Truck scale anomaly detection method, device and equipment and storage medium

Abstract

The invention relates to the technical field of data processing, and discloses a truck scale anomaly detection method, device and equipment and a storage medium, and the method comprises the steps: obtaining the operation state data and peripheral image data of a truck scale when a to-be-weighed vehicle drives onto the truck scale; processing the operation state data and the peripheral image data by using a preset target detection model, and performing cross validation and correlation analysis on a processing result to obtain a preliminary abnormality diagnosis result and a potential fault prediction result; and generating a target anomaly detection report according to the preliminary anomaly diagnosis result and the potential fault prediction result. And when the to-be-weighed vehicle is driven into the truck scale, the running state data and the peripheral image data are processed by using a preset target detection model, then cross validation and correlation analysis are performed, and a target anomaly detection report is generated. Abnormal conditions in the running process of the truck scale can be found more timely and accurately, a detailed basis is provided for subsequent maintenance and repair, and stable running of the truck scale is effectively guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of data processing, and in particular, to a method, device, equipment and storage medium for detecting abnormal conditions of a weighbridge. Background Art

[0002] A weighbridge is a large-scale measuring instrument, and its working principle is mainly based on the combination of mechanics and electronics. With the in-depth development of sensor manufacturing technology and microelectronics technology, new technologies have been continuously injected into weighbridges. At the same time, it is also necessary to handle abnormal situations that occur in weighbridges. However, the existing methods for obtaining abnormal data of weighbridges have a single dimension, and most of them only focus on obtaining the weighing data of weighbridges, and have a weak ability to synchronously obtain the operating state data and surrounding image data of weighbridges, making it difficult to comprehensively understand the overall operating conditions of weighbridges and difficult to detect potential fault hazards in advance. Summary of the Invention

[0003] The main purpose of the present application is to provide a method, device, equipment and storage medium for detecting abnormal conditions of a weighbridge, aiming to solve the technical problem that most of the existing methods for obtaining abnormal data of weighbridges only focus on obtaining the weighing data of weighbridges and do not comprehensively understand the overall operating conditions of weighbridges.

[0004] To achieve the above object, the present application proposes a method for detecting abnormal conditions of a weighbridge, and the method for detecting abnormal conditions of a weighbridge includes:

[0005] When a vehicle to be weighed drives onto the weighbridge, obtain the operating state data and surrounding image data of the weighbridge;

[0006] Use a preset target detection model to process the operating state data and the surrounding image data, and perform cross-verification and correlation analysis on the processing results to obtain a preliminary abnormal diagnosis result and a potential fault prediction result;

[0007] Generate a target abnormal detection report according to the preliminary abnormal diagnosis result and the potential fault prediction result.

[0008] Optionally, the target detection model includes a first detection model and a second detection model;

[0009] The step of using a preset target detection model to process the operating state data and the surrounding image data, and performing cross-verification and correlation analysis on the processing results to obtain a preliminary abnormal diagnosis result and a potential fault prediction result includes:

[0010] Perform data analysis on the operating state data based on the first detection model to obtain a first abnormal feature and a sensor performance index;

[0011] Performing target recognition on the peripheral image data based on the second detection model to obtain second abnormal features and mechanical structure features;

[0012] Performing cross-verification on the first abnormal features and the second abnormal features to generate a preliminary abnormal diagnosis result;

[0013] Performing correlation analysis on the sensor performance indicators and the mechanical structure features to obtain a potential fault prediction result.

[0014] Optionally, the step of performing cross-verification on the first abnormal features and the second abnormal features to generate a preliminary abnormal diagnosis result includes:

[0015] Performing feature matching on the first abnormal features and the second abnormal features to obtain a matching result;

[0016] When the matching result is consistent, determining the first abnormal feature and the second abnormal feature as definite abnormalities and generating a first-level abnormal alarm;

[0017] When the matching result is inconsistent, performing weight assignment on the first abnormal features and the second abnormal features based on the fault priority rule to generate a second-level abnormal alarm;

[0018] Determining a preliminary abnormal diagnosis result according to the first-level abnormal alarm and / or the second-level abnormal alarm.

[0019] Optionally, the step of performing correlation analysis on the sensor performance indicators and the mechanical structure features to obtain a potential fault prediction result includes:

[0020] Performing comparison processing on the time-series data of the sensor performance indicators and the spatial data of the mechanical structure features to generate a set of feature pairs with spatio-temporal correlation;

[0021] When a sensor abnormal point and / or a mechanical deformation area appears in the set of feature pairs, generating a third-level abnormal alarm;

[0022] When the number of the third-level abnormal alarms is greater than a preset abnormal threshold, performing redundant data elimination on the coverage ranges of the sensor abnormal points and the mechanical deformation areas, and retaining the target feature pairs with a confidence level higher than a preset confidence threshold in the set of feature pairs;

[0023] Outputting a potential fault prediction result based on the target feature pairs.

[0024] Optionally, the step of generating a target abnormal detection report according to the preliminary abnormal diagnosis result and the potential fault prediction result includes:

[0025] Determine the abnormal data set of the weighbridge according to the preliminary abnormal diagnosis result and the potential fault prediction result;

[0026] Match the abnormal data set with the first historical fault database, and calculate the similarity between the abnormal faults in the abnormal data set and the historical faults;

[0027] Determine the fault analysis report and repair plan of the abnormal fault according to the similarity;

[0028] Generate the target abnormal detection report of the weighbridge during the current detection process based on the abnormal data set, the fault analysis report and the repair plan.

[0029] Optionally, after the step of generating the target abnormal detection report of the weighbridge during the current detection process based on the abnormal data set, the fault analysis report and the repair plan, it further includes:

[0030] When receiving the target problem input by the target user, parse the fault description in the target problem through the natural language processing module to obtain the key fault feature vector;

[0031] Perform semantic matching between the key fault feature vector and the target abnormal detection report to determine the fault root cause node;

[0032] Generate a voice interaction instruction based on the fault root cause node, and return the fault reply corresponding to the target problem to the target user through the voice interaction instruction.

[0033] Optionally, before the step of obtaining the operation state data and surrounding image data of the weighbridge when the vehicle to be weighed drives onto the weighbridge, it further includes:

[0034] When the weighing sensors and instruments in the weighbridge receive the start signal sent by the platform, perform intelligent detection on the core components of the weighbridge through a machine learning model to obtain component state data;

[0035] Match the component state data based on the second historical fault database, and classify the matching result through deviation parameters within a preset range to obtain a classification result and corresponding abnormal feature codes;

[0036] If the classification result is the first abnormal result, match the target calibration instruction corresponding to the first abnormal result according to the preset abnormal type mapping table, and after the calibration instruction is executed, send a ready signal to the alarm module of the monitoring system through the message bus;

[0037] If the classification result is the second abnormal result, the self-check processing module of the weighbridge is used to persistently store the abnormal feature code, send a maintenance request signal with a time limit identifier to the work order module of the monitoring system according to the preset priority rule, and update the abnormal feature code after the maintenance request signal disappears.

