Abnormality identification method and system of air blower based on multi-modal data

Abstract

The invention discloses an anomaly identification method and system for an air blower based on multi-modal data, and relates to the technical field of anomaly identification. The multi-modal data corresponding to the air blower is constructed according to a multi-modal framework of the air blower and noise data and temperature data of the air blower; and the accuracy of the multi-modal data corresponding to the air blower is improved. The multiple pieces of abnormal data are determined according to detection of the multi-modal data corresponding to the air blower, the abnormal parts corresponding to the abnormal data are marked, the anomaly recognition system is determined according to the multiple pieces of abnormal data, the abnormal parts corresponding to the abnormal data and the overall form of the air blower, the accuracy of the anomaly recognition system is improved, and meanwhile the accuracy of the anomaly recognition system is improved. And determining to-be-maintained content according to the node position of each abnormal influence node, the corresponding node form and the use scene of the air blower, and determining corresponding abnormal maintenance measures based on the to-be-maintained content, the current working mode of the air blower and the corresponding maintenance system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of anomaly identification, and particularly relates to an air blower anomaly identification method and system based on multi-modal data. BACKGROUND

[0002] With the development of science and technology, air blowers are applied to industrial scenes and serve as industrial fluid mechanical devices. A plurality of working data of the air blower is collected, the working state of the air blower is determined according to the plurality of working data of the air blower, and the working content of the air blower is marked. Meanwhile, corresponding abnormal data is determined based on the detection of the working content, and the consideration of abnormal components corresponding to each abnormal data and the overall form of the air blower is ignored, which affects the accuracy of the anomaly identification system and leads to low accuracy of the abnormal maintenance measures of the air blower. SUMMARY

[0003] The present application aims to overcome the shortcomings of the prior art, and provides an air blower anomaly identification method and system based on multi-modal data.

[0004] The present application provides an air blower anomaly identification method based on multi-modal data, which comprises the following steps: When the air blower is in a working state, the real-time image of the air blower is determined based on the visual detection of the air blower, a plurality of component regions are determined based on the recognition of the real-time image of the air blower, and corresponding vibration data combinations are determined according to the vibration detection of each component region. A multi-modal framework of the air blower is determined based on the region position of each component region and the corresponding vibration data combination, and multi-modal data corresponding to the air blower is constructed according to the multi-modal framework of the air blower, noise data and temperature data of the air blower. The multi-modal framework not only contains the current vibration state, but also implies the three-dimensional geometric structure and fault importance information of the equipment. The multi-modal data contains spatial position, vibration characteristics, thermodynamic characteristics, acoustic characteristics and scene semantics. A plurality of abnormal data is determined according to the detection of the multi-modal data corresponding to the air blower, and abnormal components corresponding to each abnormal data are marked. An anomaly identification system is determined according to the plurality of abnormal data, the abnormal components corresponding to each abnormal data and the overall form of the air blower, and a plurality of abnormal regions of different abnormal influence levels are marked. The anomaly identification system is a comprehensive system that fuses the structural dimension and the working condition dimension of the air blower to realize fault grading. In each abnormal region, corresponding abnormal events are determined according to the dynamic detection of the abnormal region, a plurality of abnormal influence factors are determined based on each abnormal event, real-time working data of the air blower and the anomaly identification system, and corresponding abnormal influence routes are formed. The abnormal influence route is a panoramic multi-dimensional fault mapping graph. A plurality of abnormal influence nodes are determined based on identification of abnormal influence routes, content to be maintained is determined according to node positions of each abnormal influence node, corresponding node forms and a use scenario of the air blower, corresponding abnormal maintenance measures are determined based on the content to be maintained, a current working mode of the air blower and a corresponding maintenance system, the abnormal influence node is defined as a physical position most critical to a maintenance operation, the node position of the abnormal influence node covers corresponding installation coordinates, hierarchical relationships and an envelope boundary of the surrounding, the node form of the abnormal influence node includes corresponding form topological features, the content to be maintained refers to specific maintenance operation tasks determined to solve faults, and the maintenance system includes a mapping of fault modes and repair schemes.

[0005] The embodiment of the application provides an air blower abnormality identification system based on multi-modal data.

[0006] Compared with the prior art, the air blower abnormality identification system based on multi-modal data has the following beneficial effects: (1) The multi-modal framework of the air blower is determined based on the area positions of each component area and corresponding vibration data, the multi-modal data corresponding to the air blower is constructed according to the multi-modal framework of the air blower, noise data and temperature data of the air blower, the multi-modal framework of the air blower is introduced, the multi-modal framework of the air blower, the noise data and the temperature data of the air blower are considered, and the accuracy of the multi-modal data corresponding to the air blower is improved.

[0007] (2) A plurality of abnormal data are determined according to detection of the multi-modal data corresponding to the air blower, and each abnormal data corresponding to an abnormal component is marked, an abnormality identification system is determined according to the plurality of abnormal data, the abnormal component corresponding to each abnormal data and the overall form of the air blower, and a plurality of abnormal areas of different abnormal influence levels are marked, each abnormal data is further controlled, and the corresponding abnormal component is marked, the overall consideration of the plurality of abnormal data, the abnormal component corresponding to each abnormal data and the overall form of the air blower is realized, and the accuracy of the abnormality identification system is improved.

[0008] (3) Corresponding abnormal events are determined according to dynamic detection of the abnormal area, a plurality of abnormal influence factors are determined based on each abnormal event, real-time working data of the air blower and the abnormality identification system, and corresponding abnormal influence routes are formed, a plurality of abnormal influence nodes are determined based on identification of the abnormal influence routes, content to be maintained is determined according to node positions of each abnormal influence node, corresponding node forms and a use scenario of the air blower, corresponding abnormal maintenance measures are determined based on the content to be maintained, a current working mode of the air blower and a corresponding maintenance system, the overall consideration of the content to be maintained, the current working mode of the air blower and the corresponding maintenance system is realized, and the accuracy of the abnormal maintenance measures of the air blower is improved. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the anomaly identification method for blowers based on multimodal data in an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S11 in the anomaly identification method for a blower based on multimodal data in an embodiment of the present invention. Figure 3 This is a flowchart illustrating step S12 in the anomaly identification method for a blower based on multimodal data in an embodiment of the present invention. Figure 4 This is a flowchart illustrating step S13 in the anomaly identification method for a blower based on multimodal data in an embodiment of the present invention. Figure 5 This is a flowchart illustrating step S14 of the anomaly identification method for a blower based on multimodal data in an embodiment of the present invention. Figure 6 This is a flowchart illustrating step S15 of the anomaly identification method for a blower based on multimodal data in an embodiment of the present invention. Figure 7 This is a schematic diagram of the structural composition of the blower anomaly identification system based on multimodal data in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0011] Please see Figures 1 to 7 An anomaly identification method for blowers based on multimodal data is proposed and applied to anomaly identification scenarios. The anomaly identification method for blowers based on multimodal data includes: Step S11: When the blower is in operation, determine the real-time image of the blower based on the visual inspection of the blower, determine multiple component areas based on the recognition of the real-time image of the blower, and determine the corresponding vibration data combination based on the vibration detection of each component area. Step S12: Determine the multimodal frame of the blower based on the regional location of each component area and the corresponding vibration data combination, and construct the multimodal data corresponding to the blower based on the multimodal frame of the blower, the noise data and temperature data of the blower; Step S13: Based on the detection of multimodal data corresponding to the blower, determine multiple abnormal data and mark the abnormal components corresponding to each abnormal data. Based on the multiple abnormal data, the abnormal components corresponding to each abnormal data and the overall shape of the blower, determine the abnormal identification system and mark multiple abnormal areas with different levels of abnormal impact. Step S14: In each abnormal area, the corresponding abnormal event is determined based on the dynamic detection of the abnormal area. Based on each abnormal event, the real-time working data of the blower and the abnormal identification system, multiple abnormal influencing factors are determined, and the corresponding abnormal influence path is formed. Step S15: Based on the identification of abnormal impact routes, identify multiple abnormal impact nodes. Determine the maintenance content based on the node location, corresponding node shape, and blower usage scenario of each abnormal impact node. Based on the maintenance content, the blower's current working mode, and the corresponding maintenance system, determine the corresponding abnormal maintenance measures.

[0012] refer to Figure 2 In step S11, the specific steps are as follows: S111: Mark the location of the blower and collect multiple working data of the blower. Determine the working status of the blower based on the multiple working data and the corresponding overall shape to control the working status of the blower. At the same time, trigger the positioning and shooting of multiple cameras around the blower to perform visual inspection of the blower and collect multiple sub-images of the blower in different dimensions. Construct a real-time image of the blower based on the synthesis of multiple sub-images. S112: Dynamically identify the real-time image of the blower, and determine multiple component markers during the identification process. Based on the tracing of each component marker, determine the corresponding component area to trigger vibration detection in each component area. At the same time, determine multiple vibration data based on the vibration detection in each component area, and determine the corresponding vibration data combination based on the multiple vibration data, the corresponding vibration detection position, and the corresponding component area.

[0013] In the embodiments of this application, the precise coordinates of the blower in physical space are determined. The system collects multi-dimensional working data of the blower in real time (such as current, voltage, speed, flow rate, etc.) through an industrial Internet of Things interface. At the same time, the system performs preliminary digital modeling based on the spatial geometry of the blower and maps these real-time working data onto the equipment model, thereby comprehensively judging whether the blower is currently in a steady state, a transient state (such as the start-stop process), or an overload state.

