Washing machine fault detection method and device based on graph neural network and washing machine

Abstract

The invention provides a washing machine fault detection method and device based on a graph neural network and a washing machine, and the method comprises the steps: obtaining multi-source heterogeneous data in real time through a plurality of sensors installed at different elements of the washing machine, and obtaining first target data; constructing an element relation graph of the washing machine at least based on the first target data; and inputting the element relation graph into a target neural network to obtain a fault detection result. The multi-source heterogeneous data of the washing machine is integrated into the element relation graph to be expressed, the complex relation between the elements in the washing machine is captured through the graph convolution algorithm, potential fault detection is carried out through the relation, and the problem that in the prior art, the fault detection technology depends on single sensor data, and the fault detection efficiency is low is solved. The problem of low fault detection precision caused by the fact that multi-source heterogeneous data cannot be integrated and analyzed in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of data communication technology, and in particular to a washing machine fault detection method, device, computer-readable storage medium and washing machine based on graph neural network. Background Art

[0002] As modern families have higher requirements for quality of life, washing machines, as essential household appliances, have become more important in terms of performance and reliability. During long-term use, washing machines are bound to have various faults, such as motor failure, sensor failure, and drainage system blockage.

[0003] Traditional washing machine fault detection methods mainly rely on manual regular inspection and maintenance to ensure the normal operation of the washing machine. However, traditional fault detection methods are inefficient and can easily lead to hidden dangers not being discovered in time due to negligence of staff, affecting the normal use of the washing machine and even causing safety hazards.

[0004] In view of the shortcomings of traditional washing machine fault detection methods, the existing patent CN117992811B proposes a fault detection method and system for an IoT washing machine in a university dormitory, which specifically discloses that the multi-dimensional data collected during the laundry process is analyzed and the abnormal clusters are identified through the CURE clustering algorithm to determine the washing machine fault.

[0005] The above method realizes automatic detection of washing machine failures through simple machine learning statistical analysis. However, this method simply relies on the data itself for clustering detection, but does not consider the complex correlation between the data of different components. It lacks the ability to integrate and analyze multi-source heterogeneous data, resulting in insufficient monitoring accuracy. Summary of the invention

[0006] The main purpose of the present application is to provide a washing machine fault detection method, device, computer-readable storage medium and washing machine based on graph neural network, so as to at least solve the problem that the fault detection technology in the prior art relies on single sensor data and cannot integrate and analyze multi-source heterogeneous data, resulting in low fault detection accuracy.

[0007] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a washing machine fault detection method based on a graph neural network is provided, comprising: acquiring multi-source heterogeneous data in real time through multiple sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; constructing a component relationship graph of the washing machine based on at least the first target data, wherein the nodes of the component relationship graph are the components and sensors of the washing machine, and the edges between the nodes are used to characterize the association relationship between the nodes; inputting the component relationship graph into a target neural network to obtain a fault detection result, wherein the target neural network is used to aggregate the feature information of the neighbor nodes of the node through multi-layer convolution, and perform fault detection based on the feature information of the aggregated nodes, and the neighbor nodes are the nodes connected to the node through the edges.

[0008] Optionally, constructing the component relationship diagram of the washing machine at least based on the first target data includes: performing data cleaning on the first target data by statistical methods and anomaly detection algorithms to remove noise data and outliers in the first target data, and normalizing the first target data after data cleaning to obtain second target data; and constructing the component relationship diagram according to the second target data.

[0009] Optionally, constructing the component relationship graph according to the second target data includes: representing the component and the sensor as nodes of different colors; connecting the sensor and the component through an edge when there is a physical connection relationship between the sensor and the component; connecting the sensors through an edge when there is a logical causal relationship between the data collected by the sensor; configuring the node attributes of each node based on the association between the data in the second target data and the nodes, the node attributes including one or more of temperature data, water level data, pressure data, vibration data and current data; configuring the edge attributes of each edge based on the node attributes corresponding to the node connected to the edge, the edge attributes including one or more of Euclidean distance and cosine similarity.

[0010] Optionally, the component relationship diagram is input into the target neural network to obtain a fault detection result, including: performing feature extraction based on each of the nodes, and determining a neighbor node set corresponding to each node, performing message passing operations on the neighbor node set through each convolutional layer of the target neural network in turn, and updating the features of the node based on the message passing operation to obtain node aggregation features; when the node aggregation feature exceeds a preset range, determining that the component corresponding to the node has a fault; when the node aggregation feature is within the preset range, determining that the component corresponding to the node does not have a fault.

[0011] Optionally, feature extraction is performed based on each of the nodes, and a set of neighbor nodes corresponding to each of the nodes is determined, message passing operations are performed on the neighbor node sets through each convolutional layer of the target neural network in turn, and the features of the nodes are updated based on the message passing operations to obtain node aggregation features, including: feature initialization is performed based on the node attributes of the nodes to obtain a first feature, feature initialization is performed based on the edge attributes of the edges to obtain a second feature, and the first feature corresponding to the node in the neighbor node set is determined as a third feature; in each layer of the convolutional layer, the first feature, the second feature and the third feature are operated based on a message function to obtain a transmission message, and the first feature is operated through an activation function based on the transmission message to obtain the node aggregation feature.

[0012] Optionally, the first target data is cleaned by a statistical method and an anomaly detection algorithm to remove noise data and outliers in the first target data, and the first target data after data cleaning is normalized to obtain the second target data, including: analyzing the first target data, extracting missing data in the first target data, and filling the missing data based on the mean or median of the same type of data when the missing data is numerical data, and filling the missing data based on the mode of the same type when the missing data is categorical data; analyzing the first target data, extracting extreme values ​​in the first target data, and replacing the extreme values ​​based on the upper limit or lower limit of the same type of data; calculating the standard score of the first target data based on the Z-score algorithm, and determining the first target data as abnormal data when the standard score is greater than or equal to a first threshold; screening the first target data based on the IQR algorithm to obtain the abnormal data; clustering the first target data based on the DBSCAN algorithm, and determining the first target data corresponding to the outlier as the abnormal data; deleting all the abnormal data in the first target data, and normalizing the first target data through the Z-score algorithm to obtain the second target data.

[0013] Optionally, after inputting the component relationship diagram into the target neural network and obtaining the fault detection result, the method includes: when the fault detection result is that the washing machine is faulty, issuing a target alarm, the target alarm including the fault location and the fault troubleshooting solution; when the fault of the washing machine has been eliminated, obtaining maintenance feedback data, the maintenance feedback data at least including the deviation between the actual fault location of the washing machine and the fault location in the target alarm; and retraining the target neural network based on the maintenance feedback data.