[0038] In addition, to achieve the above object, the present application also proposes a weighbridge abnormal detection device, which includes:

[0039] A data acquisition module, configured to acquire the operation state data and the surrounding image data of the weighbridge when a vehicle to be weighed drives onto the weighbridge;

[0040] A data processing module, configured to process the operation state data and the surrounding image data by using a preset target detection model, and perform cross-verification and correlation analysis on the processing results to obtain a preliminary abnormal diagnosis result and a potential fault prediction result;

[0041] A report generation module, configured to generate a target abnormal detection report according to the preliminary abnormal diagnosis result and the potential fault prediction result.

[0042] In addition, to achieve the above object, the present application also proposes a weighbridge abnormal detection device, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the weighbridge abnormal detection method as described above.

[0043] In addition, to achieve the above object, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium, and when the computer program is executed by a processor, it implements the steps of the weighbridge abnormal detection method as described above.

[0044] In the present application, when a vehicle to be weighed drives onto the weighbridge, the operation state data and the surrounding image data of the weighbridge are acquired; the operation state data and the surrounding image data are processed by using a preset target detection model, and cross-verification and correlation analysis are performed on the processing results to obtain a preliminary abnormal diagnosis result and a potential fault prediction result; a target abnormal detection report is generated according to the preliminary abnormal diagnosis result and the potential fault prediction result. By acquiring the operation state data and the surrounding image data simultaneously when a vehicle to be weighed drives onto the weighbridge, and performing cross-verification and correlation analysis on the two types of data after processing by using a preset target detection model, and finally generating a target abnormal detection report. Compared with the traditional detection method with a single data source, it can more timely and accurately detect abnormal situations in the operation process of the weighbridge, provide a detailed basis for subsequent maintenance and repair, and effectively ensure the stable operation of the weighbridge. Brief Description of the Drawings

[0045] The drawings herein are incorporated into and form a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0046] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a schematic flowchart of the first embodiment of the vehicle scale anomaly detection method of the present application;

[0048] Figure 2 It is a structural diagram of data processing of the vehicle scale anomaly detection method of the present application;

[0049] Figure 3 It is a schematic flowchart of the second embodiment of the vehicle scale anomaly detection method of the present application;

[0050] Figure 4 It is a schematic structural diagram of the first detection model of the present application;

[0051] Figure 5 It is a schematic structural diagram of the second detection model of the present application;

[0052] Figure 6 It is a schematic flowchart of the third embodiment of the vehicle scale anomaly detection method of the present application;

[0053] Figure 7 It is a schematic module structure diagram of the vehicle scale anomaly detection device in the embodiment of the present application;

[0054] Figure 8 It is a schematic device structure diagram of the hardware operating environment involved in the vehicle scale anomaly detection method in the embodiment of the present application.

[0055] The implementation, functional features, and advantages of the present application will be further described with reference to the embodiments and the drawings. Detailed Description of the Embodiments

[0056] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0057] To better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings of the specification and the specific implementation manners.

[0058] The main solution of the embodiment of the present application is: when the vehicle to be weighed enters the truck scale, the operating status data and surrounding image data of the truck scale are obtained; the operating status data and the surrounding image data are processed using a preset target detection model, and the processing results are cross-validated and correlated to obtain preliminary abnormal diagnosis results and potential fault prediction results; a target abnormality detection report is generated based on the preliminary abnormal diagnosis results and the potential fault prediction results.

[0059] A truck scale is a large-scale measuring instrument, and its working principle is mainly based on the combination of mechanics and electronics. When vehicles and goods are parked on the load-bearing platform of the truck scale, the gravity generated will be transmitted to the weighing sensor. With the in-depth development of sensor manufacturing technology and microelectronics technology, new technologies are constantly being injected into truck scales. At the same time, there are also abnormal situations in the weighing process of truck scales, which have caused significant economic losses to enterprises and users.

[0060] An extensive investigation and analysis of the cheating forms of automobile weighing scales was conducted. The results show that cheating in the automobile weighing scale measurement process can be divided into: human cheating and technical cheating. Human cheating: refers to the form of cheating that exploits management loopholes in the automobile weighing scale measurement process. It includes vehicle following, incomplete weighing, vehicle smuggling, repeated weighing, data tampering, personnel collusion, etc. Technical cheating: refers to the cheating form of tampering with measurement data by installing cheating devices on key measurement modules such as weighing sensors, weighing instruments, junction boxes, and communication cables of automobile weighing scales. This type of cheating is highly technical, concealed, and difficult to detect.

[0061] Existing anti-cheating detection methods have a single dimension for obtaining abnormal data on truck scales. Most of them only focus on obtaining the weighing data of the truck scale. Their ability to synchronously obtain the operating status data of the truck scale and surrounding image data is weak, which makes the understanding of the overall operating status of the truck scale insufficient and makes it difficult to discover potential fault hazards in advance.

[0062] The present application provides an abnormal data detection method based on a large model. In terms of alarm statistics, it integrates the alarm information generated by instruments, sensors, scales and other equipment in various areas of the warehouse area, classifies and counts them according to dimensions such as alarm type, time, and location, and generates intuitive alarm trend charts and proportion charts to help decision-makers quickly grasp the alarm dynamics and allocate resources in a timely manner to deal with potential risks.

[0063] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, anomaly detection and intelligent response, such as an instrument equipped with intelligent AI, or an electronic device capable of realizing the above functions. The following takes the anti-cheating anomaly detection platform for automobile scales as an example to illustrate this embodiment and the following embodiments.

[0064] Based on this, the embodiments of the present application provide a method for detecting abnormalities in a weighbridge. Refer to Figure 1 , Figure 1 which is a schematic flowchart of the first embodiment of the method for detecting abnormalities in the weighbridge of the present application.

[0065] In this embodiment, the method for detecting abnormalities in the weighbridge includes:

[0066] Step S10, when the vehicle to be weighed drives onto the weighbridge, obtain the operation status data and surrounding image data of the weighbridge.

[0067] It should be noted that the operation status data is a set of physical quantity data reflecting the real-time working status of the weighbridge and its sensors, such as sensor data: sensor temperature, humidity, tilt angle, voltage, the operation status, connection status, and shell-breaking situation of each support sensor; instrument data: whether there is off-center load communication status (normal / offline), power supply voltage, filtering parameters, etc. The surrounding image data is the visual information of the weighbridge surrounding environment obtained through devices such as cameras and lidar, such as smoke detection, people at the weighbridge station, safety helmet detection, etc. This embodiment does not limit this.

[0068] It should be understood that the vehicle to be weighed can be a motor vehicle that needs to be weighed, have its status detected, or undergo other compliance inspections, such as trucks, lorries, etc. When the vehicle drives in, the detection platform automatically identifies the vehicle type (such as axle number matching) through features such as weight distribution and wheel position, and starts the corresponding data acquisition and detection program.

