An elevator fault classification method and system

Abstract

The present invention relates to the technical field of fault classification, and particularly relates to an elevator fault classification method and system, which includes: collecting multi-source heterogeneous data during the operation of the elevator in real time, including mechanical vibration signals, three-phase current waveform signals of the drive motor, car video behavior data, and control cabinet instruction streams; performing spatio-temporal alignment processing on the multi-source heterogeneous data to generate a standardized time series data matrix; extracting multi-dimensional fault features from the standardized time series data matrix; constructing a dynamic knowledge graph based on a historical fault case library, and the knowledge graph includes a three-layer topological structure of fault feature nodes, fault type nodes, and maintenance plan nodes; matching the extracted multi-dimensional fault features with the dynamic knowledge graph, and using an improved residual neural network model to output the final fault classification result and confidence level. The present invention significantly improves the accuracy, real-time responsiveness, and interpretability of fault diagnosis, and has good engineering practical value and application promotion prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of fault classification, and particularly to an elevator fault classification method and system. Background Art

[0002] With the high-density development of urban buildings and the wide deployment of intelligent buildings, the elevator system, as an important part of vertical transportation, its safety and stability have been increasingly emphasized; the existing elevator operation and maintenance rely on periodic inspections and passive response mechanisms based on abnormal logs, and it is difficult to detect potential fault risks in a timely manner. In recent years, although some systems have introduced sensors to collect signals such as vibration and current for preliminary diagnosis, most methods only rely on single-modal data or static rule bases, and it is difficult to achieve dynamic recognition of complex fault situations, especially when facing problems such as difficulties in fusing multi-source heterogeneous data, implicit expression of abnormal behaviors, and co-occurrence of multiple types of faults, and there are limitations in both accuracy and practicality.

[0003] Under this background, there is an urgent need for an intelligent fault diagnosis solution that can fuse multi-modal operation data and has the capabilities of spatio-temporal modeling and graph knowledge reasoning to achieve an end-to-end fault classification process from data perception, feature extraction to decision-making reasoning; combining knowledge graph technology and deep learning structures, historical case experience can be structurally embedded into the graph network, and semantic reasoning of the current operating state can be achieved through the matching relationship between feature nodes and graph nodes; furthermore, by introducing a residual neural network model, the expression ability of complex fault features and the classification confidence control ability can be enhanced, thereby significantly improving the accuracy, real-time performance and interpretability of elevator fault detection. Summary of the Invention

[0004] Based on the above purposes, the present invention provides an elevator fault classification method.

[0005] An elevator fault classification method includes the following steps:

[0006] S1: Real-time collect multi-source heterogeneous data of elevator operation, including mechanical vibration signals, three-phase current waveform signals of drive motors, car video behavior data, and control cabinet instruction streams;

[0007] S2: Perform spatio-temporal alignment processing on the multi-source heterogeneous data to generate a standardized time-series data matrix;

[0008] S3: Extract multi-dimensional fault features from the standardized time-series data matrix, including joint time-frequency domain features of vibration signals, harmonic distortion features of current waveforms, and abnormal behavior features of video data;

[0009] S4: Construct a dynamic knowledge graph based on a historical fault case library, and the knowledge graph includes a three-layer topological structure of fault feature nodes, fault type nodes, and maintenance plan nodes;

[0010] S5: Match the extracted multi-dimensional fault features with the dynamic knowledge graph, dynamically adjust the weights of the graph nodes according to the matching degree, and based on the optimized weights of the knowledge graph nodes, use the improved residual neural network model to output the final fault classification result and confidence level.

[0011] Further, the S1 includes:

[0012] S11, Install a triaxial acceleration sensor on the surface of the elevator traction machine housing and collect mechanical vibration signals at a sampling rate of 1 kHz;

[0013] S12, Collect the three-phase current waveform signals of the drive motor through a current transformer;

[0014] S13, Install a video acquisition device on the top of the car to collect the video behavior data of passengers in real time;

[0015] S14, Access the CAN bus interface of the elevator control cabinet to collect the control cabinet instruction stream in real time.

