Industrial equipment fault diagnosis method and equipment based on machine learning, and medium

Abstract

The invention discloses an industrial equipment fault diagnosis method and equipment based on machine learning and a medium, and the method comprises the steps: collecting operation data of target equipment, and carrying out the preprocessing of the operation data, so as to obtain monitoring data; a pre-trained fault diagnosis model is determined, the fault diagnosis model is an improved graph neural network model, and an input layer of the improved graph neural network model adopts a fuzzy reasoning mechanism; and inputting the monitoring data into the fault diagnosis model to determine the equipment state and the fault type of the target equipment. By using the graph neural network, the topological structure relation between samples can be fully mined and utilized, the method is particularly suitable for processing complex data from a multi-source sensor, interdependence between nodes can be captured, and therefore the diagnosis accuracy is improved. A fuzzy reasoning mechanism is adopted in an input layer of the graph neural network and used for processing uncertainty in data, adjustment can be carried out according to fuzziness of the data, and the capacity of processing uncertain data is improved.

Description

Technical Field

[0001] The present application relates to the field of machine learning, and specifically to a method, equipment, and medium for industrial equipment fault diagnosis based on machine learning. Background Art

[0002] In recent years, with the widespread application of industrial equipment, equipment fault diagnosis has become an important part of ensuring production safety, improving production efficiency and reducing maintenance costs.

[0003] Traditional equipment fault diagnosis methods often rely on manual experience or rule-based diagnostic models. These methods often suffer from low diagnostic accuracy, slow response, and poor adaptability to data patterns when faced with complex, nonlinear, and multivariate equipment data. Especially when processing large amounts of real-time sensor data, traditional methods are unable to fully exploit the correlations between individual equipment sensors and the uncertainty inherent in the data, resulting in limited fault detection effectiveness. Summary of the Invention

[0004] To solve the above problems, this application proposes a method, device, and medium for industrial equipment fault diagnosis based on machine learning, wherein the method includes:

[0005] Collect the operating data of the target device and preprocess the operating data to obtain monitoring data; determine a pre-trained fault diagnosis model, which is an improved graph neural network model, and the input layer of the improved graph neural network model adopts a fuzzy inference mechanism; input the monitoring data into the fault diagnosis model to determine the device status and fault type of the target device.

[0006] In one example, the operation data of the target device is collected and preprocessed to obtain monitoring data, specifically including: storing the operation data in JSON format; cleaning the operation data to obtain first intermediate data; normalizing the first intermediate data to obtain second intermediate data; and discretizing features of the second intermediate data to obtain the monitoring data.

[0007] In one example, before determining the pre-trained fault diagnosis model, the method further includes: obtaining historical operating data of the target device and annotating the pre-processed historical operating data, where the annotation content includes at least the device status and the fault type; inputting the annotated historical operating data into the initial fault diagnosis model, and training the initial fault diagnosis model by setting a dynamic damping coefficient to obtain the fault diagnosis model; determining the device information corresponding to the target device, and storing the fault diagnosis model based on the device information; the device information includes at least one of the device name, device type, device location, and device parameters.

[0008] In one example, the initial fault diagnosis model is trained by setting a dynamic damping coefficient, specifically including: determining the initial damping coefficient, the decay rate, the current iteration round of model training, and the loss change rate corresponding to the current iteration round; based on the initial damping coefficient, the decay rate, the current iteration round of model training, and the loss change rate corresponding to the current iteration round, determining the damping coefficient corresponding to the current iteration round.

[0009] In one example, the monitoring data is input into the fault diagnosis model to determine the device status and fault type of the target device, specifically including: determining the target data with a fuzzy label corresponding to the target node in the input layer of the fault diagnosis model; determining the fuzzy confidence factor corresponding to the target data based on the characteristic distribution discreteness of the target data; determining the input feature corresponding to the target node based on a preset fuzzy inference function, the original fuzzy feature of the target node, and the fuzzy confidence factor; and determining the device status and fault type of the target device based on the input feature.

[0010] In one example, the monitoring data is input into the fault diagnosis model to determine the device status and fault type of the target device, specifically including: determining the target node within the target layer, and the adjacent nodes of the target node; obtaining the first initial feature representation of the target node, the second initial feature representation of the adjacent node, and the adjacency matrix between the target node and the adjacent node; based on the adjacency matrix, fusing the first initial feature representation and the second initial feature representation as the fused feature representation of the target node.

