Industrial device and its failure root cause diagnosis method and system

Abstract

The application provides an industrial equipment and a fault root cause diagnosis method and system thereof. The method is applied to the industrial equipment and comprises the following steps: using a preset causal discovery algorithm to process historical data of different physical quantities to obtain a causal diagram; using the causal diagram as a priori to train a causal perception fusion network to obtain a fault probability reasoning model; when current detection data of the industrial equipment is monitored, inputting the current detection data into the fault probability reasoning model to obtain a current fault probability; when the current fault probability reaches or exceeds a preset threshold, adjusting each detection node in sequence to obtain intervention data; recording a probability drop value corresponding to each detection node according to the intervention data, and marking a detection node with the largest probability drop value as a fault root cause; and generating a fault transmission path corresponding to the fault root cause according to the causal diagram. The application accurately locates the root cause through the causal priori, reduces the misjudgment rate, and further reduces the shutdown loss caused by misjudgment.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis technology, and in particular to an industrial equipment and a method and system for diagnosing the root causes of its faults. Background Technology

[0002] Currently, fault diagnosis for industrial equipment such as wind turbines and pumps generally employs single-source vibration signals combined with fixed thresholds or shallow models. This involves first collecting data from sensors, then comparing it with empirical alarm values ​​or simple classifiers to output a normal or abnormal status. However, existing fault diagnosis focuses on a binary judgment of "whether there is a fault," lacking a characterization of the causal relationships between various sensor variables. It struggles to locate the physical quantity that truly triggers the fault in multidimensional heterogeneous data, and cannot provide an interpretable fault evolution path. As a result, maintenance personnel still need to rely on manual experience for secondary troubleshooting, increasing downtime and maintenance costs. Summary of the Invention

[0003] This application provides an industrial equipment and its root cause diagnosis method and system. By training a fault probability inference model through a cause-effect graph, the physical quantity that induces the fault can be accurately located in the detection data corresponding to different sensors and an interpretable fault propagation path can be given, thereby reducing misjudgment and downtime for troubleshooting.

[0004] In a first aspect, embodiments of this application provide a method for diagnosing the root cause of faults in industrial equipment. The industrial equipment acquires detection data of different physical quantities through multiple sensors. The method includes: processing historical data of the different physical quantities using a preset causal discovery algorithm to obtain a causal graph, wherein the causal graph is used to characterize the causal relationship weights and causal directions between each detection node, and each detection node corresponds one-to-one with the detection data, with each detection node corresponding to a physical quantity collected by a corresponding sensor; using the causal graph as a priori, training a causal perception fusion network to obtain a fault probability inference model; when the current detection data of the industrial equipment is detected, inputting the current detection data into the fault probability inference model to perform fault probability inference to obtain the current fault probability; when the current fault probability reaches or exceeds a preset threshold, adjusting each detection node sequentially to obtain intervention data; recording the probability decrease value corresponding to each detection node according to the intervention data, and labeling the detection node with the largest probability decrease value as the root cause of the fault; generating a fault propagation path corresponding to the root cause of the fault according to the causal graph, wherein the fault propagation path is used to connect the root cause of the fault with a preset fault phenomenon in the order of the causal graph.

[0005] Secondly, embodiments of this application provide a root cause diagnosis system for industrial equipment. The industrial equipment acquires detection data of different physical quantities through multiple sensors. The system includes a causal graph generation module, a priori training module, a fault probability acquisition module, an intervention and adjustment module, a root cause acquisition module, and a path generation module. The causal graph generation module processes historical data of the different physical quantities using a preset causal discovery algorithm to obtain a causal graph. The causal graph characterizes the causal relationship weights and causal directions between each detection node. Each detection node corresponds one-to-one with the detection data, and each detection node corresponds to a physical quantity collected by a corresponding sensor. The priori training module uses the causal graph as a priori to train a causal perception fusion network. A fault probability inference model is trained. When the current detection data of the industrial equipment is detected, the fault probability acquisition module inputs the current detection data into the fault probability inference model to obtain the current fault probability. When the current fault probability reaches or exceeds a preset threshold, the intervention adjustment module adjusts each detection node sequentially to obtain intervention data. The root cause acquisition module records the probability decrease value corresponding to each detection node based on the intervention data and marks the detection node with the largest probability decrease value as the root cause of the fault. The path generation module generates a fault propagation path corresponding to the root cause of the fault based on the cause-effect graph. The fault propagation path connects the root cause of the fault with the preset fault phenomenon in the order of the cause-effect graph.