[0069] Furthermore, in order to be able to discover potential problems of the core components in advance and ensure that the weighbridge is in good operating condition during the startup phase. Before the step S10, it further includes:

[0070] When the weighing sensors and instruments in the weighbridge receive the startup signal sent by the platform, perform intelligent detection on the core components of the weighbridge through a machine learning model to obtain component status data; match the component status data based on the second historical failure database, and classify the matching result through deviation parameters within a preset range to obtain a classification result and the corresponding abnormal feature code; if the classification result is the first abnormal result, then match the target calibration instruction corresponding to the first abnormal result according to the preset abnormal type mapping table, and after the calibration instruction is executed, send a ready signal to the alarm module of the monitoring system through the message bus; if the classification result is the second abnormal result, then persistently store the abnormal feature code based on the self-check processing module of the weighbridge, send a maintenance request signal with a time limit identifier to the work order module of the monitoring system according to the preset priority rule, and update the abnormal feature code after the maintenance request signal disappears.

[0071] It should be noted that before the weighing sensor and the instrument receive the start signal sent by the platform, all the core components in the weighbridge are in the shutdown state. After the weighing sensor and the instrument receive the start signal sent by the platform, it indicates that the weighbridge is initializing. The core components refer to the components in the weighbridge that play a key role in the weighing function, including the weighing sensor (measuring weight), the instrument (data processing and display), and the mechanical connection structure (supporting and fixing components), etc. The second historical fault database stores the historical fault cases and the corresponding component status data of the weighbridge in the startup state, and is used to match the anomalies detected during startup.

[0072] In addition, it should be noted that the first abnormal result is a fault situation that the weighbridge can handle by itself. According to the preset abnormal type mapping table, the target calibration instruction corresponding to the first abnormal result can be selected to automatically perform calibration and eliminate the anomaly. The second abnormal result is a fault situation that the weighbridge cannot handle by itself, and it is necessary to give an alarm and notify the corresponding maintenance personnel for maintenance.

[0073] It can be understood that the machine learning model is used to analyze the status data of the core components of the weighbridge during startup. It can be a long short-term memory neural network based on time series. The input data includes the initial value of the weighing sensor, the power supply voltage fluctuation curve of the instrument module, the vibration spectrum of the mechanical structure, etc. The output includes component status labels (normal / abnormal), abnormal feature codes, and various signals, etc. The training data is the historical startup data of the weighbridge.

[0074] Specifically, when the weighing sensor and the instrument in the weighbridge receive the start signal sent by the platform, it triggers the machine learning model to perform intelligent detection on the core components and obtain the component status data, such as the bridge supply voltage of the sensor, the response time of the instrument, etc. Compare the detected component status data with the cases in the second historical fault database to judge the degree of deviation of the data from the normal threshold, and then classify the matching results into "the first abnormal result" (fault that can be handled by itself) or "the second abnormal result" (fault that cannot be handled by itself), and generate the corresponding abnormal feature code. For the first abnormal result, according to the preset abnormal type mapping table, match the corresponding target calibration instruction (such as zero calibration), and after performing the calibration, send a "ready signal" to the alarm module of the monitoring system through the message bus, indicating that the fault has been eliminated. For the second abnormal result, the abnormal feature code is stored in the local storage through the self-check processing module of the weighbridge, and at the same time, a maintenance request signal with a "time limit identifier" is sent to the work order module according to the preset priority rule (such as sorting by the degree of impact of the fault). When the maintenance is completed and the signal disappears, update the status of the abnormal feature code (such as marked as "handled").

[0075] In one example, when the weighbridge operator starts the weighing scale by pressing a button on the platform at the start of each shift, the vehicle scale automatically triggers a self-check program. After the communication between the load cell, the instrument, and the platform is restored, if the machine learning model detects that the output voltage of the load cell continuously exceeds the preset range, it triggers an alarm for the load cell, matches the status data with the historical database, finds a match with the "sensor zero drift" fault, triggers an alarm, determines the deviation parameter as the first abnormal result, automatically executes the zero calibration instruction, sends a ready signal to the alarm module after calibration is completed, and the fault is resolved. If it is found during self-check that the response time of the instrument is occasionally slightly delayed but does not exceed the severe fault threshold, it is determined as the second abnormal result, the abnormal feature code is stored in the self-check module, and a maintenance request with a time limit identifier of "to be processed within 24 hours" is sent to the work order module. After the maintenance personnel repair the instrument within the time limit, the system detects that the maintenance request signal has disappeared and updates the abnormal feature code to "repaired".

[0076] Step S20: Process the operation status data and the surrounding image data by using a preset target detection model, and perform cross-validation and correlation analysis on the processing results to obtain a preliminary abnormal diagnosis result and a potential fault prediction result.

[0077] It should be noted that the target detection model includes a first detection model and a second detection model. The first detection model is an algorithm model for analyzing equipment operation status data. Based on deep learning, it extracts features from the time-series data (such as temperature, pressure, current) collected by sensors to identify abnormal patterns or evaluate sensor performance. The second detection model is a computer vision model designed for surrounding image data, which is used to identify objects, abnormal forms, or mechanical structure features in images.

[0078] It can be understood that the preliminary abnormal diagnosis result is a preliminary judgment on whether there is an abnormality and the type of abnormality in the vehicle scale currently based on data processing and analysis. The potential fault prediction result is a predicted possible future fault of the vehicle scale by analyzing data trends or historical patterns.

[0079] Step S30: Generate a target abnormal detection report according to the preliminary abnormal diagnosis result and the potential fault prediction result.

[0080] It should be noted that the target abnormal detection report can be a report presented in a structured form, or a data statistics dashboard, or a 3D-style vehicle scale interface, where each fault is displayed in the corresponding abnormal area of the vehicle scale interface. This embodiment does not limit this.

[0081] In one example, refer to Figure 2 , Figure 2This is the structural diagram of data processing for the vehicle scale anomaly detection method of this application. In the figure, the data sources cover the surrounding environment data and real-time operation data. Among them, after the surrounding environment data is obtained through the environmental data acquisition, it is stored in the database through the image collection module, and then the fault analysis module and the target recognition module are driven. The recognition results can be sent to the data warehouse for processing; the real-time operation data is captured through the real-time data acquisition interface, sorted and classified through the data queue, and stored in the data warehouse, heap operation data, latest operation data value and other storage modules in real time. The two types of data are further aggregated into the "anti-cheating detection platform", combined with the structured data of the business platform, and data mining or model calculation is carried out. The processed data passes through the data concentration module, on the one hand, supporting model optimization work such as "model or type training, model data measurement and preparation management, statistical analysis", on the other hand, serving the historical detection data analysis, and at the same time triggering real-time alarms based on the anomaly model and providing data support for the real-time monitoring or warning module, finally forming a complete closed loop from data acquisition, processing, analysis to application, ensuring the accuracy and real-time of vehicle scale anomaly detection. This solution is verified in the measured data of 500 vehicle scales. The anomaly detection accuracy reaches 95.2%, which is 16.8% higher than the traditional single model method, and the false alarm rate drops to 2.3%, a decrease of 9.8%, meeting the real-time monitoring requirements. Through multi-modal collaborative analysis of sensors and images, the model successfully detects complex anomalies such as shell-breaking cheating (confidence level 92.3%) and mechanical deformation (bounding box error less than 5 pixels). The potential fault warning coverage rate is increased to 89.7%, and maintenance suggestions are generated based on the dynamic weight strategy, with a historical fault matching success rate of 91%. Experiments show that the fusion of sensor time series features and mechanical image spatial features significantly improves the diagnostic robustness, and still maintains a detection accuracy of more than 85% in complex scenarios such as light interference and data noise. Moreover, it supports voice interaction to generate diagnostic reports, with a response delay of less than 500ms, providing efficient guidance for on-site maintenance.