[0011] In one example, after determining the device status and fault type of the target device, the method further includes: when the device status of the target device is a fault state, obtaining the fault type of the target device; based on the fault type, triggering a corresponding instruction of the target device, the corresponding instruction including at least one of a maintenance instruction or a shutdown instruction.

[0012] In one example, each piece of operating data includes at least a timestamp, a device number, operating data at the collection location, a sensor type, and a sensor measurement value; the nodes in the fault diagnosis model are used to represent data features corresponding to a piece of operating data of the target device, and the edges connecting the nodes represent the similarities between the pieces of operating data, and the data features corresponding to the nodes are updated through a graph convolution layer.

[0013] The present application also provides an industrial equipment fault diagnosis device based on machine learning, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the following: collecting operating data of the target device and preprocessing the operating data to obtain monitoring data; determining a pre-trained fault diagnosis model, wherein the fault diagnosis model is an improved graph neural network model, and the input layer of the improved graph neural network model adopts a fuzzy inference mechanism; and inputting the monitoring data into the fault diagnosis model to determine the device status and fault type of the target device.

[0014] The present application also provides a non-volatile computer storage medium storing computer-executable instructions, characterized in that the computer-executable instructions are configured to: collect operating data of a target device and pre-process the operating data to obtain monitoring data; determine a pre-trained fault diagnosis model, wherein the fault diagnosis model is an improved graph neural network model, and the input layer of the improved graph neural network model adopts a fuzzy reasoning mechanism; and input the monitoring data into the fault diagnosis model to determine the device status and fault type of the target device.

[0015] The method proposed in this application can bring the following beneficial effects:

[0016] 1. Compared with traditional machine learning methods that are usually based on independent samples, this application can fully explore and utilize the topological structure relationship between samples by using graph neural networks. It is particularly suitable for processing complex data from multi-source sensors and can capture the interdependence between nodes, thereby improving the accuracy of diagnosis.

[0017] 2. Fuzzy inference mechanisms are used at the input layer of the graph neural network to handle uncertainty in the data. Traditional machine learning methods typically assume that data is deterministic, while real-world industrial data often exhibits ambiguity and uncertainty. Fuzzy inference mechanisms can adjust based on the ambiguity of the data, improving the ability to handle uncertain data.

[0018] 3. By setting up a mechanism for dynamically adjusting the damping coefficient, the model can automatically adjust its learning strategy according to the needs of different stages during training, avoiding overfitting and improving training efficiency. This mechanism, unlike static training methods, enables the model to focus on learning the graph structure in the early stages and gradually enhance regularization in the later stages to avoid overfitting.

[0019] 4. A multi-level feature fusion module is used to fuse local features with global features through graph convolution operations, so that the model can capture multi-dimensional information of device status at different levels at the same time, improving the model's understanding and classification capabilities of data diversity. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0021] Figure 1 Schematic diagram of a process flow of an industrial equipment fault diagnosis method based on machine learning in an embodiment of the present application;

[0022] Figure 2 This is a structural diagram of an industrial equipment fault diagnosis device based on machine learning in an embodiment of the present application. DETAILED DESCRIPTION

[0023] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0024] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.

[0025] Figure 1 This is a flowchart of a machine learning-based industrial equipment fault diagnosis method provided in one or more embodiments of this specification. This method can be applied to different types of industrial equipment. The process can be executed by computing devices in the corresponding field (such as cloud servers, risk control servers, or smart mobile terminals used to control industrial equipment). Certain input parameters or intermediate results in the process can be manually adjusted to help improve accuracy.

[0026] The analysis method involved in the embodiments of the present application can be implemented by a terminal device or a server, and the present application does not impose any special restrictions on this. For ease of understanding and description, the following embodiments are described in detail using a server as an example.

[0027] It should be noted that the server can be a single device or a system composed of multiple devices, that is, a distributed server, and this application does not make any specific restrictions on this.

[0028] like Figure 1 As shown, the embodiment of the present application provides an industrial equipment fault diagnosis method based on machine learning, including:

[0029] S101: collecting operating data of a target device and preprocessing the operating data to obtain monitoring data.