[0006] Thirdly, embodiments of this application provide an industrial device, which includes multiple sensors and a main control device. The multiple sensors are used to acquire detection data corresponding to different physical quantities. The main control device includes a memory and a processor. The memory is used to store computer programs. The processor is used to execute the computer programs to implement the above-mentioned fault root cause diagnosis method for the industrial device.

[0007] The aforementioned industrial equipment and its root cause diagnosis method and system construct a causal graph through a causal discovery algorithm, use the causal graph as a prior training causal perception fusion network to fuse various detection data to output fault probabilities, and intervene in each node sequentially and reason again when a fault is judged, record the probability decrease value and mark the largest one as the root cause, and generate a fault propagation path in reverse along the causal graph, so as to realize low false alarm, traceable real-time fault root cause location and fault propagation path output under operating condition drift, and reduce downtime troubleshooting costs. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0009] Figure 1 This is a first flowchart of a method for diagnosing the root cause of industrial equipment failures provided in an embodiment of this application.

[0010] Figure 2 A flowchart of step S101 provided for an embodiment of this application.

[0011] Figure 3 A flowchart of step S102 provided in the embodiments of this application.

[0012] Figure 4 A flowchart of step S103 provided in the embodiments of this application.

[0013] Figure 5 A flowchart of step S105 provided in the embodiments of this application.

[0014] Figure 6 A flowchart of step S106 provided in the embodiments of this application.

[0015] Figure 7 This is a second flowchart of a method for diagnosing the root cause of industrial equipment failures provided in an embodiment of this application.

[0016] Figure 8 This is a structural block diagram of the industrial equipment fault root cause diagnosis system provided in the embodiments of this application.

[0017] Figure 9 This is a structural block diagram of the industrial equipment provided in the embodiments of this application.

[0018] Figure 10 A schematic diagram of the internal structure of the main control device for the fault root cause diagnosis method of industrial equipment provided in the embodiments of this application.

[0019] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0021] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar planned objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data are interchangeable where appropriate; in other words, the described embodiments are implemented according to a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, may also include other content; 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.

[0022] It should be noted that the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of a person skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0023] Please refer to Figure 1 This is a first flowchart of a fault root cause diagnosis method for industrial equipment provided in this application. This application provides a fault root cause diagnosis method for industrial equipment. The industrial equipment 1000 acquires detection data of different physical quantities through multiple sensors 200. The industrial equipment 1000 in this application includes, but is not limited to, wind turbines, centrifugal compressors, industrial pumps, gearboxes, gas turbines, paper machines, rolling mills, mine hoists, or chemical reactors. The detection data of different physical quantities include, but are not limited to, vibration acceleration / velocity / displacement signals, process temperature, pressure, flow rate, rotational speed, torque, noise intensity, and motor electrical quantities. By executing the fault root cause diagnosis method, this application can accurately locate the physical quantity that induces the fault in the detection data of different physical quantities and generate an interpretable fault transmission path, significantly reducing the false positive rate and downtime for troubleshooting. The fault root cause diagnosis method includes steps S101-S105.

[0024] Step S101: Use a preset causal discovery algorithm to process historical data of different physical quantities to obtain a causal graph.

[0025] In step S101, the preset causal discovery algorithm is a Transformer-based causal discovery algorithm. Historical data refers to data on corresponding physical quantities read within a certain period before the industrial equipment 1000 experienced a fault shutdown, such as process temperature 6 months ago, pressure 3 months ago, etc. A causal graph is used to characterize the causal relationship weights and causal directions between each detection node. Each detection node corresponds one-to-one with the detection data, and each detection node corresponds to a physical quantity collected by a corresponding sensor. In this application, causal relationship refers to the following: in the historical data corresponding to each physical quantity, when the historical data of one physical quantity (e.g., oil temperature) changes first, it can significantly lead to a measurable change in the historical data of another physical quantity (e.g., spindle vibration), rather than simply simultaneous fluctuations. Causal relationship weights are used to characterize the intensity of the change in one physical quantity that causes a measurable change in another physical quantity. Causal direction is used to clarify which physical quantity changes cause a measurable change in other physical quantities. This application can use a causal discovery algorithm combined with the corresponding time series of historical data to examine the causal weight and causal direction of each causal relationship, thereby eliminating the correlation that only has statistical association but no causal relationship, and retaining the causal relationship that has a real driving effect on the evolution of the fault, so that the subsequent fault root cause localization can still maintain low false alarm and high interpretability under the condition drift.