[0016] Further, the S2 includes:

[0017] S21, Perform timestamp normalization processing on the multi-source heterogeneous data;

[0018] S22, Construct a global unified time axis according to the complete set of timestamps in all data sources;

[0019] S23, Perform normalization processing on the alignment results of each data source;

[0020] S24, Concatenate all the normalized data sources into a standardized time series data matrix.

[0021] Further, the S3 includes:

[0022] S31, For the vibration signal, use the short-time Fourier transform to extract its time-frequency domain joint features;

[0023] S32, For the current waveform signal, use the harmonic distortion rate to extract the harmonic distortion features of the current waveform;

[0024] S33, For the video data, use the inter-frame difference method to extract the behavior anomaly features and calculate the behavior motion energy;

[0025] S34, Concatenate the time-frequency domain joint features of the vibration signal, the harmonic distortion features of the current waveform, and the behavior anomaly features of the video data to form multi-dimensional fault features.

[0026] Further, the extraction of the time-frequency domain joint features of the vibration signal in the S31 includes:

[0027] S311. Map the normalized mechanical vibration signal to the time-frequency domain using the short-time Fourier transform to obtain the energy spectral density function;

[0028] S312. Extract time-series feature indicators from the energy spectrum, including the main frequency distribution, instantaneous frequency, spectral energy concentration, and frame energy;

[0029] S313. Statistically summarize the time-series feature indicators with a sliding window length to construct a joint time-frequency domain feature vector.

[0030] Further, the S4 includes:

[0031] S41. Extract the fault feature vector, corresponding fault type label, and maintenance plan information of each case from the historical fault case database to construct a historical fault case set;

[0032] S42. Cluster the feature vectors in the historical data to generate fault feature nodes, and combine the fault type labels and maintenance plan records to construct fault type nodes and maintenance plan nodes respectively, forming a three-layer topological structure including three types of nodes;

[0033] S43. Establish directed edges in the graph according to the similarity relationship between the fault feature nodes and the fault type nodes and the co-occurrence relationship between the fault type nodes and the maintenance plan nodes, and assign weights to construct a dynamic knowledge graph.

[0034] Further, the S5 includes:

[0035] S51. Node matching and weight adjustment: Match the extracted multi-dimensional fault features with the fault feature nodes in the knowledge graph, and determine the degree of relevance of the fault type nodes according to the matching results, and dynamically adjust the weights of each fault type node in the knowledge graph;

[0036] S52. Fault classification inference output: Based on the adjusted weights of the knowledge graph nodes, fuse the current fault features with the graph weights and input them into the improved residual neural network model to output the corresponding fault classification results and their confidence levels.

[0037] Further, the S51 includes:

[0038] S511. Calculate the similarity score between the current feature vector and each fault feature node in the graph using the exponential distance function;

[0039] S512. Use the similarity score as the activation coefficient of the fault feature node, and through the graph propagation mechanism, transfer the activation value to the set of fault type nodes to obtain the updated weight of each fault type node.

[0040] Further, the S52 includes:

[0041] S521, Concatenate the current fault feature vector with the adjusted fault type node weight vector to form a joint input vector;

[0042] S522, Use the joint input vector as the input of the residual neural network, which is composed of multiple stacked residual modules;

[0043] Input the output vector of the last layer into the Softmax classifier to calculate the classification probabilities of each node in the fault type node set, and obtain the final fault classification result and confidence level.

