An intelligent fault diagnosis method and system for gantry crane

Abstract

The present application provides a method and system for intelligent fault diagnosis of gantry cranes, which can obtain the real-time working condition data and strain monitoring data of gantry cranes; pre-process the data; use a comparative analysis algorithm based on actual strain monitoring data and predicted strain data to perform trend analysis; use a comparative analysis algorithm based on actual stress and theoretical stress to perform trend analysis; establish a fault decision tree based on the above trend analysis results, artificial fault judgment and fault feature data of the historical fault library; build a fault diagnosis algorithm based on the strain signal trend analysis results, combined with the fault decision tree and the fault knowledge graph to perform main beam fault diagnosis; update the fault decision tree based on the diagnosis result data and the warning result data. The present application can monitor the equipment status of the gantry crane in real time, perform fault diagnosis and intelligent warning on the equipment according to the collected data, and repair the fault in time, thereby improving the safety of the gantry crane during production, avoiding the occurrence of gantry crane operation accidents, and improving the working efficiency of the gantry crane.

Description

Technical Field

[0001] The present invention relates to the field of special equipment operation and maintenance, and in particular to an intelligent fault diagnosis method and system for a gantry crane. Background Art

[0002] With the advancement of my country's infrastructure construction and the overall development of the heavy industry of the national economy, gantry cranes have become important equipment for the rapid development of the national economy and have been widely used in many national economic sectors. As gantry cranes develop towards large-scale, high-speed and automated directions, the safety of gantry cranes has received more and more attention.

[0003] The gantry cranes in the prior art lack effective and timely safety monitoring means, making it difficult to detect and diagnose the faults of the gantry cranes in a timely manner. Therefore, the gantry cranes cannot be repaired in a timely manner according to the faults, and the occurrence of safety accidents cannot be avoided in a timely manner.

[0004] Therefore, a solution is needed that can perform real-time fault diagnosis on gantry cranes. Summary of the invention

[0005] One purpose of the present application is to provide an intelligent fault diagnosis method and system for a gantry crane, so as to solve the problem that it is difficult to diagnose faults in the real-time working process of a gantry crane under the existing technology.

[0006] To achieve the above objectives, some embodiments of the present application provide a gantry crane intelligent fault diagnosis method, the method comprising the following steps:

[0007] Step 1. Collecting real-time operating data and strain monitoring data of the gantry crane;

[0008] Step 2. Preprocessing the collected working condition data and strain monitoring data;

[0009] Step 3. Perform trend analysis based on the strain monitoring data and the predicted strain data using a comparative analysis algorithm;

[0010] Step 4. Calculate the actual stress based on the strain monitoring data, and use a comparative analysis algorithm to compare the actual stress with the theoretical stress to perform trend analysis;

[0011] Step 5. Establish a fault decision tree based on the trend analysis results of steps 3 and 4, manual fault judgment and fault feature data of the historical fault library;

[0012] Step 6. Based on the trend analysis results and in combination with the fault decision tree and strain signal fault knowledge graph, a fault diagnosis algorithm is constructed to perform fault diagnosis;

[0013] Step 7: Update the fault decision tree based on the diagnosis result data and the warning result data.

[0014] Furthermore, collecting the strain monitoring data in step 1 further includes: distributing the measuring points of the strain sensors on the two main beams of the gantry crane, arranging 3 measuring points on each of the main beams, and arranging a total of 6 strain sensors for monitoring the stress of the main beams, and the 6 strain sensors are installed on the lower panel inside the main beam box; and, arranging 2 strain sensors for temperature compensation, and the strain sensors for temperature compensation are installed on a movable mounting block near the strain sensor at the middle position of the main beam.

[0015] Furthermore, the preprocessing in step 2 includes cleaning, correcting or filtering the data.

[0016] Furthermore, step 3 further includes: training a prediction model through the XGBoost algorithm based on pre-processed real-time operating data and strain monitoring data; processing historical operating data for strain sensor signals and using them for model training; the trained model uses data collected by a programmable logic controller (PLC) as input, finds the relationship between the PLC data and the strain sensor signal value based on the historical operating data, predicts the signal data of a specified sensor, and then calculates the predicted value with the strain monitoring data collected by the strain sensor in actual work, uses time series residuals as a judgment criterion, and issues an alarm if abnormalities continue to occur within a certain time series.

[0017] Furthermore, step 4 further includes: comparing the theoretical stress and actual stress of the gantry crane main beam to mutually verify the correctness of the numerical values; after mutual verification of rationality, the theoretical stress is used as a reference value of the actual stress to monitor and warn the main beam strain data.

[0018] Furthermore, constructing the fault decision tree in step 5 includes the following steps:

[0019] 1) Using the fault type in the historical fault data as a label and associating it with the corresponding strain monitoring data and sensor characteristic values, determining the fault type in the historical case and labeling it;

[0020] 2) Analyzing sensor data and historical fault records through the historical fault data, identifying common fault modes, determining the feature value range corresponding to the common fault modes, and summarizing these features into different fault types, and associating the extracted signal feature values ​​with the labels of the fault types;

[0021] 3) Build a fault decision tree

[0022] ① Design the hierarchical structure of the decision tree

[0023] First-layer node: Is there an abnormality?

[0024] By comparing the signal characteristic value with the standard value under normal operating conditions, it is determined whether there is an abnormality; if the amplitude or frequency of the vibration signal exceeds the normal range, it is determined as "abnormality exists", otherwise it is "normal";

[0025] Second-layer nodes: preliminary classification of fault types

[0026] Preliminarily classifying abnormal situations according to different characteristics of the sensor data;

[0027] The third layer node: refined classification

[0028] By further analyzing the specific characteristic patterns, the fault types can be subdivided;

[0029] ②Set decision conditions and thresholds

[0030] Setting feature thresholds: determining a decision criterion for each node based on the historical fault data;

[0031] Decision-making based on multi-parameter correlation: For faults with multi-parameter correlation, decision-making conditions are established by combining the changes in multiple signals;

[0032] ③Build relationships between nodes

[0033] According to the above conditions, the branch structure of the fault decision tree is gradually constructed; each node makes a judgment according to the specific characteristic value of the sensor and enters the corresponding branch;

[0034] 4) Verify and optimize the fault decision tree

[0035] ①Model verification

[0036] Input the collected historical fault data or new data into the fault decision tree model to verify the accuracy of its fault classification. If the classification accuracy is not high, it is necessary to adjust the threshold of the feature value or redesign the branch logic;

[0037] ②Optimize the fault decision tree

[0038] The fault decision tree is optimized in the following three ways: i) Dynamically adjust the threshold: According to more of the historical fault data, continuously adjust the various feature thresholds set in the fault decision tree; ii) Add new nodes: As the fault library expands, continuously add new fault types and new judgment nodes; iii) Remove redundant branches: If some branch judgments are no longer applicable or the fault mode changes, simplify or remove redundant branches.

