Fault detection method, system and device based on dynamic heterogeneous network representation learning
Based on the method of learning based on dynamic heterogeneous network representation, build a dynamic heterogeneous network of heavy machines, use random walking rules and Word2VEC model to generate vector vectors, solve the problem of the accuracy and real -time insufficient in the existing technologies, , Achieve more efficient fault recognition effects.
Patent Information
- Application Number
- CN202411257066.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-09
- Publication Date
- 2025-01-28
AI Technical Summary
The prior art is difficult to effectively process multi-source heterogeneous data generated by cranes during operation, resulting in insufficient accuracy and real-time accuracy of fault detection, and it is difficult to capture the characteristics of equipment changing over time.
Using the fault detection method based on dynamic heterogeneous network characterization learning, by constructing a dynamic heterogeneous network, selecting the time, place and operator information of the device as a node, use the random walking rules and the Word2VEC model to generate the vector of the node, and pass the pass through the pass, and pass the pass through the through. The weighted fusion and the second classification logic regression algorithm achieve fault detection.
The accuracy and real -time nature of the crane failure detection can effectively capture the characteristics of the equipment over time, and improve the effect of fault recognition.
Smart Images
Figure CN119379240A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of crane fault detection, and in particular relates to a fault detection method, system and device based on dynamic heterogeneous network representation learning. Background Art
[0002] As the core equipment in industrial production, the stability and safety of cranes directly affect production efficiency and operator safety. Due to the complex and changeable working environment, the uneven skill levels of operators, and long-term high-load operation, cranes are prone to various faults, such as damage to limiters, fatigue of metal structures caused by oblique pulling, and wear of transmission parts. If these faults cannot be detected and diagnosed in a timely and accurate manner, they will not only affect the normal operation of the equipment, but may also cause safety accidents and lead to serious economic losses. Therefore, the development of an efficient fault detection method is crucial to prevent accidents, reduce maintenance costs, and ensure production continuity.
[0003] In the daily operation of cranes, a large amount of multi-source heterogeneous data is generated, such as physical parameters monitored by sensors, operation logs, maintenance records, etc. These data include both structured data and unstructured data. In the prior art, machine learning algorithms are widely used in fault detection, but they are mainly aimed at numerical data and are difficult to be directly applied to crane data analysis containing structured and unstructured data. In addition, as a heavy equipment that works continuously, the state of the crane will change over time. Existing static analysis methods are difficult to capture these changes, resulting in insufficient accuracy and real-time performance of fault detection.
[0004] Although traditional fault detection methods can process certain numerical data and implement basic fault detection, they perform poorly when processing complex multi-source heterogeneous data. It is difficult to integrate data from different sources to form a global view, resulting in low accuracy and real-time performance of fault diagnosis. At the same time, traditional methods are usually static and difficult to capture the characteristics of equipment changing over time.
[0005] In order to solve these problems, the present invention proposes a crane fault detection method based on dynamic heterogeneous network representation learning. Summary of the invention
[0006] The purpose of the present invention is to provide a fault detection method, system and device based on dynamic heterogeneous network representation learning to solve at least one or more technical problems in the background technology.
[0007] To solve the above technical problems, the present invention is achieved through the following technical solutions:
[0008] A fault detection method based on dynamic heterogeneous network representation learning, the method comprising:
[0009] According to common equipment failures, the time, location and operator information of equipment operation are selected as nodes for building a dynamic heterogeneous network;
[0010] Represent the nodes as vectors, perform random walks on the nodes from multiple angles, and generate random walk paths;
[0011] Update parameters based on node frequency to optimize random walk path selection;
[0012] According to the frequency of the nodes, the vectors of the nodes are fused in a weighted manner to generate an overall vector representation.
[0013] Further, the method comprises:
[0014] Select node data with high relevance and frequency from the new sampled data and historical database;
[0015] Generate a large number of random walk paths using these nodes and random walk rules;
[0016] Apply random walk rules to different nodes to ensure balanced use of each rule;
[0017] The generated random walk path is used as the training data of word2vec, and the skip-gram model is used to train the vector representation of the node;
[0018] The node vector representation is obtained by maximizing the heterogeneous context conditional probability of the node.
[0019] The following method may also be adopted: a fault detection system based on dynamic heterogeneous network representation learning, the system comprising:
[0020] The data collection module is used to collect the time, location and operator information of the equipment operation and use it as a node to build a dynamic heterogeneous network;
[0021] The node representation module is used to represent nodes in vector form and perform random walks on the nodes from multiple angles to generate random walk paths;
[0022] The path optimization module is used to update parameters based on node frequency and optimize the selection of random walk paths;
[0023] The vector fusion module is used to fuse the vectors of nodes in a weighted manner according to the frequency of the nodes to generate an overall vector representation;
[0024] The fault detection module is used to detect device faults based on the overall vector representation through a binary logistic regression algorithm.