[0044] An elevator fault classification system for implementing the above-mentioned elevator fault classification method, including the following modules:

[0045] Data acquisition module: Real-time collect multi-source heterogeneous data of elevator operation, including mechanical vibration signals, three-phase current waveforms of drive motors, car video behavior data, and control cabinet instruction streams;

[0046] Spatio-temporal alignment module: Perform spatio-temporal alignment processing on the multi-source heterogeneous data to generate a standardized time series data matrix;

[0047] Feature extraction module: Extract multi-dimensional fault features from the standardized time series data matrix, where the fault features include the joint time-frequency domain features of vibration signals, the harmonic distortion features of current waveforms, and the abnormal behavior recognition features of video data;

[0048] Knowledge graph construction module: Construct a dynamic knowledge graph based on the historical fault case library, where the knowledge graph includes a three-layer topological structure of fault feature nodes, fault type nodes, and maintenance plan nodes;

[0049] Graph matching and reasoning module: Match the currently extracted multi-dimensional fault features with the dynamic knowledge graph, dynamically adjust the graph node weights according to the matching degree, and based on the optimized graph node weights, use an improved residual neural network model to output the fault classification result and its confidence level.

[0050] Advantages of the present invention:

[0051] An elevator fault classification method provided by the present invention can integrate multi-source heterogeneous operation information such as mechanical vibration signals, current waveforms, video behavior data, and control instructions, construct a standardized time series data matrix through a unified spatio-temporal alignment mechanism, and extract multi-dimensional features, including time-frequency joint features, harmonic distortion indexes, and video behavior patterns, overcoming the problems of traditional methods relying on single sensing signals and having weak recognition ability for complex abnormal behaviors, and effectively improving the comprehensiveness of feature expression and the sensitivity of fault characterization.

[0052] The three - layer dynamic knowledge graph constructed based on historical cases in the present invention can achieve the structured modeling of the semantic relationship of elevator faults. Through the matching mechanism between fault feature nodes and graph nodes, combined with graph propagation and dynamic adjustment of node weights, the fusion reasoning of prior knowledge and the current state is realized. On this basis, a residual neural network structure is introduced to enhance the deep - feature learning ability and improve the model stability. It can output classification results with confidence control, significantly improving the accuracy, real - time responsiveness and interpretability of fault diagnosis, and having good engineering practical value and application promotion prospects. Brief Description of the Drawings

[0053] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0054] Figure 1 It is the flowchart of the method for the embodiment of the present invention;

[0055] Figure 2 It is the system module diagram of the embodiment of the present invention. Detailed Embodiments

[0056] To make the objectives, technical solutions and advantages of the present invention more clear and understandable, the following further details the present invention in combination with specific embodiments.

[0057] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those of ordinary skill in the art to which the present invention belongs. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. The terms such as "connected" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "up", "down", "left", "right" are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0058] As Figure 1 shown, a method for classifying elevator faults includes the following steps:

[0059] S1: Real-time collect multi-source heterogeneous data of elevator operation, including mechanical vibration signals, three-phase current waveform signals of the drive motor, car video behavior data, and control cabinet instruction streams;

[0060] S2: Perform spatio-temporal alignment processing on the multi-source heterogeneous data to generate a standardized time-series data matrix;

[0061] S3: Extract multi-dimensional fault features from the standardized time-series data matrix, including the joint time-frequency domain features of vibration signals, the harmonic distortion features of current waveforms, and the behavior anomaly features of video data;

[0062] S4: Construct a dynamic knowledge graph based on the historical fault case library. The knowledge graph includes a three-layer topological structure of fault feature nodes, fault type nodes, and maintenance plan nodes;

[0063] S5: Match the extracted multi-dimensional fault features with the dynamic knowledge graph, dynamically adjust the weights of the graph nodes according to the matching degree, and based on the optimized knowledge graph node weights, use an improved residual neural network model to output the final fault classification result and confidence level.