[0082] In this embodiment, when the vehicle to be weighed drives onto the vehicle scale, the operation state data and surrounding image data of the vehicle scale are obtained; the preset target detection model is used to process the operation state data and the surrounding image data, and cross-validation and correlation analysis are performed on the processing results to obtain a preliminary anomaly diagnosis result and a potential fault prediction result; a target anomaly detection report is generated according to the preliminary anomaly diagnosis result and the potential fault prediction result. By obtaining the operation state data and the surrounding image data simultaneously when the vehicle to be weighed drives onto the vehicle scale, and using the preset target detection model to process the two types of data and then performing cross-validation and correlation analysis, and finally generating a target anomaly detection report. Compared with the traditional detection method with a single data source, it can more timely and accurately detect anomalies in the operation process of the vehicle scale, provide detailed basis for subsequent maintenance and repair, and effectively ensure the stable operation of the vehicle scale.

[0083] Refer toFigure 3 , Figure 3 This is a schematic flowchart of the second embodiment of the vehicle scale anomaly detection method of this application. Based on the above first embodiment, the second embodiment of the vehicle scale anomaly detection method of this application is proposed.

[0084] In the second embodiment, the step S20 includes:

[0085] Step S201, perform data analysis on the operation status data based on the first detection model to obtain the first anomaly feature and the sensor performance index.

[0086] It should be noted that the first anomaly feature is the feature that deviates from the normal operation range identified after analyzing the operation status data through the first detection model, such as the voltage value exceeding the limit instantaneously, the vibration frequency fluctuating regularly, the temperature rising continuously, etc. The sensor performance index is a key parameter used to quantitatively evaluate the working state of the sensor, including sensitivity, response time, stability, drift rate, etc. This embodiment does not limit this.

[0087] Specifically, after collecting the operation status data generated by the sensor during the operation of the vehicle scale, it is necessary to perform preprocessing such as cleaning and normalization on the operation status data; then input the preprocessed operation status data into the first detection model, and the model scans the data through algorithmic logic (such as time series analysis, anomaly detection algorithms, etc.) to identify the first anomaly feature that does not conform to the normal operation mode. For example, if it is found that the voltage value of the weighing sensor is continuously lower than the standard threshold, it is determined as a "signal anomaly feature"; at the same time, the model calculates the sensor performance index based on preset rules or a trained evaluation system. For example, by comparing the historical measurement data of the sensor with the true weight value, evaluate whether its accuracy has decreased; by monitoring the sensor data update frequency, judge whether the response time meets the standard. Finally, the model outputs the first anomaly feature reflecting the current operation anomaly and the sensor performance index for quantitatively evaluating the working state of the sensor.

[0088] In an example, refer to Figure 4 , Figure 4This is a schematic diagram of the structure of the first detection model of this application. The first detection model is a hybrid deep learning architecture designed based on the analysis of multi-modal sensor time-series data. This model combines the time-series modeling ability of the Long Short-Term Memory (LSTM) network and the spatial feature extraction advantages of the Convolutional Neural Network (CNN), and solves the problem of gradient disappearance in the training of deep networks through the Residual Network (ResNet) module, realizing precise monitoring and anomaly diagnosis of the sensor operating state. The first half is the CNN module, and the input data sequentially passes through the convolutional layer (extracting local spatial patterns), the pooling layer (compressing the feature dimension), and the compression layer (dimensionality reduction and feature fusion), abstracting spatial information layer by layer using different weights and filters; the second half is the bidirectional LSTM module, where h t-1 ,h t ,h t+1 represent the hidden states at different times, transmitting time-series information, and x t-1 ,x t ,x t+1 represent the input data at different times. LSTM represents an LSTM cell, which processes time-series data and captures long-term dependencies. The bidirectional arrow represents the bidirectional LSTM structure, which processes data from the forward direction (from x t-1 to x t+1 ) and the reverse direction (from x t+1 to x t-1 ), fusing bidirectional information to better extract time-series features. The bidirectional LSTM module consists of two groups of LSTM cells, the forward and the reverse, which capture the long-term dependencies of sensor data (such as voltage trend drift or temperature periodic fluctuations) through bidirectional time-series modeling. The spatial features extracted by the CNN and the time-series states of the LSTM are interactively fused, and finally a multi-dimensional diagnostic result with both sudden anomaly detection (such as voltage drop) and slow-varying pattern analysis (such as angle offset) is output, realizing all-round dynamic monitoring of the sensor operating state.

[0089] The model uses six-dimensional time-series data as training data and input (including voltage, temperature, and the XYZ-axis angles of the gyroscope), and captures the long-range time dependencies of sensor data through a double-layer LSTM network, such as slow-varying features like continuous voltage drop trends or angle drifts. After the high-order time-series features output by the LSTM are transformed by spatial coordinates, they are input into a one-dimensional convolutional layer to extract local spatial patterns, such as the features of sudden events like voltage drops and abnormal temperature fluctuations. The core residual connection module learns complex abstract patterns while retaining the effective information at the bottom layer through cross-layer feature fusion, significantly improving the model's ability to model multi-scale features.

[0090] At the same time, the model adopts a multi-task learning framework and outputs results in parallel through two branches: 1. The anomaly classification branch identifies 8 types of predefined anomalies (such as shell cheating, data tampering, etc.) based on the fully connected layer and the Softmax function, and outputs the confidence probability; 2. The performance evaluation branch maps the features to a health score of 0 to 100 points through the Sigmoid activation function. During the training process, the focal loss function is used to alleviate the category imbalance problem, and the Huber loss function is used to improve the regression robustness of the health score, and the dynamic weight adjustment mechanism is used to balance the optimization goals of the classification and regression tasks.

[0091] Step S202: performing target recognition on the surrounding image data based on the second detection model to obtain a second abnormal feature and a mechanical structure feature.

[0092] It should be noted that the second abnormal feature is abnormal visual information extracted from the image through target recognition, such as vehicle following, incomplete weighing, vehicle entrainment, repeated weighing, etc. The mechanical structure feature is information related to the mechanical structure of the truck scale identified from the image, including component cracks, loose screws, structural deformation, surface wear, liquid leakage, etc.

[0093] Specifically, when a vehicle enters the truck scale, the anomaly detection platform uses a camera to collect image data of the surrounding area of the truck scale in real time or regularly, covering key mechanical parts of the equipment, installation areas, personnel, etc., and preprocesses the collected images, including noise reduction, brightness adjustment, resolution optimization, etc. The preprocessed image is then input into the second detection model, which analyzes the image pixel by pixel through convolutional neural networks and other technologies to identify target objects in the image, such as weighing sensors, instrument modules, etc. During the recognition process, the model ultimately outputs the second abnormal feature (such as the vehicle is not completely on the scale) and mechanical structure features (such as a 5° deviation in the installation angle of the weighing sensor), providing an intuitive visual analysis basis for subsequent fault diagnosis and structural evaluation of the truck scale.