[0030] The operating data of the target device is collected through sensors and monitoring equipment. The data is collected by direct measurement or indirect collection. Specifically, relevant physical parameters can be collected in real time by configuring specific sensors, such as temperature sensors, pressure sensors, acceleration sensors, etc. Sensors can be installed in devices, objects or environments, and transmit data to the data collection center through wireless or wired communication. Among them, the data storage format of the operating data is JSON format, which is convenient for subsequent processing and analysis. Specifically, each piece of data contains multiple fields, mainly including timestamp, device ID, collection location, sensor type, sensor measurement value (such as temperature, pressure, etc.) and other additional information. The other additional information here refers to optional types of data such as measurement points, equipment model, equipment operation years, etc.

[0031] After obtaining the operating data, it is necessary to preprocess the operating data to obtain monitoring data that can be input into the model.

[0032] In one embodiment, during preprocessing, the operational data needs to be stored in JSON format; the operational data is then cleaned to obtain first intermediate data; the first intermediate data is normalized to obtain second intermediate data; and finally, the second intermediate data is feature discretized to obtain monitoring data. Both the first and second intermediate data are intermediate data. To facilitate distinguishing the preprocessing steps, the terms "first" and "second" are used to distinguish them.

[0033] Specifically, during data cleaning, since the raw data may contain some noise and outliers, the goal of data cleaning is to ensure data accuracy and consistency by removing outliers, filling missing values, and correcting format errors. For each piece of collected data, an outlier detection mechanism based on statistical methods, such as the standard deviation method, is used to exclude data points that fall outside the normal range. Missing values are filled through interpolation.

[0034] During data normalization, all collected numerical data is normalized to meet the training requirements of machine learning models. Because individual sensors may use different measurement units, to avoid the negative impact of inconsistent data scales, Min-Max normalization or Z-score normalization is used to map all measured data to a unified scale. For example, mapping data to the [0, 1] interval ensures that different types of data have equal influence.

[0035] Since the collected data is generally continuous numerical data, it is necessary to discretize this data to meet the requirements of the classification task. A discretization method based on the K-means clustering algorithm can be used. First, the data is grouped according to specific characteristics (such as time, spatial location, measurement value, etc.), and the data values in each group are replaced with the group's mean or median to obtain discretized features. Ultimately, after the above processing steps, the data will have the following discretization properties:

[0036] Ra represents the timestamp (normalized value), Da represents the sensor ID (discretized category), Ea represents the device state (discretized state category), Fa represents the measurement location (such as the area or component number where the sensor is located, discretized category), Ga represents the temperature value (discretized value), Ha represents the pressure value (discretized value), Ia represents the vibration value (discretized value), Ja represents the humidity value (discretized value), Ka represents the device load (discretized value), and La represents the operating frequency (discretized value). It should be noted that this embodiment is only for illustrating a data format and type of the present invention. In actual applications, the attributes of the data are usually more than 10 attributes, and the number of attributes of the data may reach dozens or even hundreds.

[0037] S102: Determine a pre-trained fault diagnosis model.

[0038] After obtaining the monitoring data, it is necessary to determine the fault diagnosis model corresponding to the target device. The fault diagnosis model here is an improved graph neural network model, and the input layer of the graph neural network model adopts a fuzzy reasoning mechanism.

[0039] In one embodiment, different types of industrial equipment correspond to different fault diagnosis model parameters. To train the fault diagnosis model parameters, the target device's historical operating data must first be acquired and annotated. The preprocessed historical operating data must include at least the device status and fault type. The device status can include normal or faulty conditions, and the fault type can include a blown fuse, a short circuit, or other faults. The annotated historical operating data is then input into an initial fault diagnosis model, which is then trained by setting a dynamic damping coefficient to obtain a fault diagnosis model. After training, the fault diagnosis model can be stored. When a fault is diagnosed again on the target device, the pre-stored model can be directly retrieved and used, avoiding the need for retraining the model. Device information corresponding to the target device must be determined, and the fault diagnosis model is stored based on this device information. The device information includes at least one of the following: device name, device type, device location, and device parameters.

