Robot fault diagnosis method and device, electronic equipment and storage medium

Abstract

The invention discloses a robot fault diagnosis method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring an abnormal description of a robot; fault reasoning is carried out on the abnormal description based on a preset fault tree and a preset reasoning rule, a first fault diagnosis result is obtained, and the first fault diagnosis result comprises the initial weight of each fault reason; based on a preset neural network model, performing mode recognition on the anomaly description to obtain a second fault diagnosis result, the second fault diagnosis result including a prediction weight of each fault cause; and carrying out weighted fusion on the initial weight and the prediction weight to obtain a comprehensive weight of each fault reason, and determining a target fault diagnosis result according to the comprehensive weight of each fault reason. According to the method, the possible fault causes and the comprehensive weight of each fault cause are deduced by combining the fault diagnosis result obtained by the preset fault tree and the fault diagnosis result obtained by the preset neural network, so that the accuracy of fault diagnosis is greatly improved.

Description

Technical Field

[0001] The present application relates to the technical field of robot fault diagnosis, and particularly relates to a robot fault diagnosis method, an electronic device, a computer-readable storage medium, and a robot fault diagnosis device. Background Art

[0002] With the wide application of robots in fields such as manufacturing and services, the operation and maintenance of robot equipment have gradually become the core technologies in various industries. However, during long-term operation, a variety of abnormalities may occur in the robot, resulting in equipment downtime or a decrease in production efficiency.

[0003] In related technologies, when an abnormality occurs in a robot, technicians conduct abnormality troubleshooting on the robot according to the operation manuals and maintenance guides provided by equipment manufacturers. However, this fault diagnosis method relies on manual experience. Due to subjectivity and limitations of the analysis method based on human judgment, it is difficult to accurately locate the causes of complex defects. At the same time, the manuals are often based on general situations and are difficult to cover all possible abnormal types in practice, reducing the accuracy of fault diagnosis. Summary of the Invention

[0004] The present application aims to at least solve one of the technical problems in related technologies to some extent. For this purpose, the first object of the present application is to propose a robot fault diagnosis method, which derives possible fault causes and the comprehensive weight of each fault cause by combining the first fault diagnosis result obtained based on a preset fault tree and the second fault diagnosis result obtained based on a preset neural network, greatly improving the accuracy of fault diagnosis and facilitating the rapid location of fault causes.

[0005] The second object of the present application is to propose an electronic device.

[0006] The third object of the present application is to propose a computer-readable storage medium.

[0007] The fourth object of the present application is to propose a robot fault diagnosis device.

[0008] To achieve the above object, an embodiment of the first aspect of the present application proposes a robot fault diagnosis method, which includes: obtaining an abnormal description of the robot; performing fault reasoning on the abnormal description based on a preset fault tree and a preset inference rule to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and the initial weight of each fault cause; performing pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and the predicted weight of each fault cause; performing weighted fusion on the initial weight and the predicted weight to obtain the comprehensive weight of each fault cause, and determining the target fault diagnosis result according to the comprehensive weight of each fault cause.

[0009] According to the robot fault diagnosis method of the embodiments of the present application, first obtain the abnormal description of the robot, perform fault reasoning on the abnormal description based on a preset fault tree and a preset inference rule to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and the initial weight of each fault cause, and perform pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and the predicted weight of each fault cause, and then perform weighted fusion on the initial weight and the predicted weight to obtain the comprehensive weight of each fault cause, so as to determine the target fault diagnosis result according to the comprehensive weight of each fault cause. Thus, this method combines the first fault diagnosis result obtained based on the preset fault tree and the second fault diagnosis result obtained based on the preset neural network to deduce possible fault causes and the comprehensive weight of each fault cause, greatly improving the accuracy of fault diagnosis and facilitating the rapid positioning of the fault cause.

[0010] In addition, the robot fault diagnosis method according to the above embodiments of the present application may further have the following additional technical features:

[0011] According to an embodiment of the present application, performing fault reasoning on the abnormal description based on a preset fault tree structure and a preset inference rule to obtain a first fault diagnosis result includes: extracting information from the abnormal description based on NLP (Natural Language Processing) to obtain abnormal information; determining an abnormal node in the preset fault tree based on the abnormal information; determining a root cause node corresponding to the abnormal node in combination with the node relationship in the preset fault tree and the preset inference rule; determining the fault cause corresponding to the abnormal description based on the root cause node, and determining the initial weight corresponding to each fault cause based on the preset weight between each root cause node and the abnormal node, so as to obtain the first fault diagnosis result.

[0012] According to an embodiment of the present application, performing pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result includes: extracting features from the abnormal description to obtain abnormal features; performing pattern recognition on the abnormal features based on the preset neural network model to obtain the fault cause and the probability value of each fault cause, and using the probability value as the predicted weight to obtain the second fault diagnosis result.

[0013] According to an embodiment of the present application, performing weighted fusion on the initial weight and the predicted weight to obtain the comprehensive weight of each fault cause includes: obtaining a first weight based on the product of the initial weight and a first preset ratio; obtaining a second weight based on the product of the predicted weight and a second preset ratio; obtaining the comprehensive weight based on the sum of the first weight and the second weight.

[0014] According to an embodiment of the present application, the robot fault diagnosis method further includes: performing abnormal processing on the robot based on the target fault diagnosis result, and determining the actual fault cause; adjusting the structure and / or weight of a preset fault tree based on the actual fault cause, and training a preset neural network model.

[0015] According to an embodiment of the present application, adjusting the structure and / or weight of a preset fault tree based on the actual fault cause includes: when the target fault diagnosis result includes the actual fault cause, adjusting the preset weight between the node corresponding to the actual fault cause and the abnormal node; when the target fault diagnosis result does not include the actual fault cause, adding a node corresponding to the actual fault cause in the preset fault tree, and setting the preset weight between the node corresponding to the actual fault cause and the abnormal node based on a preset rule.

[0016] According to an embodiment of the present application, the robot fault diagnosis method further includes: acquiring historical abnormal data of the robot, where the historical abnormal data includes abnormal phenomena and the fault causes of each abnormal phenomenon; performing NLP cleaning and standardization processing on the historical abnormal data to obtain historical abnormal data that meets a preset format; extracting and annotating information from the historical abnormal data that meets the preset format based on a convolutional neural network model to obtain a first short sentence, where the first short sentence includes an abnormal component and an abnormal phenomenon; performing machine type association on the first short sentence according to the type and attributes of the abnormal component to obtain a second short sentence; performing operation association on the first short sentence based on the operation information corresponding to the first short sentence to obtain a third short sentence; constructing fault tree nodes, the association relationships between each fault tree node, and determining the initial weight corresponding to each node in combination with the historical abnormal data based on the first short sentence, the second short sentence, and the third short sentence to obtain a preset fault tree.

[0017] To achieve the above object, an embodiment of the second aspect of the present application provides an electronic device, including a memory, a processor, and a robot fault diagnosis program stored on the memory and executable on the processor. When the processor executes the robot fault diagnosis program, the above-mentioned robot fault diagnosis method is implemented.

[0018] According to the electronic device of the embodiment of the present application, when the processor executes the robot fault diagnosis program, the above-mentioned robot fault diagnosis method is implemented. Based on the above-mentioned robot fault diagnosis method, the accuracy of fault diagnosis is greatly improved, which is conducive to quickly locating the fault cause.

[0019] To achieve the above object, an embodiment of the third aspect of the present application provides a computer-readable storage medium, on which a robot fault diagnosis program is stored. When the robot fault diagnosis program is executed by a processor, the above-mentioned robot fault diagnosis method is implemented.

[0020] A computer-readable storage medium according to an embodiment of the present application, when the robot fault diagnosis program stored thereon is executed by a processor, implements the above-mentioned robot fault diagnosis method. Based on the above-mentioned robot fault diagnosis method, the accuracy of fault diagnosis is greatly improved, which is conducive to quickly locating the cause of the fault.

[0021] To achieve the above object, an embodiment of the fourth aspect of the present application provides a robot fault diagnosis device, which includes: an acquisition module for acquiring an abnormal description of the robot; a first fault diagnosis module for performing fault reasoning on the abnormal description based on a preset fault tree and preset inference rules to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and an initial weight corresponding to each fault cause; a second fault diagnosis module for performing pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and a predicted weight corresponding to each fault cause; a weight fusion module for performing weighted fusion on the initial weight and the predicted weight to obtain a comprehensive weight corresponding to each fault cause, so as to determine a target fault diagnosis result according to the comprehensive weight of each fault cause.

