and inherent logic (for example, the data needs to be collected by the multi-modal sensor first, then the feature extraction is performed, and finally the fault warning is realized, which does not constitute a limitation on the implementation process.

[0019] Those of ordinary skill in the art can realize that the example units and algorithm steps (such as the power spectrum calculation of the noise tower, the phase space matrix construction of the vibration tower, and the solution of the connection coefficient of the fiber bundle model) described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can adopt different methods to implement the described functions for each specific application, but it should not be considered that it is beyond the scope of the method.

[0020] In several embodiments provided by the method, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-mentioned device embodiments are only schematic, and the unit division is only a logical function division (for example, the feature extraction is divided into temperature features, auxiliary features, etc. The module), and actually, there can be other division manners, such as a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling, direct coupling or communication connection shown or discussed can be implemented through an interface, and the indirect coupling or communication connection between devices or units can be electrical, mechanical or other forms (for example, the vibration signal of the device is transmitted through a mechanical structure).

[0021] The units described as separate components can be physically separated or not, and the components shown as units can be physically units or distributed to multiple network units, and some or all units can be selected according to actual needs to achieve the purpose of the embodiment (for example, only the features of the key equipment area are selected for fault warning).

[0022] In addition, each functional unit in each embodiment of the method can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit (for example, the digital twin mapping and the warning output are integrated into the same module).

[0023] If the function is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this, the contribution of the method technical solution to the prior art or part of the technical solution can be embodied as a software product. The computer software product is stored in a storage medium and contains a number of instructions for enabling a computer device (such as a personal computer, a server, a network device, etc.) to execute all or part of the steps described in the embodiments of the method (such as data acquisition, feature fusion, fault positioning, digital twin mapping, etc.). The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, and other program code storage media.

[0024] The above is only a specific implementation of the method, but the protection scope is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the method, which should be covered within the protection scope. Therefore, the protection scope of the method should be subject to the protection scope of the claims.

Claims

1. A method for thermal fault early warning of a multi-modal perception machine room inspection robot, characterized in that, The method comprises the following steps: Collecting infrared thermal images, acoustic, vibration and gas concentration data of the machine room by a robot-mounted multi-modal sensor array, generating voxelized temperature distribution by a three-dimensional temperature field reconstruction algorithm; Through wavelet transform, the voxelized temperature distribution is multi-scale decomposed, the Hurst index features of each scale are extracted by combining the fractional Brownian motion algorithm, and the wavelet energy features are fused to output the device global temperature feature set containing the device surface and internal heat distribution; The inspection robot constructs a noise tower, a device vibration tower and a gas concentration tower, outputs corresponding features and integrates them into an auxiliary judgment feature set, fuses the auxiliary judgment features with the device global temperature feature set through a non-extensive entropy algorithm and a graph attention model, and constructs a high-dimensional temperature fault feature pool; A fiber bundle model is constructed with the machine room device topology as the base space and the high-dimensional fault feature pool as the fiber vector, the model parameters are optimized by a simulated annealing algorithm, the F1 score is taken as the target, the parameters are iteratively updated by the Metropolis criterion, and the feature propagation anomaly degree is calculated according to the optimized model to determine and locate the thermal fault; Based on the alignment of the spatial physical coordinates and the digital twin three-dimensional coordinates, the optimized fiber bundle feature data is imported, the future heat diffusion trend is simulated, and a dynamic thermal map and a vibration cloud chart are generated.

2. The method for thermal fault early warning of the multimodal perception machine room inspection robot according to claim 1, characterized in that, The inspection robot constructs a noise tower, a device vibration tower and a gas concentration tower, outputs corresponding features and integrates them into an auxiliary judgment feature set, and the specific steps are as follows: The noise tower identifies the features through power spectrum and empirical mode decomposition, frames the device running noise and adds a Hanning window, performs fast Fourier transform on each frame signal by the Welch method, calculates the corresponding power spectrum, and obtains the power spectrum distribution curve by averaging multiple power spectra; the device running noise is subjected to empirical mode decomposition to obtain a plurality of intrinsic mode functions, and the noise energy proportion and instantaneous frequency of each intrinsic mode function are obtained; The device vibration tower identifies the vibration features through differential geometry manifold algorithm and quantum resonance analysis, obtains the device vibration signal, selects the embedding dimension by the false nearest neighbor method, sets the time delay, constructs the phase space matrix by the embedding dimension and the time delay, and each row of the matrix corresponds to a phase space point; the neighborhood points corresponding to each phase space point are obtained by the k-neighbor algorithm, the field points are subjected to PCA dimension reduction to obtain the local tangent space; the quantum resonance analysis algorithm performs Fourier transform on the device vibration signal to obtain the vibration frequency spectrum, and embeds the vibration frequency spectrum into the potential energy function; then the potential energy function is substituted into the Schrödinger equation to obtain the vibration Hamiltonian; the vibration feature equation is constructed based on the Hamiltonian and solved to obtain the vibration energy, which is divided by the reduced Planck constant to obtain the intrinsic frequency; the intrinsic frequency and the normal intrinsic frequency are obtained from the historical data, and the absolute value of the intrinsic frequency deviation is obtained by taking the absolute value of the intrinsic frequency and the normal intrinsic frequency; The gas concentration tower extracts gas features by fractional order diffusion field reconstruction, and obtains the gas concentration data of the machine room; Based on the fractional order diffusion equation, the fractional order diffusion field of the normal state is obtained by training the historical fault-free data, and the deviation between the real-time field and the normal field is marked as the diffusion anomaly index; the power spectrum distribution curve, noise energy ratio, instantaneous frequency, local tangent space, eigenfrequency deviation and diffusion anomaly index are integrated into the auxiliary judgment feature set.