[0040] Furthermore, when training the initial fault diagnosis model by setting the dynamic damping coefficient, it is necessary to determine the initial damping coefficient, the decay rate, the current iteration of model training, and the loss change rate corresponding to the current iteration. Based on the initial damping coefficient, the decay rate, the current iteration of model training, and the loss change rate corresponding to the current iteration, the damping coefficient corresponding to the current iteration is determined.

[0041] Among them, the loss function of the fault diagnosis model can be expressed as:

[0042]

[0043] Where L(Θ,λ(t)) is the total loss function of the fault diagnosis model, Θ is the set of all parameters in the fault diagnosis model (including the weight matrix W and the bias term b); is the final feature representation of node a at layer T; Sof() is the Softmax classifier function used to map node features to categories; y a is the true label of node a; LC() is the loss function (such as the cross entropy loss function); λ(t) is the dynamically adjusted damping coefficient, which changes as the training progresses; |||| is the L2 normal norm. Compared with the prior art, the loss function of the fault diagnosis model provided by this application has been added with more This part can be used to dynamically set the damping coefficient.

[0044] Furthermore, the dynamically adjusted damping coefficient changes with the number of iterations. Through dynamic adjustment, the model can focus on learning the graph structure in the early stage, and gradually strengthen regularization in the later stage to avoid overfitting, which can be expressed as:

[0045]

[0046] Where λ0 is the initial damping coefficient; tanh() is the hyperbolic tangent function; α is the decay rate; t is the current training iteration; λ(t) is the damping coefficient in the tth iteration; and Δ(t) is the rate of change of loss. Preferably, λ0 is set to 0.2.

[0047] Furthermore, the loss change rate is calculated as:

[0048]

[0049] Where L(t) is the loss function of the t-th iteration of the graph neural network; L(t-1) is the loss function of the t-1-th iteration of the graph neural network.

[0050] By setting up a mechanism for dynamically adjusting the damping coefficient, the model can automatically adjust its learning strategy according to the needs of different stages during training, avoiding overfitting and improving training efficiency. This mechanism, unlike static training methods, enables the model to focus on learning the graph structure in the early stages and gradually enhance regularization in the later stages to avoid overfitting.

[0051] In one embodiment, when determining the fault diagnosis model corresponding to the target device, a previously used fault diagnosis model for the target device may be used as the fault diagnosis model for the current fault diagnosis task. Furthermore, based on information such as the target device's device name, device type, device location, and device parameters, the similarity between the target device and the devices corresponding to pre-stored fault diagnosis models in the database may be determined. Based on the similarity, the device with the highest similarity to the target device is determined, and the pre-stored fault diagnosis model for the corresponding device is used as the fault diagnosis model for the target device.

[0052] S103: Input the monitoring data into the fault diagnosis model to determine the device status and fault type of the target device.

[0053] After obtaining monitoring data and determining the fault diagnosis model corresponding to the target device, the monitoring data can be input into the fault diagnosis model to determine the device status and fault type of the target device. The device status and fault type here are consistent with those described above. The device status can include normal or faulty states, and the fault type can include a blown fuse, short circuit, and other faults.

[0054] After model training is completed, the preprocessed real-time monitoring data is input into the trained graph neural network model for fault diagnosis; the model first models the uncertainty of the input features through the fuzzy inference module, combines the dynamic damping adjustment mechanism to optimize the information transmission weight between nodes, and uses graph convolution operations to aggregate the topological relationship of the multi-source sensor data of the equipment (such as temperature, pressure, vibration and other discrete features) layer by layer to generate a device status representation; the model outputs the node-level fault probability distribution, and uses the classifier to determine the current status of the equipment (normal / abnormal) and the specific fault type (such as overload, sensor failure, etc.).

[0055] In one embodiment, a fuzzy inference mechanism is integrated into the input layer of the fault diagnosis model to effectively model the fuzziness of the data from an early stage. When performing fuzzy inference, the target data corresponding to the target node, which is assigned a fuzzy label, must first be determined within the input layer of the fault diagnosis model. The fuzzy confidence factor corresponding to the target data is then determined based on the discreteness of the target data's characteristic distribution. Finally, the input features corresponding to the target node are determined based on a preset fuzzy inference function, the target node's original fuzzy features, and the fuzzy confidence factor. Finally, the input features derived from the fuzzy inference are fed into the fault diagnosis system to determine the target device's device status and fault type.