[0022] The robot fault diagnosis device according to an embodiment of the present application acquires an abnormal description of the robot through the acquisition module, performs fault reasoning on the abnormal description based on a preset fault tree and preset inference rules through the first fault diagnosis module to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and an initial weight corresponding to each fault cause, performs pattern recognition on the abnormal description based on a preset neural network model through the second fault diagnosis module to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and a predicted weight corresponding to each fault cause, and the weight fusion module performs weighted fusion on the initial weight and the predicted weight to obtain a comprehensive weight corresponding to each fault cause, so as to determine a target fault diagnosis result according to the comprehensive weight of each fault cause. Thus, the device combines the first fault diagnosis result obtained based on the preset fault tree and the second fault diagnosis result obtained based on the preset neural network to deduce possible fault causes and the comprehensive weight of each fault cause, greatly improving the accuracy of fault diagnosis and facilitating the quick location of the cause of the fault.

[0023] Additional aspects and advantages of the present application will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present application. Description of the Drawings

[0024] Figure 1 It is a flowchart of a robot fault diagnosis method according to an embodiment of the present application;

[0025] Figure 2Flowchart of a method for constructing a preset fault tree according to an embodiment of the present application;

[0026] Figure 3 Flowchart of a robot fault diagnosis method according to a specific embodiment of the present application;

[0027] Figure 4 Block diagram of an electronic device according to an embodiment of the present application;

[0028] Figure 5 Connection diagram of a robot fault diagnosis device according to an embodiment of the present application. Detailed implementation manners

[0029] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals indicate the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, and should not be construed as a limitation of the present application.

[0030] The robot fault diagnosis method, electronic device, computer-readable storage medium, and robot fault diagnosis device proposed by the embodiments of the present application will be described below with reference to the accompanying drawings.

[0031] Figure 1 Flowchart of a robot fault diagnosis method according to an embodiment of the present application.

[0032] As Figure 1 shown, the robot fault diagnosis method according to the embodiment of the present application may include:

[0033] S1. Obtain an abnormal description of the robot;

[0034] S2. Perform fault reasoning on the abnormal description based on a preset fault tree and preset inference rules to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and the initial weight of each fault cause;

[0035] S3. Perform pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and the predicted weight of each fault cause;

[0036] S4. Perform weighted fusion on the initial weight and the predicted weight to obtain the comprehensive weight of each fault cause, and determine the target fault diagnosis result according to the comprehensive weight of each fault cause.

[0037] Specifically, the abnormal description of the robot can be obtained through user input or through data interaction with the control unit. Additionally, to ensure the accuracy of fault diagnosis, the robot model and brand information can also be obtained synchronously to determine the corresponding preset fault tree and preset neural network model, so as to ensure that the fault diagnosis method can perform targeted fault analysis on the robot device. Then, obtain the abnormal description of the corresponding robot, such as "the robot cannot grasp the object" or "the robotic arm cannot move normally", as well as the specific parts where abnormalities may occur, such as "joint motor abnormality" or "control panel problem". This information will serve as the basis for subsequent analysis to ensure that the fault diagnosis method can accurately identify the cause of the fault.

[0038] The preset fault tree is obtained from the knowledge graph constructed based on the historical abnormal data of the robot. Specifically, the historical abnormal data of the robot can be refined into multiple key feature nodes to construct a fault tree structure with weight information, so as to form a graph-based association relationship in the preset fault tree to support the two-way tasks of abnormal traceability, positioning, and update. The preset fault tree can be associated with the robot model to improve the accuracy of fault diagnosis. After receiving the abnormal description, match the abnormal nodes in the preset fault tree according to the key information (such as fault phenomena, symptoms, etc.) in the abnormal description, and search downward from the abnormal nodes along the preset fault tree according to the preset inference rules to find the fault cause nodes that may cause the abnormal nodes to occur, so as to deduce the possible fault causes and the initial weights corresponding to each fault cause, and obtain the first fault diagnosis result. Among them, the initial weight of each fault cause can be determined based on the probability of fault occurrence, importance, or historical data.

[0039] In addition, after receiving the abnormal description, the standard phenomenon library can also be automatically retrieved first, and the abnormal phenomenon closest to the abnormal description can be found through a matching algorithm. Then, the preset fault tree performs fault reasoning based on the abnormal phenomenon to obtain the first fault diagnosis result. Specifically, by analyzing the keywords of the abnormal description and combining the device characteristics of the robot, possible abnormal patterns can be found to display the most matching abnormal phenomenon to ensure the accuracy of fault diagnosis. For example, if the model entered by the user is a "X100" robot of a certain brand, all specific data related to "X100", such as common abnormalities and structural characteristics, will be automatically extracted, and then compared and matched with the abnormal description to find the closest abnormal phenomenon, so as to make the subsequent diagnosis process more efficient and accurate.

[0040] The preset neural network model can be a CNN (Convolutional Neural Networks) model, which is pre-trained through the historical abnormal data of the robot. After receiving the abnormal description, the preset neural network model predicts the cause of the fault according to the abnormal description, so as to predict and generate the corresponding second fault diagnosis result based on the current abnormal description, so as to ensure efficient and accurate defect analysis in the complex and changeable robot working environment.

[0041] Then, the weighted fusion algorithm is used to integrate the weights of each fault cause in the first fault diagnosis result and the second fault diagnosis result. For example, the comprehensive weight of each fault cause is dynamically calculated by setting a proportional parameter to obtain the target fault diagnosis result, and the possible defect causes are sorted according to the calculated comprehensive weight, so as to achieve the accurate sorting and display of the defect causes.

[0042] According to the abnormal description, or further combined with multi-dimensional information such as the robot model and user operation records, this embodiment derives possible abnormal parts and root causes based on the diagnosis results of the preset fault tree and the preset neural network model, and determines the comprehensive weight of each fault cause, which can quickly identify and accurately locate the abnormality in a short time, not only improving the accuracy of fault diagnosis, but also greatly shortening the maintenance time, and providing users with convenient technical support for fault resolution.

[0043] Furthermore, multiple solutions can be generated based on the fault causes determined by the target fault diagnosis result, and they are sorted and displayed according to the priority of the solution effect of each solution. Each solution will be accompanied by detailed operation steps and required tools, and the risk assessment and implementation suggestions of each solution will be displayed. Users can choose the most suitable solution to implement according to the recommended priority or personal judgment. At the same time, the display of the solution can help users understand and perform the maintenance operation, ensuring that users can successfully complete the abnormal troubleshooting and repair.

[0044] In an embodiment of the present application, fault reasoning is performed on the abnormal description based on the preset fault tree structure and the preset reasoning rules to obtain the first fault diagnosis result, including: extracting information from the abnormal description based on NLP technology to obtain abnormal information; determining abnormal nodes in the preset fault tree based on the abnormal information; combining the node relationships in the preset fault tree and the preset reasoning rules to determine the root cause nodes corresponding to the abnormal nodes; determining the fault causes corresponding to the abnormal description based on the root cause nodes, and determining the initial weights corresponding to each fault cause based on the preset weights between each root cause node and the abnormal node, so as to obtain the first fault diagnosis result.

[0045] Specifically, during the process of obtaining the anomaly description, various types of anomaly data can be collected and processed from the robot maintenance platform as the anomaly description, such as maintenance records, alarm logs, and user operation records, etc.

[0046] After obtaining the anomaly description of the robot, first, extract the anomaly information from the input anomaly description through NLP technology, such as key information like the part (e.g., "cooling system") and the alarm code (e.g., "high temperature alarm"). Then map the extracted anomaly information to the nodes in the preset fault tree to ensure a one-to-one correspondence between the anomaly part or phenomenon and the nodes in the preset fault tree. For example, the anomaly information "cooling system anomaly" can be mapped to the node "cooling system defect" in the preset fault tree; the anomaly information "high temperature" can be mapped to the node "temperature anomaly".

[0047] Then, utilize the causal relationships in the preset fault tree to trace back the potential root causes layer by layer from the anomaly phenomenon, gradually locate the problem part, and thus determine the root cause node corresponding to the anomaly node. Additionally, during the backtracking process, verify the causal relationships between the nodes. For example, when the anomaly node is the "temperature anomaly" node, based on the node relationships in the preset fault tree and the preset inference rules, determine a root cause node as the "cooling system anomaly" node, and then conduct a causal relationship verification to confirm whether it can trigger the phenomenon of "temperature anomaly", thereby ensuring the rationality and effectiveness of the fault analysis result.

[0048] Take the anomaly phenomenon corresponding to the root cause node as the fault cause, and obtain the initial weight of the fault cause according to the preset weight between the root cause node and the anomaly node, thereby generating the first fault diagnosis result. For example, if the anomaly node is the "temperature anomaly" node, the inferred root cause node is the "cooling system anomaly" node, and the preset weight between the "temperature anomaly" node and the "cooling system anomaly" node is 0.6, then the fault cause is the cooling system anomaly, and the initial weight of the cooling system anomaly is 0.6.