3. The method of Claim 2, wherein the method further comprises: The auxiliary feature set is screened through the non-extensive entropy algorithm, specifically: Calculate the joint non-extensive entropy of the features of the device global temperature feature set and the features of the auxiliary judgment feature set; embed the joint non-extensive entropy into the mutual information formula to obtain the correlation coefficient of any feature in the auxiliary judgment feature set to the device global temperature feature set, screen the auxiliary features, and arrange the features of the auxiliary judgment feature set according to the correlation coefficient size, and retain the top m auxiliary features to obtain the screened auxiliary feature set.

4. The method for thermal fault early warning of the multimodal perception machine room inspection robot according to claim 3, characterized in that, The specific steps of outputting the high-dimensional fault feature pool through the graph attention algorithm are as follows: Take the features of the screened auxiliary feature set and the device global temperature feature set as nodes, and take the correlation coefficient as the initial edge weight to construct a feature correlation graph, input the feature correlation graph into the graph attention model and output the attention coefficient, the graph attention model specifically calculates the attention coefficient of each temperature node with all connected auxiliary nodes, thereby obtaining all attention coefficients and performing two-layer GAT stacking to obtain global correlation features, and then performing maximum pooling on the global correlation features to obtain the pooling result, and then performing concatenation on the pooling result and the feature node to obtain the high-dimensional fault feature pool.

5. The method for thermal fault early warning of the multimodal perception machine room inspection robot according to claim 1, characterized in that, A fiber bundle model is constructed with the machine room equipment topology as the base space and the high-dimensional fault feature pool as the fiber vector, specifically: Take the physical topology of the machine room equipment as the base space, and discretize it into a grid-like node, each grid node corresponds to a device component or region, and the topological relationship of the base space is represented by an adjacency matrix, if two grid points are physically connected, the corresponding adjacency matrix is 1, otherwise it is 0; set the fiber vector as the high-dimensional fault feature pool of any grid node; then obtain the connection coefficient of the base space to complete the construction of the fiber bundle model; wherein the fiber vector is obtained by mapping the high-dimensional fault feature pool to the base space.

6. The method for thermal fault early warning of the multimodal perception machine room inspection robot according to claim 5, characterized in that, Optimize the model parameters through the simulated annealing algorithm, and then calculate the feature propagation anomaly degree according to the optimized model to determine and locate the thermal fault, specifically: Simulated annealing algorithm optimizes the fiber bundle parameters, marks the initial topological constraint coefficient as the optimized parameter, the initial topological constraint coefficient is provided by historical data, marks the thermal fault prediction performance of the fiber bundle model as the thermal energy value, randomly perturbs the optimized parameter to generate a new optimized parameter, and obtains the new thermal energy value of the new optimized parameter, and the thermal energy difference is obtained by subtracting the new thermal energy value from the current thermal energy value, if the thermal energy difference is less than zero, the new optimized parameter is directly accepted, and the current parameter is updated to the new parameter; If the thermal energy difference is greater than zero, continue to perform random perturbation, and when the iteration number reaches the preset value, stop iteration and output the optimal topological constraint coefficient to the fiber bundle model; Extract the liaison coefficient of all adjacent nodes of each node in the base space to form a number of local liaison coefficient sets, calculate the liaison coefficient mean of each node, obtain the maximum deviation value of the liaison coefficient and the liaison coefficient mean of the node, mark the maximum deviation value as the thermal fault abnormality degree, and obtain the thermal fault abnormality degree of each node in the base space; extract the preset abnormality threshold in the database, and determine the node with a thermal fault abnormality degree exceeding the preset abnormality threshold as a potential fault point; Spatial clustering is performed on the potential fault points, and adjacent fault points with a physical distance less than a preset distance are aggregated as a predicted fault area.