[0056] Here is an explanation of fuzzy labels: For example, the operating data of the target device contains operating data used to characterize the severity of the device's vibration. However, whether the severity of the vibration is severe is mixed with the subjective judgment of the staff. Even if 0 represents low vibration severity, 1 represents medium vibration severity, and 2 represents high vibration severity, the operating data will still be mixed with too much human judgment. In this case, it is considered that the operating data has a fuzzy label.

[0057] The fuzzy reasoning process is described in detail below: The basic operation of the graph neural network is to update the representation of each node through graph convolution. The graph convolution operation of the fault diagnosis model can be expressed as:

[0058]

[0059] Where, is the feature representation of node a at layer t; is the feature representation of node a at layer t+1; is the set of neighbor nodes of node a; A ab is an element in the adjacency matrix, which represents the connection strength between node a and node b; W is the weight matrix of the t-th graph convolution layer, which represents the weight of each neighbor node information; b c is the bias term of the graph convolution layer; Sig() is the Sigmoid activation function.

[0060] In view of the uncertainty characteristics in the data, the fuzzy reasoning module uses fuzzy rules to make inferences and adjustments, obtains fuzzy reasoning results and uses them as the initial input of the node. Different from traditional static fuzzy rules, the fuzzy reasoning rules that are adaptive to data distribution make the reasoning results more consistent with the characteristics of the data; specifically, the fuzzy characteristics of node a are expressed as The new feature representation x′ can be obtained by calculating the fuzzy inference rules. a ,The fuzzy reasoning module fine-tunes the node features and enhances the graph neural network's ability to process fuzzy features. The reasoning process is expressed as:

[0061]

[0062] Where, is the original fuzzy feature of node a, which represents the uncertainty or fuzzy part in the data; Fe() is the fuzzy inference function; θ is the parameter of the fuzzy inference function.

[0063] Furthermore, the fusion ratio is controlled by the fuzzy confidence factor to fuse the original features and the fuzzy reasoning results. a ∈[0,1] measures the certainty level of node features, expressed as:

[0064]

[0065] Where, is the fused feature representation, which serves as the input of the graph neural network during its training. Furthermore, the fuzzy confidence factor is calculated from the discreteness of the feature distribution, and the calculation method is expressed as:

[0066]

[0067] Where x a is the original feature of node a, σ(x a ) is the standard deviation of the feature of node a, which can be obtained by calculating the standard deviation of the feature sequence value; τ is the temperature coefficient. Preferably, τ can be set to 0.3.

[0068] In an example, the calculation of the fuzzy inference function can be expressed as:

[0069]

[0070] Where μ i (x a )=soft(W μ x a ) is the activation weight of the i-th fuzzy rule; Soft() is the Softmax classifier function, which maps node features to categories; is the membership projection matrix; k is the number of preset rules; d is the number of features; is the learnable rule base matrix.

[0071] Here, we further explain the learnable rule base matrix: Assume that in a device fault diagnosis task, vibration data from a device is used as an input feature. Assume that the device's vibration signal exhibits varying degrees of ambiguity, potentially manifesting as "minor fault," "moderate fault," and "serious fault." However, the boundaries between these labels are not very clear. Fuzzy reasoning can be used to capture this ambiguity. The rule base matrix assumes three fuzzy rules within the fuzzy reasoning module: Rule 1: When the vibration signal is minor, the node feature update ratio is low; Rule 2: When the vibration signal is moderate, the node feature update ratio is moderate; Rule 3: When the vibration signal is severe, the node feature update ratio is high.

[0072] These rules are represented by a learnable rule base matrix. For example, the rule base matrix R may look like this:

[0073]

[0074] In this matrix, each column represents a rule, and the elements of the column represent the weighted influence of that rule on different feature dimensions. During training, the graph neural network automatically adjusts the weights in matrix R so that fuzzy reasoning can better adapt to the different characteristics of vibration data. For example, in the case of a "minor fault," the network may automatically adjust the rule base matrix to reduce the proportion of node feature updates; in the case of a "serious fault," the network may increase the weight of the rule, making the node feature update more significant. The fuzzy reasoning mechanism can adjust according to the ambiguity of the data, improving the ability to handle uncertain data.