[0064] S1 includes:

[0065] S11, Install a three-axis acceleration sensor on the surface of the elevator traction machine housing to collect mechanical vibration signals at a sampling rate of 1 kHz. The vector representation of the mechanical vibration signal at time is:

[0066] ;

[0067] Among them, respectively represent the instantaneous acceleration components along the axis;

[0068] S12, Collect three-phase current waveform signals of the drive motor through a current transformer. The definition of the three-phase current signal at time is expressed as:

[0069] ;

[0070] Among them, respectively represent the instantaneous current values of the motor , , for the three phases. Based on the fast Fourier transform (FFT), extract its frequency domain components to calculate the total harmonic distortion (THD), which is expressed as:

[0071] ;

[0072] Among them, is the fundamental wave amplitude, is the The amplitude of the subharmonics, is the upper limit order of harmonic analysis;

[0073] S13, collecting passenger video behavior data in real time through a video collection device installed on the top of the car;

[0074] S14, through the CAN bus interface of the elevator control cabinet, collects the control cabinet command stream in real time, which is expressed as:

[0075] ;

[0076] in, Indicates Control commands, including start and stop instructions, elevator floor target, door control status, the sampling period is less than 100ms.

[0077] S2 includes:

[0078] S21, standardize timestamps for multi-source heterogeneous data, sensors The collected original data sequence is expressed as:

[0079] ;

[0080] in, It is a sensor The original data sequence, Indicates The timestamp of each sampling point, represents the data vector corresponding to that moment, For sensor The number of sampling points;

[0081] S22, construct a global unified timeline based on the full set of timestamps in all data sources, expressed as:

[0082]

[0083] in, To unify the aligned global timeline, is the target uniform time step number, and the time interval is , mapping all data sources to a unified time axis through linear interpolation, sensors At the moment The alignment value is expressed as:

[0084] ;

[0085] in, , indicating that the encirclement is found Linear interpolation is performed on the two original time points. The sensor obtained by interpolation At time The aligned data value;

[0086] S23. Normalize the alignment results of each data source. Use the z-score normalization method, expressed as:

[0087] ;

[0088] Among them, Is the aligned and normalized sensor At time The data value, Are respectively the mean and standard deviation of all sampling points of the sensor On the aligned time axis;

[0089] S24. Concatenate all the normalized data sources into a standardized time series data matrix, expressed as:

[0090] ;

[0091] Among them, Represents the total number of data sources, and the final matrix , Is the total dimension after concatenating each data source.

[0092] S3 includes:

[0093] S31. For the vibration signal, use the short-time Fourier transform (STFT) to extract its time-frequency domain joint features, expressed as:

[0094] ;

[0095] Among them, Is the vibration signal under the unified time axis, Is the time variable used for the integral or summation of the short-time Fourier transform (STFT), Is a sliding window function (Hamming window) centered on time , Is the frequency variable, Represents the local spectral response at time Frequency ; Based on Construct a time-frequency energy spectrogram, and extract its main frequency distribution, instantaneous frequency, and energy center features as the time-frequency domain joint feature vector ;

[0096] S32. For the current waveform signal, use the total harmonic distortion (THD) to extract the harmonic distortion features of the current waveform, expressed as:

[0097] ;

[0098] Among them, is the three-phase current signal, , is the current signal The amplitude of the harmonic of the fundamental wave, obtained by FFT transformation, is the fundamental wave amplitude, ;

[0099] S33. For video data, the frame difference method is used to extract the abnormal behavior features. The gray difference between two consecutive frames of images is expressed as:

[0100] ;

[0101] Among them, is the image pixel coordinate, is the frame at the pixel position gray value, is the brightness change amount between the current frame and the previous frame at this pixel position, and are two consecutive frames of images;

[0102] Calculate the behavior movement energy , expressed as:

[0103] ;

[0104] Construct a movement energy sequence in a sliding window manner , and extract the energy mean, peak value, and mutation point as the video behavior feature vector ;

[0105] S34. Concatenate the time-frequency domain joint features of the vibration signal, the harmonic distortion features of the current waveform, and the abnormal behavior features of the video data to form multi-dimensional fault features, expressed as:

[0106] .