[0014] According to another aspect of the present application, a washing machine fault detection device based on a graph neural network is provided, and the device includes: a first acquisition unit, used to acquire multi-source heterogeneous data in real time through multiple sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; a construction unit, used to construct a component relationship graph of the washing machine based on at least the first target data, wherein the nodes of the component relationship graph are the components and sensors of the washing machine, and the edges between the nodes are used to characterize the association relationship between the nodes; a detection unit, used to input the component relationship graph into a target neural network to obtain a fault detection result, and the target neural network is used to aggregate the feature information of the neighbor nodes of the node through multi-layer convolution, and perform fault detection based on the feature information of the aggregated nodes, and the neighbor nodes are the nodes connected to the node through the edges.

[0015] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described.

[0016] According to another aspect of the present application, a washing machine is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of the methods described.

[0017] Applying the technical solution of the present application, in the above-mentioned washing machine fault detection method based on graph neural network, first, multi-source heterogeneous data are acquired in real time through multiple sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; then, a component relationship graph of the washing machine is constructed based at least on the first target data, wherein the nodes of the component relationship graph are the components and sensors of the washing machine, and the edges between the nodes are used to characterize the association relationship between the nodes; finally, the component relationship graph is input into the target neural network to obtain a fault detection result, and the target neural network is used to aggregate the feature information of the neighboring nodes of the node through multi-layer convolution, and perform fault detection according to the feature information of the aggregated nodes, wherein the neighboring nodes are the nodes connected to the node through the edges. The present application configures corresponding sensors at different positions of the washing machine to monitor characteristic parameters that can characterize the operating status of components during the operation of the washing machine, so as to obtain multi-source heterogeneous data, and then represents the multi-source heterogeneous data through a component relationship graph to explore the complex relationship between the components inside the washing machine and the association between the corresponding characteristic parameters, and performs feature extraction on the component relationship graph and washing machine fault detection through a graph convolutional neural network. That is, by combining the component relationship graph with the graph convolutional neural network, based on a comprehensive fault detection data set and an algorithm that can capture the complex associations between data, it realizes the fusion of multi-source heterogeneous data for comprehensive washing machine fault detection, and solves the problem that the fault detection technology in the prior art relies on a single sensor data and cannot realize the integrated analysis of multi-source heterogeneous data, resulting in low fault detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of a flow chart of a washing machine fault detection method based on a graph neural network provided according to an embodiment of the present application is shown;

[0019] Figure 2 A component relationship diagram provided according to an embodiment of the present application is shown;

[0020] Figure 3 A structural block diagram of a washing machine fault detection device based on a graph neural network provided according to an embodiment of the present application is shown.

[0021] The above drawings include the following reference numerals:

[0022] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0023] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] As introduced in the background technology, the washing machine fault detection algorithm in the prior art simply relies on the data itself for cluster detection, but does not consider the complex correlation between the data of different components, lacks the ability to integrate and analyze multi-source heterogeneous data, resulting in insufficient monitoring accuracy. In order to solve the problem that the fault detection technology in the prior art relies on a single sensor data and cannot integrate and analyze multi-source heterogeneous data, resulting in low fault detection accuracy, the embodiments of the present application provide a washing machine fault detection method, device, computer-readable storage medium and washing machine based on a graph neural network.

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0028] In this embodiment, a washing machine fault detection method based on a graph neural network running on a computing device is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0029] Figure 1is a flow chart of a washing machine fault detection method based on a graph neural network according to an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0030] Step S1, acquiring multi-source heterogeneous data in real time through a variety of sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor;

[0031] Specifically, the present application installs multiple sensors on multiple components of the washing machine to monitor different operating parameters and obtain the above-mentioned multi-source heterogeneous data, including the startup time (t) and the operating mode (m) of the washing machine obtained when the fault detection program is initialized, as part of the model input. At the same time, the operating parameters of different components of the washing machine are monitored in real time by sensors as another part of the model input.

[0032] In a specific embodiment, the above-mentioned sensors may include a temperature sensor for monitoring the temperature (T) of the motor, controller and washing liquid to detect faults caused by overheating, a vibration sensor for monitoring the vibration data of the components (v) to detect abnormal and unbalanced faults of the motor, etc., a current sensor for monitoring the current change (I) of the motor, etc. to detect whether the motor, etc. is working normally, a water level sensor for monitoring the water level of the washing liquid (H) to detect faults caused by abnormal water level, and a pressure sensor for monitoring the pressure change of the drainage system of the washing machine to detect blockage faults of the drainage system.

[0033] In the above embodiment, characteristic parameters of multiple components are monitored by multiple sensors, thereby achieving comprehensive monitoring of the operating status of the components, so as to subsequently fully capture the operating status and potential failure risks of the washing machine.

[0034] Step S2, constructing a component relationship graph of the washing machine based at least on the first target data, wherein the nodes of the component relationship graph are the components and the sensors of the washing machine, and the edges between the nodes are used to represent the association relationship between the nodes;

[0035] Specifically, after acquiring multi-source heterogeneous data, a component relationship graph of the laundry machine is constructed so as to be input into the graph neural network for fault detection. In the component relationship graph, the components of the washing machine, such as the motor and controller, and various sensors are all nodes of different colors, and the relationships between the above components and the above sensors are represented by edges to obtain the above component relationship graph, one of the optional forms is as follows: Figure 2 shown.

[0036] In the above embodiment, the component relationship diagram can better reflect the deep relationship between characteristic parameters and characteristic parameters and components compared with a simple data format, so as to facilitate the subsequent prediction of washing machine faults based on a neural network.

[0037] Step S3, inputting the above-mentioned component relationship diagram into the target neural network to obtain the fault detection result, the above-mentioned target neural network is used to aggregate the feature information of the neighboring nodes of the above-mentioned node through multi-layer convolution, and perform fault detection based on the feature information of the aggregated nodes, the above-mentioned neighboring nodes are the above-mentioned nodes connected to the above-mentioned nodes through the above-mentioned edges.

[0038] Specifically, after the above-mentioned component relationship diagram is constructed, the above-mentioned component relationship diagram is input into the graph neural network, and the information of the node's neighbor nodes is gradually aggregated through multi-layer convolution, and finally all the related information in the multi-source heterogeneous data is integrated, and fault detection is performed to output the final fault detection result.

[0039] In the above embodiment, replacing the neural network in the prior art with a graph convolutional neural network can intelligently identify the relationship between the components inside the washing machine without the need to aggregate parameters through other operations, and improve the detection efficiency of the fault detection process.