[0039] Furthermore, the construction process of the fault knowledge graph in step 6 includes:

[0040] S1. Knowledge acquisition and ontology model construction:

[0041] ① Knowledge acquisition: Extract knowledge from various data sources related to crane fault diagnosis, including fault type, fault handling method, maintenance experience of professionals, and equipment technical manuals. The knowledge provides detailed background and handling measures for various crane faults;

[0042] ② Ontology model construction: Construct the gantry crane fault diagnosis ontology model and define the core concepts and relationships in the graph;

[0043] ③Graph data storage: convert the ontology model into structured graph data and store it in the database of the fault knowledge graph to support subsequent fault diagnosis reasoning and knowledge retrieval;

[0044] S2. Preliminary construction and visualization of the map:

[0045] ① Relationship entity mapping: From the collected knowledge and data, fault cases, diagnostic processes, and equipment components are mapped to nodes and relationships in the knowledge graph;

[0046] ② Hierarchical relationship construction: hierarchical relationships are established in the fault knowledge graph according to the fault type, so that the fault nodes at different levels reflect the correlation relationship from fault symptoms to root causes, from treatment methods to spare parts requirements;

[0047] ③ Visualization display: Generate a visualization view of the fault knowledge graph to clearly present the core nodes and important relationships;

[0048] S3. Fault knowledge graph update mechanism:

[0049] ① Trend change determination and knowledge graph update: Operation and maintenance personnel analyze the causes of fault trend changes based on the equipment's operating trends and diagnostic data, and add the diagnostic results to the knowledge graph; by recording the cause analysis and treatment methods, the knowledge graph's ability to identify and respond to trend faults is improved;

[0050] ② Manual judgment and trend update: In the early stage, operation and maintenance experts manually judge trend changes and update the nodes or relationships of the knowledge graph;

[0051] ③ Manual and algorithmic updates: In the later stage, expert judgment is combined with diagnostic algorithms for updates. When the algorithm identifies abnormal trends, the knowledge graph is automatically updated, and nodes and relationships are improved after manual confirmation, gradually reducing manual participation;

[0052] S4. Recording and processing of diagnostic algorithm output information:

[0053] ① Information flow reception and storage: The knowledge graph framework receives the key fault information flow output by the diagnosis algorithm, including detected anomalies, predicted fault types and possible causes;

[0054] ② Graph update of important information: Map the key information output by the algorithm to the graph nodes. If a new fault mode or new processing method is detected, add or modify nodes and relationships in the graph to improve the fault diagnosis content of the graph;

[0055] ③ Automatic update of anomalies: When the algorithm identifies a new abnormal pattern, the system will automatically generate a corresponding abnormal node in the graph and establish associations with related equipment and fault types;

[0056] S5. Knowledge graph training and self-learning:

[0057] ① Automatic graph training: The updated graph data is used to train and optimize the fault diagnosis model to form a closed loop; the diagnosis algorithm will improve the recognition rate of new fault types based on the structured data of the graph;

[0058] ② Self-learning mechanism: Through the fault handling cases and experience data continuously accumulated by the graph, the diagnostic algorithm gradually achieves self-learning, which can more quickly identify new faults and adjust maintenance strategies;

[0059] ③ Manual supervision and feedback: The self-learning process of the graph still requires manual supervision, and the automatically updated content of the graph must be checked regularly to ensure the diagnostic accuracy of the model;

[0060] S6. Continuous optimization and application of knowledge graph:

[0061] ① Fault prediction and reasoning: Through the retrieval and reasoning of knowledge graphs, the system can predict potential faults, assist operation and maintenance personnel in early troubleshooting, and automatically recommend treatment methods based on historical data;

[0062] ② Intelligent operation and maintenance decision support: The knowledge graph integrates the fault information and working condition data of all crane equipment, supports the operation and maintenance team to make quick decisions, and improves the efficiency of fault handling;

[0063] ③Model visualization optimization: Through the continuous optimization of the visualization map, complex relationships can be visualized, which will help further operation and maintenance personnel training and knowledge sharing.

[0064] Furthermore, the construction of the fault diagnosis algorithm in step 6 includes: combining a Bayesian classifier, a support vector machine and a fault decision tree for fault diagnosis, wherein the combination of the three provides a basis for node selection or probability inference in the fault decision tree;

[0065] The Bayesian classifier calculates the posterior probability of a node object using the Bayesian formula through the prior probability of the node object, and selects the class with the maximum posterior probability as the class to which the object belongs. The probabilistic reasoning of the Bayesian classifier can smooth the decision process and provide a probabilistic basis for fault classification, making the fault decision tree more robust.

[0066] The support vector machine uses a multi-classification expansion algorithm to perform the fault diagnosis on multiple faults, and improves the classification accuracy of the fault decision tree by processing high-dimensional data and nonlinear relationships; the support vector machine can also reduce the risk of overfitting through maximum interval optimization.

[0067] Furthermore, updating the fault decision tree in step 7 includes the following steps:

[0068] Ss1. Structural adjustment of fault decision tree

[0069] 1) Introducing new features: Introducing new feature nodes based on the diagnosis results and warning result data;

[0070] 2) Weight adjustment: For existing decision nodes, the diagnosis results and warning result data are used to update the weight or probability of the decision nodes; the failure probability of nodes in the decision tree is adjusted based on the existing data using statistical methods based on historical data;

[0071] 3) Pruning redundant nodes: If the diagnosis results and warning result data indicate that the probability of occurrence of certain faults is extremely low, or the related faults are no longer important in decision-making, these nodes are appropriately pruned in the fault decision tree to simplify the model;

[0072] SS2. Rule Update

[0073] 1) Rule improvement based on actual fault patterns: When diagnostic data reveals new fault patterns or differences from existing fault patterns, the rules in the fault decision tree are updated according to these patterns; wherein the rules refer to the judgment criteria or conditions defined at each node of the fault decision tree, which are used to determine the branching of the path and ultimately lead to specific fault classification or decision;

[0074] 2) Threshold update guided by early warning information: The early warning result data reveals early signs of faults and is used to update the thresholds of relevant nodes in the fault decision tree so that potential faults can be identified earlier;

[0075] Ss3. Model optimization and verification

[0076] 1) Training and testing: combining the updated fault decision tree with the new diagnosis results and warning results data, training the model and testing it, and using the cross-validation method to evaluate its accuracy and robustness;

[0077] 2) Real-time update: With the continuous input of new diagnostic results and warning result data, the structure and parameters of the fault decision tree are continuously updated.

[0078] On the other hand, an intelligent fault diagnosis system for a gantry crane is provided, the system comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the system is triggered to execute the aforementioned diagnosis method. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a position arrangement diagram of strain measuring points of the main beam of the gantry crane of the present invention;

[0080] Figure 2 A schematic diagram showing a comparison between a predicted value and an actual value of a gantry crane strain sensor signal according to the present invention;

[0081] Figure 3 It is a schematic diagram comparing the theoretical stress and actual stress of the main beam of the gantry crane of the present invention. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0083] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0084] Here, the intelligent fault diagnosis method and system for gantry cranes of the embodiments of the present application are suitable for scenarios where real-time fault monitoring and diagnosis are performed on the working process of the gantry crane.