[0107] The time-frequency domain joint features extracted in S31 include:

[0108] S311. Use the short-time Fourier transform (STFT) to map the normalized mechanical vibration signal to the time-frequency domain to obtain the energy spectral density function , expressed as:

[0109] ;

[0110] S312. Extract the time-series feature metrics from the energy spectrum, including the dominant frequency distribution, instantaneous frequency, spectral energy concentration, and frame energy, which are expressed as:

[0111] 1) Dominant frequency distribution :

[0112] ;

[0113] 2) Instantaneous frequency (spectral centroid):

[0114] ;

[0115] 3) Spectral energy concentration (spectral entropy) :

[0116] ；

[0117] ;

[0118] 4) Frame energy :

[0119] ;

[0120] S313. Statistically summarize the time-series feature metrics with a sliding window length to construct a joint time-frequency domain feature vector, which is expressed as:

[0121] ;

[0122] where are the mean and standard deviation of the feature within the window, respectively.

[0123] S4 includes:

[0124] S41. Extract the fault feature vector, corresponding fault type label, and repair plan information for each case from the historical fault case library to construct a historical fault case set, which is expressed as:

[0125] ;

[0126] where represents the 'th historical fault case, including the multi-dimensional fault feature vector , the corresponding fault type label , and the repair plan description ;

[0127] S42. Cluster the feature vectors in the historical data to generate fault feature nodes. Combine the fault type labels and maintenance plan records to construct fault type nodes and maintenance plan nodes respectively, forming a three-layer topology structure including three types of nodes, specifically including:

[0128] (1) Fault feature node set :

[0129] ;

[0130] Among them, represents the center of different types of fault feature subspaces (obtained through clustering), represents the th representation vector of the feature node;

[0131] (2) Fault type node set :

[0132] ;

[0133] Among them, represents the known fault classification labels in history, represents the th representation vector of the fault type node;

[0134] (3) Maintenance plan node set :

[0135] ;

[0136] Among them, represents the standardized maintenance plan description associated with various fault types, represents the th representation vector of the maintenance plan node;

[0137] S43. Establish directed edges in the graph and assign weights according to the similarity relationship between the fault feature nodes and the fault type nodes and the co-occurrence relationship between the fault type nodes and the maintenance plan nodes, and construct a dynamic knowledge graph, expressed as:

[0138] ;

[0139] Among them, the fault type node set is , the edge set is , and the specific steps for establishing the edge set include:

[0140] (1) For the fault feature node and the fault type node , calculate their matching edge weights according to the Euclidean distance, expressed as:

[0141] ;

[0142] Among them, is the characteristic mean vector corresponding to the fault type , and is the characteristic distance scale parameter;

[0143] (2) For the fault type node and the maintenance plan node , the edge weight is normalized according to the historical co-occurrence times, expressed as:

[0144] ;

[0145] Among them, is the co-occurrence frequency of the fault type and the maintenance plan node in the case base.

[0146] S5 includes:

[0147] S51, Node matching and weight adjustment: Match the extracted multi-dimensional fault characteristics with the fault characteristic nodes in the knowledge graph, and determine the correlation degree of the fault type nodes according to the matching results, and dynamically adjust the weights of each fault type node in the knowledge graph;

[0148] S52, Fault classification inference output: Based on the adjusted knowledge graph node weights, fuse the current fault characteristics with the graph weights and input them into the improved residual neural network model to output the corresponding fault classification results and their confidence levels.

[0149] S51 includes:

[0150] S511, Calculate the similarity score between the current feature vector and each fault characteristic node in the graph using the exponential distance function, expressed as:

[0151] ;

[0152] Among them, is the matching degree scale adjustment parameter;

[0153] S512, Use the similarity score as the activation coefficient of the fault characteristic node, and through the graph propagation mechanism, transfer the activation value to the fault type node set , and obtain the updated weight of each fault type node, expressed as:

[0154] ;

[0155] Among them, Indicates a fault feature node The edge weight value between the fault type node and Indicates the dynamically adjusted weight of the fault type node after adjustment

[0156] S52 includes:

[0157] S521, concatenate the current fault feature vector with the adjusted weight vector of the fault type node to form a combined input vector , expressed as:

[0158] ;

[0159] S522, use the combined input vector as the input of the residual neural network, which consists of multiple stacked residual modules, and the output of each residual module is defined as:

[0160] ;

[0161] Among them, and are the weight matrix and bias vector of the th layer respectively, is the non-linear activation function;

[0162] Input the output vector of the last layer into the Softmax classifier to calculate the classification probability of each node in the fault type node set, expressed as:

[0163] ;

[0164] Among them, is the classification confidence of the fault type node , and are its classification weight and bias respectively;

[0165] Obtain the final fault classification result and confidence, expressed as:

[0166] ;

[0167] .