[0075] In one embodiment, in each round of training, the fault diagnosis wearability not only learns the direct relationship between nodes, but also extracts more complex and abstract features through a multi-level feature fusion module. The multi-level feature fusion module combines local node information with global information through a hierarchical graph convolution fusion strategy, thereby improving the graph neural network's ability to understand data diversity. Specifically, it is necessary to determine the target node within the target layer and the adjacent nodes of the target node; then obtain the first initial feature representation of the target node, the second initial feature representation of the adjacent nodes, and the adjacency matrix between the target node and the adjacent nodes. Then, based on the adjacency matrix, the first initial feature representation and the second initial feature representation can be fused as the fused feature representation of the target node.

[0076] Assume that the node feature calculated by node a at layer e is As the initial feature representation, the fusion strategy of local features and global features enables the model to capture local and global information at different levels, thereby improving the expressive power of the classifier, which can be expressed as:

[0077]

[0078] Where, Represents the feature fusion operation, which can be vector concatenation, addition or weighted average; is the second feature representation of neighbor node b in layer e. In the fault diagnosis model, a node represents the data feature corresponding to a piece of operating data of the target device. The edges connecting the nodes represent the similarity between the operating data. The data features corresponding to the nodes are updated through the graph convolution layer.

[0079] After several rounds of iterations of the graph neural network, the representation vector of each node is finally passed to the classifier layer for decision-making. The Softmax classifier function compares the feature vector of each node with the model center of each category to assign a category label to each sample. By combining the learning results of the graph neural network and the adjustment of fuzzy reasoning, the classifier can achieve higher classification accuracy in complex data environments. The classification decision process is expressed as:

[0080]

[0081] Where, is the predicted category label of node a. arg max() is the maximum output function.

[0082] In one embodiment, when the fault diagnosis model is trained, the learning rate can be adjusted in each round of training according to the error of the current classifier, so that the model can adjust the training strategy at different stages, thereby improving the training efficiency and classification accuracy, which is expressed as:

[0083]

[0084] Where ζ is the sensitivity coefficient; L(i) is the loss function of the GNN at the i-th iteration; L(i-1) is the loss function of the GNN at the i-1th iteration; η(t) is the learning rate of the GNN at the t-th iteration; η0 is the initial learning rate of the GNN; β is the learning rate decay rate; E(t) is the cumulative error fluctuation; and t is the current training iteration. Preferably, ζ is set to 0.01, β is set to 2, and η0 is set to 0.01.

[0085] Furthermore, based on the learning rate of the graph neural network at the tth iteration, the gradient descent method is used to perform error backpropagation to achieve iterative training of the graph neural network. The above steps are repeated until a preset stop iteration condition is met, indicating that the model training is complete. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 500.

[0086] In one embodiment, after determining the target device's device status and fault type, if the target device's device status is a fault, the fault type is obtained and, based on the fault type, a corresponding instruction is triggered for the target device. The corresponding instruction includes at least one of a maintenance instruction or a shutdown instruction. Simultaneously, diagnostic results are visually fed back to the monitoring system, and when preset alarm thresholds are triggered, maintenance recommendations or shutdown instructions are automatically generated, achieving a closed-loop response from data collection to fault decision-making. This also supports backtracking of historical diagnostic results and root cause analysis.

[0087] like Figure 2 As shown, an embodiment of the present application further provides an industrial equipment fault diagnosis device based on machine learning, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to:

[0088] Collect the operating data of the target device and preprocess the operating data to obtain monitoring data; determine a pre-trained fault diagnosis model, which is an improved graph neural network model, and the input layer of the improved graph neural network model adopts a fuzzy inference mechanism; input the monitoring data into the fault diagnosis model to determine the device status and fault type of the target device.

[0089] The embodiment of the present application further provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to:

[0090] Collect the operating data of the target device and preprocess the operating data to obtain monitoring data; determine a pre-trained fault diagnosis model, which is an improved graph neural network model, and the input layer of the improved graph neural network model adopts a fuzzy inference mechanism; input the monitoring data into the fault diagnosis model to determine the device status and fault type of the target device.