[0094] In one example, reference Figure 5 , Figure 5This is the structural schematic diagram of the second detection model of this application. The second detection model is a computer vision residual attention network, which integrates the feature extraction ability of the residual network and the context perception advantage of the spatial attention mechanism. Its core realizes multi-scale feature fusion through 1×1 convolutional layers and cross-layer feature splicing. The input image data first passes through a splicing module to integrate context information, and then undergoes feature compression and spatial enhancement in the channel dimension through 1×1 convolutional layers; the feature map dimension is dynamically adjusted through reshaping and transpose operations, and a non-linear expression ability is introduced by combining activation functions (such as ReLU). At the same time, residual skip connections are used to retain the original valid information to avoid gradient degradation. The spatial / channel dual attention mechanism in the network further calibrates the weights of the feature map - the spatial attention focuses on abnormal regions (such as structural deformations), and the channel attention suppresses background interference. Finally, a multi-task diagnosis result with both pixel-level localization (damage bounding box) and semantic classification (five types of image abnormalities) is output, significantly improving the detection robustness in complex scenarios.

[0095] This model takes RGB images with a resolution of 640×480 as input, and through multi-stage joint optimization, outputs dual results of abnormal type recognition and mechanical structure quantitative evaluation, providing high-precision visual analysis for equipment status monitoring in industrial scenarios. The model is based on a residual network as the backbone architecture. After removing the traditional global pooling layer, an atrous spatial pyramid pooling module is introduced, which uses convolutional kernels with different dilation rates to parallelly capture multi-scale context information in the image, effectively identifying diverse features from fine cracks (local scale) to overall structural deformations (global scale). To further enhance the focusing ability on key regions, the model integrates a channel-spatial dual attention mechanism: the channel attention module dynamically calibrates the importance weights of feature channels through global average pooling, suppressing irrelevant background interference; the spatial attention module generates a spatial weight heat map through 1×1 convolution, highlighting the pixel responses of abnormal regions (such as the location of foreign object intrusion or the deformed part). The two work together, enabling the model to accurately locate the target in complex environments.

[0096] For the multi-task output requirements, the model adopts a fork-like design: 1. The abnormal classification branch outputs the probability distributions of five preset abnormalities (such as foreign object intrusion, structural deformation) through a four-layer fully connected network, and uses the focal loss function to optimize the training to alleviate the problem of unbalanced sample distribution; 2. The structural regression branch parallelly outputs the component integrity score (Sigmoid activation) and the coordinates of the key component bounding box (normalized format, supervised by Smooth L1 loss). The scoring mechanism combines a non-linear decay formula to quantify the equipment health status, and the bounding box accurately frames the damaged area, providing spatial guidance for manual review.

[0097] During the training process, the model achieves stable convergence through the AdamW optimizer and the cosine annealing learning rate scheduling strategy, and uses data augmentation methods such as random erasing and multi-scale brightness and contrast perturbation to improve the generalization ability. During actual deployment, the model only takes 22 milliseconds to complete single-frame image processing on an embedded device, supports real-time analysis at 45 frames per second at 4K resolution, with an average detection accuracy of 91.4%, and the intersection over union of the structural deformation area exceeds 0.89. Its spatial attention heatmap visualization function can intuitively display the decision-making basis of the model. For example, in the detection of the weighing platform of a weighbridge, the highlighted area of the heatmap highly coincides with the deformation position manually marked, significantly improving the credibility of the results.

[0098] The second detection model and the first detection model jointly construct a multi-modal anomaly verification system. When the anomaly confidence levels of both for the same event exceed 85%, the system automatically triggers the highest-level disposal protocol, suppressing the comprehensive false alarm rate to below 0.3% and achieving all-round intelligent monitoring and early warning of mechanical structure anomalies.

[0099] Step S203: Cross-verify the first anomaly feature and the second anomaly feature to generate a preliminary anomaly diagnosis result.

[0100] It can be understood that cross-verification aligns the first anomaly feature (such as abnormal voltage of a weighing sensor) with the second anomaly feature (such as a damaged circuit shown in the sensor appearance image) to establish a spatial or logical association between the two. For example, to confirm whether the sensor with abnormal operating data and the component detected with physical damage in the image are the same object. It uses different features to supplement the limitations of single data. If the operating data only shows parameter fluctuations, but the image data further reveals that the fluctuations are caused by component wear, then complementary analysis can clarify the root cause of the anomaly.

[0101] Furthermore, to avoid misjudgments that may occur in the abnormal judgment of single data and improve the reliability of anomaly diagnosis. The step S203 may include:

[0102] Perform feature matching on the first anomaly feature and the second anomaly feature to obtain a matching result; when the matching result is consistent, then determine the first anomaly feature and the second anomaly feature as definite anomalies and generate a first-level anomaly warning; when the matching result is inconsistent, then perform weight assignment on the first anomaly feature and the second anomaly feature based on the fault priority rule to generate a second-level anomaly warning; determine the preliminary anomaly diagnosis result according to the first-level anomaly warning and / or the second-level anomaly warning.

[0103] It should be noted that feature matching is a process of comparing and analyzing the first abnormal feature and the second abnormal feature to determine whether they point to the same fault source. When the two types of abnormal features match, the type of the abnormality can be clearly determined, and the same fault can be corroborated by multi-source data. The first-level abnormal alarm indicates a high-priority alarm generated for a deterministic abnormality, prompting maintenance personnel to handle it first.

[0104] It can be understood that the fault priority rule is a pre-set fault classification standard, which divides the priority according to the abnormal influence range, urgency, etc., and is used for weight allocation. When the matching results are inconsistent, for the unmatched abnormal features, different weights need to be assigned according to the fault priority to quantitatively evaluate the comprehensive severity of the abnormality. The second-level abnormal alarm is the alarm generated based on the weight allocation result, reflecting the comprehensive risk level of the non-deterministic abnormality.

[0105] In an example, the fluctuations of sensor data and the cracks on the appearance of the sensor are matched. If the two are associated with the same fault, that is, both the data abnormality and the hardware crack point to sensor damage, the matching result is consistent, and it is determined as a deterministic abnormality, generating a first-level abnormal alarm, indicating that it needs to be urgently processed. If the matching results are inconsistent, according to the fault priority rule, the weight of the abnormality affecting the core weighing function is set to 0.7, and the weight of the minor wear of the secondary component is set to 0.3. The comprehensive risk is calculated through weights, generating a second-level abnormal alarm to reflect the overall severity of the abnormality. After a round of abnormal detection is completed (i.e., a vehicle weighing), combining multiple first-level abnormal alarms and / or multiple second-level abnormal alarms, a preliminary abnormal diagnosis result is output.

[0106] Step S204, perform a correlation analysis on the sensor performance indicators and the mechanical structure features to obtain a potential fault prediction result.

[0107] It can be understood that the correlation analysis is to analyze the correlation between the sensor performance indicators and the mechanical structure features. For example, analyze whether the decrease in sensor accuracy is caused by the wear of mechanical components. To perform a correlation analysis on the sensor performance indicators and the mechanical structure features, the mapping relationship between the sensor performance indicators and the mechanical structure features can be established, or the correlation can be determined through machine learning, or it can be determined based on the set of features with spatio-temporal correlation between the sensor performance indicators and the mechanical structure features.