[0014] Based on the determined spatial location of the blower, the system triggers the visual acquisition terminals deployed around it. This "triggering" is not a simple shooting, but an active visual guidance based on spatial geometry. The system calculates the optimal combination of cameras that can cover the key appearance features of the blower based on the blower's volume, height, and the field of view (FOV) of the surrounding cameras. These cameras perform synchronous or asynchronous positioning and shooting of the blower from different dimensions (such as front, side, and top views). The acquisition process involves optical technologies such as optical flow control and focus locking to obtain clear, unobstructed raw sub-image data.

[0015] The acquired sub-images from multiple dimensions are transmitted to the processing unit. Since these sub-images come from different perspectives and time points, the system uses feature matching algorithms in computer vision (such as SIFT, SURF, or feature extraction based on deep learning) to extract key feature points in each sub-image. By calculating the homography matrix or performing 3D reconstruction, these sub-images are geometrically corrected and stitched together to construct a high-resolution, seamlessly stitched real-time panoramic image containing complete appearance information of the blower. This image will serve as the base map for subsequent component area identification and vibration data spatial mapping.

[0016] Specifically, suppose there is a blower deployed in an industrial workshop. The equipment vibrates significantly during operation, and there are obstructions in the surrounding environment. The system uses the factory's digital twin platform to locate the blower at position 3 in area B of the workshop. At the same time, the SCADA system collects real-time data of the blower: drive end current is 45A, outlet pressure is 0.25MPa, and speed is 2950r / min. Based on these parameters and the overall operating mode of the blower, the system determines that it is currently in a "high-load stable operation" state. In order to ensure that visual inspection can capture minute deformations or features under high load, the system sets the priority of visual acquisition to the highest level and locks the current timestamp as T0.

[0017] Based on the coordinates of the blower at position 3 in area B, the system calculates the spatial coverage of the surrounding cameras C1 (located in front left of the equipment), C2 (located in the upper right of the equipment), and C3 (located at the tail of the motor). The system determines that a single camera cannot simultaneously capture the details of the impeller end and bearing housing, so it triggers the automatic positioning and shooting function of all three cameras. Camera C1 adjusts its focus to focus on the blower's air inlet housing, camera C2 focuses on the exhaust side bearing housing, and camera C3 focuses on the coupling and motor. The system obtains three sub-images in different dimensions: Img_C1 (air inlet view), Img_C2 (exhaust bearing view), and Img_C3 (transmission side view).

[0018] After receiving Img_C1, Img_C2, and Img_C3, the system uses a feature extraction algorithm to identify common features in each view (such as the corner lines of the equipment base, paint texture, connecting bolts, etc.). Through an image registration algorithm, the left area of ​​Img_C1 and the right area of ​​Img_C3 are aligned in the overlapping area. At the same time, the perspective projection of Img_C2 is corrected and integrated into the overall structure. The system generates a real-time panoramic composite image of the blower at time T0. In this image, the left side clearly shows the state of the air inlet, the middle shows the connection of the coupling, and the right side fully presents the details of the exhaust bearing seat, and all spatial coordinates are unified.

[0019] Furthermore, the real-time image of the blower is dynamically identified, and multiple component markers are determined during the identification process. The corresponding component area is determined by tracing the markers to trigger vibration detection in each component area. At the same time, multiple vibration data are determined based on the vibration detection in each component area. Based on the multiple vibration data, the corresponding vibration detection location, and the corresponding component area, a corresponding vibration data combination is determined, thus introducing vibration data combination.

[0020] At this point, deep learning object detection algorithms (such as YOLO, Mask R-CNN, or variant networks based on instance segmentation) are used to perform frame-by-frame analysis on the real-time panoramic image constructed in step S111. Since the device is in operation, the image contains blurring or slight positional shifts caused by motion, so the algorithm needs to have dynamic robustness. The system identifies the key independent components of the blower (such as impeller, bearing housing, motor housing, base, etc.) in the image and uses bounding boxes or semantic masks to accurately label these components. Each label is assigned a unique ID, forming a set of component labels.

[0021] Based on the generated component markings, the system performs mapping and tracing between the image coordinate system and the physical world coordinate system. This process uses preset camera calibration parameters (intrinsic and extrinsic parameters) to convert the two-dimensional pixel region in the image into component region coordinates in three-dimensional physical space. The system then queries the list of sensors associated with the physical region and triggers the corresponding vibration data acquisition terminal. Here, "triggering" means sending an acquisition command to a specific vibration sensor (such as a piezoelectric accelerometer), requiring it to perform synchronous sampling at a high sampling rate based on the current operating characteristics of the component (such as rotational speed frequency multiplication).

[0022] The system receives raw vibration signals (usually time-domain waveforms) from different sensors. To map these discrete data points to visual component regions, the system preprocesses the data (e.g., noise reduction and filtering) and binds the metadata of the data packets (e.g., sensor ID, acquisition timestamp) to physical location information. The system encapsulates the vibration characteristics (e.g., RMS, peak factor, kurtosis, etc.) belonging to the same component region with the spatial coordinates of that component. This encapsulation is not merely a list of values, but forms a "vibration data combination" with spatial attributes.

[0023] Specifically, the system has obtained a real-time panoramic composite image of the blower; the system runs a target detection algorithm on the real-time panoramic image of the blower; the algorithm quickly identifies three key independent objects in the image: the motor drive end, the blower impeller housing, and the gearbox bearing housing; the system generates three bounding boxes on the image and assigns them the following IDs: Part-01 (motor drive end), Part-02 (blower impeller housing), and Part-03 (gearbox bearing housing); even when the blower operates at high speed, causing slight motion blur at the image edges, the algorithm still locks the precise contours of these components through edge feature learning.

[0024] Based on the pixel coordinates of Part-03 (gearbox bearing housing) in the image and combined with the camera calibration matrix, the system calculates its physical location as 1.2 meters above the east side of the equipment. The system traces the database and finds that a triaxial vibration accelerometer with sensor number VS-003 is pre-installed at this physical location. The system sends a trigger command to VS-003, requesting it to capture data in the high-frequency band (such as 0-10kHz) to capture the high-frequency fault signal generated by gear meshing.

[0025] The VS-003 sensor returned the raw time-domain waveform data; the system performed Fast Fourier Transform (FFT) analysis on the data to extract feature values ​​(e.g., the horizontal vibration velocity is 4.5 mm / s, the vertical vibration velocity is 3.2 mm / s, and there is a significant sideband at the gear meshing frequency); the system strongly bound these vibration features to the physical location information of Part-03 (gearbox on the east side of the blower) and the image ID; the final generated "vibration data combination" not only includes the above values, but also spatial semantics: "the bearing housing area of ​​the blower gearbox exhibits high-frequency vibration characteristics".

[0026] refer to Figure 3 In step S12, the specific steps are as follows: S121: In multiple component regions, the corresponding region position is determined based on the detection of each component region, and the multimodal frame of the blower is determined according to the region position of each component region, the corresponding region priority and the corresponding vibration data combination. S122: Determine multiple environmental data based on the location detection of the blower; determine the noise and temperature data of the blower based on the filtering of multiple environmental data, and mark the usage scenario of the blower; determine primary data based on the multimodal framework of the blower and the usage scenario of the blower; and construct the corresponding multimodal data of the blower based on the primary data, the noise data and temperature data of the blower.

[0027] In the embodiments of this application, the system uses image processing technology to perform spatial analysis on the various component markers identified in step S112. By using pre-calibrated camera intrinsic parameters (focal length, principal point) and extrinsic parameters (rotation matrix, translation vector), combined with binocular vision or structured light depth estimation technology, the pixel coordinates of the components in the two-dimensional image are converted into three-dimensional spatial coordinates in the physical world. This process establishes the absolute position and relative positional relationship of each component region relative to the blower reference point (such as the center of the base), forming a spatial point set describing the geometric structure of the equipment.

[0028] Based on the determined spatial location, the system calculates priorities according to the physical properties, functional importance, and fault sensitivity of the components. This process typically employs the Analytic Hierarchy Process (AHP) or rule-based expert system logic. The system assigns different weight levels to different components. For example, high-speed rotating components and bearing housings that bear core loads are usually given the highest priority, while auxiliary connecting parts or housings are given lower priorities. In addition, the system dynamically adjusts priorities based on real-time operating conditions (such as sudden changes in rotational speed) to ensure that critical components dominate when allocating computational resources or determining anomalies.

[0029] The system uses the determined "regional location" as nodes, "regional priority" as the attribute weight of the nodes, and "vibration data combination" as the load of the nodes to construct a weighted topological network structure. In this structure, the various component regions are no longer isolated data points, but are interconnected through the geometric adjacency relationship of the mechanical transmission chain. The system strongly couples the vibration characteristics in the time domain with the spatial coordinates to generate a multimodal framework. The multimodal framework not only includes the current vibration state, but also implicitly contains the three-dimensional geometric structure of the equipment and the information on the importance of the fault.

[0030] Specifically, the system has visually identified three key components of the blower. By analyzing real-time images of the blower, the system identified the "gearbox high-speed shaft bearing housing" area. Using a pre-calibrated visual matrix, the system calculated the coordinates of the center point of this component in physical space as (X: 1.25m, Y: 0.80m, Z: 1.50m), with an offset of (+0.4m, 0m, +1.2m) relative to the base reference point of the blower. Similarly, the system determined the spatial coordinates of the "motor drive end" as (X: 0.5m, Y: 0.80m, Z: 0.5m). These precise three-dimensional coordinates were locked as spatial anchor points for subsequent data mapping.

[0031] The system evaluates the equipment based on its design parameters and fault model library. It determines that the "gearbox high-speed shaft bearing housing" is the core component responsible for power transmission and has the highest linear speed, thus posing the greatest risk of failure. Therefore, its priority is set to Level P0 (critical core level). While the "motor drive end" is important, its load is relatively stable under this operating condition, so its priority is set to Level P1 (important level). At the same time, considering that the current operating condition is at high speed, the system temporarily increases the monitoring weight coefficient of rotating components by 20%.

