Overhead crane fault prediction method, device, storage medium and electronic device

Abstract

The present application provides a method, device, storage medium and electronic device for predicting overhead crane faults, which belongs to the field of transportation control technology. The method includes: obtaining a first driving feature set of the overhead crane to be tested; calling a preset fault prediction model to predict the fault of the overhead crane to be tested, and outputting a fault prediction score; when the fault prediction score is between a first score threshold and a second score threshold, locating the target overhead rail section where the overhead crane to be tested generates a first abnormal data subset during driving; obtaining a second driving feature set, the second driving feature set including the first abnormal data subset and the second driving feature subset generated by other overhead cranes during driving on the target overhead rail section; performing a cross-device correlation analysis on the first abnormal data subset and the second driving feature subset, and calculating the abnormal score of the first abnormal data subset on the target overhead rail section; and determining whether the overhead crane to be tested has a fault risk based on the abnormal score. The present application can improve the accuracy of overhead crane fault prediction.

Description

Technical Field

[0001] The present application relates to the field of transportation control technology, and in particular to a method, device, storage medium and electronic device for predicting overhead crane faults. Background Art

[0002] In the semiconductor manufacturing industry, an overhead hoist transport (OHT) is an automated transport device operating on elevated tracks, primarily used for the precise movement of wafers, raw materials, and finished products within cleanrooms. As a core component of the automated material handling system (AMHS), the OHT works in conjunction with the track network, control system, and scheduling software to enable efficient and continuous material movement throughout the semiconductor production process. By integrating automated transport equipment with intelligent management technologies, the AMHS aims to optimize production line logistics efficiency, reduce manual intervention, and ensure process stability, making it an essential infrastructure for modern semiconductor factories.

[0003] The stability of overhead crane operation is directly related to production efficiency and product quality. Currently, the industry generally adopts regular inspection and maintenance strategies to ensure the reliability of overhead cranes, such as checking key components such as track systems, drive motors, and sensors at fixed intervals. However, this maintenance model has certain limitations:

[0004] First, scheduled maintenance relies on preset time intervals or historical experience, making it difficult to identify unexpected equipment anomalies that may occur between maintenance checks. For example, during operation, overhead cranes may experience performance degradation or localized failures due to mechanical wear, load fluctuations, or environmental factors (such as temperature and humidity changes). These dynamic issues may not be detected in a timely manner through regular inspections, ultimately causing unplanned downtime and affecting production continuity.

[0005] Secondly, existing fault detection methods are often based on single-type data analysis (such as vibration signals or current monitoring), lacking adaptability to the complex operating scenarios of overhead cranes. Overhead crane operating parameters vary significantly under different operating conditions (such as fully loaded straight-line travel on overhead rails and unloaded turning). Single-dimensional monitoring fails to fully capture the diverse state characteristics of the equipment, potentially leading to missed detection or misdiagnosis of potential faults.

[0006] Furthermore, as an integral part of a factory's logistics system, the operating status of overhead cranes is closely linked to factors such as the track environment and the coordination of multiple devices. However, existing analysis of overhead crane operating data is often limited to a single device or a local area, failing to fully integrate the dynamic relationship between the device and the environment, limiting the comprehensiveness and accuracy of fault prediction.

[0007] Overall, existing technologies still face problems such as insufficient timeliness in overhead crane fault prediction and limited multi-dimensional data integration capabilities. More efficient monitoring and analysis methods are urgently needed to improve the accuracy of equipment maintenance and ensure the efficient operation of semiconductor production. Summary of the Invention

[0008] The purpose of this application is to provide a method, device, storage medium and electronic device for predicting overhead crane faults to solve at least one of the above technical problems.

[0009] In a first aspect of the present application, a method for predicting overhead crane failures is provided, the method comprising:

[0010] Obtaining a first driving feature set of the overhead crane to be tested;

[0011] Based on the first driving feature set, calling a preset fault prediction model to perform fault prediction on the overhead crane to be tested, and outputting a fault prediction score;

[0012] When the fault prediction score is between a first score threshold and a second score threshold, locating a target overhead rail section of the overhead crane to be tested that generates a first abnormal data subset during driving based on the first driving feature set;

[0013] Acquire a second driving feature set, where the second driving feature set includes the first abnormal data subset and a second driving feature subset generated by other overhead traveling vehicles during their driving on the target overhead traveling section;

[0014] performing a cross-device correlation analysis on the first abnormal data subset and the second driving feature subset, and calculating an abnormality score of the first abnormal data subset on the target skytrain segment;

[0015] It is determined whether the overhead travelling crane to be tested has a failure risk based on the abnormality score.

[0016] Optionally, based on the first driving feature set, calling a preset fault prediction model to perform fault prediction on the overhead crane to be tested and outputting a fault prediction score includes:

[0017] Dividing the first driving feature set into a plurality of first driving feature subsets of driving types according to the driving condition of the overhead crane and / or the distribution condition of the overhead rail segments;

[0018] Calling the fault prediction model to analyze the first driving feature subset of each driving type to obtain a type score corresponding to each first driving feature subset;

[0019] The fault prediction score is calculated based on the type score.

[0020] Optionally, the second driving feature subset and the first abnormal data subset belong to the same driving type;

[0021] The overhead crane driving condition includes one or more of a loaded driving condition and an unloaded driving condition, and the overhead rail segment distribution condition includes one or more of a straight overhead rail segment, a turning overhead rail segment, an ascending overhead rail segment, and a descending overhead rail segment.

[0022] Optionally, the fault prediction model is a multimodal analysis model based on a Transformer architecture, comprising a multimodal input embedding unit, an encoding processing unit, a feature fusion unit, and a decoding unit;

[0023] The method of calling a preset fault prediction model to perform fault prediction on the overhead travelling vehicle to be tested based on the first driving feature set and outputting a fault prediction score includes:

[0024] calling the multimodal input embedding unit to convert the operating parameters of different dimensions in the first driving feature set into high-dimensional feature vectors;

[0025] Calling the encoding processing unit to perform encoding processing and feature analysis on the high-dimensional feature vector to obtain an analysis result;

[0026] Calling the feature fusion unit to fuse the analysis results to form fusion information;

[0027] The decoding unit is called to calculate a fault prediction score according to the fusion information.

[0028] Optionally, the calling of the encoding processing unit to perform encoding processing and feature analysis on the high-dimensional feature vector to obtain an analysis result includes:

[0029] Calling the spatiotemporal coding unit in the coding processing unit to convert the high-dimensional feature vector into a spatiotemporal joint coding vector;

[0030] The self-attention encoding unit in the encoding processing unit is called to extract global features and local features from the spatiotemporal joint encoding vector, and the global features and local features are analyzed to obtain analysis results.

