Equipment fault prediction method and system and electronic equipment

Abstract

The invention relates to the field of fault prediction, in particular to an equipment fault prediction method and system and electronic equipment. The method comprises the following steps: preprocessing multi-modal equipment data of various types of equipment, and performing feature extraction to obtain a plurality of equipment features; inputting the equipment features into a preset prediction model, performing position coding and time coding on the equipment features by the prediction model to obtain corresponding position coding parameters and time coding parameters, and fusing the equipment features by the prediction model in combination with the position coding parameters and the time coding parameters, fusion features are obtained; and obtaining a fault prediction result of the equipment based on the fusion feature, the prediction model and a preset knowledge graph. The method and the device are used for improving the accuracy of equipment fault prediction.

Description

Technical Field

[0001] This invention relates to the field of fault prediction, and more specifically, to a method, system, and electronic device for predicting equipment faults. Background Technology

[0002] In modern industrial systems, there exist scenarios where multiple equipment units are tightly coupled and operate collaboratively, forming a dynamically interconnected whole. However, accurate fault prediction for these devices remains a significant challenge under current technologies. This is primarily due to the complex interactions between equipment, the variability and uncertainty of operating conditions, and the subtle and concealed nature of early fault symptoms, making it difficult for traditional fault prediction methods to provide effective early warnings. Furthermore, most existing fault prediction methods remain at the post-event alarm stage, lacking true pre-fault prediction capabilities. Fault prediction analysis often relies on a single type of data source and is mostly purely data-driven, resulting in poor interpretability of fault prediction models and insufficient generalization ability under complex and changing operating conditions, leading to high false alarm and false negative rates. Summary of the Invention

[0003] This invention provides a method, system, and electronic device for predicting equipment failures, which improves the accuracy of equipment failure prediction.

[0004] According to a first aspect of this application, a method for predicting equipment failure is provided, the method comprising: After preprocessing multimodal device data from various devices, feature extraction is performed to obtain several device features; The device features are input into a preset prediction model. The prediction model performs position encoding and time encoding on the device features to obtain corresponding position encoding parameters and time encoding parameters. The prediction model then combines the position encoding parameters and time encoding parameters to fuse the device features to obtain fused features. Based on the fusion features, the prediction model, and the preset knowledge graph, the fault prediction results of the device are obtained.

[0005] Understandably, by combining prediction models with location and time coding parameters, deep fusion of multimodal device data is achieved, comprehensively capturing the spatiotemporal correlation of device status. Based on fused features, prediction models, and knowledge graphs, fault prediction results for devices are obtained, thereby transforming the operation and maintenance mode from passive response to proactive prevention and intelligent decision-making, significantly improving operation and maintenance efficiency and reliability.

[0006] Optionally, after preprocessing the multimodal device data of various devices, feature extraction is performed to obtain several device features, including: Aligning multimodal device data from multiple devices temporally and spatially yields several aligned multimodal device data sets. Feature extraction is performed on the aligned multimodal device data to obtain a feature matrix, and the feature matrix is ​​used as the corresponding device feature.

[0007] Understandably, by strictly aligning multimodal device data in both time and space, the heterogeneity problem caused by different acquisition frequencies and locations of multimodal device data is resolved, providing a unified and consistent high-quality data foundation for subsequent analysis. The feature matrix extracted from aligned multimodal device data can more accurately characterize the state changes of devices in the spatiotemporal dimensions, effectively eliminating interference from noise and misalignment, significantly improving the reliability and consistency of device features. This lays a crucial data foundation for subsequent deep fusion and accurate fault prediction, thus ensuring the accuracy and stability of device feature analysis from the source.

[0008] Optionally, the prediction model is a spatiotemporal transformation Transformer model; the fused features for: in, Indicates query parameters, Indicates key parameters, The value parameter is defined by taking the device feature as the value parameter; wherein, , Obtained based on the aforementioned device features; Indicates the position encoding parameter. Indicates time encoding parameters; This indicates the preset scaling factor. This represents the normalized exponential function.

[0009] Understandably, by incorporating location and time encoding parameters into the fusion features, the prediction model can accurately model the complex dependencies of equipment features in physical space and time series. This ensures that the generated fusion features simultaneously contain the semantic information and spatiotemporal context of multimodal data, thereby significantly improving the richness and accuracy of the fusion feature expression. This allows subsequent fault prediction and analysis to be based on deep features that better conform to the actual operating rules of the equipment, enhancing the prediction model's ability to capture temporal evolution and spatially correlated fault modes.

[0010] Optionally, the multimodal device data includes visual data, operational data, vibration data, and environmental data; the method further includes training the prediction model, wherein the loss function of the prediction model during training is... for: Among them, the This represents the data fitting loss value calculated based on historical visual data, historical operational data, historical vibration data, and historical environmental data from multiple devices. This represents the physical constraint loss value calculated based on the historical operating data. This indicates the preset weighting coefficient.

[0011] Understandably, by integrating historical multimodal equipment data such as vision, operation, vibration, and environment, a comprehensive equipment status perception system has been constructed. By introducing physical constraints based on historical operating data into the loss function and jointly optimizing them with data fitting loss, the learning process of the prediction model not only matches real equipment data but also strictly adheres to physical conservation laws such as energy and mass. This significantly improves the generalization ability of the prediction model and the physical rationality of the prediction results, thereby achieving more reliable and interpretable fault detection under complex working conditions, increasing the fault detection rate, and reducing dependence on labeled data.

[0012] Optionally, the historical operating data includes the density of the historical circulating medium of the device. The historical flow rate of the circulating medium in the device The pressure of the historical flow medium of the device The dynamic viscosity of the historical circulating medium of the device The thermodynamic energy held by the historical flow medium of the device The heat source power of the energy conversion device in the aforementioned equipment. The historical operation data corresponding to the collection time ; The physical constraint loss value for: in, The pre-defined weights representing the conservation of mass Preset weights representing the conservation of momentum. The preset weights represent the conservation of energy.