[0108] Furthermore, in order to predict faults more prospectively, reduce the probability of faults occurring, and reduce the downtime and maintenance costs caused by faults. The step S204 may include:

[0109] Perform comparison processing on the time-series data of the sensor performance indicators and the spatial data of the mechanical structure characteristics to generate a set of feature pairs with spatio-temporal correlation; when sensor anomaly points and / or mechanical deformation regions appear in the set of feature pairs, generate a third-level anomaly warning; when the number of the third-level anomaly warnings is greater than a preset anomaly threshold, perform redundant data elimination on the coverage ranges of the sensor anomaly points and the mechanical deformation regions, and retain the target feature pairs in the set of feature pairs whose confidence levels are higher than a preset confidence threshold; output a potential fault prediction result based on the target feature pairs.

[0110] It should be noted that the time-series data is a sequence of performance parameters collected by the sensor at different time points. The spatial data is information describing the spatial attributes of mechanical components, such as the three-dimensional dimensions of components, the positions and shapes of deformation regions, the loosening positions of screws, etc. The set of feature pairs is a set of "time-space-feature value" triples formed by spatio-temporally correlating the time-series data of the sensor performance indicators with the spatial data of the mechanical structure characteristics. For example, "at 10:00 on XX month XX day XX year, the accuracy of load cell A decreased by 15%, corresponding to a 2-mm deformation at mechanical support structure Y".

[0111] It can be understood that the third-level anomaly warning is a warning triggered when a sensor anomaly point or a mechanical deformation region is detected in the set of feature pairs. The preset anomaly threshold is a critical value of the warning quantity set in advance, which is used to judge whether redundant data needs to be processed, remove duplicate, low-value or low-confidence information in the set of feature pairs, and retain the core effective data to improve the analysis efficiency and accuracy. The preset confidence threshold is a confidence standard for screening feature pairs. For example, only feature pairs with a confidence level higher than 80% are retained to ensure the reliability of the analysis results. The target feature pairs are high-confidence spatio-temporal correlation feature pairs retained after redundant elimination and confidence screening, which can be used as the basis for generating potential fault prediction results.

[0112] In an example, when performing comparison processing on the fluctuation data of the sensor accuracy over time and the data of the position and degree of component deformation, first associate the time points of the sensor accuracy decrease with the deformation regions of the corresponding mechanical components to form a combination of "time-position-abnormal value". Then traverse the set of feature pairs to detect whether there is a set of feature pairs where the sensor accuracy is lower than the threshold or the structural crack length exceeds the preset value. If the above anomalies are detected, generate a feature pair showing that "the response time of sensor B at 10:30 exceeds the standard by 20%, and a 3-mm bend appears at the corresponding mechanical connecting rod C", triggering a warning. When the number of the above anomalies triggers 50 warnings within 1 hour, only retain the feature pairs with a confidence level greater than or equal to 90% to exclude accidental anomalies. Finally, based on the current sensor drift rate and mechanical deformation speed, predict that "sensor 3 will break due to excessive deformation within 24 hours and immediate maintenance is required".

[0113] In this embodiment, data analysis is performed on the operation status data based on the first detection model to obtain the first abnormal features and sensor performance indicators; object recognition is performed on the surrounding image data based on the second detection model to obtain the second abnormal features and mechanical structure features; cross-verification is performed on the first abnormal features and the second abnormal features to generate a preliminary abnormal diagnosis result; correlation analysis is performed on the sensor performance indicators and the mechanical structure features to obtain a potential fault prediction result. The object detection model is subdivided into a first detection model and a second detection model, and the operation status data and the surrounding image data are respectively processed in a targeted manner to obtain feature information in different dimensions, making the analysis of different types of data more professional and in-depth, and further improving the accuracy of abnormal detection.

[0114] Referring to Figure 6 , Figure 6 FIG.

[0115] In the third embodiment, step S30 includes:

[0116] Step S301, determining the abnormal data set of the weighbridge according to the preliminary abnormal diagnosis result and the potential fault prediction result.

[0117] Specifically, when determining the abnormal data set of the weighbridge, first, the currently identified abnormal information is extracted through the preliminary abnormal diagnosis result, including but not limited to the abnormal type, abnormal location, abnormal features, and time. Then, the fault information that may occur in the future is extracted through the potential fault prediction result, such as the predicted fault type and the possible time range. Then, the information of the preliminary diagnosis and prediction results is integrated into a unified data table according to the above fields.

[0118] Step S302, matching the abnormal data set with the first historical fault database and calculating the similarity between the abnormal faults in the abnormal data set and the historical faults.

[0119] It should be noted that the first historical fault database is a database that stores the faults that have occurred in the weighbridge system in history and their related features.

[0120] It should be understood that an appropriate algorithm needs to be selected according to the data type to calculate the similarity between the abnormal faults and the historical faults, such as using the Euclidean distance or cosine similarity, etc.

[0121] It is understandable that by quantifying the feature differences, a similarity score can be generated. The higher the score, the more similar the two faults are. According to the similarity, the type of the current anomaly can be determined. The matching strategy between the anomaly and the historical faults can be one-to-one matching. For each record in the anomaly dataset, traverse the historical fault database, calculate its similarity with all historical faults, and retain the top N records with the highest similarity; it can also be rule-based filtering and optimization. First, filter the subset of historical faults through key features (such as fault type, location), and then calculate the similarity within the subset to reduce the calculation amount; it can also be time window matching, matching the records in the historical faults whose timestamps are within a certain range before and after the anomaly time to determine the anomaly type.

[0122] In one example, for each record in the anomaly dataset, key feature vectors are extracted, such as anomaly type, location, time window, etc. The improved cosine similarity formula is used to calculate the matching degree with the records in the first historical fault database:

[0123]

[0124] Among them, S represents the similarity, and w i is the feature weight of the sensor anomaly and the mechanical structure anomaly, which can be set manually according to the operating environment during the installation of the weighbridge, or can be obtained through continuous optimization in subsequent use. x i and y i are the feature values of the anomaly fault and the historical fault respectively. After the calculation, the historical faults with a similarity greater than 80% are retained as references. If the anomaly fault matches multiple historical faults, the solution with the highest success rate of the repair plan is preferentially selected.

[0125] Some historical faults in the first historical fault database are as follows:

[0126] 1. Sensor-related

[0127] Shell-breaking cheating: It is detected that there is a shell-breaking fault;

[0128] Communication fault: Unable to communicate normally;

[0129] Data tampering: The data between the sensor and the instrument, and between the instrument and the sensor cannot be normally decrypted according to the encryption method;

[0130] Reading fault: There is a fault in the sensor data;

[0131] Acquisition interruption: The data suddenly interrupts;

[0132] Shell-breaking fault: There is shell-breaking;

[0133] Voltage change: The sensor voltage suddenly changes;

[0134] Attitude anomaly: The tilt angle of the sensor is abnormal;

[0135] Temperature anomaly: The internal temperature of the sensor is abnormal.