[0032] The system begins to construct a framework; with coordinates (1.25m, 0.80m, 1.50m) as the node center, the vibration data combination of the "gearbox high-speed shaft bearing housing" (including time-domain waveform, frequency-domain spectrum, and kurtosis index) is loaded onto this node and labeled with LevelP0 weights; simultaneously, a topological connection relationship is established between this node and the "motor drive end" node, representing the coupling between the two in mechanical transmission; the system generates a multimodal framework for the blower: a weighted point cloud network distributed in three-dimensional space with priority attributes and encapsulating real-time vibration data. This framework clearly informs the system that at the highest priority position (1.25m, 0.80m, 1.50m), a specific vibration energy characteristic is currently being carried.

[0033] Furthermore, multiple environmental data are determined based on the location of the blower. Noise and temperature data of the blower are determined by filtering these environmental data, and the blower's usage scenario is marked. Primary data is determined based on the blower's multimodal framework and usage scenario. Multimodal data corresponding to the blower is constructed based on the primary data, the blower's noise and temperature data, ensuring the accuracy of the primary data while incorporating the blower's multimodal framework, noise and temperature data, thus improving the accuracy of the multimodal data corresponding to the blower.

[0034] At this time, the system continuously collects raw environmental signals through an industrial-grade sensor array (such as a sound pressure sensor array, a non-contact infrared thermal imager, and an ambient temperature and humidity transmitter) deployed around the blower. In order to eliminate the interference of background noise, the system uses signal processing technology for filtering: for sound data, beamforming or spectrum analysis technology is used to filter out background noise from non-equipment directions (such as other work noises in the workshop) and accurately extract specific frequency band noise data originating from the blower itself; for temperature data, combined with the equipment thermal field distribution model, the region of interest (ROI) is segmented from the global thermal map of the thermal imager to eliminate the influence of environmental radiation and solar radiation and lock in the temperature data that reflects the true thermal state of the component.

[0035] While extracting pure physical quantities, the system combines the operating parameters of the blower's DCS (Distributed Control System) to perform scenario reasoning. Based on the current load rate, medium flow rate, regulating valve opening, and control commands (such as inverter frequency), the system uses clustering algorithms or rule matching to label the current operating state as a specific usage scenario (such as "heavy load full load operation", "no load start", "surge boundary operation", etc.). The system uses scenario labels to correct the multimodal framework constructed in step S121. For example, in the "heavy load" scenario, the system dynamically adjusts the baseline threshold of vibration data and removes normal physical quantity changes caused by increased load, generating "primary data" that has been filtered by scenario semantics.

[0036] The system maps the filtered noise and temperature data onto a scene-corrected primary data skeleton. This process involves strict data alignment: in the time dimension, instantaneous noise peaks and temperature changes are synchronized to vibration data slices at the same sampling time based on a unified timestamp (such as the PTP protocol); in the spatial dimension, noise signals are assigned to specific component nodes in the multimodal framework using sound source localization technology, and temperature heatmap data is pixel-level registered with the geometric regions of the components; the system ultimately constructs multimodal data, which includes spatial location, vibration characteristics, thermodynamic characteristics, acoustic characteristics, and scene semantics, providing comprehensive input for subsequent deep anomaly identification.

[0037] Specifically, the blower is located in a noisy industrial workshop. The system collects mixed sound wave signals in the workshop through a microphone array. Using an adaptive filtering algorithm, the system filters out the 2kHz cutting machine noise in the background and accurately extracts the blower's unique blade passing frequency (BPF) noise data, with a sound pressure level of 88dB. At the same time, an infrared thermal imager scans the surface of the equipment. The system uses a thermal image segmentation algorithm to eliminate the thermal radiation interference from the surrounding high-temperature pipes and locks in the true temperature data of the blower's "gearbox bearing housing" surface, which is displayed as 72℃.

[0038] The system reads the real-time operating parameters of the blower: the outlet valve opening is 100%, and the motor current reaches 95% of the rated value. Based on this, the system marks the current operating state as the "full load steady-state operation" scenario. The system retrieves the standard model under this scenario and determines that under full load, specific high-frequency components in the vibration data belong to normal physical responses. Therefore, the system corrects the multimodal framework in step S121, eliminates the structural vibration false alarm factor caused by the increase in load, and generates "primary data" that reflects the true health status of the equipment under full load.

[0039] The system injects the aforementioned noise data (88dB) and temperature data (72℃) into the primary data. Spatially, the 72℃ temperature value is bound to the "gearbox bearing housing" node with coordinates (X:1.25m, Y:0.80m, Z:1.50m) in the multimodal framework. Temporally, the 88dB noise value is timestamped and aligned with the node's current vibration acceleration RMS value (4.5mm / s). The system constructs a multimodal data object for the blower: {Node: Gearbox bearing housing | Position:(1.25,0.80,1.50) | Priority:P0 | Vibration:4.5mm / s | Temperature:72℃ | Noise:88dB | Scenario: Full load steady state}.

[0040] refer to Figure 4 In step S13, the specific steps are as follows: S131: Dynamically detect the multimodal data corresponding to the blower, identify multiple data anomaly markers during the detection process, and determine the corresponding abnormal data based on the tracing of each data anomaly marker, so as to collect multiple abnormal data. S132: Based on the detection of each abnormal data, the corresponding abnormal component is determined. The first level of abnormal identification content is determined according to the abnormal component corresponding to each abnormal data and the overall shape of the blower. The second level of abnormal identification content is determined according to each abnormal data and the current working mode of the blower. The abnormal identification system is determined according to the first level of abnormal identification content and the second level of abnormal identification content. S133: In the anomaly identification system, multiple abnormal areas are identified based on the anomaly identification system, and the corresponding anomaly impact level is determined according to the data combination of each abnormal area, the corresponding area location, and the corresponding component function.

[0041] In the embodiments of this application, the system uses a time-series sliding window algorithm to scan the multimodal data stream frame by frame. In this step, the system does not monitor the value of a single dimension, but performs dynamic detection based on a multi-parameter fusion strategy. The system performs feature-level alignment of vibration spectrum characteristics, temperature change trends, and noise sound pressure levels within the same time window. It uses machine learning models (such as autoencoders or long short-term memory networks LSTM) to calculate the "reconstruction error" or "residual" between the current input data and the standard normal model under this working condition. When the residual exceeds a dynamically set confidence threshold, or when a violation of the physical coupling law between parameters is detected (e.g., a surge in vibration energy but no corresponding increase in acoustic energy), the system immediately determines that an abnormal state has been detected and generates a "data anomaly marker".

[0042] After generating a data anomaly marker, the system uses the precise timestamp and spatial index information contained in the marker to perform a reverse retrieval in a high-speed data cache or historical database. The tracing process aims to reconstruct the complete context of the anomaly: the system not only extracts the peak data at the moment the anomaly occurred, but also extracts the "time slices" before and after the anomaly (such as data from T-5 seconds to T+5 seconds) to capture the transient characteristics of the fault evolution. At the same time, based on the spatial index, the system locates which node (such as which component region) in the multimodal framework the anomaly specifically originated from, thereby binding the marker to the specific physical entity.

[0043] Based on the tracing results, the system extracts relevant data segments from the multimodal data stream and encapsulates and collects them. This process involves not only extracting numerical values ​​but also extracting high-level feature vectors (such as kurtosis, factor, and spectral centroid) of the data segment. The system packages these raw waveforms, feature parameters, and corresponding metadata (device ID, component ID, and operating condition label) into structured "abnormal data" objects. This step ensures that subsequent fault diagnosis and analysis are based on complete, high-quality data blocks with contextual information, rather than isolated single-point alarm values.

[0044] Specifically, the blower is in "full-load steady-state operation" mode, and the system is monitoring its real-time multimodal data stream, focusing on the condition of the gearbox area. The system scans the multimodal data of the blower sequentially with a 1-second time window. At T=10:05:32, the system detected an anomaly when processing the data of the "gearbox high-speed shaft bearing housing" node: obvious asymmetric impact appeared in the time-domain waveform of the vibration signal, and the effective value of vibration acceleration (RMS) exceeded the dynamic threshold. At the same time, the spectrum analysis of the noise data showed that sidebands with an interval of revolution frequency appeared near the gear meshing frequency (500Hz). Although the temperature data only increased slightly at this time, the characteristics of "vibration impact" and "sideband" were highly coupled, which is consistent with the typical characteristics of bearing cage damage. The system determined that this was a significant abnormal event and immediately generated a "data anomaly tag Tag-01" with a timestamp of 10:05:32 and node ID (Gearbox_HighSpeed).

[0045] Based on the timestamp 10:05:32 and node ID in “Tag-01”, the system immediately backtracked in the data cache; the system located the data segment before and after that moment and found that from 10:05:28, a weak low-frequency modulation component had begun to appear in the vibration signal, and by 10:05:32, this component had evolved into a violent impact; the system confirmed that the source of the anomaly was the “gearbox high-speed shaft bearing housing” at coordinates (1.25m, 0.80m, 1.50m) in the multimodal frame.

[0046] The system packages all relevant data within the 10-second time window from 10:05:28 to 10:05:37. The collected data includes: the raw waveform data of triaxial vibration acceleration during this period, the corresponding FFT spectrum, the time series of noise sound pressure level, and the temperature sampling value at this moment. The system encapsulates this data into a standard "abnormal data" object, marked as {EventID:Tag-01,Source:Gearbox_HighSpeed,Waveform:[...],Features:{Kurtosis:5.2,Sideband_Level:High}}.