[0013] Understandably, the physical constraint loss function directly embeds the three conservation laws of mass, momentum, and energy into the prediction model training using a differentiable mathematical model, injecting inviolable prior physical knowledge into the learning process. By forcing the prediction model to adhere to these fundamental physical laws during feature fusion and prediction, the physical rationality and consistency of the output fused features are greatly enhanced. This not only significantly improves the generalization ability and robustness of the prediction model under data scarcity or noise interference, but also ensures that the final fault prediction and analysis conclusions have a solid physical basis, greatly improving the credibility and interpretability of the prediction results.

[0014] Optionally, the preset mode of the knowledge graph is as follows: Obtain the equipment entity library based on several devices; Obtain the connection relationships between the aforementioned devices, and obtain a device relationship database based on the connection relationships; A library of explicit fault propagation paths for the aforementioned devices is established; wherein, the library of explicit fault propagation paths contains several explicit fault propagation paths between faults of various devices. Historical fault data of the aforementioned devices are obtained, and the historical fault data is processed to obtain a hidden fault propagation path library for the aforementioned devices; wherein, the hidden fault propagation path library contains several hidden fault propagation paths between faults of various devices. The knowledge graph is constructed based on the device entity library, the device relationship library, the explicit fault propagation path library, and the implicit fault propagation path library.

[0015] Understandably, by integrating device entities, connection relationships, explicit fault propagation paths, and implicit fault propagation paths, fragmented maintenance information is transformed into a structured knowledge network capable of reasoning. This allows the knowledge graph to not only statically describe the connections between devices but also dynamically simulate fault propagation paths across multiple devices. Based on this knowledge graph, a robust logical reasoning foundation is provided for subsequent root cause analysis, enabling the rapid localization of predicted fault phenomena to specific source devices and the generation of maintenance recommendations that align with the actual equipment conditions. This represents a qualitative leap from data association to causal inference.

[0016] Optionally, the fault prediction result includes at least one of the following: the predicted fault of the equipment in a preset future time period, the root cause analysis report corresponding to the predicted fault, and the maintenance suggestion corresponding to the predicted fault. The step of obtaining the fault prediction result of the device based on the fused features, the prediction model, and the preset knowledge graph includes: The predicted features are predicted using the prediction model to obtain the predicted faults of the device in the future time period. The fused features and the predicted faults of the device in the future time period are input into the knowledge graph for root cause query to obtain the fault propagation path corresponding to the predicted fault. Based on the predicted faults of the device in the future time period and the fault propagation path corresponding to the predicted faults, a root cause analysis report and / or maintenance recommendations corresponding to the predicted faults are generated. The fault prediction result of the device is generated based on the predicted faults of the device in the future time period and / or the root cause analysis report and / or the maintenance suggestions corresponding to the predicted faults.

[0017] Understandably, by leveraging fusion features and knowledge graphs, a complete intelligent decision-making loop is achieved, from accurate fault warnings to in-depth root cause diagnosis and maintenance strategy generation. This not only outputs highly reliable fault predictions in advance but also utilizes the logical reasoning capabilities of knowledge graphs to automatically trace fault propagation paths. Based on predicted faults and their corresponding propagation paths in future time periods, it generates root cause analysis reports and maintenance recommendations. This changes the disconnect between early warning, diagnosis, and handling in traditional operations and maintenance, providing operations and maintenance personnel with an integrated intelligent analysis report that can directly guide their actions. This significantly shortens the decision-making chain from problem discovery to solution development, and substantially improves operations and maintenance efficiency and accuracy.

[0018] Optionally, the step of using the prediction model to predict the fused features and obtaining the predicted fault of the device in the future time period includes: In the prediction model, the predicted state value of the device in the future time period is obtained based on the fusion features; The predicted probability of the predicted state value deviating from a preset confidence interval is obtained in the prediction model; In the prediction model, it is determined whether the predicted probability is greater than the prediction threshold. If so, the device is predicted to fail, thus obtaining the predicted failure of the device in the future time period.

[0019] Understandably, by mapping the fused features to predicted state values ​​for specific future time periods and performing probabilistic evaluation based on preset confidence intervals, the fault early warning system has achieved a leap from qualitative judgment to quantitative calculation, making the fault prediction results more refined and reliable, and significantly reducing the risk of false alarms. The output fused features are transformed into clear and actionable fault early warning signals, providing key decision-making basis for taking precise maintenance measures in advance, thereby effectively supporting the transformation of the operation and maintenance model from passive response to proactive prevention.

[0020] According to a second aspect of this application, a device failure prediction system is provided, the system comprising: The feature extraction module is used to extract features from multi-modal device data of various devices after preprocessing, and obtain several device features; The feature fusion module is used to input the device features into a preset prediction model. The prediction model performs position encoding and time encoding on the device features to obtain corresponding position encoding parameters and time encoding parameters, and then combines the position encoding parameters and time encoding parameters to fuse the device features to obtain fused features. The prediction module is used to obtain the fault prediction result of the device based on the fused features, the prediction model and the preset knowledge graph.

[0021] According to a third aspect of this application, an electronic device is provided, comprising: Memory, used to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements the device fault prediction method described in the first aspect above.

[0022] According to a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the device fault prediction method described in the first aspect above.