[0025] The following method may also be adopted: a fault detection device based on dynamic heterogeneous network representation learning, the device comprising:
[0026] A processor is used to perform the following operations:
[0027] Represent the nodes in vector form, and perform random walks on the nodes from multiple angles to generate random walk paths;
[0028] Update parameters based on node frequency to optimize the selection of random walk paths;
[0029] According to the frequency of the nodes, the vectors of the nodes are fused in a weighted manner to generate the overall vector representation;
[0030] Based on the overall vector representation, the device fault detection is performed through the binary logistic regression algorithm.
[0031] Beneficial effects:
[0032] The present invention innovatively designs a dynamic node selection method, which avoids the situation of local random walk by selecting the next random walk node according to the frequency and time of node occurrence, and improves the quality of training data;
[0033] The present invention selects multiple relevant variables for common crane faults, lays a foundation for the design of random walk rules, and helps to comprehensively consider common fault types;
[0034] The present invention creatively designs multiple random walk rules to ensure that each rule is used evenly, effectively improving the effect of fault detection;
[0035] The present invention solves the problems of multi-source heterogeneity and real-time change of data during the operation of crane equipment from the two aspects of dynamics and heterogeneity, and greatly improves the effect of fault identification.
[0036] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0038] Figure 1 The present invention is a method flow chart of an embodiment of the present invention. DETAILED DESCRIPTION
[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0040] A fault detection method based on dynamic heterogeneous network representation learning includes the following steps;
[0041] Step 1: Data Collection
[0042] Collect the operation data of the crane, mainly including sensor measurement data, technician operation log, maintenance record, fault history data, etc. Divide all the data into two parts, historical data set X and test data set Y. Store the data labels in a separate file to verify the effect of the crane fault detection method proposed in this patent.
[0043] Step 2: Node selection
[0044] Common crane failures mainly involve electrical systems, mechanical structures, control systems, and other aspects. Electrical system failures include unstable power supply, motor overheating, brake system electrical control failure, etc.; mechanical structure failures include hook damage, gear wear, etc.; control system failures include computer control system software errors, position sensor, load sensor failure, and safety limit device failure. In response to the above-mentioned failures, nodes for constructing a dynamic heterogeneous graph network are selected from three perspectives: the time, location, and people of the crane equipment operation. According to the fault type and node attributes, this patent uses variables such as lifting grade, lifting height, operation time, ambient temperature, operating speed, operator, and maintenance time of major components to design random walk rules.
[0045] Step 3: Random walk rule design
[0046] The method proposed in this patent is the representation learning of dynamic heterogeneous networks, that is, each node collected by the crane is represented as a vector form, mainly relying on the word2vec method. The skip-gram method is used to train the model to obtain the vector representation of the node. In this patent, the node random walk route is used as the input of word2vec, and a graph is drawn for all nodes. The random walk starts from a node and walks to the neighboring nodes of this node with a certain probability. Finally, the random walk route is obtained according to the pre-set hyperparameters.
[0047] For example, if the power supply is unstable, the random walk rule is designed as lifting height-lifting speed-operating time-lifting speed-lifting height, and the power supply condition of the equipment is judged by the height and speed of the crane operation;
[0048] Gear wear, the random walk rule is designed as lifting level-lifting height-ambient temperature-maintenance time difference-ambient temperature-lifting height-lifting level, and the power supply condition of the equipment is judged by the lifting weight, height, ambient operating temperature, and parts replacement time;
[0049] The safety limit device fails, and the random walk rule is designed as lifting height-operator-operating time-ambient temperature-operating time-operator-lifting height.
[0050] Use metapath to update parameters and select the random walk rule for the current node to walk next, trying to consider all situations as completely as possible. The walk formula is as follows:
[0051]
[0052] Among them, Γ is the i ,node vi The random walk rule is selected, and α is a parameter that decreases with the number of times the rule is selected.
[0053] Step 4: Dynamic node acquisition
[0054] The node participation frequency is used to update the parameters. Based on the random walk rule, the position of the next walk node is selected. On the basis of taking into account the latest data, the relationship in the historical data is considered. If a node appears frequently but the subsequent nodes related to it have a low frequency, repeated walks may occur. The specific calculation formula is as follows.
[0055]
[0056] Among them, w(v i+1 ) is the frequency of appearing in the random walk path.
[0057] Step 5: Obtain training text
[0058] Step 4 selects the crane collection data that is relevant to the new sampled data and has a higher frequency in the historical database. Next, the random walk rule is used to obtain the random path based on the selected nodes. According to the two hyperparameters, the number of starting paths for each node and the length of each random walk path, a large number of random walk paths will be obtained after multiple samplings. These paths can be regarded as training texts in word2vec, and then the vector representation of each node can be obtained. In this patent, according to the different designs of random walk rules, for different nodes, the corresponding random walk rules are selected on average for walking, which can effectively capture the semantic relevance and structural relevance in the operation process of the crane equipment and improve the fault detection effect of the equipment.