[0026] Please refer to Figure 2 The flowchart below shows the sub-step S101 provided in the embodiments of this application. Steps S1011-S1014 involve processing historical data of different physical quantities using a preset causal discovery algorithm to obtain a causal graph.

[0027] Step S1011: Preprocess the historical data of each physical quantity.

[0028] In step S1011, preprocessing includes denoising, standardization, and alignment. This application performs denoising on historical data of different physical quantities to eliminate electromagnetic interference and mechanical shock glitches from the sensor 200; standardization is used to unify the dimensions of each physical quantity; and linear interpolation alignment is performed based on timestamps to ensure that historical data of all different physical quantities are on the same time grid, providing a clean, comparable, and synchronized input sequence for subsequent causal discovery.

[0029] Step S1012: Identify the corresponding historical data that changes with each historical data change after preprocessing, in order to obtain the causal relationship between different detection data.

[0030] In step S1012, a Transformer-based causal discovery algorithm is used to calculate the temporal lag information between any two detection nodes through a self-attention mechanism. Specifically, if the change in the historical data of one detection node at time (t-Δt) significantly precedes the change in the historical data of another detection node at time t, and time Δt is within the lag range allowed by device dynamics (e.g., 0.2-5 s), then a potential candidate causal relationship is determined between the two detection nodes, and the causal direction between the two detection nodes is characterized. This application uses a preset causal discovery algorithm to repeat the above scan on all detection node pairs to obtain potential candidate causal relationships between different detection data.

[0031] Step S1013: Quantify each causal relationship to obtain the causal relationship weight of the corresponding causal relationship.

[0032] In step S1013, a significance test is performed on each candidate causal relationship to calculate the causal relationship weight of the corresponding causal relationship, thereby characterizing the strength of the causal relationship between each detection node.

[0033] Step S1014: Remove the corresponding causal relationships whose causal relationship weights are lower than the preset weights, so as to obtain a causal graph based on the retained causal relationships.

[0034] In step S1014, by setting preset weights, causal relationships with weights lower than the preset weights are eliminated, retaining only those with significant causal strength. Simultaneously, the remaining causal relationships undergo directional consistency verification to eliminate bidirectional conflicts, ensuring that each retained causal relationship has a unique direction and reasonable physical meaning. This results in a weighted and directionally clear causal graph, providing a reliable causal framework for subsequent prior training and intervention reasoning. The following section will use a wind turbine as an example to illustrate how to obtain the causal graph.

[0035] First, historical data of various physical quantities within the six months prior to the wind turbine's failure and shutdown are retrieved, including main shaft vibration acceleration, gearbox oil temperature, high-speed shaft bearing temperature, generator winding temperature, impeller speed, active power, tower acceleration, and wind speed. Then, the retrieved historical data is denoised to eliminate electromagnetic interference and mechanical shock glitches from the sensors. Standardization of the physical quantities' dimensions is applied, and linear interpolation alignment is performed based on timestamps to ensure all historical data are on the same time grid. Next, the preprocessed historical data is input into a Transformer-based causal discovery algorithm. A self-attention mechanism is used to calculate the temporal lag information between any two physical quantities corresponding to the detection nodes, and potential candidate causal relationships between the detection nodes are identified. Then, a significance test is performed on each candidate causal relationship to calculate its weight, characterizing the strength of the causal relationship between the detection nodes. Finally, a preset weight is set to eliminate weak causal relationships with weights lower than the preset weight, and the remaining causal relationships undergo directional consistency verification to eliminate bidirectional conflicts, ensuring that each retained edge points uniquely and has a reasonable physical meaning. The final result is a weighted and directionally defined causal graph of the wind turbine.

[0036] Step S102: Using the causal graph as a priori, the causal perception fusion network is trained to obtain the fault probability inference model.