[0023] Based on any of the above aspects, the device fault prediction method, system, electronic device, and storage medium provided in this application embodiment can achieve the following benefits: By using a predictive model and combining location and time coding parameters, deep fusion of various equipment data in the spatiotemporal dimension can be achieved. This can fully capture the dependencies and dynamic changes of equipment status in time and space, making the fused features contain more comprehensive and coherent equipment information. This significantly improves the completeness and reliability of equipment feature expression, provides high-quality input for subsequent fault prediction, improves the fault detection rate, and reduces the false negative rate.

[0024] Introducing physical constraint loss based on the laws of conservation of mass, momentum, and energy into the training of the prediction model makes the learning process of the prediction model not only dependent on data fitting, but also guided by real physical laws. This enhances the physical rationality and consistency of the fused features, improves the generalization ability of the prediction model under complex working conditions, and makes the extracted fused features more consistent with the actual operating mechanism of the equipment. This improves the interpretability and robustness of the prediction model and helps to maintain high-precision prediction in diverse equipment failure prediction scenarios.

[0025] It can not only predict equipment failures in advance based on fused features, but also perform root cause analysis on predicted failures based on knowledge graphs, automatically generating diagnostic reports and maintenance suggestions that include the source of the failure and the propagation path. This transforms the operation and maintenance mode from post-event response to pre-event prevention, and upgrades from phenomenon identification to causal inference, thereby significantly improving the efficiency of fault diagnosis, reducing false alarms, reducing equipment downtime, shortening fault handling time, and realizing intelligent operation and maintenance decision-making from data-driven to data and knowledge-driven. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of a device fault prediction method provided in this embodiment.

[0028] Figure 2 This is a flowchart illustrating a preset method for a knowledge graph provided in this embodiment.

[0029] Figure 3 This embodiment provides a flowchart for obtaining fault prediction results.

[0030] Figure 4 This embodiment provides a flowchart for obtaining a predicted fault.

[0031] Figure 5 This is a schematic diagram of the functional modules of an equipment fault prediction system provided in this embodiment.

[0032] Figure 6 This is a schematic diagram of the structure of the electronic device provided in this embodiment. Detailed Implementation

[0033] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this application. To better illustrate the following embodiments, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0034] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0035] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0036] In modern industrial systems, due to the tight coupling of multiple devices, the variability of operating conditions, and the weak early fault characteristics, it is difficult to achieve accurate early warning using traditional technologies. In other words, existing methods are mostly post-event alarms, rely on single data, and have poor interpretability. Therefore, they lack generalization ability in complex device prediction scenarios, resulting in high false alarm and false negative rates.

[0037] This embodiment provides a technical solution that can solve the above problems. The specific implementation of this application will be described in detail below with reference to the accompanying drawings.

[0038] like Figure 1 As shown, this embodiment provides a method for predicting equipment failures, which can be further divided into the following steps: S100. After preprocessing the multimodal device data of various devices, feature extraction is performed to obtain several device features; It is understandable that in some industrial operating systems, the various devices do not operate independently; their operating states are necessarily correlated, and multiple devices together constitute a dynamically linked whole. For example, taking a pump house industrial system as an example, this application performs fault prediction on the equipment within that system. However, the method of this application is equally applicable to other industrial operating systems, and is not specifically limited here.

[0039] Understandably, with the acceleration of urbanization, the number of pumping stations for secondary water supply has increased rapidly, and the number of devices in these pumping stations that maintenance personnel need to maintain has also increased significantly. At the same time, the safe and stable operation of various devices within the pumping stations is crucial to water supply; therefore, equipment failure could trigger widespread water supply anomalies. Thus, rapid fault prediction and highly responsive maintenance of equipment within the pumping stations have become essential.

[0040] In this embodiment, the method of this application can accurately predict the equipment in the pump room and provide root cause analysis and maintenance suggestions for the failure, so as to avoid large-scale water supply anomalies as much as possible, reduce the operation and maintenance pressure, and thus solve the technical problems mentioned above.

[0041] It is understandable that a pump room may include various equipment such as water pumps, pressure gauges, flow meters, control cabinets, valves, and water tanks, all of which may malfunction, leading to abnormal water supply. In this embodiment, corresponding sensors or camera devices can be installed in the equipment to collect multimodal equipment data, serving as the data basis for predicting equipment malfunctions.

[0042] Specifically, the multimodal device data includes the device's visual data, operational data, vibration data, and environmental data.

[0043] In this embodiment, the visual data can be color images (RGB, Red, Green, Blue) and / or infrared images. The color images can be acquired using a camera installed within a first preset range of the device; the infrared images can be acquired using an infrared camera installed within the first preset range of the device. The operational data can be pressure data and / or flow data and / or temperature data of the medium flowing within a second preset range of the device during operation. The pressure data can be acquired from a pressure sensor installed within the second preset range of the device, the flow data can be acquired from a flow sensor installed within the second preset range of the device, and the temperature data can be acquired from a temperature sensor installed within the second preset range of the device. Preferably, the flowing medium can be water flowing through various devices in the pump room. The vibration data can be acceleration time-domain data and / or frequency-domain data collected during device operation. The acceleration time-domain data can be acquired from a vibration sensor installed on the device, and the frequency-domain data can be obtained by processing the time-domain signal acquired by the vibration sensor. The environmental data can be ambient temperature data and / or ambient humidity data and / or smoke concentration data and / or other gas concentration data within a third preset range of the device. The ambient temperature data can be obtained from a temperature sensor installed within a third preset range of the device. The smoke concentration data can be obtained from a smoke sensor installed within a third preset range of the device. The gas concentration data can be obtained from a gas sensor installed within a third preset range of the device.

[0044] In this embodiment, the multimodal device data corresponding to the various collected devices are preprocessed and feature extracted. The extracted device features are then input into the subsequent prediction model for fusion, thereby enabling multimodal data with different data formats to be converted into device features with the same format, thus improving the processing efficiency of the prediction model for input device features.