[0085] In this scenario, the gantry crane may have real-time faults during operation. If the faults are not handled in time, they may cause gantry crane accidents, which will not only affect the normal operation of the gantry crane, but may also cause significant personal or property losses.

[0086] The gantry crane fault diagnosis method provided in the embodiment of the present application can collect the working status data of the gantry crane in real time, pre-process the collected real-time working data, perform trend analysis on the pre-processed data, analyze the cause of the trend change based on the trend analysis result, make fault judgment according to the cause analysis result and the historical fault library, establish a fault decision tree, and perform fault diagnosis and intelligent early warning based on the fault knowledge graph and signal trend changes of the fault decision tree, thereby improving the timeliness of fault monitoring and diagnosis during the operation of the gantry crane, timely discovering the relevant faults of the gantry crane, so that countermeasures can be taken in time to the fault and the occurrence of gantry crane accidents can be avoided.

[0087] Some embodiments of the present application provide a gantry crane intelligent fault diagnosis method, the method comprising the following steps:

[0088] Step 1. Collect the real-time operating data and strain monitoring data of the gantry crane.

[0089] For example, this embodiment installs strain sensors at key positions of the main beam structure of the gantry crane to monitor the stress change trend, thereby monitoring the operating status of the gantry crane. The real-time working condition data of the gantry crane refers to the information data related to the working condition of the crane; the strain monitoring data of the gantry crane refers to the main beam working condition strain monitoring data obtained by the sensors deployed on the main beam of the gantry crane.

[0090] It is understandable that there is a relationship between the deflection of the gantry crane main beam and the stress change of the main beam panel: the deflection of the gantry crane main beam is usually caused by long-term use or overloading, which will cause the main beam to deform. This deformation will cause the stress of the main beam structure to change, resulting in stress concentration. Stress concentration may cause damage to the main beam structure or even breakage. Therefore, it is necessary to monitor the stress change trend at the key position of the gantry crane main beam structure to ensure its safe operation.

[0091] When installing strain sensors, the layout of the sensors must fully consider the stress characteristics of the main beam, and they must be installed at key locations that are prone to flexural deformation. Generally speaking, the mid-span of the main beam and near the support points at both ends are key areas for deflection monitoring. The deflection at these locations is relatively large and has a significant impact on the overall structural stability. The main beam panel is the main component that bears the load of the gantry crane, and its stress distribution is complex. Therefore, stress sensors need to be installed at multiple key locations to fully monitor its stress state. Common monitoring locations include the connection between the panel and the upper and lower flange plates of the main beam, the middle of the panel, and possible stress concentration areas.

[0092] When arranging sensors, the following points should also be noted: First, the installation location of the sensor should be convenient for data collection and monitoring, and avoid interference from other equipment or components; second, the sensor should have good durability and stability, and be able to operate stably for a long time in harsh working environments; finally, the layout of the sensor should follow certain specifications and standards to ensure the accuracy and reliability of the monitoring results.

[0093] Taking the above factors into consideration, in this embodiment, the measuring points of the strain sensors are distributed on the two main beams, such as Figure 1 As shown in the figure, 3 measuring points are arranged on each main beam, with a total of 6 strain sensors; in addition, 2 strain sensors for temperature compensation are required. The 6 strain sensors for monitoring the stress of the main beam are installed on the lower panel inside the main beam box, and the 2 strain sensors for temperature compensation are installed near the strain sensor at the middle position S / 2 of the main beam and placed on a movable mounting block. By reasonably arranging sensors, the operating status of the gantry crane can be monitored in real time, potential problems can be discovered and handled in a timely manner, and the safe and stable operation of the gantry crane can be ensured.

[0094] Step 2: Preprocess the collected operating data and monitoring data.

[0095] In some embodiments, the sampling frequency of the strain monitoring data of the main beam is 1 Hz, and the sampling frequency of the operating condition data is 1 Hz.

[0096] In the preprocessing process of the main beam strain monitoring data, a series of data cleaning and correction work is required. First, the collected raw data needs to be checked to eliminate outliers caused by equipment failure, sensor error or data transmission error. These outliers may cause deviations in data analysis, so they need to be eliminated or corrected.

[0097] Next, in order to better understand the strain characteristics of the main beam, the data was divided into different time periods according to the working status and load conditions of the crane, and statistical analysis was performed to find out the changing patterns of the main beam strain under different working conditions, providing a basis for subsequent structural analysis and optimization.

[0098] In addition, the collected data is filtered to eliminate noise and interference signals and improve the signal-to-noise ratio of the data.

[0099] Through the data preprocessing in the above steps, more accurate and reliable main beam strain data can be obtained, providing strong support for subsequent structural analysis and optimization.

[0100] Step 3. Perform trend analysis based on the actual strain monitoring data and predicted strain data using a comparative analysis algorithm.

[0101] According to the pre-processed real-time working condition data and actual strain monitoring data, the machine learning algorithm is used to realize the trend analysis of the main beam strain signal. The collected working condition data and strain monitoring data are processed and analyzed, and the machine learning algorithm is used to realize the intelligent early warning of abnormal strain signals of the gantry crane main beam.

[0102] In some embodiments, the prediction model is trained by the XGBoost algorithm based on the real-time working condition data and strain monitoring data collected on the large gantry crane. The XGBoost algorithm is a machine learning algorithm that is widely used in big data processing. The full name of XGBoost is eXtreme Gradient Boosting. The biggest feature of XGBoost is that it can automatically use the multi-threading of the CPU for parallelism, and at the same time improve the algorithm to improve the accuracy. XGBoost performs a 2nd-order Taylor expansion on the loss function, and adds a regularization term to the loss function to find the optimal solution as a whole, in order to weigh the decrease of the loss function and the complexity of the model to avoid overfitting. The XGBoost algorithm uses a 2nd-order Taylor expansion to approximate the objective function, so that it can perform leaf splitting optimization based only on the value of the input data, which improves scalability and freedom, and can construct a corresponding error function according to the specific application scenario to train a specific tree model.

[0103] The algorithm flow of XGBoost is as follows:

[0104] S1: Initialize the predicted value of each sample;

[0105] S2: Define a specific loss function;

[0106] S3: Calculate the derivative of the loss function for each sample prediction value;

[0107] S4: Build a new decision tree based on the derivative information;

[0108] S5: Use the new decision tree to predict the sample value and add it to the original value;

[0109] S6: Loop to create decision trees until the loss function meets the conditions.

[0110] For strain sensor signals, the historical operation data is processed and used for model training. The trained model uses PLC data as input, finds the relationship between PLC and strain sensor signal values ​​based on historical data, predicts the signal data of the specified sensor, and then calculates its predicted value with the data value collected by the strain sensor in actual work. The time series residual is used as the judgment standard, and an alarm is issued if an abnormality continues to occur within a certain time series. This alarm signal reflects the difference between the current equipment condition of the main beam of a large gantry crane and the equipment condition when it is healthy. Comparison between the predicted value and the actual value of the strain sensor signal Figure 2 shown.