[0037] In step S102, the fault probability inference model is used to perform fusion inference on the detection data and output the fault probability. Historical data for each different physical quantity includes a first sample representing the normal state and a second sample representing the fault state. The Causal Awareness Fusion Network (CAFNet) is a multi-channel neural network that uses the weights of each edge (i.e., causal relationship) in the causal graph as attention priors. In the feature fusion stage, it prioritizes strengthening the signals of detection nodes with high weights and suppresses redundant features without causal support, thereby outputting a fusion result that combines physical interpretability and discriminative power. After obtaining the causal graph, this application maps the weights of each causal relationship to initial attention values, naturally biasing the subsequent fusion process towards detection nodes that have a driving effect on fault evolution. The Causal Awareness Fusion Network is trained end-to-end using the first and second samples, enabling it to automatically learn the differences between the normal and fault states of industrial equipment 1000 while preserving physical causal relationships, ultimately outputting the fault probability inference model.

[0038] Please refer to Figure 3 This is a flowchart of step S102 provided in the embodiments of this application. Steps S1021-S1022 include training the causal perception fusion network to obtain a fault probability inference model using a causal graph as a priori.

[0039] Step S1021: The weights between each detection node in the causal graph are used as attention priors to initialize the causal perception fusion network, so that the causal perception fusion network is used to strengthen the detection nodes with causal relationship weights and suppress the detection nodes without causal relationship weights.

[0040] In step S1021, the detection nodes with causal relationship weights are strengthened and the detection nodes without causal relationship weights are suppressed by the causal perception fusion network, so as to embed the physical evolution logic between the corresponding physical quantities with causal relationship into the causal perception fusion network and avoid the subsequent training being dominated by pseudo-correlation.

[0041] Step S1022: The initialized causal perception fusion network is trained using the first and second samples, so that the causal perception fusion network learns the feature distribution of each detection node under normal and fault states respectively, and obtains the fault probability inference model.

[0042] In step S1022, the initialized causal perception fusion network is trained end-to-end using first and second samples from different states. This allows the causal perception fusion network to clearly define the correspondence between detection nodes corresponding to different sensors 200, and to first weight and fuse multi-source features based on causal attention, before outputting the fault probability through a fully connected layer. It is understood that normal and fault states can be represented in the form of labels in their respective samples, and each is fed back through its corresponding causal relationship. This enables the fault probability inference model to learn to distinguish between the two states of industrial equipment 1000 while maintaining sensitivity to actual fault drivers, ultimately resulting in a fault probability inference model with low false alarms and high interpretability.

[0043] Understandably, steps S101-S102 can be performed offline. Specifically, historical data is first read in all at once, and after preprocessing (denoising, standardization, alignment), a weighted causal graph is generated using a Transformer-based causal discovery algorithm. This causal graph is then used as an attention prior to train a causal-aware fusion network. After training, the causal-aware fusion network is written to the on-site memory. Subsequent online diagnosis does not require repeating steps S101-S102, thus moving the time-consuming modeling process offline. This allows subsequent online diagnosis to focus solely on inference to output fault probabilities, significantly reducing diagnostic latency and computational overhead.

[0044] Step S103: When the current detection data of the industrial equipment is detected, the current detection data is input into the fault probability inference model to perform fault probability inference and obtain the current fault probability.

[0045] Please refer to Figure 4This is a flowchart of step S103 provided in the embodiments of this application. Inputting the current detection data into the fault probability inference model to perform fault probability inference and obtain the current fault probability includes steps S1031-S1033.

[0046] Step S1031: Map the current detection data to the causal perception fusion network according to the corresponding sensor.

[0047] In step S1031, based on the correspondence between the detection nodes corresponding to different sensors 200, each current detection data is mapped to the causal perception fusion network one by one to ensure that the subsequent weighted calculation is completely consistent with the training.

[0048] Step S1032: Use the causal perception fusion network to strengthen the detection nodes with causal relationship weights and suppress the detection nodes without causal relationship weights to obtain fused data.

[0049] In step S1032, the causal perception fusion network applies a preset amplification coefficient to the detection nodes whose causal relationship weights reach or exceed the preset weights according to the weights corresponding to each causal relationship obtained in step S102, and applies a preset suppression coefficient to the detection nodes whose causal relationship weights are less than the preset weights or have no causal relationship connections. Then, it performs weighted summation along the channel dimension and outputs fused data that retains key fault-driven information while suppressing redundant noise.

[0050] Step S1033: Perform fault probability inference on the fused data to obtain the current fault probability.

[0051] In step S1033, the fused features are activated by a fully connected layer and a Sigmoid function to output the fault probability at the current time.

[0052] Step S104: When the current fault probability reaches or exceeds the preset threshold, the detection nodes are adjusted sequentially to obtain intervention data.