[0168] As Figure 2 shown, an elevator fault classification system for implementing the above elevator fault classification method includes the following modules:

[0169] Data acquisition module: Real-time acquisition of multi-source heterogeneous data of elevator operation, including mechanical vibration signals, three-phase current waveforms of drive motors, car video behavior data, and control cabinet instruction streams;

[0170] Spatio-temporal alignment module: Perform spatio-temporal alignment processing on multi-source heterogeneous data to generate a standardized time-series data matrix;

[0171] Feature extraction module: Extract multi-dimensional fault features from the standardized time-series data matrix. The fault features include joint time-frequency domain features of vibration signals, harmonic distortion features of current waveforms, and abnormal behavior recognition features of video data;

[0172] Knowledge graph construction module: Construct a dynamic knowledge graph based on the historical fault case library. The knowledge graph includes a three-layer topological structure of fault feature nodes, fault type nodes, and maintenance plan nodes;

[0173] Graph matching and reasoning module: Match the currently extracted multi-dimensional fault features with the dynamic knowledge graph, dynamically adjust the weights of graph nodes according to the matching degree, and based on the optimized weights of graph nodes, use an improved residual neural network model to output the fault classification result and its confidence level.

[0174] Those of ordinary skill in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the present invention is limited to these examples; under the concept of the present invention, the technical features in the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations in different aspects of the present invention as described above, which are not provided in detail for the sake of brevity.

Claims

1. A method for classifying elevator faults, characterized in that, The following steps are involved: S1: Real-time collection of multi-source heterogeneous data of elevator operation, including mechanical vibration signals, three-phase current waveform signals of drive motors, car video behavior data, and control cabinet command streams; S2: Performing spatiotemporal alignment processing on the multi-source heterogeneous data to generate a standardized time series data matrix; S3: Extract multi-dimensional fault features from the standardized time series data matrix, including the joint time-frequency domain features of the vibration signal, the harmonic distortion features of the current waveform, and the behavioral anomaly features of the video data; S4: construct a dynamic knowledge graph based on the historical fault case library, wherein the knowledge graph includes a three-layer topological structure of fault feature nodes, fault type nodes, and maintenance solution nodes; S5: Match the extracted multi-dimensional fault features with the dynamic knowledge graph nodes, dynamically adjust the graph node weights according to the matching degree, and use the improved residual neural network model to output the final fault classification results and confidence based on the optimized knowledge graph node weights.

2. The elevator fault classification method according to claim 1, wherein The S1 includes: S11, using a triaxial acceleration sensor installed on the surface of the elevator traction machine housing to collect mechanical vibration signals at a sampling rate of 1kHz; S12, collecting three-phase current waveform signals of the driving motor through a current transformer; S13, collecting passenger video behavior data in real time through a video collection device installed on the top of the car; S14, through accessing the CAN bus interface of the elevator control cabinet, collects the control cabinet command stream in real time.

3. A method for classifying elevator faults according to claim 2, characterized in that, The S2 includes: S21, perform timestamp standardization on multi-source heterogeneous data; S22, constructing a global unified timeline based on the full set of timestamps in all data sources; S23, normalizing the alignment results of each data source; S24, concatenate all normalized data sources into a standardized time series data matrix.

4. A method for classifying elevator faults according to claim 3, characterized in that, The S3 includes: S31, for vibration signals, short-time Fourier transform is used to extract its joint features in time and frequency domains; S32, extracting harmonic distortion characteristics of the current waveform using harmonic distortion rate for the current waveform signal; S33, for video data, using the frame difference method to extract abnormal behavior features and calculate the behavior motion energy; S34, the time-frequency domain joint features of the vibration signal, the harmonic distortion features of the current waveform, and the behavioral abnormality features of the video data are spliced to form a multi-dimensional fault feature.