[0047] Furthermore, based on the detection of each abnormal data, the corresponding abnormal component is identified. The first level of anomaly identification content is determined according to the abnormal component corresponding to each abnormal data and the overall shape of the blower. The second level of anomaly identification content is determined according to each abnormal data and the current working mode of the blower. The anomaly identification system is determined based on the first level of anomaly identification content and the second level of anomaly identification content. This system takes into account both the first level of anomaly identification content and the second level of anomaly identification content, ensuring the accuracy of the anomaly identification system.

[0048] At this point, the system performs in-depth analysis of the collected abnormal data and pinpoints the source of the fault. By analyzing the spatial coordinates of the abnormal data in the multimodal framework, the system accurately maps it to specific physical components. This process relies on matching the fault feature library: for example, a specific impact frequency is mapped to a bearing, a specific airflow noise is mapped to an impeller, and a specific temperature distribution is mapped to a winding or lubrication system. The system ultimately determines the unique physical component that generated this abnormal data and outputs the "abnormal component ID".

[0049] After identifying the abnormal component, the system performs the first level of anomaly identification by combining the overall mechanical shape and topology of the blower. This dimension focuses on the physical transmission and structural impact of the fault. The system calculates the position of the abnormal component in the overall shape (such as whether it is at the cantilever end or the core of the transmission chain) and analyzes whether its vibration mode is coupled and resonates with the surrounding components. For example, if the abnormal component is located at the base, the system will determine whether its vibration is being transmitted upward to the motor end through the frame.

[0050] For the second level of anomaly identification, the system correlates abnormal data with the blower's current operating mode (such as constant pressure control, variable frequency speed regulation, bypass adjustment, etc.). This dimension focuses on the authenticity of the fault, that is, distinguishing between a fault in the equipment itself and a normal response caused by changes in external operating conditions. The system uses a mechanism model to determine whether the current abnormal parameter changes violate physical laws under the current mode. For example, pressure fluctuations are normal at the moment of flow regulation, but if they are accompanied by specific high-frequency vibrations, then it is abnormal.

[0051] The system utilizes DS evidence theory or Bayesian inference networks to fuse the identification results of the first level of anomaly identification (structural dimension) and the second level of anomaly identification (operating condition dimension). If the first level shows "severe structural resonance" and the second level shows "stable operating condition caused by non-adjustment", the system determines it as a "severe structural fault". If the first level shows "minor local anomaly" and the second level shows "currently undergoing rapid load adjustment", the system determines it as a "condition-induced temporary anomaly". Through this cross-validation, the system constructs a highly robust anomaly identification system. This anomaly identification system is a comprehensive system that integrates the structural and operating condition dimensions of the blower to achieve fault classification.

[0052] Specifically, the system collected abnormal data at time T, and now needs to confirm the nature of the fault through a dual identification system; the blower is currently in "constant pressure variable frequency operation" mode, with the frequency set at 45Hz; the system analyzed the abnormal data and found that the dominant feature was broadband vibration at low and medium frequencies (1X-3X RPM), and the noise data was accompanied by obvious airflow turbulence; based on the spatial index (X:1.25m, Y:0.80m, Z:1.50m) of the multimodal framework and feature library matching, the system ruled out the motor and gearbox, and determined that the corresponding abnormal component was the blower's volute assembly.

[0053] The system combines the overall shape analysis of the blower: the volute is fixed to the base by four bolts; the first identification shows that the typical structural resonance frequency of the base does not appear in the spectrum of the abnormal data, indicating that the vibration mainly originates from the fluid pulsation inside the volute or its own insufficient stiffness, and has not yet formed a strong path to transmit to the base; the identification conclusion is: "The anomaly is limited to the volute body and has not yet triggered the resonance transmission of the overall structure, but there is a risk of loosening of the connecting parts under the current shape."

[0054] The system reviewed the work logs and found that between T-5 seconds and T, when the anomaly occurred, the frequency of the blower's inverter rapidly increased from 40Hz to 45Hz (i.e., it was in the step response stage). According to fluid mechanics principles, transient airflow vibrations are indeed generated inside the volute during rapid speed changes. The second identification conclusion was: "The abnormal data characteristics are consistent with 'fluid transient response under rapid load change conditions,' which is an expected physical phenomenon under this working mode, rather than equipment damage."

[0055] The system integrates the above two levels of information: Input A (first level of anomaly identification): Local vibration exists but has not propagated, indicating a moderate risk; Input B (second level of anomaly identification): The cause of the change in operating conditions is clear, and it is not a fault inherent to the system; Integration conclusion: The anomaly identification system classifies this event as "condition-induced transient fluctuation" and sets the level to "concern"; The system automatically suppresses the severe alarm for "volute failure" but generates preventive maintenance suggestions for "fatigue of connecting bolts under frequent load changes", thus achieving accurate identification.

[0056] Therefore, in the anomaly identification system, multiple abnormal areas are identified based on the anomaly identification system, and the corresponding anomaly impact level is determined according to the data combination of each abnormal area, the corresponding area location, and the corresponding component function. This system takes into account the overall consideration of the data combination, corresponding area location, and corresponding component function of each abnormal area, ensuring the accuracy of the corresponding anomaly impact level. At the same time, each abnormal data is further controlled, and the corresponding abnormal components are marked. This realizes the overall consideration of multiple abnormal data, the abnormal components corresponding to each abnormal data, and the overall shape of the blower, thus improving the accuracy of the anomaly identification system.

[0057] At this point, the system utilizes the established anomaly identification system to accurately delineate the physical scope of the fault's impact. This process is no longer a point-based localization of a single component, but rather a regionalized mapping based on data correlation. The system analyzes the propagation path of abnormal data within the multimodal framework and identifies the data clusters affected by the fault through graph theory algorithms (such as connected component analysis). The system delineates a set of components that are spatially adjacent, have highly correlated characteristic parameters, and are affected by the same fault source as one or more "abnormal regions." For example, a bearing fault not only affects the bearing housing itself but also affects the adjacent base through rigid connections. These three are spatially and logically merged and defined as an abnormal region.

[0058] For each identified anomalous region, the system uses a multi-factor weighted model to calculate its "anomaly impact level"; this model mainly quantifies and evaluates the following three core dimensions: Deviation of data combination: The deviation ratio and trend of data such as vibration RMS value, temperature gradient, and sound pressure level in the calculation area relative to the baseline (ISO10816 standard or equipment historical baseline) characterize the intensity of the fault. The criticality of the region's location: Assess the region's positional weight within the equipment based on the topology; the core drivetrain region is typically higher than the edge support region, and the high-pressure end near the outlet is typically higher than the low-pressure end near the inlet; Component functional sensitivity: Assess the impact of the functions carried by this area on the overall operation of the system; components involving power transmission, core sealing or rotor balancing are highly sensitive to their functions.

[0059] The system performs weighted calculations on the above three dimensions based on the Analytic Hierarchy Process (AHP) or fuzzy logic, and finally outputs a quantitative level of abnormal impact (usually divided into: Level I - Emergency / Critical, Level II - Important / Severe, Level III - General / Warning).

[0060] Specifically, two different types of fault symptoms were identified in the blower: one was a sudden impact on the high-speed shaft of the gearbox, and the other was a slight blockage of the blower inlet filter. It is now necessary to assess the impact level of the abnormal areas defined by these two fault sources.

[0061] Area 1 Definition: The system detected a high-frequency impact signal in the high-speed shaft bearing housing of the gearbox (coordinates 1.25m, 0.80m, 1.50m), and the signal characteristics showed a strong correlation with the adjacent coupling area (coordinates 1.10m, 0.80m, 1.50m). The identification system determined that these two physical locations had formed a connected whole under stress, and delineated them as an abnormal area, named "Power Transmission Core Area".

[0062] Zone 2 Definition: The system detected a slight fluctuation in the negative pressure data at the inlet pipe, and detected a low-frequency airflow sound (less than 200Hz) in the noise data; this signal did not propagate into the machine body, but was limited to the area around the air inlet; the system independently defined it as another abnormal area, named "Inlet Airflow Auxiliary Zone".

[0063] Assessment of the "Core Area of ​​Power Transmission": Data combination: The peak vibration acceleration reached 12 mm / s (ISO standard warning value is 7.1 mm / s), accompanied by a sharp temperature rise (+2℃ per minute), with extremely high deviation; Location of the area: Located at the power hub between the motor and the load, it is at the most critical position in the transmission chain; Function of the component: It undertakes high-speed torque transmission, and failure will directly lead to the blower shutdown, or even cause a serious accident of impeller flying off; Comprehensive assessment: The system calculates a comprehensive risk index >0.9, and the abnormal impact level of this area is judged as Level I (emergency shutdown level).

[0064] Assessment of the "Intake Airflow Auxiliary Zone": Data combination: No abnormal vibration, only a 5dB increase in noise level, negative pressure fluctuation within 3%, and low deviation; Area location: Located on a non-load-bearing structural component at the edge of the equipment; Component function: Only responsible for air filtration and airflow guidance, functional failure will only lead to a decrease in efficiency and will not immediately damage the equipment; Comprehensive assessment: The system calculates a comprehensive risk index of <0.3, classifying the abnormal impact level of this area as Level III (planned maintenance level); The system visually displays these two areas on the blower's human-machine interface (HMI) using heat maps of different colors: the gearbox area is displayed in red (Level I), and the intake area is displayed in yellow (Level III).