[0049] This embodiment gradually generates the fault diagnosis result based on the preset fault tree, which can achieve accurate test traceability and traceability analysis, ensuring the efficient diagnosis of complex anomalies.

[0050] In an embodiment of the present application, pattern recognition is performed on the anomaly description based on a preset neural network model to obtain a second fault diagnosis result, including: extracting features from the anomaly description to obtain anomaly features; performing pattern recognition on the anomaly features based on the preset neural network model to obtain the fault cause and the probability value of each fault cause, and using the probability value as the prediction weight to obtain the second fault diagnosis result.

[0051] Specifically, valuable information is first extracted from the anomaly description to obtain anomaly features, so that the subsequent preset neural network model can better identify anomaly patterns. For example, text feature extraction is performed using methods such as TF-IDF (term frequency–inverse document frequency) or Word2Vec (word to vector).

[0052] Then, the preset neural network model is used to perform pattern recognition on the extracted anomaly features to predict the cause of the fault and the probability value of each cause of the fault. The preset neural network model can be a deep learning model, such as an autoencoder, a Convolutional Neural Network (CNN), or a Recurrent Neural Network (RNN). The specific choice depends on the data type and application scenario. After the preset neural network model identifies the anomaly pattern, it outputs the probability value of each cause of the fault. These probability values can be calculated through the output layer of the model. For example, the Softmax (Normalized exponential function) function is used to normalize the output into probabilities. The probability value of each cause of the fault is used as the prediction weight to comprehensively evaluate the priority of the cause of the fault and generate the second fault diagnosis result. The second fault diagnosis result includes the specific cause of the fault that leads to the anomaly and the prediction weight of each cause of the fault.

[0053] This embodiment performs pattern recognition on the new anomaly description based on the preset neural network module to generate a probability distribution based on the existing data to predict the most likely cause of the fault.

[0054] In an embodiment of the present application, the initial weight and the prediction weight are weighted and fused to obtain the comprehensive weight of each cause of the fault, including: obtaining the first weight based on the product of the initial weight and the first preset ratio; obtaining the second weight based on the product of the prediction weight and the second preset ratio; and obtaining the comprehensive weight based on the sum of the first weight and the second weight.

[0055] That is to say, in the data fusion and weighted calculation stage, the initial weight of each fault cause determined based on the preset fault tree is fused with the predicted weight obtained from the prediction of the preset neural network model in proportion to form the comprehensive weight of each fault cause. The initial weight represents the fault result based on historical data analysis, while the predicted weight represents the pattern recognition predicted fault result based on the current abnormal description. By setting two parameters, the first preset ratio α and the second preset ratio β, the proportion of the two can be dynamically adjusted to make different data sources complement each other. For example, when the preset neural network model identifies a high probability of "sensor anomaly", even if the initial weight of this node in the preset fault tree structure is low, it can still be listed as a fault cause with a higher priority after weight fusion.

[0056] Taking the preset neural network model as a CNN (Convolutional Neural Networks) model as an example, the robot fault diagnosis process is as follows:

[0057] 1) Weight initialization: The preset fault tree generates preliminary weights for each possible fault cause through node relationships and inference rules. For example, when the abnormal description contains the phenomenon of "temperature anomaly", the fault tree structure may assign a higher initial weight to "cooling system defect" and a relatively lower weight to "sensor anomaly".

[0058] 2) CNN model prediction: Use the CNN model to perform pattern matching on the new abnormal description and generate a probability distribution based on the existing data to predict the most likely fault cause. For example, when the abnormal description contains the phenomenon of "temperature anomaly", the CNN model may generate a probability of 0.8 for "cooling system defect" and a probability of 0.6 for "sensor anomaly".

[0059] 3) Weighted fusion algorithm: Fuse the initial weight and the predicted weight based on each fault cause. Set two parameters: the first preset ratio α and the second preset ratio β, which respectively represent the proportion of the diagnosis result of the preset fault tree and the diagnosis result of the CNN model in the total weight. The calculation formula for the comprehensive weight of each fault cause is:

[0060] W final =α·W KG +β·W CNN

[0061] Where, W final represents the comprehensive weight, α represents the first preset ratio, W KG represents the initial weight, β represents the second preset ratio, and W CNN represents the predicted weight.

[0062] Assume that the initial weight of "cooling system defect" among the fault causes is 0.7, the predicted weight is 0.8, and α = 0.4 and β = 0.6 are set, then the comprehensive weight of "cooling system defect" finally is 0.76.

[0063] 4) Weight fine-tuning and sorting: Based on the comprehensive weights of each fault cause, sort each fault cause, and display the most likely fault causes to the user in descending order to output the target fault diagnosis result.

[0064] Through this weighted fusion algorithm, the knowledge graph logical reasoning advantage of the preset fault tree and the pattern recognition prediction ability of the preset neural network model are effectively combined, ensuring higher diagnostic accuracy and flexibility when dealing with new abnormal situations, and further improving the reliability of fault cause location.

[0065] In an embodiment of the present application, the robot fault diagnosis method further includes: performing abnormal processing on the robot based on the target fault diagnosis result, and determining the actual fault cause; adjusting the structure and / or weight of the preset fault tree based on the actual fault cause, and training the preset neural network model.

[0066] Specifically, when performing abnormal processing based on the target fault diagnosis result, obtain the actual fault cause that leads to the abnormality for the update and optimization of the preset fault tree and the preset neural network. For example, if "cooling system abnormality" is confirmed as the actual fault cause leading to "temperature abnormality" in fault diagnosis, the weight of this node in the preset fault tree is automatically increased to ensure a higher priority for this fault cause in subsequent similar cases; or, when the actual fault cause is determined to be a new fault cause, a corresponding node can be added to the preset fault tree.

[0067] The adjustment of the preset fault tree and the training of the preset neural network model according to the actual fault cause in this embodiment can realize the self-learning of the model architecture, ensure the response ability of the fault diagnosis method to new abnormalities, and greatly improve the diagnostic coverage and diagnostic accuracy.

[0068] In an embodiment of the present application, adjusting the structure and / or weight of the preset fault tree based on the actual fault cause includes: when the target fault diagnosis result includes the actual fault cause, adjusting the preset weight between the node corresponding to the actual fault cause and the abnormal node; when the target fault diagnosis result does not include the actual fault cause, adding a node corresponding to the actual fault cause in the preset fault tree, and setting the preset weight between the node corresponding to the actual fault cause and the abnormal node based on the preset rule.

[0069] That is, when the actual failure cause is the failure cause in the target failure diagnosis result, the preset fault tree is weight-adjusted. Specifically, the preset weight between the root cause node corresponding to this failure cause and the abnormal node corresponding to the abnormal description is increased. At the same time, the preset weight between the root cause node corresponding to other failure causes in the target failure diagnosis result and the abnormal node can be correspondingly decreased, or the preset weight between the root cause node corresponding to other failure causes in the target failure diagnosis result and the abnormal node remains unchanged.

[0070] When the actual failure cause is not in the target failure diagnosis result, it is determined that the connection relationship between this failure cause and the abnormal node is not included in the preset fault tree structure. Then, a node corresponding to the actual failure cause is added, and the connection and inference rules with the abnormal node are constructed, and the preset weight between the node corresponding to the actual failure cause and the abnormal node is set to supplement and expand the fault diagnosis inference path to meet the diversity and flexibility of robot abnormal diagnosis.

[0071] Based on the actual failure cause, this embodiment can realize the weight fine-tuning of the preset fault tree and the dynamic update of the knowledge base, ensuring that the diagnosis result can be continuously optimized to adapt to the changing abnormal environment and new abnormal types.

[0072] In an embodiment of the present application, the robot fault diagnosis method further includes: obtaining the historical abnormal data of the robot, where the historical abnormal data includes abnormal phenomena and the failure causes of each abnormal phenomenon; performing NLP cleaning and standardization processing on the historical abnormal data to obtain historical abnormal data that meets the preset format; based on the convolutional neural network model, performing information extraction and annotation on the historical abnormal data that meets the preset format to obtain the first short sentence, where the first short sentence includes abnormal components and abnormal phenomena; performing machine type association on the first short sentence according to the types and attributes of the abnormal components to obtain the second short sentence; performing operation association on the first short sentence based on the operation information corresponding to the first short sentence to obtain the third short sentence; constructing the fault tree nodes, the association relationships between each fault tree node, and determining the initial weight corresponding to each node in combination with the historical abnormal data to obtain the preset fault tree.

[0073] Specifically, the construction of the preset fault tree can be as Figure 2 shown, including the following steps:

[0074] S101. Obtain the historical abnormal data of the robot.

[0075] First, extract the required abnormal phenomenon and cause data from the structured SQL (Structured Query Language) database to obtain the historical abnormal data of the robot. These data usually include the operation logs, abnormal reports, operation records, etc. of the device.