[0111] In some embodiments, the data collected by the programmable logic controller PLC includes operating data such as the weight of the lifting mechanism, lifting height, lifting speed, trolley running speed, trolley running distance, trolley running speed, trolley running distance, etc.

[0112] Step 4. Perform trend analysis based on actual stress and theoretical stress using a comparative analysis algorithm.

[0113] Through the equipment operation status data and monitoring sensor data, the structural parts strain information under all working conditions is recorded, and a comparative analysis algorithm between the actual stress and theoretical stress of the gantry crane main beam based on sensor data is built. At the same time, big data analysis and prediction of the main beam stress change trend are carried out, and the current deformation state is judged according to the historical record data, and the main structure state is verified from two dimensions. Instant response warning is carried out for the sudden change of the corresponding variables, thereby extending the service life of the equipment and ensuring production safety.

[0114] It can be understood that the actual structural strain ε (dimensionless, usually expressed in microstrain or percentage) data measured by the strain sensor can be used to calculate the elastic modulus E (unit: Pa, or N / m 2 ) calculate the stress σ (unit: Pa, or N / m 2 ), in dynamic load monitoring, strain data is converted into stress in real time through elastic modulus for real-time stress change trend analysis. Within the elastic range, stress and strain satisfy the following relationship: σ = E·ε.

[0115] 1) Verification of theoretical value of strain on gantry crane main beam

[0116] When designing a gantry crane, the main beam stress needs to be checked according to the extreme working conditions. Therefore, it is necessary to compare the theoretical stress and actual stress of the main beam and verify the correctness of the values. After the mutual verification is reasonable, the theoretical stress can be used as a reference value for the actual stress to realize the monitoring and early warning of the main beam strain data.

[0117] 2) Actual stress of main beam

[0118] In some embodiments, a strain sensor is installed on the bottom plate of the main beam of the gantry crane along the extension direction of the main beam, and it is installed after the crane has been put into use. The strain data caused by the deadweight of the main beam cannot be measured, so the collected data is the microstrain (i.e., tensile strain) along the main beam direction caused by the moving load. After considering the influence of temperature compensation, the strain data is converted into actual stress in combination with the elastic modulus for subsequent data analysis.

[0119] 3) Theoretical stress of main beam

[0120] In some embodiments, the purpose of calculating the theoretical stress of the gantry crane main beam is to compare it with the actual stress of the crane gantry beam and verify the actual stress value. In order to keep the analysis object consistent, only the tensile stress caused by the moving load is considered (hereinafter referred to as: theoretical stress of the main beam). There are upper and lower trolleys on the main beam, and the trolley has a lifting mechanism and a hook-shifting mechanism. The load on the main beam includes the weight of the trolley and the weight of the hoisting load, which acts on the main beam through the wheels of the trolley. Therefore, the stress of the main beam is related to the working condition data such as the trolley position, lifting load, and hook-shifting distance.

[0121] 4) Verification of theoretical value and actual value

[0122] The overall trend of the theoretical stress and actual stress of the gantry crane main beam is consistent, but the two are not completely parallel. Figure 3 As shown. Check the correlation coefficient between the two and test the correlation between theory and practice, so as to achieve the purpose of mutual verification between theoretical stress and actual stress.

[0123] Step 5. Establish a fault decision tree based on the trend analysis results of steps 3 and 4, manual fault judgment and fault feature data of the historical fault library.

[0124] In some embodiments, constructing a fault decision tree comprises the following steps:

[0125] 1) The fault types in the historical fault data are used as labels and associated with the corresponding sensor data and features. Through expert experience, the fault types in some historical cases are manually determined and labeled to make up for the situation where sensor data alone cannot be fully labeled.

[0126] 2) Through historical data and expert judgment, analyze sensor data and historical fault records to identify common fault modes. Determine the range of characteristic values ​​corresponding to common fault modes, classify these characteristics into different fault types, and associate the extracted signal characteristic values ​​with the fault types.

[0127] 3) Build a fault decision tree

[0128] ① Design the hierarchical structure of the decision tree

[0129] First-layer node: Is there an abnormality?

[0130] By comparing the signal characteristic value with the standard value under normal operating conditions, it is determined whether there is an abnormality. For example, if the amplitude or frequency of the vibration signal exceeds the normal range, it is judged as "abnormal", otherwise it is "normal".

[0131] Second-layer nodes: preliminary classification of fault types

[0132] Based on the different characteristics of sensor data, the abnormal situations are preliminarily classified. For example:

[0133] Abnormal vibration → mechanical failure (such as bearings, gears, transmission mechanisms, etc.).

[0134] Abnormal current and voltage → electrical failure (such as motor, electrical control system).

[0135] Abnormal temperature → overheating or poor lubrication failure.

[0136] The third layer node: refined classification

[0137] By further analyzing the specific characteristic patterns, the fault types can be subdivided. For example:

[0138] Abnormal vibration:

[0139] High frequency vibration → bearing failure.

[0140] Low frequency, large amplitude vibration → mechanical structure is loose or unbalanced.

[0141] Electrical anomalies:

[0142] Frequent current fluctuations → motor overload.

[0143] Unstable voltage → control system failure.

[0144] ②Set decision conditions and thresholds

[0145] Set feature thresholds: Determine the decision criteria for each node based on historical data and expert experience. For example:

[0146] When the vibration amplitude exceeds a certain set value, it enters the "abnormal vibration" branch.

[0147] When the current rise rate exceeds the threshold, it enters the "Electrical Fault" branch.

[0148] Decision-making based on multi-parameter correlation: For faults associated with multiple parameters, decision-making conditions can be established by combining the changes in multiple signals. For example, if temperature rise and vibration abnormality occur at the same time, it can be further judged as "overheating and mechanical wear caused by poor lubrication."

[0149] ③Build relationships between nodes

[0150] According to the above conditions, the branch structure of the fault decision tree is gradually constructed. Each node makes a judgment based on the specific sensor feature value and enters the corresponding branch. For example:

[0151] Abnormal vibration → high frequency vibration → bearing failure.

[0152] Abnormal electrical signal → frequent current fluctuations → motor failure.

[0153] 4) Verify and optimize decision tree

[0154] ①Model verification

[0155] Input the collected historical data or new data into the decision tree model to verify the accuracy of its fault classification. If the classification accuracy is not high, it is necessary to adjust the threshold of the feature value or redesign the branch logic.

[0156] ②Optimize decision tree

[0157] Optimize decision trees by:

[0158] Dynamically adjust thresholds: Continuously adjust the feature thresholds set in the fault decision tree based on more historical data and expert feedback.

[0159] Adding new nodes: As the fault library expands, new fault types and new judgment nodes can be continuously added.