[0040] Through this embodiment, first, multi-source heterogeneous data are acquired in real time through multiple sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; then, a component relationship graph of the washing machine is constructed based at least on the first target data, wherein the nodes of the component relationship graph are the components and sensors of the washing machine, and the edges between the nodes are used to characterize the association relationship between the nodes; finally, the component relationship graph is input into the target neural network to obtain a fault detection result, and the target neural network is used to aggregate the feature information of the neighboring nodes of the node through multi-layer convolution, and perform fault detection according to the feature information of the aggregated nodes, wherein the neighboring nodes are the nodes connected to the node through the edges. The present application configures corresponding sensors at different positions of the washing machine to monitor characteristic parameters that can characterize the operating status of components during the operation of the washing machine, so as to obtain multi-source heterogeneous data, and then represents the multi-source heterogeneous data through a component relationship graph to explore the complex relationship between the components inside the washing machine and the association between the corresponding characteristic parameters, and performs feature extraction on the component relationship graph and washing machine fault detection through a graph convolutional neural network. That is, by combining the component relationship graph with the graph convolutional neural network, based on a comprehensive fault detection data set and an algorithm that can capture the complex associations between data, it realizes the fusion of multi-source heterogeneous data for comprehensive washing machine fault detection, and solves the problem that the fault detection technology in the prior art relies on a single sensor data and cannot realize the integrated analysis of multi-source heterogeneous data, resulting in low fault detection accuracy.

[0041] In order to ensure the accuracy of fault detection, in an optional implementation, the above step S2 includes:

[0042] Step S21, performing data cleaning on the first target data by using a statistical method and an anomaly detection algorithm to remove noise data and outliers in the first target data, and normalizing the first target data after data cleaning to obtain second target data;

[0043] Specifically, since the data collected by different sensors may have inconsistent dimensions, and noise and outliers may be introduced during the sensor parameter collection process, thereby affecting the results of washing machine fault detection, the present application is configured to use statistical methods and outlier detection to remove data in the first target data that affects the accuracy of washing machine fault detection before performing washing machine fault detection. Furthermore, for the convenience of calculation, the present application is also configured to normalize the parameters collected by different sensors through data normalization to obtain the above-mentioned second target data.

[0044] Step S22, constructing the component relationship diagram according to the second target data.

[0045] Specifically, after obtaining the preprocessed multi-source heterogeneous data, the component relationship diagram is constructed based on the processed second target data, thereby ensuring the accuracy of the component relationship diagram and improving the accuracy of washing machine fault detection.

[0046] In order to construct the above-mentioned component relationship diagram, in an optional implementation manner, the above-mentioned step S22 includes:

[0047] Step S221, representing the above-mentioned components and the above-mentioned sensors as the above-mentioned nodes of different colors;

[0048] Specifically, in the process of constructing the above-mentioned component relationship diagram, the present application represents different components and sensors with different colors, and represents the same sensor with the same color, such as Figure 2 As shown, the colors of current sensor 1, current sensor 2 and current sensor 3 are the same color to facilitate recognition by the graph neural network.

[0049] Step S222, in the case where there is a physical connection relationship between the sensor and the component, connecting the sensor and the component through an edge;

[0050] Specifically, the relationship between the components or sensors can be represented by an edge, including connecting the sensor and the component through an edge when there is a physical connection relationship between the sensor and the component, such as Figure 2 As shown, for example, the current sensor 2 and the motor, the current sensor 2 is installed on the motor, and the current sensor 2 and the motor are connected through an edge.

[0051] Step S223, in the case where there is a logical causal relationship between the data collected by the above sensors, connecting the above sensors through edges;

[0052] Specifically, the relationship between components or sensors can be represented by edges. In the case where there is a logical causal relationship between the data collected by the above sensors, the above sensors are connected by edges, such as Figure 2 As shown, for example, the water level sensor and the current sensor 1 installed in the washing tub, the parameters monitored by them will be affected by the washing liquid, and the current sensor 1 and the water level sensor are connected through an edge.

[0053] Step S224, configuring node attributes of each of the nodes based on the association between the data in the second target data and the nodes, the node attributes including one or more of temperature data, water level data, pressure data, vibration data and current data;

[0054] Specifically, after the connection relationship between the nodes is determined, the node attributes of the nodes are configured in the above-mentioned node relationship diagram to associate the above-mentioned multi-source heterogeneous data. For example, the node attributes of the temperature sensor include but are not limited to average temperature, maximum temperature, minimum temperature and temperature change rate.

[0055] Step S225 , configuring edge attributes of each edge based on the node attributes corresponding to the node connected to the edge, wherein the edge attributes include one or more of Euclidean distance and cosine similarity.

[0056] Specifically, after the node attributes of the node are configured, the edge attributes of the edge are configured in the above node relationship diagram to characterize the relationship between different parameters of the node. The edge attributes are determined by the attributes of the left and right points, such as calculating the Euclidean distance or cosine similarity between the features of two nodes.

[0057] In a specific embodiment, a component relationship diagram, such as Figure 2 shown, including:

[0058] Node 1: current sensor 1; Node 2: water level sensor; Node 3: pressure sensor; Node 4: current sensor 2; Node 5: motor; Node 6: temperature sensor; Node 7: vibration sensor; Node 8: current sensor 3; Node 9: current sensor 4. Among them, the node attributes are the parameter values ​​extracted after preprocessing, including: temperature sensor: average temperature, maximum temperature, minimum temperature and temperature change rate; current sensor: average current, maximum current, minimum current and current change rate; vibration sensor: vibration amplitude, vibration frequency and vibration energy; motor: speed, load and temperature; water level sensor: average water level, maximum water level, minimum water level and water level change rate. For the connection between nodes and nodes, including: there is a clear physical connection and control relationship between the various components inside the washing machine, and the relationship is the edge. For example: node 1 depends on the data of nodes 2 and 3, there is an edge connection, node 5 affects nodes 4, 6, 7, 8, 9, there is an edge connection, node 2 and node 6 have no relationship, so there is no edge connection. The attributes of the edge are determined by the attributes of the left and right points.

[0059] It is understandable that the process of building a graph is to serve the subsequent graph neural network calculation, and it reflects the relationship between different types of data. Specifically, the description of the data transmission connection between A and B, or A can control the start and stop of B, or A feeds back the results to B, can all be converted into edge relationships, which are direct and explicit relationships; some implicit edges can also be realized through some relationship extraction strategies, such as determining the correlation between different sensor data through correlation analysis (such as Pearson correlation coefficient, mutual information, etc.). Usually, explicit edges are used first.

[0060] In order to perform fault detection on the washing machine, in an optional implementation, the above step S3 includes:

[0061] Step S31, extracting features based on each of the above nodes, and determining a neighbor node set corresponding to each of the above nodes, performing message passing operations on the neighbor node sets through each convolutional layer of the target neural network in turn, and updating the features of the above nodes based on the message passing operations to obtain node aggregation features;

[0062] Specifically, the network uses message passing extremes, and each node aggregates the messages of its neighbor nodes and updates its own status. That is, the above-mentioned message passing operations and update operations are performed through multiple convolutional layers to gradually aggregate the multi-order neighbor messages of the node until all relevant messages in multi-source heterogeneous data are integrated to obtain the above-mentioned node aggregation features.