[0045] Specifically, the preprocessing of multimodal device data from various devices followed by feature extraction yields several device features, including: Aligning multimodal device data from multiple devices temporally and spatially yields several aligned multimodal device data sets. In this embodiment, since the data from each modal device is collected based on the corresponding sensor or camera, the data collected from multiple sensors will have slight differences in time and space. Therefore, it is necessary to align the multimodal device data in time and space. Preferably, in this embodiment, a conventional alignment method is used to align the multimodal device data, which will not be described in detail here.

[0046] Feature extraction is performed on the aligned multimodal device data to obtain a feature matrix, and the feature matrix is ​​used as the corresponding device feature.

[0047] In this embodiment, feature extraction is performed on the aligned multimodal device data. Specifically, feature extraction is performed on the aligned visual data to obtain a visual feature matrix, which is then used as the visual device feature. Feature extraction is performed on the aligned operational data to obtain an operational feature matrix, which is then used as the operational device feature. Feature extraction is performed on the aligned vibration data to obtain a vibration feature matrix, which is then used as the vibration device feature. Feature extraction is performed on the aligned environmental data to obtain an environmental feature matrix, which is then used as the environmental device feature. The visual device feature, the operational device feature, the vibration feature matrix, and the environmental device feature constitute the device feature.

[0048] S200. The device features are input into a preset prediction model. The prediction model performs position encoding and time encoding on the device features to obtain corresponding position encoding parameters and time encoding parameters. The prediction model then combines the position encoding parameters and time encoding parameters to fuse the device features to obtain fused features. For example, the device features, which consist of the visual device features, the operating device features, the vibration device features, and the environmental device features, can be input into the prediction model. The prediction model performs positional and temporal encoding on the visual device features, the operating device features, the vibration device features, and the environmental device features based on a pre-integrated self-attention mechanism to obtain positional encoding parameters and temporal encoding parameters. The device features are then fused by combining the positional encoding parameters and temporal encoding parameters to obtain fused features. This can more comprehensively capture the spatial structural relationships and temporal dynamic changes of device features. This not only optimizes the fusion alignment quality of multimodal device features but also improves the computational efficiency and prediction accuracy of the prediction model.

[0049] Preferably, the prediction model can be a Spatio-Temporal Transformer Model.

[0050] Specifically, the fusion feature for: in, Indicates query parameters, Indicates key parameters, The value parameter is defined by taking the device feature as the value parameter; wherein, , Obtained based on the aforementioned device features; Indicates the position encoding parameter. Indicates time encoding parameters; This indicates the preset scaling factor. This represents the normalized exponential function.

[0051] Specifically, the method further includes training the prediction model; In this embodiment, the prediction model needs to be trained in advance, and the trained prediction model is then used directly to predict equipment failures, thereby improving the accuracy and efficiency of the prediction.

[0052] Understandably, historical multimodal device data can be collected and labeled based on the historical operating conditions of the devices, thus using the labeled historical multimodal device data as a sample set. The prediction model is then trained based on this sample set and optimized using a preset loss function to obtain an optimized prediction model.

[0053] The loss function of the prediction model during training and optimization for: Among them, the This represents the data fitting loss value calculated based on historical visual data, historical operational data, historical vibration data, and historical environmental data from multiple devices. This represents the physical constraint loss value calculated based on the historical operating data. This indicates the preset weighting coefficient.

[0054] In this embodiment, the data fitting loss value can measure the deviation between the optimized predicted value and the true labeled value in the prediction model. By minimizing the data fitting loss value in the optimization, the prediction model can learn the accurate mapping relationship from multimodal device features to device state, ensuring that the prediction result is as close as possible to the actual labeled data.

[0055] Specifically, the data fitting loss value is calculated using the Mean Squared Error (MSE) method. Preferably, for a preset future time... Prediction, data fitting loss value It can be represented as: in, Indicates a preset future time period. This indicates the collection time corresponding to the historical operation data. Relative to The predicted device state values ​​at future moments. Relative to The actual device state value at a future moment. It can be calculated based on the device's historical visual data, historical operating data, historical vibration data, historical environmental data, and corresponding real labeled values.

[0056] It is understood that the historical operating data also includes the density of the historical circulating medium of the device. The historical flow rate of the circulating medium in the device The pressure of the historical flow medium of the device The dynamic viscosity of the historical circulating medium of the device The thermodynamic energy held by the historical flow medium of the device The heat source power of the energy conversion device in the aforementioned equipment. The historical operation data corresponding to the collection time ; Specifically, the physical constraint loss value for: in, The pre-defined weights representing the conservation of mass Preset weights representing the conservation of momentum. The preset weights represent the conservation of energy.

[0057] It is understood that when this embodiment is applied to a water supply pumping station, the circulating medium can be water. For example, the density of water is... The density can be obtained by consulting a standard substance density table; the water flow rate in the device can be obtained based on historical water flow data and the corresponding pipe area. Specifically, the water flow rate... The historical flow rate is calculated by dividing the pipe area. The historical flow rate is obtained through a flow sensor located within the second preset range of the device. The pipe area is the pipe area at the location where the flow sensor is installed, and is obtained by consulting a pipe specification table or design drawings. The historical water pressure is obtained through a pressure sensor located within the second preset range of the device. The dynamic viscosity of the water can be obtained by consulting a standard substance dynamic viscosity table. The historical thermodynamic energy held by the water is obtained using a conventional energy calculation formula, and the calculation parameters for the energy capacity calculation method can be obtained from the historical operating data of the device. The energy conversion device can be a water pump and / or heating equipment in a pump room, and the heat source power of the energy conversion device can be obtained by consulting the instruction manual of the energy conversion device.