[0160] Remove redundant branches: If certain branch decisions are no longer applicable or the failure mode changes, redundant branches can be simplified or removed.

[0161] ③Expert adjustment and feedback

[0162] Regularly communicate with experts to provide feedback on the diagnosis results in the decision tree. Combined with the fault types and characteristic manifestations in actual cases, further improve the decision tree structure.

[0163] Step 6. Based on the strain signal trend analysis results, combined with the fault decision tree and strain signal fault knowledge graph, build a fault diagnosis algorithm for fault diagnosis.

[0164] The process of building a crane fault diagnosis knowledge graph includes multiple steps to manage fault knowledge in a systematic and structured manner and to achieve automatic updating and intelligent application. The following is a more detailed construction process:

[0165] (1) Knowledge acquisition and ontology model construction

[0166] Knowledge collection: Extract knowledge from various data sources related to crane fault diagnosis, including fault type, fault handling method, maintenance experience of professionals, equipment technical manuals and other diagnostic related documents. This knowledge provides detailed background and handling measures for various crane faults.

[0167] Ontology model construction: Build a gantry crane fault diagnosis ontology model and define the core concepts and relationships in the graph. For example, establish basic entities such as "fault type", "fault cause", "fault symptom", "treatment method", "spare parts", and define their relationships (such as "fault type-corresponding-treatment method", "treatment method-need-spare parts", etc.).

[0168] Graph data storage: Convert the ontology model into structured graph data and store it in the knowledge graph database to support subsequent fault diagnosis reasoning and knowledge retrieval.

[0169] (2) Preliminary construction and visualization of the map

[0170] Relational entity mapping: From the collected knowledge and data, fault cases, diagnostic processes, equipment components, etc. are mapped to nodes and relationships in the knowledge graph. Typical nodes include "equipment components" (such as pulleys, bearing seats), "fault results" (such as wear, loose bolts), etc.

[0171] Hierarchical relationship construction: Establish hierarchical relationships in the graph according to fault types, so that fault nodes at different levels can reflect the relationship from fault symptoms to root causes, from treatment methods to spare parts requirements.

[0172] Visualization: Generate a visual view of the knowledge graph, clearly present the core nodes and important relationships, and assist operation and maintenance personnel in understanding failure modes and common treatment methods.

[0173] (3) Knowledge graph update mechanism

[0174] Trend change determination and knowledge graph update: Operation and maintenance personnel analyze the causes of fault trend changes based on the equipment's operating trends and diagnostic data, and add the diagnostic results to the knowledge graph. By recording the cause analysis and treatment methods, the knowledge graph's ability to identify and respond to trend faults is improved.

[0175] Manual judgment and trend update: In the early stage, operation and maintenance experts manually judge trend changes and update the nodes (such as new fault causes) or relationships (such as new fault handling methods) of the knowledge graph.

[0176] Combination of manual and algorithmic updates: In the later stage, expert judgment is combined with diagnostic algorithms for updates. When the algorithm identifies abnormal trends, the knowledge graph is automatically updated, and the nodes and relationships are improved after manual confirmation, gradually reducing manual participation.

[0177] (4) Recording and processing of diagnostic algorithm output information

[0178] Information flow reception and storage: The knowledge graph framework receives the key fault information flows output by two diagnostic algorithms (such as machine learning-based and rule-based diagnostic algorithms), including detected anomalies, predicted fault types, and possible causes.

[0179] Graph update of important information: Map the key information output by the algorithm to the graph nodes. If a new failure mode or new processing method is detected, add or modify nodes and relationships in the graph to improve the fault diagnosis content of the graph.

[0180] Automatic update of anomalies: When the algorithm identifies a new abnormal pattern (such as a new vibration frequency anomaly in the lifting mechanism), the system will automatically generate a corresponding abnormal node in the graph and establish associations with related equipment and fault types.

[0181] (5) Knowledge graph training and self-learning

[0182] Automatic graph training: The updated graph data is used to train and optimize the fault diagnosis model to form a closed loop. The diagnosis algorithm will improve the recognition rate of new fault types based on the structured data of the graph.

[0183] Self-learning mechanism: Through the fault handling cases and experience data continuously accumulated by the graph, the diagnostic algorithm gradually achieves self-learning, which can more quickly identify new faults and adjust maintenance strategies.

[0184] Manual supervision and feedback: The self-learning process of the graph still requires manual supervision, and the automatically updated content of the graph must be checked regularly to ensure the diagnostic accuracy of the model.

[0185] (6) Continuous optimization and application of knowledge graphs

[0186] Fault prediction and reasoning: Through the retrieval and reasoning of knowledge graphs, the system can predict potential faults, assist operation and maintenance personnel in early inspection, and automatically recommend processing methods based on historical data.

[0187] Intelligent operation and maintenance decision support: The knowledge graph integrates the fault information and working condition data of all crane equipment, supports the operation and maintenance team to make quick decisions, and improves fault handling efficiency.

[0188] Model visualization optimization: Through the continuous optimization of the visualization map, complex relationships are made intuitive, which helps further operation and maintenance personnel training and knowledge sharing.

[0189] Through the above steps, the crane fault knowledge graph has achieved closed-loop management from construction, updating to intelligent application, greatly improving the efficiency and accuracy of fault diagnosis.

[0190] Crane equipment failure is a complex event, which is affected by many factors such as mechanical components, electrical components, control systems, servo systems, etc. For a crane failure case event, its various failure influencing factors and the case event itself form a complex network relationship, and the same set of events belonging to the same failure phenomenon constitutes a huge knowledge network system. As a direct representation and organization of knowledge relationships, knowledge graphs are more conducive to the accumulation and reuse of knowledge, and effectively analyze specific potential failures in complex relationships.

[0191] Combined with the data sources and characteristics of the crane failure field, a knowledge graph construction technology combining "top-down" and "bottom-up", a data-driven incremental ontology modeling method and a pattern-based knowledge mapping mechanism are adopted to organize rich diagnosis-related knowledge and complete the construction and application of the crane failure knowledge graph.

[0192] Based on the expert knowledge related to cranes and historical failure cases, the failure results, failure phenomena, failure modes, failure causes, impact range, etc. are built into the knowledge graph.

[0193] For example, in the fault knowledge graph, the fault decision tree can be used as a reasoning path in the graph. The entities and relationships in the knowledge graph can be simplified through the hierarchical logic of the fault decision tree.

[0194] It can be understood that the fault decision tree is a hierarchical structure based on logical branches, which is used to gradually narrow down the possible fault range and finally determine the cause of the fault. Each of its nodes represents a decision, and each branch represents the possible result of the decision. The structure of the fault decision tree is a tree structure, the root node represents the initial state of the system, and the leaf node is the final fault diagnosis result. Through a series of conditional judgments, the data is gradually divided until a certain type of fault is determined.