[0031] Optionally, performing cross-device correlation analysis on the first abnormal data subset and the second driving feature subset to calculate an abnormality score of the first abnormal data subset on the target ATL segment includes:

[0032] Calculating the correlation value between the first abnormal data subset and each second driving feature subset having the same driving type according to a preset correlation calculation model;

[0033] Performing a weighted sum on each calculated correlation value, and using the obtained weighted sum value as the anomaly score;

[0034] The determining whether the overhead travelling crane to be tested has a failure risk based on the abnormality score includes: determining that the overhead travelling crane to be tested has a failure risk when the abnormality score exceeds a third score threshold.

[0035] Optionally, the method further includes: when the fault prediction score is less than a first score threshold, determining that the overhead travelling crane to be tested does not have a fault risk;

[0036] When the fault prediction score is greater than a second score threshold, it is determined that the overhead travelling crane to be tested has a fault risk.

[0037] In a second aspect of the present application, a device for predicting a failure of an overhead crane is provided, the device comprising:

[0038] A first feature acquisition module is used to acquire a first travel feature set of the overhead crane to be tested;

[0039] a first scoring module, configured to call a preset fault prediction model to perform fault prediction on the overhead travelling crane to be tested based on the first driving feature set, and output a fault prediction score;

[0040] a target overhead rail segment locating module, configured to locate, based on the first driving feature set, a target overhead rail segment of the overhead crane to be tested that generates a first abnormal data subset during driving when the fault prediction score is between a first scoring threshold and a second scoring threshold;

[0041] a second feature acquisition module, configured to acquire a second driving feature set, wherein the second driving feature set includes the first abnormal data subset and a second driving feature subset generated by the other overhead traveling vehicle during its driving on the target overhead traveling section;

[0042] a second scoring module, configured to perform a cross-device correlation analysis on the first abnormal data subset and the second driving feature subset, and calculate an abnormality score of the first abnormal data subset on the target ATL segment;

[0043] A fault prediction module is used to determine whether the overhead travelling crane to be tested has a fault risk based on the abnormality score.

[0044] In a third aspect of the present application, a computer-readable storage medium is provided, on which executable instructions are stored. When the executable instructions are executed by a processor, the processor executes the method described in any embodiment of the present application.

[0045] In a fourth aspect of the present application, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method as described in any one of the embodiments of the present application.

[0046] The overhead crane fault prediction method, device, storage medium, and electronic device in the present application predict faults based on a first driving feature set of the overhead crane to be tested. When the fault prediction score based on the first driving feature set is between a first score threshold and a second score threshold, a first abnormal data subset that can reflect the abnormal condition of the overhead crane to be tested is further determined from the first driving feature set, and the target overhead rail section corresponding to the first abnormal data subset is located. The first abnormal data subset and the second driving feature subset of other overhead cranes when traveling on the target overhead rail section are then combined to perform a correlation analysis on the two to obtain an abnormal score; and thereby ultimately determine whether the overhead crane to be tested has a fault risk. The present application locates the target overhead rail section where the overhead crane to be tested is operating (the overhead rail section where the overhead crane travels that can reflect the abnormal condition of the overhead crane to be tested), and incorporates factors such as the coordinated operation of multiple devices (reflected by the driving data of other overhead cranes on the overhead rail section) into the analysis scope, thereby achieving a more comprehensive and accurate fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope of the present application.

[0048] Figure 1 A schematic diagram of a scenario of a method for predicting overhead crane failures in one embodiment;

[0049] Figure 2 A schematic flow chart of a method for predicting overhead crane failures in one embodiment;

[0050] Figure 3 A schematic diagram of a process for calling a preset fault prediction model to perform fault prediction on an overhead crane to be tested based on a first driving feature set and outputting a fault prediction score in one embodiment;

[0051] Figure 4 A schematic diagram of a process for calling a preset fault prediction model to perform fault prediction on an overhead crane to be tested based on a first driving feature set and outputting a fault prediction score in another embodiment;

[0052] Figure 5 2 is a schematic structural diagram of an overhead crane fault prediction device in one embodiment;

[0053] Figure 6 FIG. 1 is a schematic structural diagram of an electronic device in an embodiment. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0055] All terms (including technical and scientific terms) used in this application have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0056] For example, the terms "first," "second," etc. used in this application may be used herein to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish a first element from another element.

[0057] For example, the terms "include", "comprising", etc. used in this application indicate the existence of features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0058] The overhead crane fault prediction method in this application can be applied to Figure 1 In the scenario shown. Combined Figure 1 As shown, in the AMHS, numerous overhead cranes 120 travel on the overhead rails to deliver items (such as wafer boxes) to the corresponding destinations. The overhead rails can be divided into straight overhead rail sections 110-A and curved overhead rail sections 110-B according to their distribution, and can be further divided into ascending overhead rail sections and descending overhead rail sections. The overhead crane 120 can travel both unloaded and loaded on the overhead rails. Corresponding sensing devices are distributed on the overhead rails and / or the overhead cranes, which can locate the overhead crane's position on the overhead rails in real time and measure relevant driving data such as the overhead crane's speed, power consumption, and vibration at various locations on the overhead rails.

[0059] In one embodiment, a method for predicting a crane failure is provided, which can be applied to Figure 1 In the scenario shown. Figure 2 As shown, the method includes:

[0060] Step 210: Obtain a first driving feature set of the overhead crane to be tested.

[0061] In this embodiment, the first driving feature set is a set of all or part of the driving features generated by the overhead crane under test during its operation within a preset time period. The preset time period can be from the last fault detection to the present, or any suitable time period such as the last day, two days, or a week.

[0062] These driving characteristics include one or more information such as the driving speed, power consumption, vibration, friction coefficient, load status, etc. when the overhead crane is driving at various positions on the overhead rail. The power consumption can be one or more information such as the power and current of the overhead crane.

[0063] The overhead crane can be integrated with one or more detection devices such as vibration sensors, speed sensors, acceleration sensors, temperature sensors, current sensors, optical encoders, and laser rangefinders. Each detection device can collect the aforementioned vibration data, driving speed, power consumption, friction coefficient, and other driving data at a corresponding preset collection frequency, and these driving data can be transmitted to the electronic equipment in the AMHS. The electronic equipment can specifically be a relevant server. For the relevant driving data received within a preset time period, it can be segmented according to a preset sliding time window to form initial driving features, and these initial driving features are pre-processed such as normalization. The pre-processed initial driving features are then aggregated and summarized to form a first driving feature set.