[0058] In this embodiment, physical constraints are introduced into the loss function of the prediction model, specifically weighted conservation of mass, momentum, and energy in fluids. This embeds known physical laws or prior knowledge into the data-driven learning process, significantly improving the physical rationality and predictive reliability of the prediction model. This not only guides the model's output to conform to objective laws and avoids results that violate common sense, but also enhances the model's generalization ability and robustness in sparse or noisy environments. Simultaneously, physical constraints, as additional supervisory signals, reduce reliance on large-scale labeled data and promote the collaborative fusion of features among multimodal devices, enabling the prediction model to learn more fundamental relationships.

[0059] S300. Based on the fusion features, the prediction model, and the preset knowledge graph, obtain the fault prediction result of the device.

[0060] In this embodiment, the device features are fused based on the trained prediction model to obtain fused features, and the device faults are predicted based on the fused features and the preset knowledge graph, thereby obtaining the fault prediction results of each device. This can improve the situation where the device fails at a preset future time, thereby enabling the device to perform pre-fault prevention or troubleshooting, reducing the wide-ranging impact of the device failure, and improving the reliability of the device operation.

[0061] Understandably, the knowledge graph can describe the global connectivity between devices, providing prior knowledge of device topology information and fault causal relationships for device fault prediction. This enables the acquisition of root cause analysis reports corresponding to predicted faults based on the knowledge graph and fused features, thereby allowing maintenance personnel to better understand the occurrence of predicted faults and providing necessary guidance for maintenance and troubleshooting work.

[0062] Understandably, the knowledge graph needs to be pre-defined so that it can quickly obtain the fault propagation path in practical applications, thereby generating a root cause analysis report based on the fault propagation path and the predicted fault, thus improving the accuracy and efficiency of generating the root cause analysis report.

[0063] Specifically, such as Figure 2 As shown, the preset method of the knowledge graph may include the following steps: A1. Obtain the equipment entity library based on several devices; In this embodiment, the pump room contains numerous devices, each corresponding to a device entity in the knowledge graph. Collecting devices in the pump room that affect fault prediction as device entities provides a more comprehensive and accurate data foundation for subsequent root cause analysis reports. All device entities are aggregated to form a device entity library.

[0064] A2. Obtain the connection relationships between the devices, and obtain a device relationship database based on the connection relationships; In this embodiment, the devices are connected via pipes and / or electrical and / or dependent connections. It is necessary to obtain the connection relationships between the devices to make accurate root cause inferences in the subsequent root cause analysis report. For example, if the devices are connected via pipes, the corresponding connection relationship is: the first water pump is connected to the first water tank via a first pipe and a first valve; if the devices are connected via electrical connections, the corresponding connection relationship is: the control cabinet is electrically connected to the first water pump; if the devices are connected via dependent connections, the corresponding connection relationship is: the first pressure sensor depends on the first water pump, and the first water pump depends on the first pressure sensor to measure its pressure data. All connection relationships are summarized to form a device relationship database.

[0065] A3. Establish a library of explicit fault propagation paths for the aforementioned devices; wherein the library of explicit fault propagation paths contains several explicit fault propagation paths between faults of various devices. In this embodiment, the explicit fault propagation path library reflects the explicit propagation path rules of equipment faults. This library is based on historical data or expert experience. The explicit fault propagation path is a direct and easily understood causal chain of equipment faults, typically reflecting a clear fault transmission logic at the physical level or in the operational process, and is a direct cause of equipment faults. The explicit fault propagation path has high interpretability and is often used for preliminary fault diagnosis analysis. For example, the first explicit fault propagation path is: a sudden increase in the pressure of the flowing medium leads to stress concentration in the corresponding pipe, resulting in pipe rupture and ultimately water leakage. The second explicit fault propagation path is: blockage of the first water pump causes a decrease in the flowing medium at the pump outlet, leading to abnormal flow of the subsequent flowing medium and ultimately insufficient water supply. The third explicit fault propagation path is: jamming of the first valve causes a failure in its control, causing the control cabinet to issue abnormal overload commands to other equipment to ensure normal water supply, ultimately damaging some equipment. All explicit fault propagation paths are summarized to form the explicit fault propagation path library.

[0066] A4. Obtain historical fault data of the aforementioned devices, and process the historical fault data to obtain a hidden fault propagation path library for the aforementioned devices; wherein, the hidden fault propagation path library contains several hidden fault propagation paths between faults of various devices. In this embodiment, historical fault data of each device in the pump room needs to be collected in advance. Preferably, the historical fault data is input into a preset algorithm for processing to obtain hidden fault propagation paths in the devices. The preset algorithm can be a preset association rule mining algorithm and causal inference algorithm. In this embodiment, the preset algorithm can be set based on conventional mining and inference algorithms, which will not be elaborated here. The preset algorithm can identify non-intuitive or hidden device fault associations from historical fault data and obtain the hidden fault propagation paths based on the device fault associations. The hidden fault propagation paths can reveal indirect, multi-factor, or cross-device fault causal relationships, and these fault causal relationships often exceed conventional experience and cognition, but are of great significance and guiding significance for a deeper understanding of the root causes of complex device faults. For example, a hidden fault propagation path can be: excessive water pressure in the water pump caused the pipe to rupture because the second valve upstream of the water pump malfunctioned. All hidden fault propagation paths are summarized to form the hidden fault propagation path library.

[0067] Understandably, each hidden fault propagation path has a corresponding causal confidence level. This causal confidence level is updated in real time by learning the causal relationships of historical fault data through a preset algorithm. Specifically, the causal confidence level... The updated formula can be: Among them, Indicates the time when historical fault data was collected. This represents the causal confidence level at the previous moment of the latent fault propagation path. This indicates the preset learning rate. This represents the increment of causal confidence learned from historical fault data.