[0136] 2. Related to the instrument

[0137] Zero point anomaly: The data at the zero point is abnormal;

[0138] Parameter backup: Comparing with historical data, there is illegal scale calibration;

[0139] Instrument offline: It goes offline during use;

[0140] Instrument case opened: The instrument case has been opened, and cheating may occur after opening.

[0141] 3. Scanning faults

[0142] Health anomaly: Under the empty scale state, by analyzing the real-time acquisition of the coaxial sensor AD, there is a scale body deviation;

[0143] Non-empty scale on the scale: The scale body is not zeroed and there is a situation of getting on the scale;

[0144] Offload detection: Under the non-empty scale state, there is an offload in the force of the coaxial sensor.

[0145] Through multi-modal data fusion (sensor and image), fine-grained model design (LSTM and residual attention network), and dynamic weight allocation strategy, this solution reduces the false alarm rate to 2.3%, which is 5 times higher than the traditional method. At the same time, it supports early warning of 83% of potential faults 24 hours in advance.

[0146] Step S303: Determine the fault analysis report and repair plan of the abnormal fault according to the similarity.

[0147] It can be understood that after obtaining the corresponding abnormality, the preliminary repair strategy can be formulated according to the solution of the historical fault with the highest similarity. At the same time, analyze the differences between the current abnormality and historical faults, and optimize the repair steps.

[0148] Step S304: Generate the target abnormal detection report of the weighbridge in the current detection process based on the abnormal data set, the fault analysis report, and the repair plan.

[0149] Furthermore, in order to facilitate users to obtain accurate fault solutions through an intelligent user interaction method, improve the problem-solving efficiency, and enhance the user experience. After the step S304, it further includes:

[0150] When receiving the target problem input by the target user, parse the fault description in the target problem through the natural language processing module to obtain the key fault feature vector; perform semantic matching between the key fault feature vector and the target anomaly detection report to determine the fault root cause node; generate a voice interaction instruction based on the fault root cause node, and return the fault response corresponding to the target problem to the target user through the voice interaction instruction.

[0151] It should be noted that the natural language processing module is responsible for parsing the target problem input by the user, extracting the semantic information therein (such as fault type, location, phenomenon, etc.), and converting it into structured data that can be processed by a computer. The format of the target problem input by the user can be text, image, voice, etc., and this embodiment does not limit this.

[0152] It can be understood that the key fault feature vector is the core fault information extracted from the target problem by the natural language processing module. Semantic matching is to compare the similarity of two texts (such as the user problem and the content of the target anomaly detection report) at the semantic level to determine whether they describe the same fault or related faults. The fault root cause node is the fault description or the root cause node with the highest semantic matching degree with the user problem in the target anomaly detection report. For example, if the user problem points to "abnormal sensor voltage", the fault root cause node can be the abnormal record of "the sensor casing is broken resulting in low voltage" in the report.

[0153] It should be understood that after determining the fault root cause node through semantic matching, an audio reply instruction can be generated according to the fault root cause node, and the text information can be converted into voice through text-to-speech technology and returned to the target user to achieve human-machine voice interaction.

[0154] In this embodiment, determine the abnormal data set of the weighbridge according to the preliminary abnormal diagnosis result and the potential fault prediction result; match the abnormal data set with the first historical fault database, and calculate the similarity between the abnormal faults in the abnormal data set and the historical faults; determine the fault analysis report and repair plan of the abnormal faults according to the similarity; generate the target anomaly detection report of the weighbridge in the current detection process based on the abnormal data set, the fault analysis report and the repair plan. When an anomaly is detected, execute the solution generation process driven by machine learning. The maintenance personnel can quickly perform maintenance according to the detailed fault analysis and repair plan in the report, improve the maintenance efficiency, shorten the equipment downtime, and reduce the enterprise operation cost.

[0155] It should be noted that the above examples are only for understanding this application and do not constitute a limitation to the weighbridge anomaly detection method of this application. Based on this technical concept, more forms of simple transformations are within the protection scope of this application.

[0156] This application also provides a vehicle scale anomaly detection device. Please refer to Figure 7 , the vehicle scale anomaly detection device includes:

[0157] A data acquisition module 10, configured to acquire the operating status data and surrounding image data of the vehicle scale when a vehicle to be weighed drives onto the vehicle scale;

[0158] A data processing module 20, configured to process the operating status data and the surrounding image data by using a preset target detection model, and perform cross-verification and correlation analysis on the processing results to obtain a preliminary anomaly diagnosis result and a potential fault prediction result;

[0159] A report generation module 30, configured to generate a target anomaly detection report according to the preliminary anomaly diagnosis result and the potential fault prediction result.

[0160] The vehicle scale anomaly detection device provided by this application adopts the vehicle scale anomaly detection method in the above embodiment, and can solve the technical problem that most of the existing methods for obtaining vehicle scale anomaly data only focus on obtaining the weighing data of the vehicle scale, and the understanding of the overall operating condition of the vehicle scale is not comprehensive enough. Compared with the prior art, the beneficial effects of the vehicle scale anomaly detection device provided by this application are the same as those of the vehicle scale anomaly detection method provided by the above embodiment, and the other technical features in the vehicle scale anomaly detection device are the same as those disclosed in the method of the above embodiment, and will not be elaborated here.

[0161] This application provides a vehicle scale anomaly detection device. The vehicle scale anomaly detection device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the vehicle scale anomaly detection method in the first embodiment above.

[0162] Next, refer to Figure 8 , which shows a schematic structural diagram of a vehicle scale anomaly detection device suitable for implementing the embodiments of this application. The vehicle scale anomaly detection device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 8The illustrated truck scale anomaly detection device is merely an example and should not impose any limitations on the functions and usage scope of the embodiments of the present application.

[0163] As Figure 8 shown, the truck scale anomaly detection device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory 1002 or the program loaded from the storage device 1003 into the random access memory 1004. In the random access memory 1004, various programs and data required for the operation of the truck scale anomaly detection device are also stored. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. The input / output interface 1006 is also connected to the bus. Generally, the following systems may be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the truck scale anomaly detection device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a truck scale anomaly detection device with various systems, it should be understood that it is not required to implement or have all the shown systems. More or fewer systems may be implemented or had alternatively.

[0164] Specifically, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device, or installed from the storage device 1003, or installed from the read-only memory 1002. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0165] The vehicle scale anomaly detection device provided by this application adopts the vehicle scale anomaly detection method in the above-mentioned embodiment, which can solve the technical problem that most of the existing methods for obtaining vehicle scale anomaly data only focus on obtaining the weighing data of the vehicle scale and lack a comprehensive understanding of the overall operating condition of the vehicle scale. Compared with the prior art, the beneficial effects of the vehicle scale anomaly detection device provided by this application are the same as those of the vehicle scale anomaly detection method provided by the above-mentioned embodiment, and other technical features in this vehicle scale anomaly detection device are the same as those disclosed in the method of the previous embodiment, which will not be elaborated here.

[0166] It should be understood that each part disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0167] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0168] This application provides a computer-readable storage medium with computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the vehicle scale anomaly detection method in the above-mentioned embodiment.