[0064] The sliding time window can be fixed or variable, for example, with default settings of 3 seconds, 5 seconds, or other default lengths. The length of each sliding time window can also be adjusted in real time based on one or more of the following factors: the overhead crane's delivery status, travel speed, and the distribution of overhead rails. For example, if the base window length is set to 5 seconds, when the overhead crane's travel speed is greater than 2 m / s, the window is shortened to 3 seconds to capture rapid changes; when the speed is less than 0.5 m / s (such as during start-stop phases), the window is extended to 10 seconds to improve data stability. By adjusting the length of each sliding time window in real time, the amount of data in each set of driving data remains consistent.

[0065] This initial driving data can be synchronized and aligned through methods such as timestamp calibration and spatial coordinate matching to ensure that each data point corresponds to an accurate spatial location (such as track segment number and 3D coordinates). Furthermore, through noise filtering and signal enhancement, normalization, missing data processing, and outlier removal, the pre-processed initial driving characteristics are obtained.

[0066] Step 220 : Based on the first driving feature set, a preset fault prediction model is called to perform fault prediction on the overhead crane to be tested, and a fault prediction score is output.

[0067] In this embodiment, the fault prediction model can be a pre-trained model for overhead crane fault prediction. Specifically, the model can be a combination of one or more models such as support vector machines, random forests, long short-term memory networks, and transformer architectures. The fault prediction model analyzes and calculates the input first driving feature set and outputs a fault prediction score based on its internal algorithm and trained parameters.

[0068] The Fault Prediction Score is a numerical indicator used to quantify the likelihood of a potential overhead crane failure. A higher score indicates a higher probability of failure. This score is based on various overhead crane driving characteristic data, such as speed, acceleration, vibration, temperature, power consumption, and load, combined with the data patterns and regularities learned by the fault prediction model under normal operating and fault conditions, to assess how close the overhead crane's current operating state is to a faulty state.

[0069] The scoring range can be any appropriate numerical value, such as 0-100, with higher scores indicating a greater likelihood of an overhead crane failure. This scoring allows staff to quickly understand the potential risk of an overhead crane failure and take appropriate measures, such as further inspection, maintenance, or pre-prepared spare parts, to ensure the safe operation of the overhead crane and reduce production interruptions and safety incidents caused by failures.

[0070] Specifically, the electronic device can iteratively train the model using a pre-collected set of normal overhead crane driving characteristics (negative samples) and a set of driving characteristics of faulty overhead cranes (positive samples), ultimately forming a trained fault prediction model. This fault prediction model can output a fault prediction score reflecting the likelihood of an overhead crane failure, as well as one or more driving characteristics from the first driving characteristics set that best indicate an abnormality in the overhead crane (i.e., abnormal data).

[0071] Taking a fault prediction model as a multimodal analysis model based on a Transformer architecture as an example, the model may include a multimodal input embedding unit, an encoding processing unit, and a decoding unit.

[0072] The multimodal input embedding layer is used to convert each piece of driving feature data into a feature vector through one-dimensional convolution and LSTM; the encoding processing unit can perform time encoding and spatial encoding on the formed feature vector, and perform feature analysis to obtain the analysis results; the decoding unit can encode the analysis results and ultimately obtain the corresponding fault prediction score.

[0073] Step 230 : When the fault prediction score is between the first score threshold and the second score threshold, locate the target overhead rail section where the overhead crane to be tested generates the first abnormal data subset during its driving process based on the first driving feature set.

[0074] In this embodiment, the electronic device pre-sets a first scoring threshold and a second scoring threshold. The first scoring threshold and the second scoring threshold can be custom values ​​set based on historical fault data or manually set based on operating experience. If the first scoring threshold is less than the second scoring threshold, and the fault prediction score of the overhead crane under test falls within the interval formed by the first scoring threshold and the second scoring threshold, it indicates that the overhead crane under test may have a potential fault risk and requires further analysis.

[0075] Specifically, the first and second scoring thresholds define the range within which the overhead crane's failure prediction score falls, thereby determining the degree of potential failure risk. The first scoring threshold is a relatively low score limit, such as 40 points. When the failure prediction score rises above this threshold, it indicates that the overhead crane's operating status has begun to deviate from the normal range and enter an area where potential failure risk may exist. This serves as an early warning signal, alerting personnel that the overhead crane has exhibited certain signs of concern and requires further evaluation and analysis, but it does not necessarily indicate a serious failure.

[0076] The second scoring threshold is a relatively high score limit, such as 70 points. When the fault prediction score is lower than this threshold, it indicates that while the overhead crane has potential risks, it has not yet reached a high-risk state with a high probability of failure. If the score is between the first and second scoring thresholds, such as a fault prediction score of 60 points, the overhead crane is in a fuzzy zone in the middle. Subsequent steps, such as locating the target overhead rail segment and obtaining and analyzing the second driving feature set, are needed to further accurately determine its failure risk and determine whether maintenance measures are necessary.

[0077] When in this interval, the fault prediction model further outputs one or more driving features in the first driving feature set that best reflect the suspected fault of the overhead crane to be tested. The set consisting of the output one or more driving features (i.e., abnormal data) is the first abnormal data subset described below.

[0078] For each piece of abnormal data output, the corresponding overhead rail segment (i.e., the target overhead rail segment) can be determined by combining the corresponding timestamp and / or location data. The first driving feature set is divided by the overhead crane's travel trajectory and time. Using data analysis methods such as outlier detection, the corresponding abnormal data is located. The overhead rail segment associated with this abnormal data becomes the target overhead rail segment. For example, if a piece of abnormal data is located as being generated by the overhead crane during travel on overhead rail segment A-5, then A-5 is designated as the target overhead rail segment. After determining the target overhead rail segment, further analysis of the abnormal data on this segment can be performed, combined with data from other overhead cranes traveling in this segment, to provide a basis for accurately determining the risk of overhead crane failure.

[0079] In one embodiment, when the fault prediction score is less than a first score threshold, it is determined that the overhead crane to be tested has no fault risk; when the fault prediction score is greater than a second score threshold, it is determined that the overhead crane to be tested has a fault risk.

[0080] Specifically, if the output fault prediction score is less than the first scoring threshold, it means that the risk probability of the overhead crane under test being at fault is low, and it can be considered that there is no fault risk. If it is higher than the second scoring threshold, it means that the risk probability of the overhead crane under test being at fault is very high, and it can be directly determined that there is a fault risk.

[0081] Step 240: Acquire a second driving feature set.