[0065] refer to Figure 5 In step S14, the specific steps are as follows: S141: Monitor each abnormal area in real time and mark the abnormal time period of each abnormal area. Dynamically identify the abnormal time period of each abnormal area and determine the multiple work contents of the abnormal area during the abnormal time period during the identification process. Determine the corresponding abnormal event based on the area shape of the abnormal area, the corresponding multiple work contents and the priority of the corresponding component. S142: Collect real-time operating data of the blower, determine the first level of impact based on various abnormal events and the real-time operating data of the blower, determine the second level of impact based on the real-time operating data of the blower and the anomaly identification system, and determine multiple abnormal influencing factors based on the first and second level of impact. S143: Mark the affected components of each abnormal influencing factor, determine multiple sub-influence routes based on each affected component, the corresponding abnormal influencing factor, and the current working mode of the blower, and form the corresponding abnormal influence route based on the multiple sub-influence routes, the current working route of the blower, and the corresponding data transmission route.

[0066] In the embodiments of this application, each abnormal area is monitored in real time, and the abnormal time period of each abnormal area is marked. The abnormal time period of each abnormal area is dynamically identified, and multiple work contents of the abnormal area during the abnormal time period are determined during the identification process. The corresponding abnormal event is determined according to the area shape of the abnormal area, the corresponding multiple work contents and the corresponding component priority, thereby improving the accuracy of abnormal events.

[0067] At this point, the system initiates a high-frequency sampling monitoring stream for the locked abnormal area and uses a state machine model or change point detection algorithm to accurately define the start and end times of the abnormality. This process not only records a single point in time but also marks continuous "abnormal time periods." Within this time window, the system performs streaming processing on multimodal data to capture the dynamic evolution trend of signal characteristics. For example, it distinguishes between "transient pulses" (such as instantaneous impacts) and "continuous drifts" (such as slow temperature increases). The system uses these marked time periods as the basic context for subsequent behavioral analysis.

[0068] The system uses pattern recognition algorithms to analyze the dynamic behavioral characteristics of the area and transform them into specific "working content." Here, "working content" does not refer to the normal operation of the equipment, but rather to the physical behavior patterns exhibited by the faulty components at the microscopic level. The system extracts feature sequences through time-frequency analysis (such as Short-Time Fourier Transform (STFT) or Wavelet Transform) to identify specific behavioral patterns, such as: "the appearance and gradual increase of subsynchronous frequency components" (manifested as friction behavior), "the amplitude increases with rotational speed in a quadratic curve" (manifested as unbalanced behavior), and "the dense emergence of high-frequency envelope demodulation spectrum" (manifested as peeling behavior). These behavioral descriptions constitute the "working content" of the abnormal area.

[0069] The system semantically fuses the identified "work content" with the "regional morphology" and "component priority" of the area to define the final "abnormal event"; it assesses the impact of physical structure on behavior; for example, "cantilever structure" amplifies vibration amplitude, and "enclosed cavity" causes noise standing waves; it qualitatively characterizes the nature of the event by combining P0 and P1 levels; minor behaviors of high-priority components are defined as serious events; the system maps [work content + regional morphology + component priority] into standardized abnormal event labels through a rule engine or expert system; for example, "high-frequency impact behavior" + "high-speed rotational morphology" + "P0-level component" is defined as a "precursor event of catastrophic failure of high-speed shaft bearing".

[0070] Specifically, the system has marked the "gearbox high-speed shaft area" of the blower as an abnormal area (Level I), and the system maintains real-time monitoring of the "gearbox high-speed shaft area" at a sampling rate of 20kbps. At 10:15:30, the vibration RMS value exceeded the threshold, and the system started a timer. In subsequent monitoring, the signal did not fall back and was accompanied by intermittent pulse peaks until the system triggered a safety logic shutdown at 10:17:00. The system accurately marked the abnormal time period of this area as [10:15:30, 10:17:00], with a duration of 90 seconds. The system extracted all vibration waveforms, temperature curves, and audio clips within this time period.

[0071] During this 90-second abnormal period, the system performed dynamic time-frequency analysis on the collected data: it identified obvious modulation phenomena in the vibration signal, namely amplitude modulation with the frequency conversion as the carrier and the fault frequency as the sideband; it identified non-stationary impact characteristics, with the impact intervals being relatively regular in the early stages, but then becoming chaotic; it identified high-frequency energy release, with a continuous resonant demodulation spectrum appearing in the range of 2000Hz-4000Hz; based on the above analysis, the system determined that the working content of this area during the abnormal period was: "irregular mechanical impact and rolling element slippage friction behavior".

[0072] The system further combines the attributes of the area for comprehensive judgment: Area morphology: This area belongs to a high-speed heavy-load gear transmission structure and is located at a rigidly connected transmission chain node. Any slight instability will be amplified by the structure; Component priority: This component is a P0-level core component (high-speed shaft bearing), and its failure will directly lead to downtime; Event generation: The system puts "irregular impact and slippage behavior" into "high-speed heavy-load morphology" and superimposes the high-risk weight of "P0 level"; The system determines the performance during this abnormal time period as an abnormal event: "catastrophic friction event caused by high-speed bearing cage fracture". This event name contains richer physical meaning and maintenance guidance value than simply "large vibration".

[0073] Furthermore, real-time operating data of the blower is collected. Based on various abnormal events and the real-time operating data of the blower, the first level of impact is determined. Based on the real-time operating data of the blower and the anomaly identification system, the second level of impact is determined. Based on the first and second levels of impact, multiple abnormal influencing factors are determined, which takes into account the overall consideration of the first and second levels of impact and ensures the accuracy of multiple abnormal influencing factors.

[0074] At this point, the system collects real-time operating data of the blower via high-frequency industrial bus, including current, voltage, inlet and outlet pressure, flow rate, and valve opening. The system uses Pearson correlation coefficient or Granger causality test to compare the characteristics of the "abnormal events" (such as vibration peaks and sudden changes) determined in the previous steps with the time series of these real-time operating data point by point. This process aims to find the synchronicity between abnormal events and changes in external operating conditions. For example, is the outbreak of an abnormal event synchronized with a sudden increase in current? Is the change in vibration frequency consistent with the pace of flow regulation? If strong synchronicity exists, the first layer of influence will point to "externally induced" factors. If the operating data remains stable when an abnormal event occurs, the first layer of influence will point to "internally spontaneous" factors.

[0075] The system inputs the collected real-time working data into the benchmark model or digital twin in the "anomaly identification system"; the model stores the physical coupling relationship between various parameters of the equipment in a healthy state (such as pressure-flow curve, speed-vibration threshold curve).

[0076] The system calculates the residual between real-time data and model output. By analyzing the residual distribution, it determines whether the current operating state deviates from the design conditions or health baseline. For example, under the same load, is the current power consumption abnormally high? Does the vibration spectrum violate the specific area limits in the ISO10816 standard? The second layer of impact mainly answers whether the current operating state triggers the failure mode defined by the system.

[0077] The system utilizes decision-level fusion logic (such as DS evidence theory or fuzzy logic reasoning) to cross-verify and synthesize the first and second levels of influence. If the first level shows "stable operating conditions" and the second level shows "severe deviation from the healthy model," the judgment factor is "component performance degradation." If the first level shows "severe fluctuations in operating conditions" and the second level shows "model deviation within the allowable range," the judgment factor is "external process disturbance." If both show abnormalities, the judgment factor is "complex coupling fault." By eliminating interference items (such as coincidental fluctuations in operating conditions) through this logic, the system finally outputs a conclusive and physically meaningful "abnormal influencing factor."

[0078] Specifically, at 10:15:30, a catastrophic friction event caused by the fracture of the high-speed bearing cage occurred in the blower. The system extracted real-time operating data during the abnormal time period [10:15:30, 10:17:00]. Data analysis showed that during this period, the blower motor current remained around 120A (rated value 125A), with a fluctuation range of less than ±1%; the outlet air pressure remained at 45kPa, with no obvious pressure pulses; the inlet valve opening was locked at 75% and did not move. Related conclusions: When the abnormal event occurred, the external load and process parameters were extremely stable, and there was no overload or transient fluid impact. First impact content: It was determined that the abnormality was unrelated to the external operating conditions, pointing to "spontaneous mechanical failure inside the equipment".

[0079] The system compares real-time vibration data with the "gearbox health baseline model" in the anomaly identification system; model calculation: at the current speed, the normal vibration RMS threshold of the bearing housing should be 2.5 mm / s; however, the actual collected data shows that the vibration value soars to over 8.0 mm / s, and the high-frequency demodulation spectrum amplitude exceeds the baseline by 400%; system judgment: this serious deviation exceeds the "aging" or "wear" range allowed by the model and enters the characteristic region of "sudden fracture"; second impact content: it is determined that the current state has triggered the "catastrophic structural failure mode" in the system.

[0080] The system integrates two factors for decision-making: Input A (first level of influence): Stable external environment, no precipitating factors > eliminates "improper operation" or "overload"; Input B (second level of influence): Significant deviation from the health model, characteristics consistent with fracture > confirms "internal failure"; Comprehensive judgment: Combining the bearing's operating time (reached the end of its design life) and vibration characteristics (random impact of cage fracture), the system ultimately determined the abnormal influencing factor to be: "structural fracture of the cage caused by the exhaustion of the high-speed shaft bearing's fatigue life". This precise identification of the factor provides a direct basis for the subsequent development of a maintenance plan.

[0081] Therefore, the affected components of each abnormal influencing factor are marked, and multiple sub-influence routes are determined based on each affected component, the corresponding abnormal influencing factor, and the current operating mode of the blower. Based on the multiple sub-influence routes, the current operating route of the blower, and the corresponding data transmission route, a corresponding abnormal influence route is formed. This approach takes into account the overall consideration of each affected component, the corresponding abnormal influencing factor, and the current operating mode of the blower, ensuring the accuracy of the multiple sub-influence routes.

[0082] At this point, the system performs fault tree analysis (FTA) or topology mapping on the identified abnormal influencing factors to accurately pinpoint the specific components where the fault occurs at the physical level, i.e., the "affected components." This process not only marks the directly failed components (such as the source component) but also identifies the secondary components that are directly related to or subject to force transmission based on the mechanical structure of the equipment. For example, if the factor is "lubricating oil contamination," the affected components are not only the "oil pump" but also include the lubricated "gears" and "bearings." The system highlights these components in the topology map as the set of nodes for subsequent route construction.