[0063] Step S32, when the node aggregation feature exceeds a preset range, determining that the component corresponding to the node is faulty;

[0064] Step S33: When the node aggregation characteristic is within the preset range, it is determined that the component corresponding to the node is not faulty.

[0065] Specifically, fault detection based on the values ​​in the node aggregation features can include sensor data thresholds, and an upper and lower limit can be set to alarm when the predicted value exceeds the range, for example: a warning is triggered when the temperature sensor exceeds 60°C or is lower than 5°C; a warning is triggered when the current sensor exceeds 3A or is lower than 0.5A; a warning is triggered when the vibration intensity exceeds a certain value, etc. It can also include: by comparing the fluctuations of current data with historical data, setting a threshold for abnormal fluctuations, for example, triggering an alarm when the current data fluctuates by more than 20% compared with historical data. It can also include: inference based on the confidence of the model prediction, for example, triggering an alarm when the probability of failure exceeds 80%. Otherwise, no alarm is triggered.

[0066] In order to complete the above-mentioned message transmission operation and update operation, in an optional implementation manner, the above-mentioned step S31 includes:

[0067] Step S311, performing feature initialization based on the node attribute of the node to obtain a first feature, performing feature initialization based on the edge attribute of the edge to obtain a second feature, and determining the first feature corresponding to the node in the neighbor node set as a third feature;

[0068] Specifically, the node features are initialized as These features can be node attributes or randomly initialized vectors. The set of neighbor nodes is initialized as Then, the edge features and node features in the neighbor node set are initialized based on the neighbor node set.

[0069] In step S312, at each convolution layer, the first feature, the second feature and the third feature are operated based on a message function to obtain a transfer message, and the first feature is operated through an activation function based on the transfer message to obtain the node aggregation feature.

[0070] Specifically, the first thing to do is message passing. At each layer, each node v collects messages from its neighbor nodes N(v). The formula is as follows:

[0071]

[0072] in, is the message of node v at layer L+1, M is the message function, e uv is the feature of edge (u, v). is the message of node u at layer L.

[0073] Then perform an update operation to aggregate the features Used to update the feature representation of node v. The formula is as follows:

[0074]

[0075] in, is the updated feature. If l+1 is the last layer, it is the aggregate node feature. and b are learnable parameters, and σ is the activation function (such as ReLU).

[0076] In the above embodiment, the neighbor information of the point is aggregated into the features of the point itself after the first layer of the graph neural network, and the second layer aggregates the information of the neighbors of the neighbors, and so on, finally achieving feature fusion, and then the model uses the last layer of comprehensive feature vectors to predict the information of the point, such as the temperature of the temperature sensor, the current of the current sensor, etc., to perform fault detection. This process can be repeated for multiple layers to enrich the node features, and finally the model gives the fault detection results based on the high-dimensional features of the nodes.

[0077] In order to ensure the accuracy of the second target data, in an optional implementation, the step S21 includes:

[0078] Step S211, analyzing the first target data, extracting missing data in the first target data, and filling the missing data based on the mean or median of the same type of data when the missing data is numerical data, and filling the missing data based on the mode of the same type when the missing data is categorical data;

[0079] Specifically, one of the data cleaning methods is to fill in the mean / median, that is, for missing values, the mean or median of the feature is used to fill, which is suitable for numerical data. One of the data cleaning methods is to fill in the mode, that is, for categorical data, the mode of the feature can be used to fill.

[0080] Step S212, analyzing the first target data, extracting extreme values ​​in the first target data, and replacing the extreme values ​​based on upper limits or lower limits of data of the same type;

[0081] Specifically, one of the data cleaning methods is upper and lower limit processing, that is, for extreme values, a reasonable upper and lower limits can be set, and values ​​outside the range can be replaced with upper or lower limit values.

[0082] Step S213, calculating the standard score of the first target data based on a Z-score algorithm, and determining the first target data as abnormal data when the standard score is greater than or equal to a first threshold;

[0083] Specifically, one of the anomaly detection algorithms may be a Z-score algorithm, which calculates a standard score for each parameter, and data points exceeding a certain threshold are considered outliers.

[0084] Step S214, screening the first target data based on the IQR algorithm to obtain the abnormal data;

[0085] Specifically, one of the anomaly detection algorithms may be an IQR algorithm, which uses quartiles (IQR) to detect outliers, calculates the first quartile (Q1) and the third quartile (Q3), and parameters falling outside Q1-1.5*IQR and Q3+1.5*IQR are considered outliers.

[0086] Step S215, clustering the first target data based on the DBSCAN algorithm, and determining the first target data corresponding to the outliers as the abnormal data;

[0087] Specifically, one of the anomaly detection algorithms may be a density-based clustering algorithm, which can identify noise points and outliers.

[0088] Step S216, deleting all the abnormal data in the first target data, and normalizing the first target data by using a Z-score algorithm to obtain the second target data.

[0089] Specifically, the present application simultaneously adopts the above three algorithms to remove outliers, further ensuring the accuracy of the predicted data and improving the fault detection accuracy of the graph neural network.

[0090] In order to ensure the accuracy of the graph neural network, in an optional implementation, after the component relationship diagram is input into the target neural network and the fault detection result is obtained, the method includes:

[0091] Step S4, when the fault detection result indicates that the washing machine has a fault, issuing a target alarm, wherein the target alarm includes fault location and a fault elimination solution;

[0092] Specifically, when the graph neural network detects a potential fault in the washing machine, it will generate the target alert to provide fault location and solutions. Before the washing process ends, the system will continuously monitor the data of multiple sensor modules on the washing machine and issue warnings for abnormal prediction values ​​in real time.

[0093] Step S5, when the fault of the washing machine has been eliminated, obtaining maintenance feedback data, the maintenance feedback data at least including a deviation between the actual fault location of the washing machine and the fault location in the target alarm;

[0094] Specifically, after the washing machine fault troubleshooting is completed, maintenance feedback information is collected based on the deviation between the detection results of the graph neural network and the maintenance feedback information.

[0095] Step S6: retraining the target neural network based on the maintenance feedback data.

[0096] Specifically, the fault detection algorithm and the corresponding model structure of the graph neural network are further optimized based on the maintenance feedback information.

[0097] In a specific embodiment, the model update and training process of the present application can be completed in the cloud, that is, the above-mentioned maintenance feedback information of the mobile phone is transmitted to the cloud for training, and after each time the washing machine is started, it is determined whether the model needs to be updated based on the local model version and the cloud model version.

[0098] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the washing machine fault detection method based on graph neural network of the present application will be described in detail below in combination with specific embodiments.