[0082] In this embodiment, the second driving feature set includes the first abnormal data subset and other second driving feature subsets generated during the driving process of the overhead crane in the target overhead rail section.

[0083] For the identified target overhead rail segment, the electronic device can obtain driving characteristic data generated by other overhead cranes in the AHSM while traveling in the target overhead rail segment, aggregate and summarize the data to form a second driving characteristic subset. The second driving characteristic subset and the first abnormal data subset together constitute the second driving characteristic set. Optionally, the data in the second driving characteristic subset and the data in the first driving characteristic set were both generated when the overhead crane traveled in the target overhead rail segment, and both were generated within the same time period.

[0084] After locating the target overhead rail segment, the corresponding second time period is further determined based on the time corresponding to the driving characteristic data in the first abnormal data subset, and the driving characteristic data of other overhead cranes driving under the target overhead rail segment during the second time period are obtained and aggregated into a second driving characteristic subset.

[0085] Specifically, for each piece of abnormal data, the corresponding target overhead rail segment and the corresponding timestamp are determined. Based on the timestamp, the corresponding second time period is determined. Driving characteristic data of other overhead cranes (excluding the overhead crane under test) traveling under the target overhead rail segment during the second time period is obtained. After obtaining the driving characteristic data of other overhead cranes corresponding to each piece of abnormal data, the data is aggregated to form a second driving characteristic subset.

[0086] Furthermore, the load conditions of other overhead cranes can be considered to ensure that their load conditions are consistent with the load condition of the overhead crane under test. For example, if one piece of abnormal data of the overhead crane under test is generated during a certain time period while traveling unloaded on overhead rail segment A-1, then driving characteristic data generated by other overhead cranes traveling unloaded on overhead rail segment A-1 during a second time period can be obtained and used as the second driving characteristic data corresponding to the abnormal data.

[0087] Step 250 : Perform a cross-device correlation analysis on the first abnormal data subset and the second driving feature subset, and calculate an abnormality score of the first abnormal data subset on the target skytrain segment.

[0088] In this embodiment, the correlation between the driving feature data in the first abnormal data subset and the driving feature data in the second driving feature subset can be analyzed through a relevant correlation calculation model or a cluster analysis model. The correlation is used to reflect the degree of difference between the two. The smaller the correlation, the greater the difference between the two.

[0089] When in the same time period and the same overhead rail section, the greater the difference between the first abnormal data subset and the second driving feature subset, the greater the difference in driving data between the overhead crane to be tested and other overhead cranes, and the greater the possibility that the overhead crane to be tested has an abnormality.

[0090] Optionally, a correlation analysis can be performed on the driving characteristic data of other overhead cranes corresponding to each abnormal data, and a corresponding abnormality score can be calculated based on each correlation analysis. Optionally, a corresponding correlation value can be calculated for each correlation analysis, and a corresponding abnormality score can be finally calculated based on each correlation value.

[0091] Step 260: Determine whether the overhead travelling crane to be tested has a failure risk based on the abnormality score.

[0092] In this embodiment, after obtaining the anomaly score and the fault prediction score, the two can be combined to comprehensively determine whether the overhead crane under test has a fault risk. For example, when the fault prediction score is within the above range, if the anomaly score is higher than a certain value, it can be determined that the overhead crane under test has a fault risk. If the anomaly score is lower than a certain value, it can be determined that the overhead crane under test does not have a fault risk.

[0093] The overhead crane fault prediction method in this application can continuously monitor the operating status of the overhead crane by acquiring the first driving feature set of the overhead crane to be tested in real time, covering multi-dimensional real-time data such as speed, acceleration, vibration, and temperature. Based on these real-time data, the preset fault prediction model is called to predict the fault and output the fault prediction score, which can quickly determine whether the overhead crane currently has a potential fault risk. Compared with traditional regular maintenance, this method can detect anomalies in the early stages before the fault occurs, greatly improving the timeliness of fault prediction, effectively avoiding unplanned downtime caused by failure to detect potential fault hazards in time, and effectively ensuring the continuity of production.

[0094] Secondly, when obtaining the first driving feature set, the present application integrates measurement data of multiple dimensions to fully reflect the operating status of the overhead crane under different working conditions. For example, when the overhead crane is running in a straight line with full load or turning without load, parameters such as speed and vibration will change significantly, and these multi-dimensional data are all included in the first driving feature set. When the fault prediction score is in a specific interval, the second driving feature set is further obtained, including the first abnormal data subset of the overhead crane to be tested and the second driving feature subset generated when other overhead cranes are running in the target overhead rail section, and a cross-device correlation analysis is performed on them. This in-depth integration and analysis of multi-dimensional data can comprehensively cover the diverse status characteristics of the overhead crane, greatly reduce the missed detection or misjudgment of potential faults, and significantly improve the accuracy of fault prediction.

[0095] In addition, when determining the target overhead rail section, by comprehensively considering the abnormal data generated by the overhead crane to be tested when it is traveling on the overhead rail section and the driving data of other overhead cranes on the same target overhead rail section, factors such as the track environment in which the overhead crane to be tested is operating (such as the characteristics of different overhead rail sections) and the coordinated operation of multiple devices (reflected by the driving data of other overhead cranes on the overhead rail section) are included in the analysis scope, thereby achieving more comprehensive and accurate fault prediction, effectively overcoming the shortcomings of existing technologies in the dynamic correlation analysis between equipment and environment.

[0096] In one embodiment, Figure 3 As shown, step 220 includes:

[0097] Step 310 : Divide the first driving feature set into a plurality of first driving feature subsets of driving types according to the overhead travelling crane driving condition and / or the overhead rail segment distribution condition.

[0098] In this embodiment, during operation, the overhead crane can experience different driving states, including fully loaded, unloaded, accelerated, decelerated, high-speed, and low-speed. Based on their shape and function, overhead rail segments can be divided into straight, curved, ascending, and descending types. The type of overhead rail segment the overhead crane is in is determined using the overhead crane's position sensor and map information from the overhead rail system. The type of driving state the overhead crane is in can also be determined using the overhead crane's speed and / or weight sensors.

[0099] For example, when the speed is between 0 and 1 m / s and the weight sensor indicates a load of 80% or more of the overhead crane's rated capacity, it is considered fully loaded and traveling at low speed. When the speed is between 1 and 3 m / s and the load is less than 20% of the rated capacity, it is considered unloaded and traveling at medium speed. When the position sensor data combined with map information indicates that the overhead crane is traveling on a section of track with no curvature change, it is considered to be in a straight track section. When continuous angular changes in the direction of travel of the overhead crane are detected, it is considered to be in a curved track section. Based on these conditions, various driving types can be classified, each corresponding to a specific condition (one driving type corresponds to the same overhead crane driving condition and track section distribution).