[0083] The system combines the blower's "current operating mode" (such as variable frequency regulation mode, constant voltage mode, bypass return mode) to analyze the transmission logic of fault energy or abnormal signals between components. For physical faults, it analyzes how mechanical vibration, thermal stress, or fluid impact is transmitted from one component to another (e.g., rotor imbalance > bearing > bearing housing > foundation). For operating condition effects, it analyzes how the feedback loop of the control system is affected under specific modes (e.g., surge occurs > flow rate decreases > PLC issues command to increase inverter frequency > motor overload). The system defines these local, unidirectional transmission paths as "sub-influence routes," with each route representing an independent causal dimension.

[0084] The system aligns and merges the physical-level "sub-influence routes" with the information-level "data transmission routes" (sensors > gateway > cloud platform) and the control-level "current working routes" (given value > controller > actuator) in time and space. The system constructs a multi-dimensional directed graph, where nodes represent components or data points and edges represent influence relationships. The resulting "abnormal influence routes" serve as a panoramic multi-dimensional fault mapping map, which not only shows where the fault is, but also presents the entire process from the physical occurrence of the fault to the system response.

[0085] Specifically, the system has identified the abnormal influencing factor of the blower as "high-speed shaft bearing cage fracture"; the current operating mode is "constant pressure variable frequency control mode"; based on the influencing factor of "cage fracture", the system performs fault source tracing and propagation analysis in the digital twin model of the blower; direct marking: the core influencing component is identified as "high-speed shaft bearing (component ID: B-002)"; associated marking: analysis found that the severe vibration generated by the fracture will be directly transmitted to "gearbox housing (component ID: G-001)" and transmitted to "motor drive end (component ID: M-001)" through the coupling; the system marks these three components as the key influencing nodes of this abnormal event.

[0086] The system, combined with the "constant pressure variable frequency control" mode, derives two key sub-influence routes: Sub-influence route A (mechanical physical chain): high-speed shaft bearing (breakage) > high-frequency impact > causing severe vibration of the gearbox housing > transmitted to the motor shaft through the coupling > inducing abnormal vibration of the generator rotor; Sub-influence route B (operating condition control chain): impeller dynamic imbalance due to bearing damage > causing air pressure fluctuation at the outlet > pressure sensor detects pressure drop > PID controller (in constant pressure mode) outputs an increase frequency command > frequency converter increases motor speed > vibration further intensifies (positive feedback vicious cycle).

[0087] The system integrates the above sub-routes with the data flow to form a panoramic anomaly impact route: [Physical Propagation Layer]: High-speed shaft bearing (B-002) — (vibration and shock) —> Gearbox (G-001) — (torque fluctuation) —> Motor (M-001); [Data Sensing Layer]: Vibration sensor (VS-03) — (high-frequency signal) —> Edge gateway — (data packet) —> Anomaly identification engine; [Control Response Layer]: Pressure sensor (PT-01) — (low-pressure signal) —> PLC controller — (frequency upsampling command) —> Inverter (Inv-01) — (high-frequency carrier) —> Motor winding; [Fusion Conclusion]: The anomaly impact route map clearly shows that the bearing fracture caused physical vibration. This vibration was captured by the sensor and triggered an alarm (data chain). On the other hand, it interfered with the constant pressure control, causing the system to erroneously upsampling (control chain), ultimately forming a vicious closed loop of continuously expanding vibration.

[0088] refer to Figure 6 In step S15, the specific steps are as follows: S151: In the abnormal impact route, multiple abnormal impact road segments are identified based on the identification of the abnormal impact route. Each abnormal impact road segment corresponds to at least one abnormal component in the blower. The corresponding abnormal impact node is determined based on the road segment shape and the corresponding abnormal component of each abnormal impact road segment, so as to collect multiple abnormal impact nodes. S152: Determine the corresponding node location based on the detection of each abnormally affected node, mark the node shape of each abnormally affected node, determine the maintenance area based on the node location of each abnormally affected node and the usage scenario of the blower, and determine the maintenance content based on the maintenance area and the node shape of each abnormally affected node. S153: Based on the tracing of past maintenance events of the blower, determine the corresponding maintenance system, determine the first maintenance coefficient according to the maintenance system and the content to be maintained, determine the second maintenance coefficient according to the maintenance system and the corresponding abnormal parts, and determine the corresponding abnormal maintenance measures based on the mapping relationship table of the first maintenance coefficient, the second maintenance coefficient and abnormal maintenance measures.

[0089] In the embodiments of this application, the system performs graph theory-level logical segmentation on the constructed "abnormal impact route". Since the impact route usually covers a continuous process of energy flow, mechanical transmission or fluid transmission, the system divides the entire route into several "abnormal impact segments" based on the discontinuity of physical connection or the logical level of signal transmission. Each segment represents the transmission interval of fault energy between two key components. The division principle follows the "change of transmission medium" or "change of connection method". For example, from "rotor" to "bearing" is one segment (mechanical contact), and from "bearing seat" to "foundation" is another segment (rigid connection). In this way, the complex long chain of faults is decoupled into multiple easily manageable local segments, and each segment necessarily maps to at least one specific "abnormal component" in the blower's physical structure.

[0090] For each defined road segment affected by anomalies, the system performs multi-dimensional feature analysis to determine monitoring nodes; road segment morphology analysis: assesses the physical connection characteristics of the road segment, such as rigid connection, flexible connection, or fluid coupling; rigid connection road segments typically require monitoring vibration transmissibility, while flexible connection road segments require monitoring displacement or alignment; abnormal component attribute analysis: combines the material properties (such as metal, rubber) and motion properties (such as rotation, stationary) of the abnormal components corresponding to the road segment; the system integrates these two dimensions to determine "anomaly impact nodes" at the geometric boundaries or sensitive points of the road segment; anomaly impact nodes are typically defined as the most critical physical locations for maintenance operations, such as bolt fastening surfaces, sealing grooves, centerlines, or sensor mounting bases.

[0091] After identifying all nodes affected by anomalies, the system initiates a targeted data acquisition mechanism. This is no longer a general acquisition of all equipment, but rather a "focused acquisition" targeting specific nodes. Based on the physical environment of the nodes (such as high temperature, high pressure, and strong electromagnetic interference), the system dynamically calls upon suitable sensor resources or adjusts sampling parameters (such as increasing the sampling rate and enabling anti-aliasing filtering). The acquired data includes the physical state parameters of the nodes (displacement, stress, temperature), environmental parameters, and relative position data between nodes. The system encapsulates this data into digital twin images of the nodes, providing a quantitative basis for subsequent precise maintenance.

[0092] Specifically, the "abnormal impact route" of the blower has been determined as follows: high-speed shaft bearing (fault source) > gearbox housing > coupling > motor shaft; based on the mechanical connection relationship, the system divides the entire impact route into two main abnormal impact segments: Segment 1: from the high-speed shaft bearing to the lower gearbox housing, this segment belongs to the internal transmission structure, and the corresponding abnormal components include the "high-speed shaft bearing" and the "lower gearbox housing"; Segment 2: from the gearbox output shaft flange to the motor shaft flange, this segment belongs to the external transmission connection, and the corresponding abnormal components include the "diaphragm coupling" and the "motor input shaft".

[0093] For section one: the system analysis shows its section morphology as "high-speed, heavy-load, enclosed rigid connection," with the abnormal component being a precision bearing. To capture internal damage and impact, the system identifies the "outer vertical radial surface of the gearbox high-speed shaft bearing housing" as the abnormal impact node N-01, a "critical sensitive point" on the vibration transmission path that best reflects the internal state of the bearing. For section two: the system analysis shows its section morphology as "flexible torque transmission," with the abnormal component including the coupling. To assess the impact of the fault on transmission accuracy, the system identifies the "horizontal alignment reference plane between the coupling cover and the motor base" as the abnormal impact node N-02, a key location for detecting whether coaxiality has failed due to vibration.

[0094] Node N-01 data acquisition: The system instructs the piezoelectric vibration sensor located at this position to acquire acceleration waveforms at a high sampling rate of 20kHz, focusing on capturing the high-frequency envelope signal of bearing failure, while simultaneously acquiring temperature data at this point; Node N-02 data acquisition: The system uses a laser alignment instrument or displacement sensor to acquire the radial runout and axial displacement of this node, quantifying the degree of deformation of road segment two; The system obtained node N-01 (high-frequency impact data) and node N-02 (deformation data).

[0095] Furthermore, the corresponding node locations are determined based on the detection of each abnormally affected node, and the node shape of each abnormally affected node is marked. The maintenance area is determined based on the node locations of each abnormally affected node and the usage scenario of the blower. The maintenance content is determined based on the maintenance area and the node shape of each abnormally affected node. This overall consideration of the maintenance area and the node shape of each abnormally affected node ensures the accuracy of the maintenance content.

[0096] At this point, the system uses the equipment's three-dimensional digital twin model to map each abstract anomaly-affecting node onto a physical coordinate system, determining its precise "node location." This process involves resolving the node's installation coordinates, hierarchical relationship (e.g., whether it's located at the bottom or top layer of the equipment), and surrounding envelope boundaries. Based on the geometric features and physical properties of the components at the node, the system marks the node with a "node morphology." This morphology marking is not limited to simple names; the node morphology of anomaly-affecting nodes includes topological features such as "rotational mating surface," "cantilever support structure," "flange connection interface," or "closed fluid cavity." Thus, the node location of the anomaly-affecting node encompasses its corresponding installation coordinates, hierarchical relationship, and surrounding envelope boundaries.