[0076] S102. Perform NLP cleaning and standardization on historical abnormal data to obtain historical abnormal data that meets the preset format.

[0077] To ensure the accuracy and integrity of the data, the extracted historical abnormal data is first preprocessed, including format conversion, data normalization, noise data filtering, etc. After processing, a structured data set is obtained, laying a foundation for the subsequent construction of the knowledge graph.

[0078] In addition, during the data preprocessing process, the fields in the data are also initially cleaned to remove irrelevant information, ensuring that the abnormal phenomenon and cause data can be efficiently processed subsequently. If the database contains a large amount of data, a batch loading and preprocessing method can be adopted to ensure the stability and efficiency of data processing.

[0079] After the data preprocessing is completed, word segmentation and cleaning operations are performed on the abnormal phenomenon and cause text data. First, stop word removal is performed to eliminate words that are meaningless for information analysis, such as common words and function words. At the same time, the cleaned text will remove redundant symbols, useless characters, etc., to ensure the standardization of the data. Then, a custom dictionary is used to annotate special terms to ensure that the word segmentation results meet the professional requirements in the fields of robots and equipment. For example, words such as "motor abnormality" and "temperature abnormality" are accurately marked for subsequent analysis. Finally, the text content is segmented into multiple short sub-clauses according to semantic logic. These short sub-clauses usually have clear information such as abnormal phenomena, components, and operations, facilitating independent extraction and analysis in the subsequent steps of the system.

[0080] S103. Perform information extraction and annotation on the historical abnormal data that meets the preset format based on the convolutional neural network model to obtain the first short sentence (k3 sentences).

[0081] That is to say, in the information extraction stage, a convolutional neural network (CNN) model is used to analyze the text data to identify key contents such as alarm words and location words in the abnormal description. Among them, the CNN model is trained to be able to extract keywords related to abnormal phenomena from the text to ensure the accuracy and integrity of the extraction.

[0082] For complex abnormal descriptions, the abnormal phenomena are further split to generate multiple short sentences (k3 sentences) containing key information, and according to features such as alarm information and abnormal components, they are labeled as different abnormal phenomenon nodes. For example, when describing "temperature sensor alarm, motor shutdown", it can automatically identify "temperature sensor" and "motor" as component words, and "alarm" and "shutdown" as abnormal phenomenon words, and form corresponding k3 sentences. These sentences are used as basic information nodes in the subsequent construction of the knowledge graph to represent specific abnormal phenomena.

[0083] After the extraction and annotation of abnormal phenomena, the data is further classified and categorized, and abnormal type and operation extraction are performed (Steps S104 and S105).

[0084] S104, associate the first short sentence with a machine type according to the type and attributes of the abnormal component to obtain a second short sentence (k1 sentence).

[0085] That is to say, according to the type and attributes of the device, each piece of data is associated with the corresponding machine type to generate a k1 sentence, so that each node in the knowledge graph has a clear type identifier.

[0086] S105, perform operation association on the first short sentence based on the operation information corresponding to the first short sentence to obtain a third short sentence (k2 sentence);

[0087] That is to say, extract the specific operations performed by the user on the device when the abnormality occurs to generate a k2 sentence, and these operation information will be used as the operation nodes associated with the abnormal phenomena in the knowledge graph to help the system more accurately locate the cause of the failure.

[0088] For example, when recording "After the user turns off the device power, the system does not restart", the system can extract "turn off the device power" as the operation sentence (k2 sentence) and associate it with the corresponding abnormal phenomenon node. Through the joint analysis of the operation sentence (k2 sentence) and the abnormal phenomenon sentence (k3 sentence), the system can more effectively infer the possible causes of the abnormality.

[0089] S106, construct the fault tree nodes, the association relationships between each fault tree node, and determine the initial weight corresponding to each node in combination with historical abnormal data based on the first short sentence (k3 sentence), the second short sentence (k1 sentence), and the third short sentence (k3 sentence) to obtain a preset fault tree.

[0090] In this step, various association relationships of the knowledge graph are constructed based on the previously extracted k1 sentence, k2 sentence, and k3 sentence to obtain the corresponding preset fault tree. First, based on the k1 sentence, establish the causal or association relationships between abnormal phenomena, and form a series of abnormal chains by analyzing the relevance between different abnormal phenomena. Then, based on the k2 sentence, associate the operations with the abnormal phenomenon nodes to describe which operations will cause which abnormal phenomena to occur. Finally, based on the k3 sentence, construct the association relationship between the alarm code and the abnormal phenomenon through the relationship between the alarm word and the abnormal component.

[0091] For example, abnormal phenomenon A may lead to abnormal phenomenon B, and the system will record the direct or indirect relationship between the phenomena. For another example, a certain operation of the device (such as "restart") may eliminate or trigger a certain abnormal phenomenon, and the system will track and predict the abnormality through the two-way association between the operation and the abnormal phenomenon.

[0092] S107. Storing and exporting the preset fault tree.

[0093] After the preset fault tree is built, it is stored in RDF (Resource Description Framework) format to facilitate subsequent rapid query and reasoning. Data in RDF format has good semantic description capabilities and is suitable for representing node and edge information in the knowledge graph. At the same time, the RDF data is imported into the Neo4j graph database, which has the advantages of a graph structure, can efficiently store and display the graph structure, and supports complex queries and graph data analysis.

[0094] In the database, each node of the preset fault tree (abnormal phenomena, operations, alarm codes, etc.) and the relationship between them are clearly stored, forming a complete and searchable knowledge network, providing users with visual fault diagnosis information support.

[0095] S108. Dynamic update of the preset fault tree.

[0096] After the preset fault tree is built, as abnormal data accumulates during the use and maintenance of the robot equipment, the knowledge graph of the existing preset fault tree is dynamically updated. For example, when new abnormal data is detected, data preprocessing and word segmentation cleaning will be performed first, and the corresponding k1, k2, and k3 sentences will be generated according to the above steps. Next, new nodes and relationships are added to the knowledge graph of the existing preset fault tree to ensure the real-time and accuracy of the knowledge graph.

[0097] At the same time, the difference between new data and existing knowledge is analyzed. If new abnormal patterns or cause relationships are found, these relationships will be automatically added to the Neo4j graph database. This dynamic update mechanism enables the knowledge graph of the preset fault tree to have self-learning capabilities, which can be continuously enriched and expanded during operation, providing users with more accurate fault diagnosis support.

[0098] By building a preset fault tree through the above steps, potential defects in equipment or products can be accurately identified, thereby assisting technicians to solve problems efficiently. This process not only improves the stability and maintenance efficiency of robot equipment, but also provides efficient technical support for abnormal management of equipment.

[0099] In order to further improve the robot's abnormality diagnosis performance, the following improvements can be made:

[0100] Enhanced Fault Diagnosis Model Training: By introducing more abnormal case data and optimizing the training algorithm, further improve the accuracy and adaptability of abnormal diagnosis, enabling it to better cover a wide range of fault types.

[0101] Multi-modal Data Fusion: Combine multi-modal data such as images and videos to enhance the fault analysis ability of the preset neural network model. For example, in certain abnormal scenarios, the image or video information of the robotic device can be combined for fault identification and location, thereby providing more detailed analysis results.

[0102] User Feedback Mechanism: Establish a user feedback mechanism that allows technicians to provide feedback on the test traceability results. Based on the user feedback, the fault diagnosis model can be optimized, thereby continuously improving its self-adaptability and location accuracy.

[0103] Dynamic Knowledge Update: Through an automated data update mechanism, maintain the latest status of abnormal patterns and abnormal information in the knowledge graph, enabling timely adjustment to follow device updates or environmental changes to adapt to the ever-changing diagnostic requirements.

[0104] These improvement measures will further enhance the intelligence and effectiveness of the fault diagnosis method, enabling it to provide higher-quality diagnostic technical support in more complex scenarios.

[0105] As a specific embodiment of this application, as Figure 3 shown, the robotic fault diagnosis method may include the following steps:

[0106] S201, Data Collection and Preprocessing.

[0107] 1. Multi-source Data Acquisition: Collect abnormal and maintenance information related to the robot from multiple data sources to ensure the diversity and comprehensiveness of the data, including but not limited to the following types of data:

[0108] (1) Maintenance Platform Data: Obtain regular maintenance and emergency repair records from the robot maintenance platform, including information such as operation history and abnormal descriptions, providing detailed operation records of the device in different states;

[0109] (2) Sensor Data: Collect real-time operation data from sensors on the robotic device, such as key parameters like temperature, pressure, and current, to help the system analyze the state changes of the machine before and after an abnormality occurs;

[0110] (3) Operation Logs: Record the operation instructions of the operator during the operation of the robot, such as start / stop operations and mode switches, reflecting the state changes during the human-machine interaction process;

[0111] (4) User feedback: Collect subjective feedback and anomaly reports from users during the use of the robot, provide anomaly phenomena observed from the user's perspective, and increase the dimension of anomaly descriptions.