[0068] A5. Construct the knowledge graph based on the device entity library, the device relationship library, the explicit fault propagation path library, and the implicit fault propagation path library.

[0069] In this embodiment, a knowledge graph corresponding to the pump room is constructed based on the acquired equipment entity library, equipment relationship library, explicit fault propagation path library, and implicit fault propagation path library. Specifically, based on the explicit fault propagation path library and the implicit fault propagation path library, equipment fault entities and the relationships between each equipment fault entity are constructed. This allows the knowledge graph to visually display the causal relationships between equipment entities, between equipment fault entities, and between equipment entities, thereby providing equipment topology information and causal prior knowledge for fault prediction, enabling accurate subsequent root cause analysis.

[0070] Specifically, the fault prediction results include at least one of the following: predicted faults of the equipment in a preset future time period, root cause analysis reports corresponding to the predicted faults, and maintenance suggestions corresponding to the predicted faults. like Figure 3 As shown, obtaining the fault prediction result of the device based on the fused features, the prediction model, and the preset knowledge graph may include the following steps: S310. Use the prediction model to predict the fused features and obtain the predicted fault of the device in the future time period; In this embodiment, after obtaining the fused features based on the self-attention mechanism of the prediction model, the fused features are then predicted based on the prediction mechanism of the prediction model, thereby enabling the acquisition of the predicted faults of the device in the future time period. It is understood that in this embodiment, the predicted faults can be visualized and displayed to maintenance personnel in the form of tables or other data. Preferably, the predicted faults include one or more of the following: predicted fault level, name of the faulty device, warning time of the fault occurrence, predicted status value of the device, and fault confidence level. For example, in a fault prediction item, the predicted fault level can be high, medium, or low; the name of the faulty device can be the name of the pump room equipment; the warning time of the fault occurrence can be a future time relative to the current time; the predicted status value of the device can be a value output by the prediction model to represent the health status of the device; and the fault confidence level can be a value output by the prediction model to represent the reliability of the fault.

[0071] Specifically, such as Figure 4 As shown, using the prediction model to predict the fused features and obtain the predicted fault of the device in the future time period may include the following steps: S311. Obtain the predicted state value of the device in the future time period based on the fusion features in the prediction model; S312. Obtain the predicted probability that the predicted state value deviates from the preset confidence interval in the prediction model; S313. In the prediction model, determine whether the prediction probability is greater than the prediction threshold. If so, predict that the device will fail, and obtain the predicted failure of the device in the future time period.

[0072] Preferably, the predicted fault can be predicted based on the following formula: in, This indicates the acquisition time of the multimodal device data corresponding to the aforementioned device characteristics. Relative to The predicted device state values ​​at future moments. To pre-set the signal interval, The predicted probability is the deviation of the predicted state value from a preset confidence interval. The prediction threshold is used. If the predicted probability of the predicted state value deviating from the preset confidence interval is greater than the prediction threshold, it indicates that the device is predicted to malfunction, thus obtaining the predicted malfunction of the device in the future time period.

[0073] S320. Input the fused features and the predicted faults of the device in the future time period into the knowledge graph to perform root cause query and obtain the fault propagation path corresponding to the predicted fault. S330. Based on the predicted faults of the device in the future time period and the fault propagation path corresponding to the predicted faults, generate a root cause analysis report and / or maintenance recommendations corresponding to the predicted faults. In this embodiment, the fused features and the predicted faults of the device in the future time period are input into the knowledge graph for root cause querying to obtain the fault propagation path corresponding to the predicted fault. The fault propagation path can reflect the fault propagation path based on the predicted fault, and can also obtain the associated faults of other devices and the root causes caused by the predicted fault, thereby providing an effective data foundation for subsequent root cause analysis.

[0074] In this embodiment, a root cause analysis report for the predicted fault is generated based on the predicted fault of the device in a future time period and the corresponding fault propagation path. The root cause analysis report may include the symptoms of the fault, the root cause location of the fault, the causal effect quantification information of the fault, and the fault propagation path. The causal effect quantification information includes factors obtained through a knowledge graph and corresponding calculated causal effect values. The factors obtained through the knowledge graph include factors that the predicted fault of the device may cause, and the causal effect values ​​reflect the degree of influence of each factor.

[0075] For example, the symptoms of the fault could be an abnormal drop in the outlet pressure of the first water pump, reduced flow, and increased vibration. The root cause could be aging of the sealing ring of the first valve downstream of the first water pump, leading to internal leakage in the first valve and a drop in pressure in the related pipeline. The factors could include: a first factor: aging of the first valve sealing ring, with causal effect A; a second factor: abnormal output from the control cabinet, with causal effect B; and a third factor: pipeline blockage, with causal effect C. The fault propagation path is from the first valve to the first water pump to the control cabinet to the pipeline.

[0076] Preferably, the knowledge graph also needs to be updated periodically with fault causal relationships. Understandably, when a fault is predicted, the fault causal relationships and causal confidence levels related to the fault can be updated immediately; and / or the fault causal relationships and causal confidence levels related to the fault can be adjusted based on the predicted fault at preset short time intervals; and / or the knowledge graph can be re-optimized based on newly added historical fault data at preset long time intervals.

[0077] In this embodiment, maintenance suggestions are also required based on the fusion features and the predicted faults of the device in the future time period, so as to reduce the technical threshold for operation and maintenance personnel and improve the efficiency of operation and maintenance and troubleshooting.

[0078] S340. Generate a fault prediction result for the device based on the predicted faults of the device in the future time period and / or the root cause analysis report and / or the maintenance recommendations corresponding to the predicted faults.