7. The method for thermal fault early warning of the multimodal perception machine room inspection robot according to claim 6, characterized in that, The base space physical coordinates are aligned with the digital twin three-dimensional coordinates, the future thermal diffusion trend is simulated, and a dynamic thermal map and a vibration cloud chart are generated, specifically: The base space physical coordinates are accurately aligned with the three-dimensional coordinates of the digital twin model, the optimized fiber cluster feature data including abnormality and liaison coefficient are imported; the future thermal diffusion trend is simulated through the digital twin to generate a visual dynamic thermal map and a vibration cloud chart; the inspection robot sends the predicted fault area, the dynamic thermal map and the vibration cloud chart to the client. 8.The method of claim 1, wherein the method further comprises: determining a temperature of the machine room based on the sensed temperature; and determining whether the temperature of the machine room is higher than a threshold temperature. The infrared thermal image, acoustic, vibration and gas concentration data of the machine room are collected by the multi-modal sensor array carried by the robot, specifically: The inspection robot moves according to the preset inspection path, and the multi-modal sensor array starts collecting data when it reaches each monitoring point. The modal sensor array includes an infrared thermal imager, an acoustic sensor, a vibration sensor and a gas sensor, which can obtain real-time machine room equipment data sets, including infrared images, equipment operating noise, equipment vibration signals and gas concentration around the equipment. The collected multiple infrared images are arranged to obtain an infrared image sequence. 9.The thermal fault pre-warning method of the multimodal perception machine room inspection robot according to claim 8, characterized in that, Obtain the structured light depth map, and generate the voxelized temperature distribution using the three-dimensional temperature field reconstruction algorithm, specifically: A binocular structured light camera scheme is adopted, which consists of two infrared cameras and an infrared projector. The infrared projector emits a preset pattern of infrared light, and the camera and the projector are installed at the front end of the robot. The infrared projector projects the pre-encoded structured light pattern onto the surface of the cabinet. The light is deformed due to the shape or distance of the equipment surface. Two infrared cameras synchronously collect images with deformed patterns. Each time an image is collected, the left and right cameras generate one image each, forming a binocular image pair. The collected binocular images are subjected to epipolar rectification processing. Through camera calibration parameters, the image coordinates are converted to the epipolar constraint plane to ensure that the corresponding point search is only performed in the same row. A block matching algorithm is used to search for matching points in the epipolar rectified binocular images. Taking a certain pixel point in the left image as the center, a certain size of image block is taken, and the image block with the highest similarity in the corresponding row of the right image is searched to determine the matching point pair to generate a disparity image. According to the principle of triangulation, combined with the baseline distance, camera focal length and disparity image obtained by camera calibration, the depth values of each point on the surface of the cabinet are calculated pixel by pixel, and finally a structured light depth image is generated. The structured light depth image and the infrared image sequence are input into the U-Net-LSTM network, the encoder extracts multi-scale temperature features through convolution, and outputs a temperature feature map, the LSTM layer captures the temperature change line in the time dimension through a bidirectional LSTM, and the decoder outputs a voxelized temperature distribution.

10. The method for thermal fault early warning of the multimodal perception machine room inspection robot according to claim 1, characterized in that, The process of outputting the device global temperature feature set containing the device surface and internal heat distribution is as follows: The voxelized temperature distribution is decomposed into multiple layers through low-pass filters and high-pass filters through a wavelet algorithm, and the decomposition scales of each filter are set as microscopic, mesoscopic and macroscopic scales; the low-frequency quantity of the decomposition scale is obtained through the decomposition calculation logic of the low-pass filter recursive rule; The high-frequency quantity is obtained through the decomposition calculation logic of the high-pass filter recursive rule; The cabinet surface and internal area are marked as area and area respectively, the Hurst index is calculated at k scales under wavelet decomposition, the voxelized temperature distribution is in the spatial grid at scale k, for any spatial position, the time series of the spatial position is extracted; the mean value of the temperature in the time dimension at the spatial position is calculated, and the cumulative deviation of the temperature sequence is obtained based on the mean value; the range is obtained by subtracting the maximum value from the minimum value of the subsequence length, and the standard deviation of the subsequence length is calculated; the range is divided by the standard deviation to obtain the re-standardized range; the least square method is used for fitting; the above steps are repeated for each scale to obtain the Hurst index of each scale; The instantaneous thermal mutation features at each scale are extracted through the high-frequency quantity, the energy of the high-frequency quantity at each scale is calculated, the modulus maximum points in the high-frequency quantity are located, the hotspot coordinates and the corresponding temperature peak value are extracted; the voxelized temperature distribution, fractional Brownian motion features and wavelet features are cross-scale fused to construct the device global temperature feature set.

Images