[0112] (5) Alarm data: The system records the alarm codes and specific descriptions each time an anomaly is triggered, providing important clues for subsequent traceability analysis.

[0113] 2. NLP cleaning and standardization: To ensure the accuracy and consistency of data, perform natural language processing (NLP) cleaning and standardization on the collected raw data:

[0114] (1) Noise removal: Delete redundant symbols, special characters, and irrelevant text in the anomaly text data to simplify the data content and remove unnecessary information.

[0115] (2) Data format standardization: Uniformly convert data from different sources into a predetermined structured format, such as converting data such as temperature and pressure into standard unit formats.

[0116] (3) Unified coding of anomaly records: According to the robot maintenance standard, file the collected anomaly records according to a unified coding system, that is, achieve unified coding formats for subsequent data management and search.

[0117] 3. Semantic parsing and feature extraction: Use natural language processing and semantic analysis technologies to extract structured key information from anomaly descriptions, including alarm codes, anomaly locations, anomaly phenomena, etc., to ensure the comprehensive and accurate construction of knowledge graph nodes:

[0118] (1) Part-of-speech tagging: Perform part-of-speech tagging on each word in the anomaly description to identify parts of speech such as nouns, verbs, and adjectives, thereby helping the system understand important entities and actions in the description;

[0119] (2) Syntax analysis: Analyze the syntactic structure of the anomaly description, extract the main information in the description (such as "temperature anomaly" or "sensor failure"), and extract key causal relationships from the sentence structure;

[0120] (3) Key information extraction: Through named entity recognition (NER) and keyword extraction algorithms, capture specific attributes in the anomaly description, such as anomaly location (e.g., "cooling system"), anomaly phenomenon (e.g., "temperature is too high"), etc.;

[0121] (4) Node information generation: Generate node names based on key information extraction.

[0122] 4. Label Data Generation: Provide clear initial label data for subsequent knowledge graph construction and fault tree analysis. Generate labels through supervised learning and annotate the data in a standardized manner:

[0123] (1) Data Annotation: Conduct supervised annotation on historical abnormal data, indicating the abnormal type, the subsystems involved, and the possible causes, providing clear labels for subsequent data analysis;

[0124] (2) Category Label Generation: Based on expert knowledge and standard definitions in the robotics field, classify the fault causes into several categories (such as electrical abnormalities, mechanical abnormalities, etc.) based on data annotation, and generate category labels for abnormal descriptions;

[0125] (3) Standardized Data Structure: Map the extracted key information such as abnormal records and operation logs into a standardized structure format, such as "abnormal phenomenon + abnormal location + operation process", for subsequent data processing;

[0126] (4) Initial Weight Setting: Based on the collected data and / or expert experience, assign preliminary weights to different fault causes to represent the likelihood weights of these fault causes in the existing knowledge base, providing a basic basis for subsequent analysis.

[0127] The purpose of this step is to transform the original abnormal data from different channels into structured information through a systematic data collection, cleaning, and feature extraction process, providing a high-quality data foundation for the construction of the knowledge graph and the fault tree.

[0128] S202, Knowledge Graph and Fault Tree Construction.

[0129] 1. Multi-level Node Definition: According to the attributes of different abnormal information, divide the nodes of the knowledge graph into multiple types to refine the hierarchical structure of the fault tree:

[0130] (1) Node Type Definition: Define node types such as "abnormal phenomenon", "component location", "alarm code", "operation process", etc. to distinguish different types of information;

[0131] (2) Hierarchical Organizational Structure: Organize the nodes according to the hierarchical structure of the equipment to ensure the clarity of the fault tree topology. For example, divide the nodes according to the "system - subsystem - component" hierarchy, so that the abnormal information is distributed at different levels for traceability analysis;

[0132] (3) Attribute Annotation: Attach descriptive attributes to each node, such as abnormal type, priority, associated system, etc., to ensure that the nodes can provide comprehensive information during the reasoning process.

[0133] 2. Initialization of Fault Tree Weights: Used to assign initial weights to each node and its associated relationships in the knowledge graph, representing the relative importance of different nodes, and constructing a preset fault tree. The initial weights can be set based on historical data or expert experience to ensure the accuracy of reasoning and analysis.

[0134] (1) Node Weights: Assign initial weights to each fault cause node. For example, the weight of the "Cooling System Abnormality" node can be set to 0.8, and the weight of the "Sensor Abnormality" node can be set to 0.6 to reflect their importance in different scenarios.

[0135] (2) Relationship Weights: Attach weights to the causal relationships between nodes, representing the correlation strength between abnormal phenomena and potential causes. For example, the weight between "High Temperature Alarm" and "Cooling System Abnormality" can be set to 0.9.

[0136] (3) Priority Assignment: Set the priorities of different abnormal types according to historical abnormal data and expert suggestions to ensure that common or severe abnormalities are preferentially recognized in reasoning. For example, when abnormal types A and B occur simultaneously, the abnormal diagnosis of abnormal type A is carried out first according to the type priority, and then the abnormal diagnosis of abnormal type B is carried out.

[0137] 3. Node Instantiation and Storage: Instantiate specific abnormal information into knowledge graph nodes and store them in a graph database (such as Neo4j) for efficient storage and query.

[0138] (1) Instantiation Process: Specify the abnormal description and attributes into node instances, such as "Cooling System Abnormality - 2023 / 11 / 02".

[0139] (2) Graph Database Storage: Store the instantiated nodes in the graph database to enable structured information to support complex queries and relationship parsing.

[0140] (3) Index Mechanism: Establish indexes for node types and relationships in the database to improve the query efficiency of the system for quickly finding relevant abnormal information.

[0141] 4. Multi-dimensional Relationship Annotation: The relationships between nodes in the knowledge graph can be multi-dimensional. By annotating the relationships between nodes, a multi-dimensional relationship network that supports reasoning is generated.

[0142] (1) Causal Relationship Annotation: Annotate the causal relationship between abnormal phenomena and causes (such as "Cooling System Abnormality" causing "Temperature Abnormality") to support the system for traceability reasoning.

[0143] (2) Trigger Relationship Annotation: Annotate the trigger relationship. For example, "Sensor Alarm Code" triggers "Temperature Abnormality" to ensure that the system can identify related events.

[0144] (3) Priority Relationship Annotation: Annotate the priority among nodes to ensure that high-risk events are preferentially displayed in case of multiple exceptions. For example, label "safety-related exception" with a higher priority;

[0145] (4) Logic Chain Construction: Construct the logic chain between the cause and phenomenon of the fault according to different relationship types. For example, "Operation A → Phenomenon B → Alarm C", providing a basis for multi-dimensional reasoning.

[0146] 5. Inference Rule Definition and Graph Verification: To support automatic reasoning analysis, define inference rules in the knowledge graph and verify the rationality of the graph through data testing.

[0147] (1) Inference Rule Setting: Set inference rules, such as "If Operation A causes Phenomenon B, then Phenomenon B may trigger Alarm C". These rules are used to identify potential causes in complex exceptions and locate them according to the logic chain;

[0148] (2) Verification and Testing: Verify the knowledge graph, simulate actual exception scenarios, and ensure the accuracy of node relationships and inference rules. For example, test the association relationships in the knowledge graph through historical exception data to see if it can accurately locate the common fault causes;

[0149] (3) Dynamic Adjustment: During the verification process, if it is found that the inference rules in the graph are imperfect, fine-tune the inference rules and relationship weights according to the feedback information.

[0150] This step generates a fault tree structure containing weight information through the knowledge graph construction algorithm to ensure that it can support exception traceability and reasoning analysis in actual use. Through the integration of the knowledge graph and the fault tree, complex causal relationship chains can be established to achieve multi-dimensional and hierarchical exception analysis, providing a solid foundation for subsequent test traceability, analysis, and update mechanisms.

[0151] S203, Test Traceability and Traceability Analysis.

[0152] 1. Exception Description Parsing: The system extracts key exception information from the exception description input by the user and maps it to the nodes of the knowledge graph, providing a basis for traceability analysis.

[0153] (1) Exception Information Extraction: Use natural language processing technology to extract fault information from the input text, such as key information like the location (e.g., "cooling system") and alarm code (e.g., "high temperature alarm");

[0154] (2) Matching Nodes in the Preset Fault Tree: The extracted exception information is mapped to the nodes in the preset fault tree to ensure a one-to-one correspondence between the fault location or phenomenon and the nodes. For example, the exception description "cooling system failure" can be mapped to the node "cooling system defect" in the graph;

[0155] (3) Abnormal description mapping: Map the abnormal description to a specific node instance, create a fault instance record, and determine the location in combination with the current scenario to provide necessary context support for traceability reasoning.