[0195] In addition, the fault knowledge graph is a network structure based on knowledge representation and semantic association, which contains entities (such as equipment components, fault types, signal characteristics) and the relationships between them. It is used to capture the associations between different concepts in the system and support complex fault reasoning and knowledge query. The structure of the fault knowledge graph is a graph structure, where nodes represent entities in the knowledge (such as equipment, fault phenomena, fault causes, characteristic signals, etc.), and edges represent the relationships between them (such as "cause", "dependency", etc.).

[0196] The entities and relationships in the fault knowledge graph are simplified through the hierarchical logic of the fault decision tree, for example:

[0197] Fault feature nodes such as "abnormal vibration", "current fluctuation", "temperature increase" in the knowledge graph may be connected to specific "bearing failure" or "motor failure" through the logical branches of the decision tree.

[0198] Decision trees provide a fast reasoning method based on conditional judgments, which can be embedded in a more complex reasoning framework of the knowledge graph to handle structured and standardized fault decision-making processes.

[0199] You can first use the fault decision tree to quickly and preliminarily screen the faults and identify the most likely fault type. Then, based on the knowledge graph, you can perform more in-depth reasoning and verification on the screened faults and use the multi-source knowledge in the graph to infer the root cause behind the fault. For example:

[0200] The decision tree determined the vibration anomaly by analyzing the sensor data and preliminarily judged it to be a "mechanical failure."

[0201] Next, the knowledge graph further associates historical data and equipment maintenance records to infer that the mechanical failure may be caused by long-term wear or poor lubrication.

[0202] For example, suppose a crane vibrates abnormally and the temperature rises. The system triggers the fault decision tree through sensor data and makes a preliminary judgment that it may be a "mechanical failure". Then, the knowledge graph further infers and finds that "mechanical failure" in historical data is often caused by "bearing wear" or "insufficient lubrication". The system finally gives the suggestion of "checking bearing wear or adding lubricant" and records this case path in the knowledge graph for subsequent reasoning and verification of similar faults.

[0203] This integrated model based on decision trees and knowledge graphs improves the efficiency and accuracy of intelligent fault diagnosis through rapid fault screening and multi-level reasoning, providing systematic support for the operation and maintenance of gantry cranes.

[0204] Based on the trend analysis results of the strain signal of the main structure (main beam), combined with the knowledge graph of the strain signal, a fault diagnosis algorithm based on the Bayesian network is built to infer the causes and corresponding probabilities of trend changes, output the diagnosis results, and feedback the key information flow of the fault to the knowledge graph. The intelligent early warning and diagnosis function of the strain signal of the main beam of a large crane is realized, and the diagnosis results push the abnormal part, abnormal time, signal evaluation value, early warning status, fault result, failure mode, failure cause, fault phenomenon, treatment measures, and impact range.

[0205] In some embodiments, the process of building a fault diagnosis algorithm based on a Bayesian network is as follows:

[0206] Fault diagnosis can actually be seen as a pattern recognition problem, judging whether the equipment is in normal state or has a fault; judging the fault form and location of the faulty equipment, in a sense, is a classification problem. By combining Bayesian classifiers, support vector machines and decision trees for fault diagnosis, the robustness and accuracy of classification can be improved, especially in scenarios with more uncertainty and noise. The combination of the three provides a basis for node selection or probability inference in the decision tree.

[0207] (1) Bayesian Classifier

[0208] The Bayesian classifier is the classifier with the lowest probability of classification error or the lowest average risk under a given cost. Its design method is a basic statistical classification method.

[0209] The principle of Bayesian classification is to use the prior probability of an object and the Bayesian formula to calculate its posterior probability, that is, the probability that the object belongs to a certain class, and select the class with the largest posterior probability as the class to which the object belongs. In short, it is to comprehensively evaluate the most likely result.

[0210] (2) Support Vector Machine

[0211] Support Vector Machines (SVM) is a binary classification model. Its purpose is to find a hyperplane to segment samples. The principle of segmentation is to maximize the interval, which is ultimately converted into a convex quadratic programming problem to solve.

[0212] Support vector machines generally deal with binary classification problems, but there are usually more than two faults. Therefore, when support vector machines are used for fault diagnosis, it is necessary to consider the expansion algorithm of multi-classification problems.

[0213] The advantages of decision tree combined with Bayesian classifier and SVM:

[0214] ① Improve classification accuracy: SVM can improve classification accuracy by processing high-dimensional data and nonlinear relationships, while the Bayesian classifier can provide a probabilistic basis for fault classification, making the decision tree more robust.

[0215] ② Handling complex fault modes: In fault diagnosis scenarios, fault modes often have complex feature spaces. SVM and Bayesian reasoning can help handle these complex modes.

[0216] ③Avoid overfitting: The probabilistic reasoning of the Bayesian classifier can smooth the decision-making process, while SVM reduces the risk of overfitting through maximum margin optimization.

[0217] In actual applications, the diagnostic result return form is shown in Table 1, where location: strain sensor No. 1; time: 2024 / 10 / 18, 13:00-13:09; signal evaluation value: 1.61871904034377; warning status: abnormal.

[0218] Table 1 Main beam strain intelligent diagnosis results

[0219]

[0220]

[0221] Step 7: Update the fault decision tree based on the diagnosis result data and the warning result data.

[0222] In some embodiments, the warning result data relies on health assessment and trend prediction algorithms. The data is obtained through evaluation methods such as trend analysis and remaining life prediction. It warns of possible failures of the crane structure or important components within a period of time and intervenes before the failure occurs, such as arranging preventive maintenance, planned maintenance, spare parts replacement, etc., thereby effectively extending the life of the equipment and avoiding sudden downtime.

[0223] (1) Structural adjustment of decision tree

[0224] 1) Introducing new features: Based on the diagnosis results and warning result data, new feature nodes can be introduced. For example, if the warning result data indicates that the possibility of a certain fault has increased, the relevant nodes in the fault decision tree can add the warning feature.

[0225] 2) Weight adjustment: For existing decision nodes, the diagnosis and warning results can be used to update the weight or probability of the node. Statistical methods based on historical data (such as Bayesian reasoning) can adjust the failure probability of nodes in the decision tree based on existing data.

[0226] 3) Prune redundant nodes: If the warning results or diagnosis results show that the probability of certain faults is extremely low, or the related faults are no longer important in decision-making, these nodes can be appropriately pruned in the decision tree to simplify the model.

[0227] (2) Rule Updates

[0228] 1) Rule improvement based on actual fault patterns: Diagnostic data may reveal new fault patterns or differences from existing fault patterns, and the rules in the fault decision tree need to be updated based on these patterns. "Rule" refers to the judgment criteria or conditions defined at each node of the decision tree, which are used to determine the branching of the path and ultimately lead to specific fault classification or decision. Each rule corresponds to a certain judgment criterion, such as a threshold range, a specific signal feature, or a data indicator.

[0229] 2) Threshold update guided by early warning information: The early warning result data may reveal early signs of failure, which can be used to update the thresholds of relevant nodes in the fault decision tree, enabling it to identify potential failures earlier.