[0099] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0100] The embodiment of the present application also provides a washing machine fault detection device based on a graph neural network. It should be noted that the washing machine fault detection device based on a graph neural network in the embodiment of the present application can be used to execute the washing machine fault detection method based on a graph neural network provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and those that have been described will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0101] The following is an introduction to the washing machine fault detection device based on graph neural network provided in an embodiment of the present application.

[0102] Figure 3 This is a structural block diagram of a washing machine fault detection device based on a graph neural network according to an embodiment of the present application.

[0103] like Figure 3 As shown, the device comprises:

[0104] A first acquisition unit 10 is used to acquire multi-source heterogeneous data in real time through a variety of sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor;

[0105] Specifically, the present application installs multiple sensors on multiple components of the washing machine to monitor different operating parameters and obtain the above-mentioned multi-source heterogeneous data, including the startup time (t) and the operating mode (m) of the washing machine obtained when the fault detection program is initialized, as part of the model input. At the same time, the operating parameters of different components of the washing machine are monitored in real time by sensors as another part of the model input.

[0106] In a specific embodiment, the above-mentioned sensors may include a temperature sensor for monitoring the temperature (T) of the motor, controller and washing liquid to detect faults caused by overheating, a vibration sensor for monitoring the vibration data of the components (v) to detect abnormal and unbalanced faults of the motor, etc., a current sensor for monitoring the current change (I) of the motor, etc. to detect whether the motor, etc. is working normally, a water level sensor for monitoring the water level of the washing liquid (H) to detect faults caused by abnormal water level, and a pressure sensor for monitoring the pressure change of the drainage system of the washing machine to detect blockage faults of the drainage system.

[0107] In the above embodiment, characteristic parameters of multiple components are monitored by multiple sensors, thereby achieving comprehensive monitoring of the operating status of the components, so as to subsequently fully capture the operating status and potential failure risks of the washing machine.

[0108] A construction unit 20 is used to construct a component relationship graph of the washing machine based on at least the first target data, wherein the nodes of the component relationship graph are the components and the sensors of the washing machine, and the edges between the nodes are used to represent the association relationship between the nodes;

[0109] Specifically, after acquiring multi-source heterogeneous data, a component relationship graph of the laundry machine is constructed so as to be input into the graph neural network for fault detection. In the component relationship graph, the components of the washing machine, such as the motor and controller, and various sensors are all nodes of different colors, and the relationships between the above components and the above sensors are represented by edges to obtain the above component relationship graph, one of the optional forms is as follows: Figure 2 shown.

[0110] In the above embodiment, the component relationship diagram can better reflect the deep relationship between characteristic parameters and characteristic parameters and components compared with a simple data format, so as to facilitate the subsequent prediction of washing machine faults based on a neural network.

[0111] The detection unit 30 is used to input the above-mentioned component relationship diagram into the target neural network to obtain a fault detection result. The above-mentioned target neural network is used to aggregate the characteristic information of the neighboring nodes of the above-mentioned node through multi-layer convolution, and perform fault detection based on the characteristic information of the aggregated nodes. The above-mentioned neighboring nodes are the above-mentioned nodes connected to the above-mentioned nodes through the above-mentioned edges.

[0112] Specifically, after the above-mentioned component relationship diagram is constructed, the above-mentioned component relationship diagram is input into the graph neural network, and the information of the node's neighbor nodes is gradually aggregated through multi-layer convolution, and finally all the related information in the multi-source heterogeneous data is integrated, and fault detection is performed to output the final fault detection result.

[0113] In the above embodiment, replacing the neural network in the prior art with a graph convolutional neural network can intelligently identify the relationship between the components inside the washing machine without the need to aggregate parameters through other operations, and improve the detection efficiency of the fault detection process.

[0114] Through this embodiment, the first acquisition unit acquires multi-source heterogeneous data in real time through multiple sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; the construction unit constructs a component relationship graph of the washing machine based at least on the first target data, wherein the nodes of the component relationship graph are the components and sensors of the washing machine, and the edges between the nodes are used to characterize the association relationship between the nodes; the detection unit inputs the component relationship graph into the target neural network to obtain a fault detection result, wherein the target neural network is used to aggregate feature information of neighbor nodes of the node through multi-layer convolution, and perform fault detection according to the feature information of the aggregated nodes, and the neighbor nodes are the nodes connected to the node through the edges. The present application configures corresponding sensors at different positions of the washing machine to monitor characteristic parameters that can characterize the operating status of components during the operation of the washing machine, so as to obtain multi-source heterogeneous data, and then represents the multi-source heterogeneous data through a component relationship graph to explore the complex relationship between the components inside the washing machine and the association between the corresponding characteristic parameters, and performs feature extraction on the component relationship graph and washing machine fault detection through a graph convolutional neural network. That is, by combining the component relationship graph with the graph convolutional neural network, based on a comprehensive fault detection data set and an algorithm that can capture the complex associations between data, it realizes the fusion of multi-source heterogeneous data for comprehensive washing machine fault detection, and solves the problem that the fault detection technology in the prior art relies on a single sensor data and cannot realize the integrated analysis of multi-source heterogeneous data, resulting in low fault detection accuracy.

[0115] In order to ensure the accuracy of fault detection, in an optional implementation manner, the above-mentioned construction unit includes:

[0116] A first processing module is used to perform data cleaning on the first target data by using a statistical method and an anomaly detection algorithm to remove noise data and outliers in the first target data, and normalize the first target data after data cleaning to obtain second target data;

[0117] Specifically, since the data collected by different sensors may have inconsistent dimensions, and noise and outliers may be introduced during the sensor parameter collection process, thereby affecting the results of washing machine fault detection, the present application is configured to use statistical methods and outlier detection to remove data in the first target data that affects the accuracy of washing machine fault detection before performing washing machine fault detection. Furthermore, for the convenience of calculation, the present application is also configured to normalize the parameters collected by different sensors through data normalization to obtain the above-mentioned second target data.

[0118] A construction module is used to construct the above-mentioned component relationship diagram according to the above-mentioned second target data.

[0119] Specifically, after obtaining the preprocessed multi-source heterogeneous data, the component relationship diagram is constructed based on the processed second target data, thereby ensuring the accuracy of the component relationship diagram and improving the accuracy of washing machine fault detection.

[0120] In order to construct the above-mentioned component relationship diagram, in an optional implementation manner, the above-mentioned construction module includes:

[0121] A first processing submodule, used for representing the above-mentioned elements and the above-mentioned sensors as the above-mentioned nodes of different colors;

[0122] Specifically, in the process of constructing the above-mentioned component relationship diagram, the present application represents different components and sensors with different colors, and represents the same sensor with the same color, such as Figure 2 As shown, the colors of current sensor 1, current sensor 2 and current sensor 3 are the same color to facilitate recognition by the graph neural network.