[0100] In one embodiment, the overhead crane driving condition includes one or more of a loaded driving condition and an unloaded driving condition, and the rail segment distribution condition includes one or more of a straight rail segment, a curved rail segment, an ascending rail segment, and a descending rail segment. For example, according to the overhead crane driving condition, it can be divided into two conditions, namely, a loaded driving condition and an unloaded driving condition. According to the rail segment distribution condition, it can be divided into four conditions, namely, a straight rail segment, a curved rail segment, an ascending rail segment, and a descending rail segment. Combining the division conditions based on the above two dimensions, 8 driving types can be obtained in the end. For example, some of the driving types are a loaded driving type under a straight rail segment, an unloaded driving type under a straight rail segment, a loaded driving type under a curved rail segment, an unloaded driving type under a curved rail segment, a loaded driving type under an ascending rail segment, an unloaded driving type under an ascending rail segment, a loaded driving type under a descending rail segment, and an unloaded driving type under a descending rail segment.

[0101] Driving data generated during a complete delivery process can be used to categorize travel status. A complete delivery process can be defined as the process from the crane picking up a package to its destination and then traveling empty from the destination to the next pickup point. Based on the real-time driving data, the data generated during a complete delivery process can be classified into one or more of the eight aforementioned driving types, depending on the route and load. For each driving type, the data is segmented according to the corresponding sliding time window to generate multiple sets of initial driving data.

[0102] Correspondingly, the second driving feature subset and the first abnormal data subset belong to the same driving type.

[0103] For example, when the driving type of a certain abnormal data in the first abnormal data subset is a load-driven driving type on a straight overhead rail section, the driving characteristic data of other overhead cranes in a loaded state on the target overhead rail section (the target overhead rail section belongs to a straight overhead rail section) in the corresponding time period are found, and these data are used as the first abnormal data subset corresponding to the abnormal data.

[0104] By combining the overhead crane's driving conditions and / or the distribution of overhead rail sections, the data in the first driving feature set corresponding to different overhead rail section types and / or different driving conditions are segmented to form first driving feature subsets based on the overhead rail section distribution. Each first driving feature subset corresponds to a driving type. The driving feature data in each first driving feature subset can also be segmented according to the aforementioned sliding time window. Each segmented driving data set can also include the aforementioned relevant data such as driving speed, power consumption, and vibration.

[0105] Step 320 : Calling the fault prediction model to analyze the first driving feature subset of each driving type to obtain a type score corresponding to each first driving feature subset.

[0106] In this embodiment, the fault prediction model may independently analyze the first driving feature subset of each driving type to obtain a type score for the corresponding driving type.

[0107] Step 330 : Calculate a fault prediction score based on the type score.

[0108] The type score for each driving type can be weighted and summed according to the preset type score weights to ultimately calculate the corresponding fault prediction score. The weights for each driving type can be the same or different, and the electronic device can set the weights based on actual conditions.

[0109] In one embodiment, the fault prediction model is a multimodal analysis model based on the Transformer architecture, including a multimodal input embedding unit, an encoding processing unit, a feature fusion unit, and a decoding unit. Figure 4 As shown, step 220 includes:

[0110] Step 410 : Calling a multimodal input embedding unit to convert the operating parameters of different dimensions in the first driving feature set into a high-dimensional feature vector.

[0111] In this embodiment, the first driving feature set includes one or more types of operating parameters such as vibration data, driving speed, power consumption, friction, load, temperature, etc. For these multiple types of operating parameters, the multimodal input embedding unit uses a specific embedding function to map each type of operating parameter to a vector space of corresponding dimensions. For example, for speed data, each speed value is mapped to a 100-dimensional vector space so that it carries more semantic information. Assume that the speed value 1.2 is converted by the embedding function to obtain the vector [0.23, 0.12, -0.05,…, 0.08] (100-dimensional vector example). Similarly, similar embedding operations are performed on acceleration, vibration, temperature and other data to convert operating parameters of different dimensions into high-dimensional feature vectors for subsequent processing.

[0112] Step 420: Call the encoding processing unit to perform encoding processing and feature analysis on the high-dimensional feature vector to obtain an analysis result.

[0113] The encoding processing unit is based on the encoder module in the Transformer architecture. It consists of multiple multi-head attention mechanisms and feedforward neural network layers. Taking the high-dimensional feature vector sequence of velocity data as an example, the multi-head attention mechanism can simultaneously focus on vector information at different positions, capturing both global and local characteristics of velocity changes. Assuming the velocity vector sequence is [V1, V2, …, V10] (each Vi is a 100-dimensional vector), the multi-head attention mechanism allows different heads to focus on the relationship between vectors at different positions. For example, head 1 focuses on the changing trends of adjacent vectors, while head 2 focuses on the differences between two vectors.

[0114] The data processed by the multi-head attention mechanism is then passed through a feedforward neural network layer for nonlinear transformations to further extract features. For example, the feedforward neural network layer performs a series of linear transformations and activation function operations (such as the ReLU function) on the velocity vector processed by the attention mechanism to obtain a new feature representation. Similar encoding and feature analysis are performed on the high-dimensional feature vectors of other dimensional data, such as vibration, power consumption, friction, load, and temperature, ultimately yielding analysis results for each dimensional data.

[0115] Specifically, the spatiotemporal coding unit in the coding processing unit is called to convert the high-dimensional feature vector into a spatiotemporal joint coding vector; the self-attention coding unit in the coding processing unit is called to extract global features and local features from the spatiotemporal joint coding vector, and the global features and local features are analyzed to obtain analysis results.

[0116] The spatiotemporal coding unit uses a pre-defined encoding algorithm, such as one based on convolutional neural networks, to construct a spatiotemporal coding structure. For a velocity vector sequence, the vector at each time step (for example, velocity data collected once per second corresponds to one time step) is considered a "position" in space. A two-dimensional convolution kernel is designed, whose width corresponds to the dimensions of the high-dimensional feature vector (100 dimensions) and whose length corresponds to the number of time steps (here, 10 time steps). The convolution kernel slides over the velocity vector sequence, performing a convolution operation on the vector at each time step. Assuming the convolution kernel weight is K and the bias is b, for the vector Vi at the i-th time step, the convolution operation yields a new feature vector Si. For example, if the ReLU function is used as the activation function, Si = ReLU (K × Vi + b). This new feature vector Si not only incorporates the characteristics of the velocity data at that time step but also includes correlation information from the velocity data at adjacent time steps, achieving information integration across the temporal dimension.