[0230] (3) Model optimization and verification

[0231] 1) Training and testing: The updated decision tree is used in conjunction with new diagnosis and warning result data to train and test the model, using methods such as cross-validation to evaluate its accuracy and robustness.

[0232] 2) Real-time update: As new diagnostic results and warning information are continuously input, the structure and parameters of the decision tree will be continuously updated.

[0233] Some embodiments of the present application also provide an intelligent fault diagnosis system for a gantry crane, the system comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the system is triggered to execute the aforementioned intelligent fault diagnosis method for a gantry crane.

[0234] In summary, the solution provided by the present application can obtain the real-time working condition data and strain monitoring data of the gantry crane; and pre-process the data; use a comparative analysis algorithm based on the actual strain monitoring data and the predicted strain data to perform trend analysis; use a comparative analysis algorithm based on the actual stress and the theoretical stress to perform trend analysis; establish a fault decision tree based on the above trend analysis results, human fault judgment and fault feature data of the historical fault library; based on the strain signal trend analysis results, combined with the fault decision tree and the fault knowledge graph, build a fault diagnosis algorithm to perform main beam fault diagnosis; update the fault decision tree based on the diagnosis result data and the warning result data. The present application can monitor the equipment status of the gantry crane in real time, perform fault diagnosis and intelligent warning on the equipment according to the collected data, and repair the fault in a timely manner, thereby improving the safety of the gantry crane during production, avoiding the occurrence of gantry crane operation accidents, and improving the working efficiency of the gantry crane.

[0235] It should be noted that the present application can be implemented in software and / or a combination of software and hardware, for example, can be implemented using an application specific integrated circuit (ASIC), a general purpose computer or any other similar hardware device. In one embodiment, the software program of the present application can be executed by a processor to implement the steps or functions described above. Similarly, the software program of the present application (including relevant data structures) can be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and similar devices. In addition, some steps or functions of the present application can be implemented using hardware, for example, as a circuit that cooperates with a processor to perform each step or function.

[0236] In a typical configuration of the present application, the terminal and the network device each include one or more processors (CPU), input / output interface, network interface and memory.

[0237] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0238] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0239] In addition, a part of the present application may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present application through the operation of the computer. The program instruction for calling the method of the present application may be stored in a fixed or removable recording medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium, and / or stored in a working memory of a computer device that runs according to the program instruction. Here, according to an embodiment of the present application, a device is included, the device including a memory for storing computer program instructions and a processor for executing program instructions, wherein, when the computer program instruction is executed by the processor, the device is triggered to run the method and / or technical solution based on the aforementioned multiple embodiments according to the present application.

[0240] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the spirit or basic features of the present application. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present application is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present application. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the device claim can also be implemented by one unit or device through software or hardware.

Claims

1. An intelligent fault diagnosis method for a gantry crane, characterized in that: The following steps are involved: Step 1. Collect real-time working condition data and strain monitoring data; Step 2. Preprocess the collected working condition data and strain monitoring data; Step 3. Perform trend analysis based on the strain monitoring data and predicted strain data using a comparative analysis algorithm; Step 4. Calculate the actual stress based on the strain monitoring data, and use a comparative analysis algorithm to compare the actual stress with the theoretical stress to perform trend analysis; wherein, the comparative analysis algorithm is used to compare the actual stress with the theoretical stress to perform trend analysis, including: comparing the actual stress and the theoretical stress of the gantry crane main beam, and mutually verifying the correctness of the numerical values; after mutual verification, the theoretical stress is used as a reference value of the actual stress to monitor and warn the main beam strain monitoring data; Step 5. Construct a fault decision tree based on the trend analysis results of steps 3 and 4, manual fault judgment, and fault feature data of the historical fault library; Step 6. Based on the trend analysis results, fault decision tree and strain signal fault knowledge graph, build a fault diagnosis algorithm to perform fault diagnosis; Step 7. Update the fault decision tree based on the diagnosis result data and the warning result data; Wherein, constructing a fault decision tree comprises: 1) The fault types in the historical fault data are used as labels, associated with the corresponding strain monitoring data and sensor feature values, and the fault types in the historical cases are determined and labeled; 2) Through historical fault data, analyze sensor data and historical fault records, identify common fault modes, determine the feature value range corresponding to common fault modes, and classify these features into different fault types, and associate the extracted signal feature values ​​with the labels of the fault types; 3) Build a fault decision tree ① Design the hierarchical structure of the fault decision tree ②Set decision conditions and thresholds Set feature thresholds: Determine the decision criteria for each node based on historical failure data; Decision-making based on multi-parameter correlation: For faults with multi-parameter correlation, decision-making conditions are established by combining the changes in multiple signals; ③Build relationships between nodes According to the decision conditions, the branch structure of the fault decision tree is gradually constructed; each node makes a judgment based on the specific sensor feature value and enters the corresponding branch.

2. The method according to claim 1, characterized in that Collecting the strain monitoring data in the step 1 further includes: distributing the measuring points of the strain sensors on the two main beams of the gantry crane, arranging 3 measuring points on each main beam, and arranging a total of 6 strain sensors for monitoring the stress of the main beams, and the 6 strain sensors are installed on the lower panel inside the main beam box; and, arranging 2 strain sensors for temperature compensation, and the strain sensors for temperature compensation are installed on a movable mounting block near the strain sensor at the middle position of the main beam.

3. The method according to claim 1, characterized in that The preprocessing in step 2 includes cleaning, correcting or filtering the data.

4. The method according to claim 1, characterized in that The step 3 further includes: training a prediction model through the XGBoost algorithm based on the pre-processed real-time operating data and strain monitoring data; processing the historical operating data for the strain sensor signal and using it for model training; the trained model uses the data collected by the programmable logic controller (PLC) as input, finds the relationship between the data collected by the PLC and the strain sensor signal value based on the historical operating data, predicts the signal data of the specified strain sensor, and then calculates the predicted value with the strain monitoring data collected by the strain sensor in actual work, uses the time series residual as the judgment standard, and issues an alarm if the abnormality continues to occur within a certain time series.

5. The method according to claim 1, characterized in that The hierarchical structure of the design fault decision tree in step 5 specifically includes: First-layer node: Is there an abnormality? By comparing the signal characteristic value with the standard value under normal operating conditions, it is determined whether there is an abnormality; if the amplitude or frequency of the vibration signal exceeds the normal range, it is determined as "abnormality exists", otherwise it is "normal"; Second-layer nodes: preliminary classification of fault types Preliminarily classifying abnormal situations according to different characteristics of the sensor data; The third layer node: refined classification By further analyzing the specific characteristic patterns, the fault types can be subdivided.

6. The method according to claim 1, characterized in that The construction of the fault decision tree in step 5 further includes the following steps: Verify and optimize the fault decision tree: ①Model verification Input the collected historical fault data or new data into the fault decision tree to verify the accuracy of its fault classification. If the classification accuracy is not high, it is necessary to adjust the threshold of the feature value or redesign the branch logic; ②Optimize the fault decision tree The fault decision tree is optimized in the following three ways: i) Dynamically adjust the threshold: According to more of the historical fault data, continuously adjust the various feature thresholds set in the fault decision tree; ii) Add new nodes: As the fault library expands, continuously add new fault types and new judgment nodes; iii) Remove redundant branches: If some branch judgments are no longer applicable or the fault mode changes, simplify or remove redundant branches.