[0123] A second processing submodule, configured to connect the sensor and the element through an edge when there is a physical connection relationship between the sensor and the element;

[0124] Specifically, the relationship between the components or sensors can be represented by an edge, including connecting the sensor and the component through an edge when there is a physical connection relationship between the sensor and the component, such as Figure 2 As shown, for example, the current sensor 2 and the motor, the current sensor 2 is installed on the motor, and the current sensor 2 and the motor are connected through an edge.

[0125] A third processing submodule is used to connect the sensors through edges when there is a logical causal relationship between the data collected by the sensors;

[0126] Specifically, the relationship between components or sensors can be represented by edges. In the case where there is a logical causal relationship between the data collected by the above sensors, the above sensors are connected by edges, such as Figure 2 As shown, for example, the water level sensor and the current sensor 1 installed in the washing tub, the parameters monitored by them will be affected by the washing liquid, and the current sensor 1 and the water level sensor are connected through an edge.

[0127] a fourth processing submodule, configured to configure node attributes of each of the nodes based on the association between the data in the second target data and the nodes, wherein the node attributes include one or more of temperature data, water level data, pressure data, vibration data and current data;

[0128] Specifically, after the connection relationship between the nodes is determined, the node attributes of the nodes are configured in the above-mentioned node relationship diagram to associate the above-mentioned multi-source heterogeneous data. For example, the node attributes of the temperature sensor include but are not limited to average temperature, maximum temperature, minimum temperature and temperature change rate.

[0129] The fifth processing submodule is used to configure the edge attributes of each edge based on the node attributes corresponding to the nodes connected to the edge, wherein the edge attributes include one or more of Euclidean distance and cosine similarity.

[0130] Specifically, after the node attributes of the node are configured, the edge attributes of the edge are configured in the above node relationship diagram to characterize the relationship between different parameters of the node. The edge attributes are determined by the attributes of the left and right points, such as calculating the Euclidean distance or cosine similarity between the features of two nodes.

[0131] In a specific embodiment, a component relationship diagram, such as Figure 2 shown, including:

[0132] Node 1: current sensor 1; Node 2: water level sensor; Node 3: pressure sensor; Node 4: current sensor 2; Node 5: motor; Node 6: temperature sensor; Node 7: vibration sensor; Node 8: current sensor 3; Node 9: current sensor 4. Among them, the node attributes are the parameter values ​​extracted after preprocessing, including: temperature sensor: average temperature, maximum temperature, minimum temperature and temperature change rate; current sensor: average current, maximum current, minimum current and current change rate; vibration sensor: vibration amplitude, vibration frequency and vibration energy; motor: speed, load and temperature; water level sensor: average water level, maximum water level, minimum water level and water level change rate. For the connection between nodes and nodes, including: there is a clear physical connection and control relationship between the various components inside the washing machine, and the relationship is the edge. For example: node 1 depends on the data of nodes 2 and 3, there is an edge connection, node 5 affects nodes 4, 6, 7, 8, 9, there is an edge connection, node 2 and node 6 have no relationship, so there is no edge connection. The attributes of the edge are determined by the attributes of the left and right points.

[0133] It is understandable that the process of building a graph is to serve the subsequent graph neural network calculation, and it reflects the relationship between different types of data. Specifically, the description of the data transmission connection between A and B, or A can control the start and stop of B, or A feeds back the results to B, can all be converted into edge relationships, which are direct and explicit relationships; some implicit edges can also be realized through some relationship extraction strategies, such as determining the correlation between different sensor data through correlation analysis (such as Pearson correlation coefficient, mutual information, etc.). Usually, explicit edges are used first.

[0134] In order to perform fault detection on the washing machine, in an optional implementation, the detection includes:

[0135] A second processing module is used to extract features based on each of the above nodes, determine a neighbor node set corresponding to each of the above nodes, perform message passing operations on the neighbor node sets through each convolutional layer of the target neural network in turn, and update the features of the above nodes based on the message passing operations to obtain node aggregation features;

[0136] Specifically, the network uses message passing extremes, and each node aggregates the messages of its neighbor nodes and updates its own status. That is, the above-mentioned message passing operations and update operations are performed through multiple convolutional layers to gradually aggregate the multi-order neighbor messages of the node until all relevant messages in multi-source heterogeneous data are integrated to obtain the above-mentioned node aggregation features.

[0137] A first determination module is used to determine that the component corresponding to the node is faulty when the node aggregation feature exceeds a preset range;

[0138] The second determination module is used to determine that there is no fault in the component corresponding to the node when the node aggregation feature is within the preset range.

[0139] Specifically, fault detection based on the values ​​in the node aggregation features can include sensor data thresholds, and an upper and lower limit can be set to alarm when the predicted value exceeds the range, for example: a warning is triggered when the temperature sensor exceeds 60°C or is lower than 5°C; a warning is triggered when the current sensor exceeds 3A or is lower than 0.5A; a warning is triggered when the vibration intensity exceeds a certain value, etc. It can also include: by comparing the fluctuations of current data with historical data, setting a threshold for abnormal fluctuations, for example, triggering an alarm when the current data fluctuates by more than 20% compared with historical data. It can also include: inference based on the confidence of the model prediction, for example, triggering an alarm when the probability of failure exceeds 80%. Otherwise, no alarm is triggered.

[0140] In order to complete the above-mentioned message transmission operation and update operation, in an optional implementation manner, the above-mentioned second processing module includes:

[0141] a sixth processing submodule, configured to perform feature initialization based on the node attribute of the node to obtain a first feature, perform feature initialization based on the edge attribute of the edge to obtain a second feature, and determine the first feature corresponding to the node in the neighbor node set as a third feature;

[0142] Specifically, the node features are initialized as These features can be node attributes or randomly initialized vectors. The set of neighbor nodes is initialized as Then, the edge features and node features in the neighbor node set are initialized based on the neighbor node set.

[0143] The seventh processing submodule is used to operate the first feature, the second feature and the third feature based on the message function in each convolution layer to obtain the transmission message, and operate the first feature through the activation function based on the transmission message to obtain the node aggregation feature.

[0144] Specifically, the first thing to do is message passing. At each layer, each node v collects messages from its neighbor nodes N(v). The formula is as follows:

[0145]

[0146] in, is the message of node v at layer L+1, M is the message function, e uv is the feature of edge (u, v). is the message of node u at layer L.

[0147] Then perform an update operation to aggregate the features Used to update the feature representation of node v. The formula is as follows:

[0148]

[0149] in, is the updated feature. If l+1 is the last layer, it is the aggregate node feature. and b are learnable parameters, and σ is the activation function (such as ReLU).