[0156] 2. Multi-level traceability reasoning: Trace back to the root cause of the fault through the node relationships of the preset fault tree, analyze layer by layer in depth, and form a recommended list of solutions with priority ranking.

[0157] (1) Hierarchical backtracking based on the preset fault tree: Utilize the causal relationships in the preset fault tree to trace back the potential root causes layer by layer from abnormal phenomena. The backtracking at each layer locates the problem area step by step according to the upstream and downstream relationships of the nodes;

[0158] (2) Causal relationship verification: During the backtracking process, verify the causal relationships between nodes. For example, when backtracking to the "cooling system failure" node, confirm whether it can cause the phenomenon of "abnormal temperature" to ensure the rationality and effectiveness of the analysis;

[0159] (3) Solution generation and ranking: Generate solutions step by step through the preset fault tree, and rank each solution according to the priority of the potential root causes to ensure that users can obtain efficient solutions first.

[0160] 3. Recommended solutions: Based on the results of fault tree reasoning, recommend possible solutions to users, and preferentially display the most likely causes and detailed solution steps according to the weights.

[0161] (1) Analysis of reasoning results: Synthesize the results of multi-level traceability reasoning, obtain the most likely cause of the fault, and generate effective solutions accordingly;

[0162] (2) Priority ranking of recommended solutions: Adjust according to the weighted weights, arrange the recommended solutions in priority, and display the most effective solutions to users first;

[0163] (3) Suggestions on detailed solution steps: Provide specific operation steps for each recommended solution, including troubleshooting methods, equipment inspection processes, etc., to ensure that users can execute the solutions quickly and accurately.

[0164] 4. Test traceability: Combine the traceability results and the fault tree reasoning path to finally determine the specific location of the fault.

[0165] (1) Feedback of location results: Combine the traceability analysis and the fault tree backtracking path to determine the exact location of the fault, and record the location results in the knowledge base for future traceability and update;

[0166] (2) Real-time location update: Through user feedback and system self-learning adjustment, continuously optimize the location process to make it more in line with the actual application scenario requirements, and improve the accuracy and stability of fault diagnosis.

[0167] Through these four steps, precise test traceability and traceability analysis are achieved. The intelligent recommendation combining the knowledge graph and the fault tree structure ensures the efficient resolution of complex faults. Based on the dynamic optimization of the inference path and weight assignment, accurate and reliable fault solutions can be provided under different working conditions, providing users with powerful technical support and an optimized experience.

[0168] S204, weighted fusion and feedback update.

[0169] 1. User feedback collection: The goal of this step is to evaluate the effectiveness of the recommended solution through user feedback and collect new abnormal information for continuous improvement and updating of the knowledge base.

[0170] (1) Record user feedback: A feedback mechanism can be designed to allow users to evaluate the recommended solution. For example, users can mark whether the solution is effective, whether it solves the fault problem, or provide more detailed feedback (such as "the solution did not solve the problem" or "the solution solved the problem but there are minor issues");

[0171] (2) Annotate the effectiveness of the solution: The effectiveness annotation of user feedback is crucial for subsequent weight updates. The system uses these annotations to judge the correctness and rationality of the recommended solution. By counting the proportion of effective and ineffective solutions, the system can identify which solutions are more reliable in the fault location process;

[0172] (3) Collect new abnormal information: During the execution of the solution, users may encounter new abnormal situations. The system collects these new abnormal information regularly or in real-time and uses it as new data input for subsequent updates and optimizations.

[0173] 2. Weighted fusion: Combining the probability distribution of defect causes generated by the CNN model, the node weights in the knowledge graph are adjusted according to preset parameters, and the fault causes are ranked in the form of comprehensive weights. When further implementing the inference model, the prediction results of the knowledge graph and the convolutional neural network (CNN) model are combined, and the accuracy of defect cause location is improved through a weighted fusion algorithm. The weighted fusion process is as follows:

[0174] (1) Weight initialization: The knowledge graph generates initial weights for each possible defect cause through node relationships and inference rules.

[0175] (2) CNN model prediction: The system uses CNN to perform pattern matching on the new abnormal description and generates a probability distribution based on the existing data to predict the most likely defect cause.

[0176] (3) Weighted fusion algorithm: Fuse the weights of the knowledge graph and CNN. Set two parameters α and β, representing the proportions of the knowledge graph and CNN in the total weight respectively. The formula for calculating the fused weighted weight is:

[0177] W final = α·W KG + β·W CNN

[0178] For example, if the weight of the knowledge graph for "cooling system defect" is 0.7 and the CNN prediction is 0.8, and α = 0.4 and β = 0.6 are set, then the final weighted weight is 0.76.

[0179] (4) Weight fine-tuning and sorting: Based on the fused weights, sort each defect cause and display the most likely defect causes to the user in descending order.

[0180] Through this weighted fusion algorithm, the logical reasoning advantage of the knowledge graph and the pattern recognition ability of CNN are effectively combined, ensuring higher diagnostic accuracy and flexibility when dealing with new abnormal situations, and further enhancing the reliability of defect cause location.

[0181] 3. Node weight integration and dynamic adjustment: According to user feedback and fault data, perform weighted fusion and dynamic adjustment on the node weights in the knowledge graph and fault tree:

[0182] (1) Feedback after fault tracing: After completing a fault tracing through the knowledge graph and reasoning model, adjust the weights based on the difference between the actual fault diagnosis result and the target fault diagnosis result;

[0183] (2) Adaptive optimization loop: After each tracing and diagnosis, dynamically adjust the node weights based on the feedback. This loop helps to gradually optimize the model to adapt to new abnormal types and fault scenarios. For example, when a certain fault mode appears for the first time, the initial weight may be low, but after dealing with this fault mode multiple times, the weight will gradually increase until accurate diagnosis is possible;

[0184] (3) Dynamically adjust node weights: Adjust the weights of each node in the knowledge graph according to user feedback and new fault cases. For example, if a certain fault cause is reported as a high-frequency or common problem, the weight of this node can be increased;

[0185] (4) Adjust the priority of feedback influence: Re-evaluate the priorities of different fault nodes according to user feedback. For example, if certain fault types show a higher repair success rate or shorter repair time, the node weights of these nodes will be enhanced accordingly;

[0186] (5) Knowledge graph optimization: As new data is input, the knowledge graph will undergo an optimization process. The optimization includes not only weight adjustment but also possible addition, deletion, or merging of nodes. New nodes or associations will be added to the graph to support new fault inference paths;

[0187] (6) Precise positioning in complex scenarios: Through continuous weight feedback and optimization, it can achieve higher positioning accuracy in complex or rare fault scenarios. Especially in complex fault situations where multiple factors act together, its recognition and diagnosis capabilities can be effectively improved through multiple feedback optimizations.

[0188] 4. Fine-tuning and optimization of inference rules: Based on newly added data and user feedback, automatically fine-tune existing inference rules or introduce new rules to improve the adaptability and accuracy of the model in dealing with new fault scenarios.

[0189] (1) Fine-tuning existing rules: Analyze user feedback and newly added fault data to identify potential blind spots or inapplicable situations in the current inference rules. For example, some rules may not provide accurate results in certain fault scenarios. Then, adjust the existing rules based on the analysis to optimize their scope of application and accuracy;

[0190] (2) Introducing new inference rules: According to the newly collected data, new inference rules may be introduced. For example, if it is found through new data that there is a stronger causal relationship between "abnormal high temperature" and "cooling system failure", then a new rule is added to capture this pattern;

[0191] (3) Testing and verification of inference rules: Newly added or optimized inference rules need to be tested and verified. For example, use historical data or simulated fault scenarios to test the effectiveness of the new rules. If the rules pass the test, they will be integrated into the existing inference process.

[0192] Through the four major steps of user feedback collection, weighted fusion of weights, integration and dynamic adjustment of node weights, and fine-tuning and optimization of inference rules, its fault positioning and inference capabilities are continuously strengthened. The design of each step complements each other, from obtaining user feedback, updating weights, optimizing the knowledge graph to improving inference rules, ensuring that the fault diagnosis method has high adaptability and high-precision diagnostic capabilities for new abnormal scenarios. In addition, the combination of the knowledge graph and the fault tree and the weighted fusion of the CNN prediction model enable the fault diagnosis to have the ability to efficiently handle complex and changing fault scenarios, forming an intelligent and precise test traceability process.

[0193] S205, Update and adaptive improvement of the knowledge base.

[0194] 1. Automatic data crawling and new data annotation: New solutions and fault cases can be regularly crawled from external data sources (such as databases, online resources, etc.) according to a preset cycle, and then undergo data cleaning, annotation, and integration to update and expand the knowledge graph.

[0195] (1) Data source crawling: New fault cases and solutions are automatically obtained from external data sources (such as device logs, maintenance manuals, technical forums, etc.) by setting up crawlers or automated interfaces. These data sources can include: historical fault records, expert forums or technical support articles, updated device documents, and user feedback data.