[0091] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device and medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0092] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0093] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0094] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0095] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0096] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0097] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0098] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0099] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0100] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0101] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for industrial equipment fault diagnosis based on machine learning, characterized in that: include: Collecting operating data of the target device and preprocessing the operating data to obtain monitoring data; Determine a pre-trained fault diagnosis model, where the fault diagnosis model is an improved graph neural network model, and the input layer of the improved graph neural network model adopts a fuzzy inference mechanism; The monitoring data is input into the fault diagnosis model to determine the device status and fault type of the target device.

2. The method according to claim 1, characterized in that The collecting of the operating data of the target device and preprocessing the operating data to obtain monitoring data specifically includes: Storing the running data in JSON format; performing data cleaning on the operation data to obtain first intermediate data; Normalizing the first intermediate data to obtain second intermediate data; Feature discretization is performed on the second intermediate data to obtain the monitoring data.

3. The method according to claim 1, characterized in that Before determining the pre-trained fault diagnosis model, the method further includes: Obtaining historical operating data of the target device and annotating the preprocessed historical operating data, wherein the annotated content includes at least the device status and the fault type; Inputting the annotated historical operating data into an initial fault diagnosis model, and training the initial fault diagnosis model by setting a dynamic damping coefficient to obtain the fault diagnosis model; Determine device information corresponding to the target device, and store the fault diagnosis model based on the device information; the device information includes at least one of a device name, a device type, a device location, and device parameters.

4. The method according to claim 3, characterized in that The training of the initial fault diagnosis model by setting the dynamic damping coefficient specifically includes: Determining an initial damping coefficient, a decay rate, a current iteration of model training, and a rate of change of loss corresponding to the current iteration; Based on the initial damping coefficient, the decay rate, the current iteration round of model training, and the loss change rate corresponding to the current iteration round, the damping coefficient corresponding to the current iteration round is determined.

5. The method according to claim 1, wherein Inputting the monitoring data into the fault diagnosis model to determine the device status and fault type of the target device specifically includes: In the input layer of the fault diagnosis model, determining target data with fuzzy labels corresponding to the target node; Determining a fuzzy confidence factor corresponding to the target data based on the characteristic distribution dispersion of the target data; Determining the input feature corresponding to the target node based on a preset fuzzy inference function, the original fuzzy feature of the target node, and the fuzzy confidence factor; Based on the input features, a device status and a fault type of the target device are determined.

6. The method according to claim 1, characterized in that Inputting the monitoring data into the fault diagnosis model to determine the device status and fault type of the target device specifically includes: Determine a target node in a target layer, and adjacent nodes of the target node; Obtaining a first initial feature representation of the target node, a second initial feature representation of the adjacent node, and an adjacency matrix between the target node and the adjacent node; Based on the adjacency matrix, the first initial feature representation and the second initial feature representation are fused as a fused feature representation of the target node.

7. The method according to claim 1, characterized in that After determining the device status and fault type of the target device, the method further includes: When the device state of the target device is a fault state, obtaining the fault type of the target device; Based on the fault type, a corresponding instruction of the target device is triggered, where the corresponding instruction includes at least one of a maintenance instruction or a shutdown instruction.

8. The method according to claim 1, characterized in that Each piece of said operation data includes at least a timestamp, a device number, a collection location operation data, a sensor type, and a sensor measurement value; The nodes in the fault diagnosis model are used to represent data features corresponding to a piece of operating data of the target device, and the edges connecting the nodes represent the similarities between the operating data. The data features corresponding to the nodes are updated through the graph convolution layer.

9. An industrial equipment fault diagnosis device based on machine learning, characterized in that: include: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, wherein the instructions are executed by the at least one processor to enable the at least one processor to perform: Collecting operating data of the target device and preprocessing the operating data to obtain monitoring data; Determine a pre-trained fault diagnosis model, where the fault diagnosis model is an improved graph neural network model, and the input layer of the improved graph neural network model adopts a fuzzy inference mechanism; The monitoring data is input into the fault diagnosis model to determine the device status and fault type of the target device.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured to: Collecting operating data of the target device and preprocessing the operating data to obtain monitoring data; Determine a pre-trained fault diagnosis model, wherein the fault diagnosis model is an improved graph neural network model, and the input layer of the improved graph neural network model adopts a fuzzy inference mechanism; The monitoring data is input into the fault diagnosis model to determine the device status and fault type of the target device.

Images