[0059] Step 6: Vector acquisition
[0060] Based on the random walk route obtained above as the training text of word2vec, the skip-gram model is used to obtain the vector representation of each node. The main goal of the solution is to maximize the conditional probability of heterogeneous context under the condition of a given node v, so as to learn the vector representation of the node.
[0061]
[0062] Where N t (v) refers to the t-th type of neighbor node of node v, p(c t |v;θ) refers to the softmax function, and the calculation formula is shown in (5).
[0063]
[0064] Where W v is the vector representation of node v.
[0065] The negative sampling method is also used to effectively optimize the objective function, reduce the computational complexity, and use a binary classification model instead of the softmax function.
[0066] Step 7: Towards Fusion
[0067] The frequency of each node is used as the weight matrix to fuse the vector representation of the node. The calculation formula is shown in (1).
[0068] Step 8: Crane Fault Detection
[0069] According to the obtained node vector representation, step seven uses the node frequency as a weight to integrate each sample, obtains a vector representation of the entire crane collection data, and then uses a binary classification logistic regression algorithm to realize the fault detection of the crane equipment.
[0070] In some specific implementations, the input data includes: historical crane data X, test auto insurance data Y, multiple random walk rules Γ, the number of each node as a starting node n, the length l of the random walk sequence, the embedding dimension d, the number of iterations t, and the window size m;
[0071] Output data: vector representation V of test data, test crane data label label.
[0072] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does 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.
[0073] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. A fault detection method based on dynamic heterogeneous network representation learning, characterized in that: The method comprises: According to common equipment failures, the time, location and operator information of equipment operation are selected as nodes for building a dynamic heterogeneous network; Represent the nodes as vectors, perform random walks on the nodes from multiple angles, and generate random walk paths; Update parameters based on node frequency to optimize random walk path selection; According to the frequency of the nodes, the vectors of the nodes are fused in a weighted manner to generate an overall vector representation.
2. A fault detection method based on dynamic heterogeneous network representation learning according to claim 1, characterized in that: The method comprises: Select node data with high correlation and frequency from the resampled data and historical database; Generate a large number of random walk paths using these nodes and random walk rules; Apply random walk rules to different nodes to ensure balanced use of each rule; The generated random walk path is used as the training data of word2vec, and the skip-gram model is used to train the vector representation of the node; The node vector representation is obtained by maximizing the heterogeneous context conditional probability of the node.
3. A fault detection method based on dynamic heterogeneous network representation learning according to claim 1, characterized in that: The frequency of the node is the number of occurrences of each node in the random walk path.
4. A fault detection method based on dynamic heterogeneous network representation learning according to claim 1, characterized in that: The overall vector representation is obtained by weighted summing of the vector representations of the nodes, where the weights are determined by the frequencies of the nodes.
5. A fault detection method based on dynamic heterogeneous network representation learning according to claim 1, characterized in that: The fault detection step includes detecting device faults through a binary logistic regression algorithm based on the overall vector representation.
6. A fault detection method based on dynamic heterogeneous network representation learning according to claim 1, characterized in that: During the random walk, metapath is used to update parameters and a random walk rule for the current node to walk next is selected.
7. A fault detection method based on dynamic heterogeneous network representation learning according to claim 1, characterized in that: The equipment involved in the method is a crane.
8. A fault detection method based on dynamic heterogeneous network representation learning according to claim 7, characterized in that: The random walk rules of the nodes include the following: The power supply is unstable. The random walk rule is designed as lifting height-lifting speed-operating time-lifting speed-lifting height. The power supply condition of the equipment is judged by the height and speed of the crane. Gear wear, the random walk rule is designed as lifting level-lifting height-ambient temperature-maintenance time difference-ambient temperature-lifting height-lifting level, and the power supply condition of the equipment is judged by the lifting weight, height, ambient operating temperature, and parts replacement time; The safety limit device fails, and the random walk rule is designed as lifting height-operator-operating time-ambient temperature-operating time-operator-lifting height.
9. A fault detection system based on dynamic heterogeneous network representation learning, characterized in that: The system comprises: The data collection module is used to collect the time, location and operator information of the equipment operation and use it as a node to build a dynamic heterogeneous network; The node representation module is used to represent nodes in vector form and perform random walks on the nodes from multiple angles to generate random walk paths; The path optimization module is used to update parameters based on node frequency and optimize the selection of random walk paths; The vector fusion module is used to fuse the vectors of nodes in a weighted manner according to the frequency of the nodes to generate an overall vector representation; The fault detection module is used to detect device faults based on the overall vector representation through a binary logistic regression algorithm.
10. A fault detection device based on dynamic heterogeneous network representation learning, characterized in that: The device comprises: A processor is used to perform the following operations: Represent the nodes in vector form, and perform random walks on the nodes from multiple angles to generate random walk paths; Update parameters based on node frequency to optimize the selection of random walk paths; According to the frequency of the nodes, the vectors of the nodes are fused in a weighted manner to generate the overall vector representation; Based on the overall vector representation, the device fault detection is performed through the binary logistic regression algorithm.