[0196] (2) Data cleaning: The collected external data is cleaned to remove redundant, irrelevant, or incorrect data. For example, by removing incorrectly formatted or duplicate records, the data quality is improved.

[0197] (3) New data annotation: The cleaned data needs to be annotated to indicate the relevance, fault cause, repair method, etc. of each data point. The annotation process can rely on manual annotation, rule-based automatic annotation, or automated annotation based on existing models.

[0198] (4) Knowledge graph expansion: The annotated data is integrated into the knowledge graph as new nodes or relationships. Through these new data, the graph can cover more fault modes and solutions, thus providing more support for subsequent reasoning.

[0199] 2. Model verification and self-learning calibration: To ensure the stability of the model in a new data environment, regular verification and calibration are required to ensure accurate predictions under different working conditions.

[0200] (1) Regularly verify the model: Conduct verification tests regularly to check its performance with new data. This includes using historical fault data, simulated data, or real-time data to evaluate the accuracy of the model. During the verification process, compare the model's prediction results with the actual fault results and calculate metrics such as accuracy and recall.

[0201] (2) Self-learning calibration: To handle new fault modes and data and achieve self-learning capabilities, automatically adjust its model parameters after each verification. This calibration is not limited to adjusting the weights of nodes but can also include fine-tuning of inference rules and the introduction of new inference logics. Through self-learning, experience can be accumulated from the constantly changing data to optimize the fault diagnosis ability.

[0202] (3) Ensure stability: The goal of self-learning calibration is to maintain the stability of fault diagnosis under various working conditions, enabling it to continue to operate efficiently without complete reconstruction whether it is due to equipment upgrades, environmental changes, or new fault types.

[0203] 3. Multimodal data fusion: This step integrates different types of data (such as text, images, videos, etc.) to enhance the ability of fault detection and analysis and improve the accuracy of fault location.

[0204] (1) Text data processing: Analyze the input text data such as abnormal descriptions, alarm logs, or maintenance reports through natural language processing (NLP) techniques, and extract key information (such as abnormal phenomena, alarm codes, equipment status, etc.). This information will be used as the input for the nodes in the knowledge graph to support subsequent reasoning and analysis.

[0205] (2) Image data analysis: Integrate image processing techniques, obtain image data of equipment components through cameras or image sensors, and combine computer vision (CV) algorithms to identify damages, abnormalities, or other fault signs on the equipment surface or components. For example, use image recognition to analyze whether there is physical damage, part deformation, etc.

[0206] (3) Video data integration: For some complex faults, video data can provide more abundant information. By combining video analysis, the dynamic behavior of the equipment during operation can be captured, such as vibration, temperature changes, abnormal movements, etc. These information help to further accurately locate the fault source.

[0207] (4) Fusion analysis: Conduct fusion analysis on different modal data (text, images, videos), synthesize various signals and information, and provide multi-angle support for fault location. This multimodal data fusion improves the parsing ability of the pre-set neural network for complex faults. Especially when fault information is presented in different forms in different modalities, fusion analysis can effectively integrate this information and improve the diagnostic accuracy.

[0208] 4. Optimization of the user feedback mechanism: To improve the user experience and system adaptability, continuously optimize the weights and reasoning strategies according to the user feedback on fault location and solutions.

[0209] (1) Collect user feedback: Users can evaluate the recommended fault causes and solutions, so as to understand the needs and expectations of users based on the user feedback. The feedback content includes but is not limited to whether the solution solves the problem, whether the recommended priority is reasonable, and other problems encountered by users during the operation process.

[0210] (2) Model weight adjustment: According to the user feedback, adjust the weights in the reasoning model. For example, if a certain type of fault solution is frequently rated as ineffective by users, its weight will be correspondingly reduced, or the reasoning rules will be adjusted so that this type of fault no longer appears in the priority position of the recommended list.

[0211] (3) Inference strategy optimization: Optimize the inference strategy based on the user's feedback. The user's feedback helps identify problems with the adaptability of fault diagnosis in specific scenarios and optimize the inference path and priorities to improve the practicality and accuracy of the fault diagnosis method.

[0212] (4) Improve user experience: By continuously optimizing the user feedback mechanism, it is possible to better align with user needs and provide personalized solutions. This not only improves the accuracy of fault location but also enhances the user's trust and satisfaction with the fault diagnosis method.

[0213] Through these steps, multi-faceted improvements in fault diagnosis capabilities are achieved. Automatic data crawling and new data annotation ensure the continuous update of the knowledge base, providing fresh and extensive case support; model verification and self-learning calibration ensure the stability and adaptability of the model in a new data environment; multi-modal data fusion utilizes complementary information in different data forms to enhance the accuracy of fault location; the optimized user feedback mechanism helps the system continuously adjust according to user needs, thus enhancing the practicality and reliability of the inference model. Through dynamic optimization and self-learning capabilities, it is possible to gradually enhance the accuracy and efficiency of fault detection and diagnosis in complex and changing working conditions.

[0214] The robot fault diagnosis method of this embodiment can efficiently perform test traceability in a complex and changing environment and use the self-learning mechanism to dynamically update the knowledge base and the training of the preset neural network to achieve accurate traceability and adaptive improvement.

[0215] In summary, for the robot fault diagnosis method according to the embodiments of the present application, first obtain the abnormal description of the robot, perform fault inference on the abnormal description based on the preset fault tree and the preset inference rules to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and the initial weight of each fault cause, and perform pattern recognition on the abnormal description based on the preset neural network model to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and the predicted weight of each fault cause, and then perform weighted fusion on the initial weight and the predicted weight to obtain the comprehensive weight of each fault cause, so as to determine the target fault diagnosis result according to the comprehensive weight of each fault cause. Thus, this method combines the first fault diagnosis result obtained by preset fault tree reasoning and the second fault diagnosis result predicted by the preset neural network to deduce possible fault causes and the comprehensive weight of each fault cause, greatly improving the accuracy of fault diagnosis and facilitating the rapid location of the fault cause.

[0216] Corresponding to the above embodiments, the present application also proposes an electronic device.

[0217] As Figure 4As shown in the figure, the electronic device 100 according to the embodiment of the present application includes a memory 110, a processor 120, and a robot fault diagnosis program stored on the memory 110 and operable on the processor 120. When the processor 120 executes the robot fault diagnosis program, the above-mentioned robot fault diagnosis method is implemented.

[0218] According to the electronic device of the embodiment of the present application, when the processor executes the robot fault diagnosis program, the above-mentioned robot fault diagnosis method is implemented. Based on the above robot fault diagnosis method, the accuracy of fault diagnosis is greatly improved, which is conducive to quickly locating the cause of the fault.

[0219] Corresponding to the above embodiment, the present application also proposes a computer-readable storage medium.

[0220] The computer-readable storage medium according to the embodiment of the present application stores a robot fault diagnosis program thereon. When the robot fault diagnosis program is executed by a processor, the above-mentioned robot fault diagnosis method is implemented.

[0221] According to the computer-readable storage medium of the embodiment of the present application, when the robot fault diagnosis program stored thereon is executed by a processor, the above-mentioned robot fault diagnosis method is implemented. Based on the above robot fault diagnosis method, the accuracy of fault diagnosis is greatly improved, which is conducive to quickly locating the cause of the fault.

[0222] Corresponding to the above embodiment, the present application also proposes a robot fault diagnosis device.

[0223] As Figure 5 shown, the robot fault diagnosis device according to the embodiment of the present application includes: an acquisition module 10, a first fault diagnosis module 20, a second fault diagnosis module 30, and a weight fusion module 40.

[0224] Among them, the acquisition module 10 is used to acquire the abnormal description of the robot. The first fault diagnosis module 20 is used to perform fault reasoning on the abnormal description based on a preset fault tree and a preset inference rule to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and an initial weight corresponding to each fault cause. The second fault diagnosis module 30 is used to perform pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and a predicted weight corresponding to each fault cause. The weight fusion module 40 is used to perform weighted fusion on the initial weight and the predicted weight to obtain a comprehensive weight corresponding to each fault cause, so as to determine a target fault diagnosis result according to the comprehensive weight of each fault cause.

[0225] According to an embodiment of the present application, the first fault diagnosis module 20 performs fault reasoning on the abnormal description based on a preset fault tree structure and preset inference rules to obtain a first fault diagnosis result, and specifically is used for: extracting information from the abnormal description based on NLP technology to obtain abnormal information; determining abnormal nodes in the preset fault tree based on the abnormal information; combining the node relationships in the preset fault tree and the preset inference rules to determine the root cause nodes corresponding to the abnormal nodes; determining the fault causes corresponding to the abnormal description based on the root cause nodes, and determining the initial weights corresponding to each fault cause based on the preset weights between each root cause node and the abnormal node, so as to obtain a first fault diagnosis result.