[0117] Furthermore, considering the crane's position on the track (assuming the track is divided into 10 distinct zones, with the crane's position detected in real time by position sensors), this position information can also be encoded into the feature vector. For example, the zone number of the track where the crane is currently located is converted into a 10-dimensional one-hot encoding vector P, which is then concatenated with Si to obtain the final spatiotemporal joint encoding vector Ti = [Si, P]. Similar methods are used to perform spatiotemporal encoding on high-dimensional feature vectors of other dimensional data, yielding their corresponding spatiotemporal joint encoding vectors.

[0118] The self-attention encoding unit is based on the self-attention mechanism in the Transformer architecture. The self-attention mechanism calculates attention scores between vectors at different positions to determine the degree to which each vector pays attention to other vectors. Again, using the spatiotemporal joint encoding vector sequence [T1, T2, …, T10] of velocity data as an example, the specific calculation process is as follows:

[0119] First, each spatiotemporal joint coding vector Ti is passed through three linear transformation matrices Wq, Wk, and Wv respectively to obtain the query vector Qi = Ti × Wq, the key vector Ki = Ti × Wk, and the value vector Vi = Ti × Wv.

[0120] Next, we calculate the attention score. For any two vectors Ti and Tj, the attention score eij = (Qi × Kj^T) / sqrt (d), where d is the dimension of the query and key vectors (here, we assume 50 dimensions), and Kj^T represents the transpose of Kj. The attention score eij reflects the degree of attention that vector Ti pays to vector Tj.

[0121] Next, the attention score is normalized by Softmax to obtain the normalized attention score hij=Softmax(eij).

[0122] Finally, the output vector is calculated based on the normalized attention score. For the i-th vector Ti, its output vector O i = Σ(h ij ×Vj), j = 1, 2, ... 10. In this way, the self-attention mechanism can capture the global characteristics of the velocity data at different time steps and orbit positions.

[0123] To extract local features, we can limit the scope of attention based on the self-attention mechanism. For example, we can only calculate the attention score between two adjacent time step vectors. That is, for the i-th vector Ti, we only calculate eij (j = i-1, i, i + 1), and then follow the above steps to calculate the local feature vector Li.

[0124] After obtaining the global feature vector Oi and the local feature vector Li, they are further analyzed. For example, a fully connected neural network layer can be used to concatenate the global feature vector Oi and the local feature vector Li as input. After a series of linear transformations and activation function operations (such as the ReLU function), the analysis results for the speed data are obtained. The same self-attention encoding unit operation is also used for the spatiotemporal joint encoding vectors of other dimensional data to extract global and local features and analyze them. Ultimately, the analysis results for each dimensional data are obtained. These analysis results will serve as input to the subsequent feature fusion unit for further calculation of the fault prediction score.

[0125] Step 430: Call the feature fusion unit to fuse the analysis results to form fusion information.

[0126] The feature fusion unit fuses the analysis results of the different dimensional data obtained by the encoding processing unit. Optionally, the fusion is performed in a splicing manner, where the feature vectors are sequentially spliced ​​into a new vector F, whose dimension can be the sum of the dimensions of the data of each dimension.

[0127] By performing feature fusion, the overhead crane operation status information contained in data of different dimensions can be integrated to form more comprehensive fusion information, providing a basis for the subsequent accurate calculation of fault prediction scores.

[0128] Step 440: Call the decoding unit to calculate the fault prediction score based on the fusion information.

[0129] The decoding unit is also based on the decoder module in the Transformer architecture. It receives the fused information vector F output by the feature fusion unit. The decoding unit contains a fully connected neural network layer. The fused information vector F is input into the fully connected neural network layer, which performs linear transformations and activation function operations (such as the sigmoid function) through a series of weight matrices and bias terms.

[0130] Assume that the weight matrix W and bias term b obtained after training the fully connected neural network layer are calculated (for example, score = Sigmoid (W × F + b)) to ultimately output a fault prediction score. For example, the calculated fault prediction score is 60 (assuming a full score of 100 indicates the highest probability of a fault and 0 indicates no fault). This score reflects the probability of a fault in the current operating state of the overhead crane and is used to determine whether the overhead crane requires further inspection and maintenance.

[0131] In one embodiment, when the first driving feature set is divided into multiple first driving feature subsets, each first driving feature subset can be used as input to the multimodal analysis model based on the Transformer architecture. Each first driving feature subset is analyzed according to the process from steps 410 to 440 described above. In step 440, a type score corresponding to each first driving feature subset is output. Each type score is then combined to ultimately calculate a fault prediction score. This will not be further described.

[0132] In one embodiment, step 250 includes: calculating the correlation value between the first abnormal data subset and each second driving feature subset having the same driving type according to a preset correlation calculation model; performing a weighted summation on each calculated correlation value, and using the obtained weighted summation value as the abnormality score.

[0133] In this embodiment, for each piece of abnormal data, its corresponding second driving feature subset is determined, and the correlation value between it and the corresponding second driving feature subset is calculated. After calculating the correlation value corresponding to each piece of abnormal data, a weighted sum is performed according to a preset weight, and the summed value is used as the corresponding abnormality score.

[0134] Specifically, the correlation value can be calculated using the Pearson correlation coefficient calculation model. Taking the data of load driving type under a certain straight line overhead rail section as an example, assuming that a piece of abnormal data in the first abnormal data subset is a sequence A=[a1,a2,……,an], and its corresponding second driving feature subset is a sequence B=[b1,b2,……,bn]. The calculation formula of the Pearson correlation coefficient is

[0135] in, is the mean value of sequence A, is the average value of sequence B. This formula is used to calculate the correlation value r between the abnormal data and the corresponding second driving feature subset. The final abnormality score can be obtained by taking a weighted sum of the correlation values ​​r corresponding to each piece of abnormal data.

[0136] Based on the importance of different travel types in the operation of the overhead crane and analysis of historical fault data, weights can be assigned to the relevance values ​​of each travel type. For example, when the overhead crane is fully loaded, the load is greater and the risk of failure is relatively high. Therefore, the weights for loaded and curved travel are set higher, while the weights for unloaded and straight travel are set relatively lower.

[0137] In one embodiment, when sequence A and sequence B have different lengths, a dynamic time warping (DTW) algorithm can be used to calculate the correlation value. The DTW algorithm elastically aligns the two sequences along the time axis to find the optimal alignment path between the two sequences, minimizing the sum of the distances between the two sequences along this path. This allows measuring the similarity or correlation between the two sequences without requiring the sequences to be of equal length.