[0150] In the above embodiment, the neighbor information of the point is aggregated into the features of the point itself after the first layer of the graph neural network, and the second layer aggregates the information of the neighbors of the neighbors, and so on, finally achieving feature fusion, and then the model uses the last layer of comprehensive feature vectors to predict the information of the point, such as the temperature of the temperature sensor, the current of the current sensor, etc., to perform fault detection. This process can be repeated for multiple layers to enrich the node features, and finally the model gives the fault detection results based on the high-dimensional features of the nodes.

[0151] In order to ensure the accuracy of the second target data, in an optional implementation manner, the first processing module includes:

[0152] an eighth processing submodule, configured to analyze the first target data, extract missing data from the first target data, and fill in the missing data based on the mean or median of the same type of data if the missing data is numerical data, or fill in the missing data based on the mode of the same type if the missing data is categorical data;

[0153] Specifically, one of the data cleaning methods is to fill in the mean / median, that is, for missing values, the mean or median of the feature is used to fill, which is suitable for numerical data. One of the data cleaning methods is to fill in the mode, that is, for categorical data, the mode of the feature can be used to fill.

[0154] a ninth processing submodule, configured to analyze the first target data, extract extreme values ​​from the first target data, and replace the extreme values ​​based on upper or lower limits of data of the same type;

[0155] Specifically, one of the data cleaning methods is upper and lower limit processing, that is, for extreme values, a reasonable upper and lower limits can be set, and values ​​outside the range can be replaced with upper or lower limit values.

[0156] a tenth processing submodule, configured to calculate a standard score of the first target data based on a Z-score algorithm, and determine the first target data as abnormal data when the standard score is greater than or equal to a first threshold;

[0157] Specifically, one of the anomaly detection algorithms may be a Z-score algorithm, which calculates a standard score for each parameter, and data points exceeding a certain threshold are considered outliers.

[0158] An eleventh processing submodule is used to screen the first target data based on an IQR algorithm to obtain the abnormal data;

[0159] Specifically, one of the anomaly detection algorithms may be an IQR algorithm, which uses quartiles (IQR) to detect outliers, calculates the first quartile (Q1) and the third quartile (Q3), and parameters falling outside Q1-1.5*IQR and Q3+1.5*IQR are considered outliers.

[0160] A twelfth processing submodule is used to cluster the first target data based on the DBSCAN algorithm, and determine the first target data corresponding to the outlier as the abnormal data;

[0161] Specifically, one of the anomaly detection algorithms may be a density-based clustering algorithm, which can identify noise points and outliers.

[0162] The thirteenth processing submodule is used to delete all the above-mentioned abnormal data in the above-mentioned first target data, and normalize the above-mentioned first target data through a Z-score algorithm to obtain the above-mentioned second target data.

[0163] Specifically, the present application simultaneously adopts the above three algorithms to remove outliers, further ensuring the accuracy of the predicted data and improving the fault detection accuracy of the graph neural network.

[0164] In order to ensure the accuracy of the graph neural network, in an optional implementation, the above-mentioned device includes:

[0165] an alarm unit, configured to, after inputting the component relationship diagram into a target neural network and obtaining a fault detection result, issue a target alarm if the fault detection result indicates that the washing machine has a fault, wherein the target alarm includes a fault location and a fault elimination solution;

[0166] Specifically, when the graph neural network detects a potential fault in the washing machine, it will generate the target alert to provide fault location and solutions. Before the washing process ends, the system will continuously monitor the data of multiple sensor modules on the washing machine and issue warnings for abnormal prediction values ​​in real time.

[0167] a second acquisition unit, configured to acquire maintenance feedback data when the fault of the washing machine has been eliminated, the maintenance feedback data at least including a deviation between an actual fault location of the washing machine and the fault location in the target alarm;

[0168] Specifically, after the washing machine fault troubleshooting is completed, maintenance feedback information is collected based on the deviation between the detection results of the graph neural network and the maintenance feedback information.

[0169] A training unit is used to retrain the target neural network based on the maintenance feedback data.

[0170] Specifically, the fault detection algorithm and the corresponding model structure of the graph neural network are further optimized based on the maintenance feedback information.

[0171] In a specific embodiment, the model update and training process of the present application can be completed in the cloud, that is, the above-mentioned maintenance feedback information of the mobile phone is transmitted to the cloud for training, and after each time the washing machine is started, it is determined whether the model needs to be updated based on the local model version and the cloud model version.

[0172] The above-mentioned washing machine fault detection device based on graph neural network includes a processor and a memory. The above-mentioned first acquisition unit, construction unit and detection unit are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize the corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned modules are located in different processors in any combination.

[0173] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the accuracy of washing machine fault detection can be improved by adjusting kernel parameters.

[0174] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0175] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the washing machine fault detection method based on graph neural network.

[0176] An embodiment of the present invention provides a processor, which is used to run a program, wherein the program executes the above-mentioned washing machine fault detection method based on graph neural network when running.

[0177] An embodiment of the present invention provides a washing machine, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements at least the steps of a washing machine fault detection method based on a graph neural network.

[0178] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes at least the steps of a washing machine fault detection method based on a graph neural network.

[0179] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0180] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may 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 may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0181] 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 generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. 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.

[0182] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate 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 A function specified in one or more boxes.

[0183] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0184] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0185] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0186] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, 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 temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0187] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0188] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0189] 1) The washing machine fault detection method based on graph neural network of the present application first obtains multi-source heterogeneous data in real time through multiple sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; then, a component relationship graph of the washing machine is constructed based at least on the first target data, wherein the nodes of the component relationship graph are the components and sensors of the washing machine, and the edges between the nodes are used to characterize the association relationship between the nodes; finally, the component relationship graph is input into the target neural network to obtain the fault detection result, and the target neural network is used to aggregate the feature information of the neighboring nodes of the node through multi-layer convolution, and perform fault detection according to the feature information of the aggregated nodes, wherein the neighboring nodes are the nodes connected to the node through the edges. The present application configures corresponding sensors at different positions of the washing machine to monitor characteristic parameters that can characterize the operating status of components during the operation of the washing machine, so as to obtain multi-source heterogeneous data, and then represents the multi-source heterogeneous data through a component relationship graph to explore the complex relationship between the components inside the washing machine and the association between the corresponding characteristic parameters, and performs feature extraction on the component relationship graph and washing machine fault detection through a graph convolutional neural network. That is, by combining the component relationship graph with the graph convolutional neural network, based on a comprehensive fault detection data set and an algorithm that can capture the complex associations between data, it realizes the fusion of multi-source heterogeneous data for comprehensive washing machine fault detection, and solves the problem that the fault detection technology in the prior art relies on a single sensor data and cannot realize the integrated analysis of multi-source heterogeneous data, resulting in low fault detection accuracy.