[0079] In this embodiment, the device's fault prediction result can be generated by selecting one or more of the following according to actual needs: the predicted fault of the device in the future time period, the root cause analysis report corresponding to the predicted fault, and maintenance suggestions. Preferably, the predicted fault of the device in the future time period, the root cause analysis report corresponding to the predicted fault, and the maintenance suggestions can all be output as the device's fault prediction result, thereby providing maintenance personnel with comprehensive equipment fault prediction information and improving the accuracy, timeliness, and overall efficiency of fault prediction and fault maintenance.

[0080] like Figure 5 As shown in the illustration, this application also provides a device failure prediction system. Optionally, the system includes: Feature extraction module 411, feature fusion module 412, prediction module 413, wherein: The feature extraction module 411 is used to extract features from multi-modal device data of various devices after preprocessing, and to obtain several device features; In this embodiment, the feature extraction module 411 can be used to perform... Figure 1 For a detailed description of the feature extraction module 411, please refer to the description of step S100 shown.

[0081] The feature fusion module 412 is used to input the device features into a preset prediction model. The prediction model performs position encoding and time encoding on the device features to obtain corresponding position encoding parameters and time encoding parameters, and then combines the position encoding parameters and time encoding parameters to fuse the device features to obtain fused features. In this embodiment, the feature fusion module 412 can be used to perform... Figure 1 For a detailed description of the feature fusion module 412, please refer to the description of step S200 shown.

[0082] The prediction module 413 is used to obtain the fault prediction result of the device based on the fusion features, the prediction model and the preset knowledge graph.

[0083] In this embodiment, the prediction module 413 can be used to perform... Figure 1 For a detailed description of the prediction module 413, please refer to the description of step S300 shown.

[0084] Specifically, the prediction module 413 can be used to predict the fused features using the prediction model to obtain the predicted faults of the device in the future time period; perform root cause query to obtain the fault propagation path corresponding to the predicted fault; generate a root cause analysis report and / or a maintenance suggestion corresponding to the predicted fault based on the predicted faults of the device in the future time period and the fault propagation path corresponding to the predicted faults; and generate a fault prediction result of the device based on the predicted faults of the device in the future time period and / or the root cause analysis report and / or the maintenance suggestion corresponding to the predicted faults.

[0085] In this embodiment, the prediction module 413 can also be used to perform... Figure 3 For a detailed description of the prediction module 413, please refer to the description of steps S310-S340 shown.

[0086] Specifically, the prediction module 413 can be used to obtain the predicted state value of the device in the future time period based on the fusion features in the prediction model; obtain the prediction probability of the predicted state value deviating from a preset confidence interval in the prediction model; determine whether the prediction probability is greater than the prediction threshold in the prediction model, and if so, predict that the device will malfunction, thereby obtaining the predicted malfunction of the device in the future time period.

[0087] In this embodiment, the prediction module 413 can also be used to perform... Figure 4 For a detailed description of the prediction module 413, see steps S311-S313 shown below. For further details on steps S311-S313, please refer to the description of steps S311-S313.

[0088] The system further includes a knowledge graph pre-setting module, which is used to obtain a device entity library based on several devices; obtain the connection relationships between the several devices, and obtain a device relationship library based on the connection relationships; set an explicit fault propagation path library for the several devices; wherein the explicit fault propagation path library contains several explicit fault propagation paths between faults of various devices; obtain historical fault data of the devices, process the historical fault data to obtain a implicit fault propagation path library for the devices; wherein the implicit fault propagation path library contains several implicit fault propagation paths between faults of various devices; and construct the knowledge graph based on the device entity library, the device relationship library, the explicit fault propagation path library, and the implicit fault propagation path library.

[0089] In this embodiment, the knowledge graph preset module can be used to execute Figure 2 For a detailed description of the knowledge graph preset module shown in steps A1-A5, please refer to the description of steps A1-A5.

[0090] This application also provides an electronic device, the structure of which is as follows: Figure 6 As shown, the electronic device includes a memory 511, a processor 512, a communication module 513, and an input / output interface 514, etc. Optionally, the memory 511, the processor 512, the communication module 513, and the input / output interface 514 can be connected and communicate with each other through a bus 515.

[0091] The memory 511 is used to store one or more computer programs and to transfer the code of the computer programs to the processor 512; when the one or more computer programs are executed by the processor 512, a device fault prediction method in this application embodiment is implemented.

[0092] Optionally, the electronic device can be connected to a network via communication module 513 to exchange data with other devices, such as terminals or servers, through communication over the network. The electronic device can be various forms of digital computers, exemplarily such as desktop computers, servers, workbenches, mainframes, or other types of computers. The electronic device can also be various forms of mobile terminals, exemplarily such as smartphones, tablets, wearable devices (such as helmets, glasses, watches, etc.), and other similar mobile terminals.

[0093] Optionally, the electronic device can connect to desired input / output devices, such as a keyboard or display device, via the input / output interface 514. The electronic device itself may have a display device, and other display devices can also be connected externally via the input / output interface 514. Optionally, a storage device, such as a hard disk, can also be connected via the input / output interface 514 to store data from the electronic device, read data from the storage device, or store data from the storage device in the memory 511. It is understood that the input / output interface 514 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to the input / output interface 514 can be a component of the electronic device or an external device connected to the electronic device when needed.

[0094] Optionally, the memory 511 may be a volatile memory and / or a non-volatile memory. The volatile memory may be a random access memory, etc., and the non-volatile memory may be a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, or a flash memory, etc.

[0095] Optionally, the computer program stored in the memory 511 can be divided into one or more modules, which are stored in the memory 511 and executed by the processor 512 to perform the method provided in this embodiment. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device.

[0096] Optionally, the processor 512 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 512 include, but are not limited to, a central processing unit, a graphics processing unit, a digital signal processor, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, and can also be any suitable controller, microcontroller, processor, etc. The processor 512 executes the various methods and processes of this embodiment, exemplarily, such as a device fault prediction method according to an embodiment of this application.