[0138] Specifically, we first create a two-dimensional matrix D of size m×n, where m is the length of sequence A and n is the length of sequence B. Each element D(i,j) of the matrix represents the distance between the i-th element of sequence A and the j-th element of sequence B. Euclidean distance or Manhattan distance is usually used to calculate the distance between two elements.

[0139] Dynamic programming calculates the shortest path: Starting from the upper left corner of the matrix D(1,1), the shortest path to the lower right corner D(m,n) is calculated using dynamic programming. For each position D(i,j), its value is calculated using the following formula:

[0140] Where d(i,j) is the distance between the i-th element of sequence A and the j-th element of sequence B, In this way, by continuously comparing and selecting the minimum distance, we can find the shortest path from D(1,1) to D(m,n).

[0141] The sum of the distances on the shortest path is the DTW distance between sequence A and sequence B. In order to obtain the correlation value, the DTW distance can be normalized, for example, using the formula The correlation value obtained in this way ranges from 0 to 1, and the larger the value, the more similar or correlated the two sequences are.

[0142] In one embodiment, step 260 includes: when the abnormality score exceeds a third score threshold, determining that the overhead travelling crane to be tested has a failure risk.

[0143] The third scoring threshold can be an appropriate value set based on relevant experience. For example, the third scoring threshold can be set to 0.5. When the calculated anomaly score S = 0.39 is less than the third scoring threshold of 0.5, it is determined that the overhead crane under test currently has no failure risk. If the anomaly score exceeds the third scoring threshold, it is determined that the overhead crane under test has a failure risk. In this case, maintenance personnel should be arranged to inspect and maintain the overhead crane in a timely manner to prevent failure.

[0144] In one embodiment, Figure 5 As shown, a device for predicting a fault of an overhead crane is provided, the device comprising:

[0145] The first feature acquisition module 510 is configured to acquire a first travel feature set of the overhead crane to be tested.

[0146] The first scoring module 520 is configured to call a preset fault prediction model to perform fault prediction on the overhead crane to be tested based on the first driving feature set, and output a fault prediction score.

[0147] The target overhead rail segment positioning module 530 is configured to locate the target overhead rail segment where the overhead crane to be tested generates the first abnormal data subset during driving based on the first driving feature set when the fault prediction score is between the first scoring threshold and the second scoring threshold.

[0148] The second feature acquisition module 540 is configured to acquire a second driving feature set, where the second driving feature set includes the first abnormal data subset and other second driving feature subsets generated during the overhead travelling vehicle's driving on the target overhead travelling section.

[0149] The second scoring module 550 is configured to perform cross-device correlation analysis on the first abnormal data subset and the second driving feature subset, and calculate an abnormality score of the first abnormal data subset on the target skytrain segment.

[0150] The fault prediction module 560 is used to determine whether the overhead travelling crane to be tested has a fault risk based on the abnormality score.

[0151] In one embodiment, the first scoring module 520 is further used to divide the first driving feature set into first driving feature subsets of multiple driving types according to the overhead crane driving condition and / or the overhead rail segment distribution condition; call the fault prediction model to analyze the first driving feature subset of each driving type to obtain a type score corresponding to each type of first driving feature subset; and calculate the fault prediction score based on the type score.

[0152] In one embodiment, the second driving feature subset and the first abnormal data subset belong to the same driving type.

[0153] In one embodiment, the overhead crane driving condition includes one or more of a loaded driving condition and an unloaded driving condition, and the overhead rail segment distribution condition includes one or more of a straight overhead rail segment, a turning overhead rail segment, an ascending overhead rail segment, and a descending overhead rail segment.

[0154] In one embodiment, the fault prediction model is a multimodal analysis model based on the Transformer architecture, including a multimodal input embedding unit, an encoding processing unit, a feature fusion unit, and a decoding unit.

[0155] The first scoring module 520 is also used to call the multimodal input embedding unit to convert the operating parameters of different dimensions in the first driving feature set into a high-dimensional feature vector; call the encoding processing unit to perform encoding processing and feature analysis on the high-dimensional feature vector to obtain an analysis result; call the feature fusion unit to fuse the analysis results to form fusion information; and call the decoding unit to calculate the fault prediction score based on the fusion information.

[0156] In one embodiment, the first scoring module 520 is also used to call the spatiotemporal coding unit in the coding processing unit to convert the high-dimensional feature vector into a spatiotemporal joint coding vector; call the self-attention coding unit in the coding processing unit to extract global features and local features from the spatiotemporal joint coding vector, analyze the global features and local features, and obtain analysis results.

[0157] In one embodiment, the second scoring module 550 is further used to calculate the correlation value between the first abnormal data subset and each second driving feature subset having the same driving type according to a preset correlation calculation model; perform weighted summation on each calculated correlation value, and use the obtained weighted summed value as the abnormality score; and determine whether the overhead crane to be tested has a failure risk based on the abnormality score, including: when the abnormality score exceeds a third scoring threshold, determining that the overhead crane to be tested has a failure risk.

[0158] In one embodiment, the fault prediction module 560 is further configured to determine that the overhead travelling crane under test has no fault risk when the fault prediction score is less than a first score threshold, and to determine that the overhead travelling crane under test has a fault risk when the fault prediction score is greater than a second score threshold.

[0159] In one embodiment, a computer-readable storage medium is provided, on which executable instructions are stored. When the instructions are executed by a processor, the processor executes the steps in the above-mentioned method embodiments.

[0160] In one embodiment, an electronic device is provided, comprising one or more processors and a memory storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are caused to perform the steps of the above-described method embodiments. The electronic device may be a device equipped with the above-described distribution control system, such as a backend server that communicates with the overhead crane, controls the operation of the overhead crane, transmits a travel route to the overhead crane, and so on.

[0161] In one embodiment, Figure 6 , which shows a schematic diagram of the structure of an electronic device for implementing an embodiment of the present application. Electronic device 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage unit 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of electronic device 600. CPU 601, ROM 602, and RAM 603 are connected to each other via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0162] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, mouse, and the like; an output section 607 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 608 including devices such as a hard disk; and a communication section 609 including a network interface card such as a LAN card or a modem. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. Removable media 611, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 610 as needed, so that computer programs read from the removable media can be installed in the storage section 608 as needed.