[0190] 2) In the washing machine fault detection device based on graph neural network of the present application, the first acquisition unit acquires multi-source heterogeneous data in real time through multiple sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; the construction unit constructs a component relationship graph of the washing machine based at least on the first target data, wherein the nodes of the component relationship graph are the components and sensors of the washing machine, and the edges between the nodes are used to characterize the association relationship between the nodes; the detection unit inputs the component relationship graph into the target neural network to obtain a fault detection result, wherein the target neural network is used to aggregate the feature information of the neighboring nodes of the node through multi-layer convolution, and perform fault detection according to the feature information of the aggregated nodes, and the neighboring nodes are the nodes connected to the node through the edges. The present application configures corresponding sensors at different positions of the washing machine to monitor characteristic parameters that can characterize the operating status of components during the operation of the washing machine, so as to obtain multi-source heterogeneous data, and then represents the multi-source heterogeneous data through a component relationship graph to explore the complex relationship between the components inside the washing machine and the association between the corresponding characteristic parameters, and performs feature extraction on the component relationship graph and washing machine fault detection through a graph convolutional neural network. That is, by combining the component relationship graph with the graph convolutional neural network, based on a comprehensive fault detection data set and an algorithm that can capture the complex associations between data, it realizes the fusion of multi-source heterogeneous data for comprehensive washing machine fault detection, and solves the problem that the fault detection technology in the prior art relies on a single sensor data and cannot realize the integrated analysis of multi-source heterogeneous data, resulting in low fault detection accuracy.

[0191] The above description is only the preferred 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 modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A washing machine fault detection method based on graph neural network, characterized in that: include: Acquire multi-source heterogeneous data in real time through multiple sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; constructing a component relationship graph of the washing machine based at least on the first target data, wherein the nodes of the component relationship graph are the components and the sensors of the washing machine, and the edges between the nodes are used to represent the association relationship between the nodes; The component relationship diagram is input into a target neural network to obtain a fault detection result. The target neural network is used to aggregate feature information of neighbor nodes of the node through multi-layer convolution, and perform fault detection based on the feature information of the aggregated nodes. The neighbor nodes are the nodes connected to the node through the edge.

2. The method according to claim 1, characterized in that Constructing a component relationship diagram of the washing machine based at least on the first target data, comprising: Performing data cleaning on the first target data by using a statistical method and an anomaly detection algorithm to remove noise data and outliers in the first target data, and normalizing the first target data after data cleaning to obtain second target data; The component relationship diagram is constructed according to the second target data.

3. The method according to claim 2, characterized in that Constructing the component relationship diagram according to the second target data includes: representing the elements and the sensors as the nodes of different colors; In the case where there is a physical connection relationship between the sensor and the component, connecting the sensor and the component through an edge; In the case where there is a logical causal relationship between the data collected by the sensors, the sensors are connected through edges; Based on the association between the data in the second target data and the nodes, configuring node attributes of each of the nodes, the node attributes including one or more of temperature data, water level data, pressure data, vibration data, and current data; The edge attributes of each edge are configured based on the node attributes corresponding to the nodes connected to the edge, and the edge attributes include one or more of Euclidean distance and cosine similarity.

4. The method according to claim 3, characterized in that The component relationship diagram is input into the target neural network to obtain fault detection results, including: Based on each of the nodes, feature extraction is performed respectively, and a neighbor node set corresponding to each of the nodes is determined, a message passing operation is performed on the neighbor node set through each convolutional layer of the target neural network in turn, and a feature of the node is updated based on the message passing operation to obtain a node aggregation feature; When the node aggregation characteristic exceeds a preset range, determining that the component corresponding to the node has a fault; When the node aggregation feature is within the preset range, it is determined that the component corresponding to the node does not have a fault.

5. The method according to claim 4, characterized in that Based on each of the nodes, feature extraction is performed respectively, and a neighbor node set corresponding to each of the nodes is determined, a message passing operation is performed on the neighbor node set through each convolutional layer of the target neural network in turn, and a feature of the node is updated based on the message passing operation to obtain a node aggregation feature, including: Initialize features based on the node attribute of the node to obtain a first feature, initialize features based on the edge attribute of the edge to obtain a second feature, and determine the first feature corresponding to the node in the neighbor node set as a third feature; In each convolution layer, the first feature, the second feature and the third feature are operated based on the message function to obtain the transfer message, and the first feature is operated through the activation function based on the transfer message to obtain the node aggregation feature.

6. The method according to claim 2, characterized in that The first target data is cleaned by using a statistical method and an anomaly detection algorithm to remove noise data and outliers in the first target data, and the first target data after data cleaning is normalized to obtain second target data, including: Analyze the first target data, extract missing data in the first target data, and fill in the missing data based on the mean or median of the same type of data when the missing data is numerical data, and fill in the missing data based on the mode of the same type when the missing data is categorical data; Analyze the first target data, extract extreme values ​​in the first target data, and replace the extreme values ​​based on upper limits or lower limits of data of the same type; Calculating a standard score of the first target data based on a Z-score algorithm, and determining the first target data as abnormal data when the standard score is greater than or equal to a first threshold; Screening the first target data based on an IQR algorithm to obtain the abnormal data; Clustering the first target data based on the DBSCAN algorithm, and determining the first target data corresponding to the outliers as the abnormal data; All the abnormal data in the first target data are deleted, and the first target data is normalized by using a Z-score algorithm to obtain the second target data.

7. The method according to claim 1, characterized in that After the component relationship diagram is input into the target neural network to obtain the fault detection result, the method includes: When the fault detection result indicates that the washing machine has a fault, a target alarm is issued, wherein the target alarm includes fault location and a fault elimination solution; When the fault of the washing machine has been eliminated, obtaining maintenance feedback data, the maintenance feedback data at least including a deviation between an actual fault location of the washing machine and the fault location in the target alarm; The target neural network is retrained based on the maintenance feedback data.

8. A washing machine fault detection device based on graph neural network, characterized in that: The device comprises: A first acquisition unit is used to acquire multi-source heterogeneous data in real time through a plurality of sensors installed on different components of the washing machine to obtain first target data, wherein the multi-source heterogeneous data includes at least one operating parameter of different components, and one component corresponds to at least one sensor; a construction unit, configured to construct a component relationship graph of the washing machine based at least on the first target data, wherein the nodes of the component relationship graph are the components and the sensors of the washing machine, and the edges between the nodes are used to represent the association relationship between the nodes; A detection unit is used to input the component relationship diagram into a target neural network to obtain a fault detection result. The target neural network is used to aggregate the feature information of the neighboring nodes of the node through multi-layer convolution, and perform fault detection based on the feature information of the aggregated nodes. The neighboring nodes are the nodes connected to the node through the edge.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. A washing machine, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of claims 1 to 7.

Images