[0226] According to an embodiment of the present application, the second fault diagnosis module 30 performs pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result, and specifically is used for: extracting features from the abnormal description to obtain abnormal features; performing pattern recognition on the abnormal features based on the preset neural network model to obtain the fault causes and the probability values of each fault cause, and using the probability values as prediction weights to obtain a second fault diagnosis result.

[0227] According to an embodiment of the present application, the weight fusion module 40 performs weighted fusion on the initial weights and the prediction weights to obtain the comprehensive weights of each fault cause, and specifically is used for: obtaining a first weight based on the product of the initial weight and a first preset ratio; obtaining a second weight based on the product of the prediction weight and a second preset ratio; obtaining a comprehensive weight based on the sum of the first weight and the second weight.

[0228] According to an embodiment of the present application, the robot fault diagnosis method further includes a control module, and the control module is used for: performing abnormal processing on the robot based on the target fault diagnosis result and determining the actual fault cause; adjusting the structure and / or weights of the preset fault tree based on the actual fault cause, and training the preset neural network model.

[0229] According to an embodiment of the present application, the control module adjusts the structure and / or weights of the preset fault tree based on the actual fault cause, and specifically is used for: when the target fault diagnosis result includes the actual fault cause, adjusting the preset weight between the node corresponding to the actual fault cause and the abnormal node; when the target fault diagnosis result does not include the actual fault cause, adding a node corresponding to the actual fault cause in the preset fault tree, and setting the preset weight between the node corresponding to the actual fault cause and the abnormal node based on a preset rule.

[0230] According to an embodiment of the present application, the control module is further configured to: obtain historical abnormal data of the robot, where the historical abnormal data includes abnormal phenomena and the fault causes of each abnormal phenomenon; perform NLP cleaning and standardization processing on the historical abnormal data to obtain historical abnormal data that meets a preset format; perform information extraction and annotation on the historical abnormal data that meets the preset format based on a convolutional neural network model to obtain a first short sentence, where the first short sentence includes an abnormal component and an abnormal phenomenon; perform machine type association on the first short sentence according to the type and attributes of the abnormal component to obtain a second short sentence; perform operation association on the first short sentence based on the operation information corresponding to the first short sentence to obtain a third short sentence; construct fault tree nodes, the association relationships between each fault tree node, and determine the initial weight corresponding to each node in combination with the historical abnormal data according to the first short sentence, the second short sentence, and the third short sentence to obtain a preset fault tree.

[0231] It should be noted that for the details not disclosed in the robot fault diagnosis device in the embodiment of the present application, please refer to the details disclosed in the robot fault diagnosis method in the above embodiment of the present application, and will not be elaborated here specifically.

[0232] For the robot fault diagnosis device according to the embodiment of the present application, an abnormal description is obtained through an acquisition module, and a first fault diagnosis result is obtained through a first fault diagnosis module based on a preset fault tree and a preset inference rule for fault inference on the abnormal description, where the first fault diagnosis result includes at least one fault cause and the initial weight corresponding to each fault cause. A second fault diagnosis result is obtained through a second fault diagnosis module based on a preset neural network model for pattern recognition of the abnormal description, where the second fault diagnosis result includes at least one fault cause and the predicted weight corresponding to each fault cause. A weight fusion module performs weighted fusion on the initial weight and the predicted weight to obtain the comprehensive weight corresponding to each fault cause, so as to determine the target fault diagnosis result according to the comprehensive weight of each fault cause. Thus, the device combines the first fault diagnosis result obtained by reasoning with the preset fault tree and the second fault diagnosis result predicted by the preset neural network to deduce possible fault causes and the comprehensive weight of each fault cause, greatly improving the accuracy of fault diagnosis and facilitating the rapid positioning of fault causes.

[0233] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion having one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing as appropriate, and then stored in a computer memory.

[0234] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or combinations thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), and the like.

[0235] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0236] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0237] In this application, unless otherwise clearly defined and limited, terms such as "installed", "connected", "joined", "fixed", etc. shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the internal connection of two components or the interaction relationship between two components, unless otherwise clearly limited. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0238] Although the embodiments of this application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting this application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A robot fault diagnosis method, characterized in that, The method includes: Obtaining an abnormal description of the robot; Performing fault reasoning on the abnormal description based on a preset fault tree and preset inference rules to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and an initial weight of each fault cause; Performing pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and a predicted weight of each fault cause; Performing weighted fusion on the initial weight and the predicted weight to obtain a comprehensive weight of each fault cause, and determining a target fault diagnosis result according to the comprehensive weight of each fault cause.

2. The robot fault diagnosis method according to claim 1, wherein, The performing fault reasoning on the abnormal description based on a preset fault tree structure and preset inference rules to obtain a first fault diagnosis result includes: Performing information extraction on the abnormal description based on NLP technology to obtain abnormal information; Determining an abnormal node in the preset fault tree based on the abnormal information; Combining the node relationship in the preset fault tree and the preset inference rules to determine a root cause node corresponding to the abnormal node; Determining a fault cause corresponding to the abnormal description based on the root cause node, and determining an initial weight corresponding to each fault cause based on a preset weight between each root cause node and the abnormal node, so as to obtain the first fault diagnosis result.

3. The robot fault diagnosis method according to claim 1, wherein The performing pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result includes: Performing feature extraction on the abnormal description to obtain abnormal features; Performing pattern recognition on the abnormal features based on the preset neural network model to obtain a fault cause and a probability value of each fault cause, and using the probability value as the predicted weight to obtain the second fault diagnosis result.

4. The robot fault diagnosis method according to claim 1, wherein The performing weighted fusion on the initial weight and the predicted weight to obtain a comprehensive weight of each fault cause includes: Obtaining a first weight based on the product of the initial weight and a first preset ratio; Obtaining a second weight based on the product of the predicted weight and a second preset ratio; Obtaining the comprehensive weight based on the sum of the first weight and the second weight.

5. The robot fault diagnosis method according to claim 1, wherein The method further includes: Performing abnormal processing on the robot based on the target fault diagnosis result and determining an actual fault cause; Adjusting the structure and / or weight of the preset fault tree based on the actual fault cause, and training the preset neural network model.

6. The robot fault diagnosis method according to claim 5, wherein The adjusting the structure and / or weight of the preset fault tree based on the actual fault cause includes: When the target fault diagnosis result includes the actual fault cause, adjusting the preset weight between the node corresponding to the actual fault cause and the abnormal node; When the target fault diagnosis result does not include the actual fault cause, adding a node corresponding to the actual fault cause in the preset fault tree, and setting the preset weight between the node corresponding to the actual fault cause and the abnormal node based on a preset rule.

7. The robot fault diagnosis method according to claim 1, wherein, The method further includes: Obtain the historical abnormal data of the robot, where the historical abnormal data includes abnormal phenomena and the fault causes of each abnormal phenomenon; Perform NLP cleaning and standardization processing on the historical abnormal data to obtain historical abnormal data that meets the preset format; Based on the convolutional neural network model, extract and annotate the information from the historical abnormal data that meets the preset format to obtain the first short sentence, where the first short sentence includes abnormal components and abnormal phenomena; Perform machine type association on the first short sentence according to the types and attributes of the abnormal components to obtain the second short sentence; Perform operation association on the first short sentence based on the operation information corresponding to the first short sentence to obtain the third short sentence; Construct the fault tree nodes, the association relationships between each fault tree node, and determine the initial weight corresponding to each node in combination with the historical abnormal data according to the first short sentence, the second short sentence, and the third short sentence to obtain the preset fault tree.

8. An electronic device, characterized in that, It includes a memory, a processor, and a robot fault diagnosis program stored on the memory and executable on the processor. When the processor executes the robot fault diagnosis program, it implements the robot fault diagnosis method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, A robot fault diagnosis program is stored thereon, and when the robot fault diagnosis program is executed by the processor, it implements the robot fault diagnosis method according to any one of claims 1-7.

10. A robot fault diagnosis device, characterized in that, The device includes: An acquisition module, configured to acquire the abnormal description of the robot; A first fault diagnosis module, configured to perform fault reasoning on the abnormal description based on a preset fault tree and preset inference rules to obtain a first fault diagnosis result, where the first fault diagnosis result includes at least one fault cause and the initial weight of each fault cause; A second fault diagnosis module, configured to perform pattern recognition on the abnormal description based on a preset neural network model to obtain a second fault diagnosis result, where the second fault diagnosis result includes at least one fault cause and the predicted weight of each fault cause; A weight fusion module, configured to perform weighted fusion on the initial weight and the predicted weight to obtain the comprehensive weight of each fault cause, so as to determine the target fault diagnosis result according to the comprehensive weight of each fault cause.

Images