[0097] The system introduces the parameter of "blower usage scenario", including personnel accessibility, safe working distance, environmental interference factors (such as high temperature area, high pressure pipeline area) and disassembly and assembly logic (such as whether the top cover needs to be hoisted); the system uses spatial clustering algorithm to merge abnormal impact nodes that are spatially adjacent and have the same maintenance scenario attributes (for example, both need to open the same maintenance port, or both need the same protective equipment); the merged spatial set is defined as "area to be maintained".

[0098] For each identified maintenance area, the system analyzes the "node morphology" of all nodes within that area and, combined with the fault type (such as wear, fracture, misalignment), derives specific "maintenance content." The content definition follows standardized operational language. For example, for a rotating node with a "wear morphology," the content is defined as "measuring and repairing mating surface dimensions"; for a fastener node with a "fracture morphology," the content is defined as "removal of debris and rethreading of threads"; and for a connection node with an "offset morphology," the content is defined as "geometric alignment verification (laser alignment)." The maintenance content forms a detailed work order for that area. Here, the maintenance content refers to the specific maintenance tasks determined to resolve the fault. The maintenance system includes a mapping between fault modes and repair solutions.

[0099] Specifically, the system has identified two key abnormal impact nodes of the blower: N-01 (gearbox high-speed shaft bearing housing) and N-02 (coupling cover side); Node N-01: The system locates its coordinates in the digital model as inside the gearbox, vertically at the input end of the high-speed shaft; based on its characteristics of bearing high-speed rotating parts and being enclosed by the housing, its node form is marked as "precision rotary support cavity (closed type)"; Node N-02: The system locates its coordinates as between the output end of the gearbox and the motor, on the outer circumference of the coupling flange; based on its characteristics of connecting two shafts and being exposed, its node form is marked as "open coaxial connection end face".

[0100] Scenario Analysis: The blower is currently shut down and awaiting maintenance, with an ambient temperature of 50℃. N-01 is located inside the gearbox, requiring the removal of the upper cover for access, and involves precision components, necessitating a dust-free environment. N-02 is located externally and can be directly accessed, but the space is narrow and limited by surrounding pipes. Area Division: Considering the enclosed nature of N-01 and the difficulty of disassembly (requiring hoisting), the system designates the area around N-01 and its associated gearbox upper cover (within a 1-meter radius) as "Maintenance Area A: Core Transmission Precision Area". Considering the external location of N-02 and the alignment requirements, the system designates the area around the coupling as "Maintenance Area B: Transmission Alignment Area".

[0101] For area A (node ​​N-01): Based on the node's "closed precision cavity" shape and the fault being a broken bearing, the system determines the maintenance content to be performed as follows: "Perform gearbox cover opening operation, clean metal residue from the cavity, and replace the high-speed shaft bearing assembly"; For area B (node ​​N-02): Based on the node's "open connection end face" shape and the offset caused by vibration, the system determines the maintenance content to be performed as follows: "Remove the coupling cover, clean the flange, perform laser alignment, and check the wear condition of the elastomer pins".

[0102] Therefore, a corresponding maintenance system is determined based on the tracing of past maintenance events of the blower. A first-level maintenance coefficient is determined based on this maintenance system and the content to be maintained. A second-level maintenance coefficient is determined based on this maintenance system and the corresponding abnormal components. The corresponding abnormal maintenance measures are determined based on the mapping relationship table between the first-level maintenance coefficient, the second-level maintenance coefficient, and the abnormal maintenance measures. This approach takes into account the overall consideration of the mapping relationship table between the first-level maintenance coefficient, the second-level maintenance coefficient, and the abnormal maintenance measures, ensuring the accuracy of the corresponding abnormal maintenance measures. At the same time, it realizes the overall consideration of the content to be maintained, the current working mode of the blower, and the corresponding maintenance system, thereby improving the accuracy of the abnormal maintenance measures of the blower.

[0103] At this point, the system accesses the Equipment Lifecycle Management Database (PLM / EAM) to perform in-depth tracing of past maintenance records for this model of blower and related equipment. The system utilizes Natural Language Processing (NLP) and association rule mining techniques to extract key entities and relationships from historical work orders, fault reports, and maintenance logs to construct a dynamic "maintenance system." This system not only includes the mapping between fault modes and maintenance solutions but also includes historical success rates, mean time to repair (MTTR), spare parts consumption patterns, and the coupling characteristics of "people-machine-material-method-environment." Through this system, the system can identify the optimal handling path for specific faults in different scenarios, providing data support for subsequent coefficient calculations.

[0104] The system inputs the "maintenance items" (such as "bearing replacement", "laser alignment", and "dynamic balancing verification") determined in the previous steps into the maintenance system. The system assesses the technical complexity of these items, including the precision level of the required tools, the topological complexity of the disassembly steps, and the skill and qualification requirements for maintenance personnel. The system calculates the "first-level maintenance coefficient" by combining the fluctuation rate of operation time and the first-time repair rate for such items in historical records. This coefficient mainly reflects the execution difficulty and resource consumption intensity of the maintenance operation itself. For example, for precision fitting operations that require heated disassembly and cold assembly, the coefficient will be significantly higher than that of conventional bolt tightening operations.

[0105] The system inputs "abnormal components" (such as P0-level high-speed shafts and P1-level impellers) into the maintenance system. Based on the component's BOM (Bill of Materials) level and reliability data, the system assesses the component's functional criticality, the severity of failure consequences (Safety Criticism), and economic value within the system. Combining the component's mean time between failures (MTBF) and spare parts supply cycle in historical records, the system calculates a "second-level maintenance coefficient." This coefficient mainly reflects the risk weight of component failure on production and the urgency of replacement / repair. For example, failure of a core load-bearing component can lead to a complete machine shutdown, and its coefficient is higher than that of auxiliary cooling system components.

[0106] The system establishes a high-dimensional "abnormal maintenance measure mapping table," which categorizes maintenance measures into multiple levels, such as "post-event maintenance (BM), preventive maintenance (PM), predictive maintenance (PdM), improvement maintenance (IM), or proactive reset (AR)." The system utilizes weighted decision-making algorithms (such as the Analytic Hierarchy Process (AHP) or fuzzy comprehensive evaluation) to fuse the "first-level maintenance coefficient" (execution difficulty) and the "second-level maintenance coefficient" (component importance) to generate a comprehensive decision value. The system then retrieves the entry in the mapping table that best matches this decision value and outputs the final "abnormal maintenance measure." This measure includes not only action instructions (such as "immediate shutdown") but also auxiliary resource suggestions (such as "a 300-ton crane needs to be dispatched" and "spare part A-002 needs to be used").

[0107] Specifically, the blower's maintenance requirements have been identified as "gearbox cover opening and high-speed shaft bearing replacement (high difficulty)" and "coupling alignment (medium difficulty)". The abnormal component is "high-speed shaft and bearing (P0 core component)". A system search of the blower's maintenance records for the past 5 years revealed 12 historical events related to "gearbox high-speed shaft" failures. System analysis showed that 8 cases used a "bearing replacement only" approach, with 3 of these failing again within six months due to gear damage. 4 cases used a "bearing replacement + gear flaw detection + shaft system repair" approach, extending the average service life by 200%. Based on this, the maintenance system confirms that for this part of the failure, simple replacement is a "high-risk, low-return" strategy, while systematic repair is a "high-reliability strategy".

[0108] The system analyzes the "maintenance content": the content includes "gearbox opening" and "precision bearing replacement". Historical data shows that this type of operation takes an average of 24 hours and requires a constant temperature environment and hydraulic puller tools. According to the complexity assessment model of the maintenance system (reference value is 0.0-1.0), this type of operation is judged to be extremely complex. The system calculates the first maintenance coefficient K1=0.9 (close to 1, indicating that the operation is extremely difficult and resource consumption is high).

[0109] The system analyzes the "abnormal components": the high-speed shaft bearing is a P0-level core component, and its direct failure will cause the production line to stop, resulting in a loss of hundreds of thousands of yuan per hour; the spare parts are long-cycle imported parts (procurement cycle of 8 weeks); according to the maintenance system's FMEA (Failure Mode and Effects Analysis) model, the risk priority (RPN) of this type of component failure is the highest; the system calculates the second maintenance coefficient K2=1.0 (the maximum value, indicating that the component is extremely critical and the consequences of failure are extremely serious).

[0110] The system calculates the comprehensive decision value: V=α⋅K1+β⋅K2 (assuming each weight is 0.5, V=0.5×0.9+0.5×1.0=0.95); the system queries the "Abnormal Maintenance Measures Mapping Relationship Table" and finds that the measure corresponding to the interval [0.85,1.0][0.85,1.0] is "Level I Emergency Overhaul (Full Restoration)"; the final output is: the system determines the abnormal maintenance measure as "Immediately execute a shutdown overhaul: the entire gearbox unit is lifted off the ground and returned to the factory or the expert group performs a full disassembly and inspection, replaces the entire high-speed shaft system assembly, and performs magnetic particle testing on all related gears. It is strictly forbidden to only replace the bearings partially."