[0163] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present application include a computer program product comprising a computer-readable medium carrying instructions. In such embodiments, the instructions can be downloaded and installed from a network via communication portion 609 and / or installed from removable media 611. When the instructions are executed by central processing unit (CPU) 601, the various method steps described in this application are performed.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

[0165] Furthermore, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of this application and to form different embodiments. For example, all of the above embodiments may be used in any combination. The information disclosed in this background section is intended solely to enhance understanding of the overall background of this application and should not be construed as an admission or any form of implication that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for predicting overhead crane failures, characterized in that: The method comprises: Obtaining a first driving characteristic set of the overhead crane to be tested, where the first driving characteristic set is a collection of all or part of the driving characteristics generated during the driving process of the overhead crane to be tested within a preset time period, where the driving characteristics include one or more of driving speed, power consumption, vibration, friction coefficient, and load state; Based on the first driving feature set, calling a preset fault prediction model to perform fault prediction on the overhead crane to be tested, and outputting a fault prediction score and a first abnormal data subset, where the first abnormal data subset is a collection of one or more abnormal data in the first driving feature set that best reflects a suspected fault of the overhead crane to be tested; When the fault prediction score is between the first scoring threshold and the second scoring threshold, for each piece of output abnormal data, a corresponding target overhead rail segment and a corresponding timestamp are determined, and a corresponding second time period is determined based on the timestamp, and a second driving feature subset is obtained, where the second driving feature subset is a collection of driving feature data of other overhead cranes, excluding the overhead crane to be tested, that travel under the corresponding target overhead rail segment during the second time period; Acquire a second driving feature set, where the second driving feature set includes the first abnormal data subset and the second driving feature subset; performing a cross-device correlation analysis on the first abnormal data subset and the second driving characteristic subset to calculate an anomaly score for the first abnormal data subset on the target ATL segment, including: calculating a correlation value between the first abnormal data subset and each second driving characteristic subset having the same driving type according to a preset correlation calculation model, performing a weighted summation on each calculated correlation value, and using the obtained weighted summation as the anomaly score; Determining whether the overhead crane to be tested has a failure risk based on the abnormality score includes: determining that the overhead crane to be tested has a failure risk when the abnormality score exceeds a third score threshold.

2. The method for predicting overhead crane failure according to claim 1, characterized in that: The method of calling a preset fault prediction model to perform fault prediction on the overhead travelling vehicle to be tested based on the first driving feature set and outputting a fault prediction score includes: Dividing the first driving feature set into a plurality of first driving feature subsets of driving types according to the driving condition of the overhead crane and / or the distribution condition of the overhead rail segments; Calling the fault prediction model to analyze the first driving feature subset of each driving type to obtain a type score corresponding to each first driving feature subset; The fault prediction score is calculated based on the type score.

3. The method for predicting overhead crane failure according to claim 2, characterized in that: The second driving feature subset and the first abnormal data subset belong to the same driving type; The overhead crane driving condition includes one or more of a loaded driving condition and an unloaded driving condition, and the overhead rail segment distribution condition includes one or more of a straight overhead rail segment, a turning overhead rail segment, an ascending overhead rail segment, and a descending overhead rail segment.

4. The method for predicting overhead crane failure according to claim 1, characterized in that: The fault prediction model is a multimodal analysis model based on the Transformer architecture, which includes a multimodal input embedding unit, an encoding processing unit, a feature fusion unit, and a decoding unit; The method of calling a preset fault prediction model to perform fault prediction on the overhead travelling vehicle to be tested based on the first driving feature set and outputting a fault prediction score includes: calling the multimodal input embedding unit to convert the operating parameters of different dimensions in the first driving feature set into high-dimensional feature vectors; Calling the encoding processing unit to perform encoding processing and feature analysis on the high-dimensional feature vector to obtain an analysis result; Calling the feature fusion unit to fuse the analysis results to form fusion information; The decoding unit is called to calculate a fault prediction score according to the fusion information.

5. The method for predicting overhead crane failure according to claim 4, characterized in that: The calling of the encoding processing unit to perform encoding processing and feature analysis on the high-dimensional feature vector to obtain an analysis result includes: Calling the spatiotemporal coding unit in the coding processing unit to convert the high-dimensional feature vector into a spatiotemporal joint coding vector; The self-attention encoding unit in the encoding processing unit is called to extract global features and local features from the spatiotemporal joint encoding vector, and the global features and local features are analyzed to obtain analysis results.

6. The method for predicting overhead crane failure according to any one of claims 1 to 5, characterized in that: The method further comprises: When the fault prediction score is less than a first score threshold, it is determined that the overhead travelling crane to be tested does not have a fault risk; When the fault prediction score is greater than a second score threshold, it is determined that the overhead travelling crane to be tested has a fault risk.

7. A crane fault prediction device, characterized in that: The device comprises: a first feature acquisition module, configured to acquire a first driving feature set of the overhead crane to be tested, wherein the first driving feature set is a collection of all or part of the driving features generated during the driving process of the overhead crane to be tested within a preset time period, wherein the driving features include one or more of driving speed, power consumption, vibration, friction coefficient, and load state; a first scoring module configured to perform fault prediction on the overhead crane under test by calling a preset fault prediction model based on the first driving feature set, and output a fault prediction score and a first abnormal data subset, wherein the first abnormal data subset is a set of one or more driving features in the first driving feature set that best reflect a suspected fault of the overhead crane under test; a target overhead rail segment positioning module, configured to, when the fault prediction score is between a first scoring threshold and a second scoring threshold, determine, for each output driving feature, a corresponding target overhead rail segment and a corresponding timestamp, determine a corresponding second time period based on the timestamp, and obtain a second driving feature subset, the second driving feature subset being a collection of driving feature data of overhead cranes other than the overhead crane to be tested that traveled under the corresponding target overhead rail segment during the second time period; a second feature acquisition module, configured to acquire a second driving feature set, wherein the second driving feature set includes the first abnormal data subset and the second driving feature subset; a second scoring module, configured to perform a cross-device correlation analysis on the first abnormal data subset and the second driving characteristic subset, and calculate an anomaly score for the first abnormal data subset on the target ATL segment, including: calculating a correlation value between the first abnormal data subset and each second driving characteristic subset having the same driving type according to a preset correlation calculation model, performing a weighted summation on each calculated correlation value, and using the obtained weighted summation as the anomaly score; The fault prediction module is used to determine whether the overhead travelling vehicle to be tested has a fault risk based on the abnormality score, including: when the abnormality score exceeds a third score threshold, determining that the overhead travelling vehicle to be tested has a fault risk.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores executable instructions, which, when executed by a processor, enable the processor to perform the method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs, which, when executed by the one or more processors, causes the one or more processors to perform the method according to any one of claims 1 to 6.

Images