[0169] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. The program code contained on the computer-readable storage medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0170] The above computer-readable storage medium can be included in the vehicle scale anomaly detection device; or it can exist independently without being assembled into the vehicle scale anomaly detection device.

[0171] The above computer-readable storage medium carries one or more programs. When the above one or more programs are executed by the vehicle scale anomaly detection device, the vehicle scale anomaly detection device is caused to execute the vehicle scale anomaly detection method described above.

[0172] Computer program code for performing the operations of this application can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages - such as Java, Smalltalk, C++; and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network - including a local area network (LAN) or a wide area network (WAN) - or can be connected to an external computer (e.g., by using an Internet service provider to connect through the Internet).

[0173] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0174] The modules described in the embodiments of the present application can be implemented in software or in hardware. Among them, the name of the module does not constitute a limitation on the unit itself in some cases.

[0175] The readable storage medium provided by the present application is a computer-readable storage medium. The computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned vehicle scale anomaly detection method, which can solve the technical problem that most of the existing methods for obtaining vehicle scale anomaly data only focus on obtaining the weighing data of the vehicle scale and do not comprehensively understand the overall operating condition of the vehicle scale. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the vehicle scale anomaly detection method provided by the above embodiments, and will not be elaborated here.

[0176] The above are only some embodiments of the present application, and do not limit the patent scope of the present application. All equivalent structural transformations made under the technical concept of the present application by using the content of the specification and drawings of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for detecting abnormalities in a weighbridge, characterized in that, The abnormal detection method for the weighbridge includes: When a vehicle to be weighed drives onto the weighbridge, obtaining the operation status data and surrounding image data of the weighbridge; Using a preset target detection model to process the operation status data and the surrounding image data, and performing cross-verification and correlation analysis on the processing results to obtain a preliminary abnormal diagnosis result and a potential fault prediction result; Generating a target abnormal detection report according to the preliminary abnormal diagnosis result and the potential fault prediction result.

2. The vehicle scale anomaly detection method according to claim 1, wherein The target detection model includes a first detection model and a second detection model; The step of using a preset target detection model to process the operation status data and the surrounding image data, and performing cross-verification and correlation analysis on the processing results to obtain a preliminary abnormal diagnosis result and a potential fault prediction result includes: Performing data analysis on the operation status data based on the first detection model to obtain a first abnormal feature and a sensor performance index; Performing target recognition on the surrounding image data based on the second detection model to obtain a second abnormal feature and a mechanical structure feature; Performing cross-verification on the first abnormal feature and the second abnormal feature to generate a preliminary abnormal diagnosis result; Performing correlation analysis on the sensor performance index and the mechanical structure feature to obtain a potential fault prediction result.

3. The vehicle scale anomaly detection method according to claim 2, characterized in that, The step of performing cross-verification on the first abnormal feature and the second abnormal feature to generate a preliminary abnormal diagnosis result includes: Performing feature matching on the first abnormal feature and the second abnormal feature to obtain a matching result; When the matching result is consistent, determining the first abnormal feature and the second abnormal feature as definite abnormalities and generating a first-level abnormal alarm; When the matching result is inconsistent, performing weight assignment on the first abnormal feature and the second abnormal feature based on the fault priority rule to generate a second-level abnormal alarm; Determining a preliminary abnormal diagnosis result according to the first-level abnormal alarm and / or the second-level abnormal alarm.

4. The vehicle scale anomaly detection method according to claim 2, characterized in that, The step of performing correlation analysis on the sensor performance index and the mechanical structure feature to obtain a potential fault prediction result includes: Performing comparison processing on the time-series data of the sensor performance index and the spatial data of the mechanical structure feature to generate a set of feature pairs with spatio-temporal correlation; When a sensor abnormal point and / or a mechanical deformation area appear in the set of feature pairs, generating a third-level abnormal alarm; When the number of the third-level abnormal alarms is greater than a preset abnormal threshold, performing redundant data elimination on the coverage ranges of the sensor abnormal points and the mechanical deformation areas, and retaining the target feature pairs with a confidence level higher than a preset confidence threshold in the set of feature pairs; Outputting a potential fault prediction result based on the target feature pairs.

5. The abnormal detection method of the weighbridge according to claim 1, characterized in that, The step of generating a target abnormal detection report according to the preliminary abnormal diagnosis result and the potential fault prediction result includes: Determining an abnormal data set of the weighbridge according to the preliminary abnormal diagnosis result and the potential fault prediction result; Matching the abnormal data set with a first historical fault database and calculating the similarity between the abnormal faults in the abnormal data set and the historical faults; Determine the fault analysis report and repair plan of the abnormal fault according to the similarity; Generate a target abnormal detection report of the weighbridge during the current detection process based on the abnormal data set, the fault analysis report, and the repair plan.

6. The method for detecting the abnormality of the weighbridge according to claim 5, characterized in that, After the step of generating the target abnormal detection report of the weighbridge during the current detection process based on the abnormal data set, the fault analysis report, and the repair plan, it further includes: When receiving a target problem input by a target user, parse the fault description in the target problem through a natural language processing module to obtain a key fault feature vector; Semantically match the key fault feature vector with the target abnormal detection report to determine the fault root cause node; Generate a voice interaction instruction based on the fault root cause node, and return the fault reply corresponding to the target problem to the target user through the voice interaction instruction.

7. The method for detecting abnormalities of a weighbridge according to any one of claims 1 to 6, characterized in that, Before the step of obtaining the operation state data and surrounding image data of the weighbridge when a vehicle to be weighed drives onto the weighbridge, it further includes: When the weighing sensors and instruments in the weighbridge receive a start signal sent by the platform, intelligently detect the core components of the weighbridge through a machine learning model to obtain component state data; Match the component state data based on a second historical fault database, and classify the matching result through a deviation parameter within a preset range to obtain a classification result and a corresponding abnormal feature code; If the classification result is a first abnormal result, match the target calibration instruction corresponding to the first abnormal result according to a preset abnormal type mapping table, and after the calibration instruction is executed, send a ready signal to the alarm module of the monitoring system through a message bus; If the classification result is a second abnormal result, persistently store the abnormal feature code based on the self-check processing module of the weighbridge, send a time-limited maintenance request signal with a time limit identifier to the work order module of the monitoring system according to a preset priority rule, and update the abnormal feature code after the maintenance request signal disappears.

8. An abnormal detection device for a weighbridge, characterized in that, The device includes: A data acquisition module, configured to obtain the operation state data and surrounding image data of the weighbridge when a vehicle to be weighed drives onto the weighbridge; A data processing module, configured to process the operation state data and the surrounding image data by using a preset target detection model, and perform cross-validation and correlation analysis on the processing results to obtain a preliminary abnormal diagnosis result and a potential fault prediction result; A report generation module, configured to generate a target abnormal detection report according to the preliminary abnormal diagnosis result and the potential fault prediction result.

9. An abnormal detection device for a weighbridge, characterized in that, The device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the computer program is configured to implement the steps of the weighbridge abnormal detection method according to any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium, and when the computer program is executed by a processor, it implements the steps of the weighbridge abnormal detection method according to any one of claims 1 to 7.

Images