[0111] Please see Figure 7 The anomaly identification system for blowers based on multimodal data includes: The vibration data module 21 is used to determine the real-time image of the blower based on the visual detection of the blower when the blower is in operation, to determine multiple component areas based on the recognition of the real-time image of the blower, and to determine the corresponding vibration data combination based on the vibration detection of each component area. The multimodal data module 22 is used to determine the multimodal frame of the blower based on the regional location of each component area and the corresponding vibration data combination, and to construct the multimodal data corresponding to the blower based on the multimodal frame of the blower, the noise data and temperature data of the blower. The anomaly identification module 23 is used to determine multiple abnormal data based on the detection of multimodal data corresponding to the blower, and mark the abnormal components corresponding to each abnormal data. Based on the multiple abnormal data, the abnormal components corresponding to each abnormal data and the overall shape of the blower, an anomaly identification system is determined, and multiple abnormal areas with different anomaly impact levels are marked. The abnormal impact route module 24 is used to determine the corresponding abnormal events in each abnormal area based on the dynamic detection of the abnormal area, determine multiple abnormal impact factors based on each abnormal event, the real-time working data of the blower and the abnormal identification system, and form the corresponding abnormal impact route. The anomaly maintenance module 25 is used to identify multiple anomaly impact nodes based on the identification of anomaly impact routes, determine the maintenance content based on the node location, corresponding node shape and blower usage scenario of each anomaly impact node, and determine the corresponding anomaly maintenance measures based on the maintenance content, the blower's current working mode and the corresponding maintenance system.

[0112] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for identifying an anomaly of a blower based on multi-modal data, the method comprising: The method comprises the following steps: determining a real-time image of the blower based on visual detection of the blower when the blower is in operation, determining a plurality of component regions based on recognition of the real-time image of the blower, and determining a corresponding vibration data combination according to vibration detection of each component region; determining a multi-modal framework of the blower based on the region position of each component region and the corresponding vibration data combination, and constructing multi-modal data corresponding to the blower according to the multi-modal framework of the blower, noise data and temperature data of the blower; the multi-modal framework not only contains the current vibration state, but also implies the three-dimensional geometric structure and fault importance information of the equipment; the multi-modal data contains spatial position, vibration characteristics, thermodynamic characteristics, acoustic characteristics and scene semantics; determining a plurality of abnormal data according to detection of the multi-modal data corresponding to the blower, and marking an abnormal component corresponding to each abnormal data, determining an abnormal recognition system according to the plurality of abnormal data, the abnormal component corresponding to each abnormal data and the overall form of the blower, and marking a plurality of abnormal regions of different abnormal influence levels, the abnormal recognition system being a comprehensive system for fault grading by double fusion of the structural dimension and the working condition dimension of the blower; in each abnormal region, determining a corresponding abnormal event according to dynamic detection of the abnormal region, determining a plurality of abnormal influence factors based on each abnormal event, real-time working data of the blower and the abnormal recognition system, and forming a corresponding abnormal influence route, the abnormal influence route being a panoramic multi-dimensional fault mapping graph; determining a plurality of abnormal influence nodes based on recognition of the abnormal influence route, determining a maintenance content according to the node position, the corresponding node form and the use scene of the blower of each abnormal influence node, determining a corresponding abnormal maintenance measure based on the maintenance content, the current working mode of the blower and the corresponding maintenance system, and the abnormal influence node being defined as the most critical physical position of the maintenance operation; the node position of the abnormal influence node covers the corresponding installation coordinates, hierarchical relationship and surrounding envelope boundary; the node form of the abnormal influence node contains the corresponding form topological characteristics; the maintenance content refers to a specific maintenance task determined to solve the fault; the maintenance system contains the mapping of the fault mode and the repair scheme. 2.The method of claim 1, wherein, The method comprises the following steps: marking the position of the blower, collecting a plurality of working data of the blower, determining the working state of the blower according to the plurality of working data of the blower and the corresponding overall form, controlling the working state of the blower, triggering the positioning shooting of a plurality of cameras around the position of the blower to visually detect the blower, and collecting a plurality of sub-images of the blower in different dimensions, and constructing the real-time image of the blower according to the synthesis of the plurality of sub-images; The real-time image of the air blower is dynamically identified, and a plurality of component marks are determined in the identification process. The corresponding component area is determined according to the traceability of each component mark, so as to trigger the vibration detection of each component area. At the same time, a plurality of vibration data are determined according to the vibration detection of each component area. Based on the plurality of vibration data, the corresponding vibration detection position and the corresponding component area, the corresponding vibration data combination is determined. 3.The method of claim 1, wherein, The multi-modal framework of the air blower is determined based on the area position of each component area and the corresponding vibration data combination. The multi-modal data corresponding to the air blower is constructed according to the multi-modal framework of the air blower, the noise data and the temperature data of the air blower, including: In the plurality of component areas, the corresponding area position is determined based on the detection of each component area. The multi-modal framework of the air blower is determined according to the area position of each component area, the corresponding area priority and the corresponding vibration data combination. A plurality of environmental data are determined based on the detection of the location of the air blower. The noise data and the temperature data of the air blower are determined according to the screening of the plurality of environmental data, and the use scene of the air blower is marked. The primary data is determined according to the multi-modal framework of the air blower and the use scene of the air blower, and the multi-modal data corresponding to the air blower is constructed based on the primary data, the noise data and the temperature data of the air blower. 4.The method of claim 1, wherein, A plurality of abnormal data are determined according to the detection of the multi-modal data corresponding to the air blower, and the abnormal components corresponding to each abnormal data are marked. The abnormal recognition system is determined according to the plurality of abnormal data, the abnormal components corresponding to each abnormal data and the overall form of the air blower, and a plurality of abnormal areas of different abnormal influence levels are marked, including: The multi-modal data corresponding to the air blower is dynamically detected, and a plurality of data abnormality marks are determined in the detection process. The corresponding abnormal data is determined according to the traceability of each data abnormality mark, so as to collect a plurality of abnormal data. 5.The method of claim 4, wherein, A plurality of abnormal data are determined according to the detection of the multi-modal data corresponding to the air blower, and the abnormal components corresponding to each abnormal data are marked. The abnormal recognition system is determined according to the plurality of abnormal data, the abnormal components corresponding to each abnormal data and the overall form of the air blower, and a plurality of abnormal areas of different abnormal influence levels are marked, including: The corresponding abnormal component is determined based on the detection of each abnormal data. The first abnormal recognition content is determined according to the abnormal component corresponding to each abnormal data and the overall form of the air blower. The second abnormal recognition content is determined according to each abnormal data and the current working mode of the air blower. The abnormal recognition system is determined according to the first abnormal recognition content and the second abnormal recognition content. In the abnormal recognition system, a plurality of abnormal areas are determined based on the identification of the abnormal recognition system, and the corresponding abnormal influence level is determined according to the data combination of each abnormal area, the corresponding area position and the corresponding component function. 6.The method of identifying an abnormality of a blower based on multi-modal data according to claim 1, wherein, In each abnormal area, the corresponding abnormal event is determined according to the dynamic detection of the abnormal area. A plurality of abnormal influence factors are determined based on each abnormal event, the real-time working data of the air blower and the abnormal recognition system, and the corresponding abnormal influence route is formed, including: Real-time monitoring of each abnormal area, and marking the abnormal time period of each abnormal area, dynamically identifying the abnormal time period of each abnormal area, and determining the working content of the abnormal area in the abnormal time period in the identification process, determining the corresponding abnormal event according to the area form of the abnormal area, the corresponding working content and the corresponding component priority. 7.The method of claim 6, wherein, According to the dynamic detection of the abnormal area, the corresponding abnormal event is determined, and based on each abnormal event, the real-time working data of the blower and the abnormal identification system, a plurality of abnormal influence factors are determined, and a corresponding abnormal influence route is formed. Collecting real-time working data of the blower, determining the first heavy influence content according to each abnormal event and the real-time working data of the blower, determining the second heavy influence content according to the real-time working data of the blower and the abnormal identification system, and determining a plurality of abnormal influence factors based on the first heavy influence content and the second heavy influence content; Marking the influence components of each abnormal influence factor, determining a plurality of sub-influence routes according to each influence component, the corresponding abnormal influence factor and the current working mode of the blower, and forming a corresponding abnormal influence route based on the plurality of sub-influence routes, the current working route of the blower and the corresponding data transmission route. 8.The method of identifying an abnormality of a blower based on multi-modal data according to claim 1, wherein, According to the identification of the abnormal influence route, a plurality of abnormal influence nodes are determined, the maintenance content is determined according to the node position of each abnormal influence node, the corresponding node form and the use scene of the blower, and the corresponding abnormal maintenance measure is determined based on the maintenance content, the current working mode of the blower and the corresponding maintenance system, including: In the abnormal influence route, a plurality of abnormal influence sections are determined according to the identification of the abnormal influence route, each abnormal influence section corresponds to at least one abnormal component in the blower, the corresponding abnormal influence node is determined according to the section form of each abnormal influence section and the corresponding abnormal component, and a plurality of abnormal influence nodes are collected; According to the detection of each abnormal influence node, the node position of each abnormal influence node is determined, and the node form of each abnormal influence node is marked, the maintenance area is determined according to the node position of each abnormal influence node and the use scene of the blower, and the maintenance content is determined according to the maintenance area and the node form of each abnormal influence node. 9.The method of claim 8, wherein, According to the identification of the abnormal influence route, a plurality of abnormal influence nodes are determined, the maintenance content is determined according to the node position of each abnormal influence node, the corresponding node form and the use scene of the blower, and the corresponding abnormal maintenance measure is determined based on the maintenance content, the current working mode of the blower and the corresponding maintenance system, including: Based on the traceability of the past maintenance events of the blower, the corresponding maintenance system is determined, the first heavy maintenance coefficient is determined according to the maintenance system and the maintenance content, the second heavy maintenance coefficient is determined according to the maintenance system and the corresponding abnormal component, and the corresponding abnormal maintenance measure is determined based on the mapping relationship table of the first heavy maintenance coefficient, the second heavy maintenance coefficient and the abnormal maintenance measure.

10. A blower abnormality identification system based on multi-modal data, characterized by, The abnormal identification system of the blower based on multi-modal data is applied to the abnormal identification method of the blower based on multi-modal data as claimed in any one of claims 1-9.

Images