[0097] Optionally, the bus 515 may include a path for transmitting information. Depending on its function, the bus 515 may be classified as an address bus, a data bus, a control bus, etc.

[0098] In an optional implementation, this application embodiment also provides a computer storage medium storing a computer program thereon, which, when executed by a computer, enables the computer to perform the methods described in the above-described method embodiments. Part or all of the computer program can be loaded and / or installed on the memory 511 of an electronic device. When the computer program is executed by the processor 512, one or more steps of a device fault prediction method according to an embodiment of this application can be performed.

[0099] Optionally, the computer-readable storage medium may be a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, etc.

[0100] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for predicting equipment failure, characterized in that, The method includes: After preprocessing multimodal device data from various devices, feature extraction is performed to obtain several device features; The device features are input into a preset prediction model. The prediction model performs position encoding and time encoding on the device features to obtain corresponding position encoding parameters and time encoding parameters. The prediction model then combines the position encoding parameters and time encoding parameters to fuse the device features, resulting in fused features. Based on the fusion features, the prediction model, and the preset knowledge graph, the fault prediction results of the device are obtained.

2. The method according to claim 1, characterized in that, The process of preprocessing multimodal device data from various devices and then extracting features yields several device features, including: Aligning multimodal device data from multiple devices temporally and spatially yields several aligned multimodal device data sets. Feature extraction is performed on the aligned multimodal device data to obtain a feature matrix, and the feature matrix is ​​used as the corresponding device feature.

3. The method according to claim 1, characterized in that, The prediction model is a spatiotemporal transformation Transformer model; the fused features for: in, Indicates query parameters, Indicates key parameters, The value parameter is represented by the device feature; wherein, , Obtained based on the aforementioned device features; Indicates the position encoding parameter. Indicates time encoding parameters; This indicates the preset scaling factor. This represents the normalized exponential function.

4. The method according to claim 1, characterized in that, The multimodal device data includes visual data, operational data, vibration data, and environmental data; the method further includes training the prediction model, wherein the loss function of the prediction model during training is... for: Among them, the This represents the data fitting loss value calculated based on historical visual data, historical operational data, historical vibration data, and historical environmental data from multiple devices. This represents the physical constraint loss value calculated based on the historical operating data. This indicates the preset weighting coefficients.

5. The method according to claim 4, characterized in that, The historical operating data includes the density of the historical circulating medium of the device. The historical flow rate of the circulating medium in the device The pressure of the historical flow medium of the device The dynamic viscosity of the historical circulating medium of the device The thermodynamic energy held by the historical flow medium of the device The heat source power of the energy conversion device in the aforementioned equipment. The historical operation data corresponding to the collection time ; The physical constraint loss value for: in, The pre-defined weights representing the conservation of mass Preset weights representing the conservation of momentum. The preset weights represent the conservation of energy.

6. The method according to any one of claims 1 to 5, characterized in that, The preset method for the knowledge graph is as follows: Obtain the equipment entity library based on several devices; Obtain the connection relationships between the aforementioned devices, and obtain a device relationship database based on the connection relationships; A library of explicit fault propagation paths for the aforementioned devices is established; wherein, the library of explicit fault propagation paths contains several explicit fault propagation paths between faults of various devices. Historical fault data of the aforementioned devices are obtained, and the historical fault data is processed to obtain a hidden fault propagation path library for the aforementioned devices; wherein, the hidden fault propagation path library contains several hidden fault propagation paths between faults of various devices. The knowledge graph is constructed based on the device entity library, the device relationship library, the explicit fault propagation path library, and the implicit fault propagation path library.

7. The method according to any one of claims 1 to 5, characterized in that, The fault prediction results include at least one of the following: predicted faults of the equipment in a preset future time period, root cause analysis reports corresponding to the predicted faults, and maintenance suggestions corresponding to the predicted faults. The step of obtaining the fault prediction result of the device based on the fused features, the prediction model, and the preset knowledge graph includes: The predicted features are predicted using the prediction model to obtain the predicted faults of the device in the future time period. The fused features and the predicted faults of the device in the future time period are input into the knowledge graph for root cause query to obtain the fault propagation path corresponding to the predicted fault. Based on the predicted faults of the device in the future time period and the fault propagation path corresponding to the predicted faults, a root cause analysis report and / or maintenance recommendations corresponding to the predicted faults are generated. The fault prediction result of the device is generated based on the predicted faults of the device in the future time period and / or the root cause analysis report and / or the maintenance suggestions corresponding to the predicted faults.

8. The method according to claim 7, characterized in that, The step of using the prediction model to predict the fused features and obtaining the predicted fault of the device in the future time period includes: In the prediction model, the predicted state value of the device in the future time period is obtained based on the fusion features; The predicted probability of the predicted state value deviating from a preset confidence interval is obtained in the prediction model; In the prediction model, it is determined whether the predicted probability is greater than the prediction threshold. If so, the device is predicted to fail, thus obtaining the predicted failure of the device in the future time period.

9. A device failure prediction system, characterized in that, The system includes: The feature extraction module is used to extract features from multi-modal device data of various devices after preprocessing, and obtain several device features; The feature fusion module is used to input the device features into a preset prediction model. The prediction model performs position encoding and time encoding on the device features to obtain corresponding position encoding parameters and time encoding parameters, and then combines the position encoding parameters and time encoding parameters to fuse the device features to obtain fused features. The prediction module is used to obtain the fault prediction result of the device based on the fused features, the prediction model and the preset knowledge graph.

10. An electronic device, characterized in that, include: Memory, used to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements a device failure prediction method as described in any one of claims 1-8.

Images