5. A method for classifying elevator faults according to claim 4, characterized in that, Extracting the time-frequency domain joint features of the vibration signal in S31 includes: S311, mapping the normalized mechanical vibration signal to the time-frequency domain using short-time Fourier transform to obtain an energy spectral density function; S312, extracting time series characteristic indicators from the energy spectrum, including main frequency distribution, instantaneous frequency, spectrum energy concentration, and frame energy; S313, statistically summarize the time series feature indicators based on the sliding window length to construct a joint feature vector in the time and frequency domains.

6. A method for classifying elevator faults according to claim 5, characterized in that, The S4 includes: S41, extracting the fault feature vector, the corresponding fault type label and the maintenance plan information of each case based on the historical fault case library, and constructing a historical fault case set; S42. Cluster the feature vectors in the historical data to generate fault feature nodes, and combine the fault type labels and maintenance plan records to construct fault type nodes and maintenance plan nodes respectively, forming a three-layer topological structure including three types of nodes; S43. Establish directed edges in the graph and assign weights according to the similarity relationship between the fault feature nodes and the fault type nodes and the co-occurrence relationship between the fault type nodes and the maintenance plan nodes to construct a dynamic knowledge graph.

7. A method for classifying elevator faults according to claim 6, characterized in that, The S5 includes: S51. Node matching and weight adjustment: Match the extracted multi-dimensional fault features with the fault feature nodes in the knowledge graph, and determine the correlation degree of the fault type nodes according to the matching results, and dynamically adjust the weights of each fault type node in the knowledge graph; S52. Fault classification reasoning output: Based on the adjusted knowledge graph node weights, fuse the current fault features with the graph weights and input them into the improved residual neural network model to output the corresponding fault classification results and their confidence levels.

8. A method for classifying elevator faults according to claim 7, characterized in that, The S51 includes: S511. Calculate the similarity scores between the current fault feature vector and each fault feature node in the graph using the exponential distance function; S512. Take the similarity scores as the activation coefficients of the fault feature nodes, and through the graph propagation mechanism, transfer the activation values to the set of fault type nodes to obtain the updated weights of each fault type node.

9. A method for classifying elevator faults according to claim 7, characterized in that, The S52 includes: S521. Concatenate the current fault feature vector with the adjusted fault type node weight vector to form a combined input vector; S522. Use the combined input vector as the input of the residual neural network, and the network is composed of multiple stacked residual modules; Input the last layer output vector into the Softmax classifier to calculate the classification probabilities of each node in the fault type node set, and obtain the final fault classification result and confidence level.

10. An elevator fault classification system for implementing an elevator fault classification method according to any one of claims 1-9, characterized in that, It includes the following modules: Data acquisition module: Real-time collect multi-source heterogeneous data of elevator operation, including mechanical vibration signals, three-phase current waveforms of drive motors, car video behavior data, and control cabinet instruction flows; Spatio-temporal alignment module: Perform spatio-temporal alignment processing on the multi-source heterogeneous data to generate a standardized time series data matrix; Feature extraction module: Extract multi-dimensional fault features from the standardized time series data matrix, and the fault features include the joint time-frequency domain features of vibration signals, the harmonic distortion features of current waveforms, and the abnormal behavior recognition features of video data; Knowledge graph construction module: Construct a dynamic knowledge graph based on the historical fault case library, and the knowledge graph includes a three-layer topological structure of fault feature nodes, fault type nodes, and maintenance plan nodes; Graph matching and reasoning module: Match the currently extracted multi-dimensional fault features with the dynamic knowledge graph, dynamically adjust the graph node weights according to the matching degree, and use the improved residual neural network model to output the fault classification results and their confidence levels based on the optimized graph node weights.

Images