7. The method according to claim 1, characterized in that The construction process of the fault knowledge graph in step 6 includes: S1. Knowledge acquisition and ontology model construction: ① Knowledge acquisition: Extract knowledge from various data sources related to gantry crane fault diagnosis, including fault type, fault handling method, maintenance experience of professionals, and equipment technical manuals. The knowledge provides detailed background and handling measures for various gantry crane faults; ② Ontology model construction: Construct a gantry crane fault diagnosis ontology model and define the core concepts and relationships in the fault knowledge graph; ③Graph data storage: convert the ontology model into structured graph data and store it in the database of the fault knowledge graph to support subsequent fault diagnosis reasoning and knowledge retrieval; S2. Preliminary construction and visualization of fault knowledge graph: ① Relationship entity mapping: From the collected knowledge and data, fault cases, diagnostic processes, and equipment components are mapped into nodes and relationships in the fault knowledge graph; ② Hierarchical relationship construction: hierarchical relationships are established in the fault knowledge graph according to the fault type, so that the fault nodes at different levels reflect the correlation relationship from fault symptoms to root causes, from treatment methods to spare parts requirements; ③ Visualization display: Generate a visualization view of the fault knowledge graph to clearly present the core nodes and important relationships; S3. Fault knowledge graph update mechanism: ① Trend change determination and fault knowledge graph update: Operation and maintenance personnel analyze the causes of fault trend changes based on the equipment's operating trends and diagnostic data, and add the diagnostic results to the fault knowledge graph; by recording the cause analysis and treatment methods, the fault knowledge graph's ability to identify and respond to trend faults is improved; ② Manual judgment and trend update: In the early stage, operation and maintenance experts manually judge trend changes and update the nodes or relationships of the fault knowledge graph; ③ Manual and algorithm combined update: In the later stage, expert judgment is combined with the diagnostic algorithm for update. When the algorithm identifies an abnormal trend, the fault knowledge graph is automatically updated, and the nodes and relationships are improved after manual confirmation, gradually reducing manual participation; S4. Recording and processing of diagnostic algorithm output information: ① Information flow reception and storage: The fault knowledge graph framework receives the key fault information flow output by the diagnosis algorithm, including detected anomalies, predicted fault types and possible causes; ② Update the fault knowledge graph of important information: Map the key information output by the algorithm to the fault knowledge graph nodes. If a new fault mode or new processing method is detected, add or modify nodes and relationships in the fault knowledge graph to improve the fault diagnosis content of the fault knowledge graph; ③ Automatic update of anomalies: When the algorithm identifies a new anomaly pattern, the system will automatically generate a corresponding anomaly node in the fault knowledge graph and associate it with related equipment and fault types; S5. Fault knowledge graph training and self-learning: ① Automatic training of fault knowledge graph: The updated data of fault knowledge graph is used to train and optimize the fault diagnosis model to form a closed loop; the diagnosis algorithm is based on the structured data of fault knowledge graph to improve the recognition rate of new fault types; ② Self-learning mechanism: Through the fault handling cases and experience data continuously accumulated by the fault knowledge graph, the diagnostic algorithm gradually achieves self-learning, which can more quickly identify new faults and adjust maintenance strategies; ③ Manual supervision and feedback: The self-learning process of the fault knowledge graph still requires manual supervision, and the automatically updated content of the fault knowledge graph must be checked regularly to ensure the diagnostic accuracy of the model; S6. Continuous optimization and application of fault knowledge graph: ① Fault prediction and reasoning: Through the retrieval and reasoning of the fault knowledge graph, the system can predict potential faults, assist operation and maintenance personnel in early troubleshooting, and automatically recommend treatment methods based on historical data; ② Intelligent operation and maintenance decision support: The fault knowledge graph integrates the fault information and working condition data of all gantry crane equipment, supports the operation and maintenance team to make quick decisions, and improves the efficiency of fault handling; ③ Model visualization optimization: Through the continuous optimization of the visual fault knowledge graph, complex relationships are visualized, which is conducive to further training and knowledge sharing of operation and maintenance personnel.

8. The method according to claim 7, characterized in that The fault diagnosis algorithm in step 6 includes: combining a Bayesian classifier, a support vector machine and a fault decision tree for fault diagnosis, wherein the combination of the three provides a basis for node selection or probability inference in the fault decision tree; The Bayesian classifier calculates the posterior probability of a node object using the Bayesian formula through the prior probability of the node object, and selects the class with the maximum posterior probability as the class to which the node object belongs. The probabilistic reasoning of the Bayesian classifier can smooth the decision process and provide a probabilistic basis for fault classification, making the fault decision tree more robust. The support vector machine uses a multi-classification expansion algorithm to perform the fault diagnosis on multiple faults, and improves the classification accuracy of the fault decision tree by processing high-dimensional data and nonlinear relationships; the support vector machine can also reduce the risk of overfitting through maximum interval optimization.

9. The method according to claim 8, characterized in that The updating of the fault decision tree in step 7 includes the following steps: Ss1. Structural adjustment of fault decision tree 1) Introducing new features: Introducing new feature nodes based on the diagnosis results and warning result data; 2) Weight adjustment: For existing decision nodes, the diagnosis results and warning result data are used to update the weight or probability of the decision nodes; the failure probability of nodes in the fault decision tree is adjusted based on the existing data using a statistical method based on historical data; 3) Pruning redundant nodes: If the diagnosis results and warning result data indicate that the probability of occurrence of certain faults is extremely low, or the related faults are no longer important in decision-making, these nodes are appropriately pruned in the fault decision tree to simplify the model; SS2. Rule Update 1) Rule improvement based on actual fault patterns: When diagnostic data reveals new fault patterns or differences from existing fault patterns, the rules in the fault decision tree are updated according to these patterns; wherein the rules refer to the judgment criteria or conditions defined at each node of the fault decision tree, which are used to determine the branching of the path and ultimately lead to specific fault classification or decision; 2) Threshold update guided by early warning information: The early warning result data reveals early signs of faults and is used to update the thresholds of relevant nodes in the fault decision tree so that potential faults can be identified earlier; Ss3. Model optimization and verification 1) Training and testing: combining the updated fault decision tree with the new diagnosis results and warning results data, training the model and testing it, and using the cross-validation method to evaluate its accuracy and robustness; 2) Real-time update: With the continuous input of new diagnostic results and warning result data, the structure and parameters of the fault decision tree are continuously updated.

10. An intelligent fault diagnosis system for a gantry crane, the system comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein: When the computer program instructions are executed by the processor, the system is triggered to execute the method according to any one of claims 1 to 9.

Images