[0053] In step S104, the intervention data of each detection node is used to examine the degree of influence of the corresponding detection node on the failure probability.

[0054] Step S105: Record the probability decrease value corresponding to each detection node based on the intervention data, and mark the detection node with the largest probability decrease value as the root cause of the fault.

[0055] Please refer to Figure 5 This is a flowchart of step S105 provided in the embodiments of this application. Steps S1051-S1054 include recording the probability decrease value corresponding to each detection node based on the intervention data.

[0056] Step S1051: Select any detection node as the target node, replace the current detection data corresponding to the target node with the preset reference data under normal conditions, and leave the other detection nodes unchanged to form intervention data.

[0057] In step S1051, any detection node is selected as the target node for intervention, and its current detection data is replaced with preset reference data under normal conditions to simulate a virtual scenario of restoring the physical quantity to the healthy range, ensuring that the intervention is a single variable and eliminating interference from other factors.

[0058] Step S1052: Input the intervention data and the remaining unchanged current detection data into the fault probability inference model to obtain the fault probability after intervention.

[0059] In step S1052, the fault probability inference model performs fault probability inference again based on the intervention data and the remaining unchanged current detection data, and outputs the fault probability after intervention to reflect the fault risk of industrial equipment 1000 when only the target node is adjusted, and then verifies one by one which physical quantity will have a significant impact on the fault risk of industrial equipment 1000.

[0060] Step S1053: Calculate the difference between the current failure probability and the failure probability after intervention to obtain the probability reduction value of the target node.

[0061] In step S1053, the greater the probability decrease value, the more significant the impact of the target node on the current fault, and the degree of fault of each physical quantity can be quantified accordingly.

[0062] Step S1054: Restore the intervention data to the corresponding current detection data before replacement, and select any detection node from the remaining detection nodes as the target node until the probability decrease value corresponding to each detection node is obtained.

[0063] In step S1054, the intervention data is first restored to the corresponding current detection data before the replacement. Then, S1051-S1053 are executed repeatedly on the remaining detection nodes to complete a single intervention for all detection nodes in sequence. The results are summarized to obtain a complete list of probability decrease values, which provides a quantitative basis for subsequently identifying the node with the largest decrease value as the root cause of the fault.

[0064] Step S106: Generate the fault propagation path corresponding to the root cause of the fault based on the cause-effect graph.

[0065] In step S106, the root cause of the fault may be that certain physical quantities measured by the sensor 200 exceed the numerical range of the industrial equipment 1000 under normal conditions. For example, the root cause may be that the gearbox inlet oil temperature exceeds 80°C, or the outer ring temperature of the main shaft bearing exceeds 65°C. The fault propagation path is used to connect the root cause of the fault with the preset fault phenomenon in a cause-effect diagram sequence, forming a detachable propagation chain from the root cause of the fault to the preset fault phenomenon. The preset fault phenomenon may be the phenomenon that occurs in the industrial equipment 1000 when different physical quantities are outside the preset reference data of the normal state, such as obvious noise in the high-speed shaft bearing or excessively high gearbox inlet oil temperature. For example, when the gearbox inlet oil temperature exceeds 80°C, the fault propagation path may be "excessively high gearbox inlet oil temperature → excessively low lubricating oil viscosity → excessively high vibration of the high-speed shaft bearing → excessively violent fluctuation of generator current → wind turbine generator failure".

[0066] Please refer to Figure 6 This is a flowchart of step S106, a sub-step provided in the embodiments of this application. Generating the fault propagation path corresponding to the root cause of the fault based on the cause-effect graph includes steps S1061-S1062.

[0067] Step S1061: Starting from the detection node identified as the root cause of the fault, search for downstream detection nodes in the cause-effect graph in the direction of decreasing causal relationship weight until the target node representing the preset fault phenomenon is reached, forming a root cause node sequence.

[0068] In step S1061, starting from the detection node identified as the root cause of the fault, the weighted causal relationship in the weighted causal graph is traversed downstream using a causal relationship weight decreasing priority strategy: each time, the causal relationship with the largest weight that has not been visited is selected, and the downstream detection nodes are pushed into the sequence in turn until the corresponding node of the corresponding preset fault phenomenon (such as unit shutdown, current over-limit, etc.) is encountered, and then the process stops, forming an ordered root cause node sequence from the root cause of the fault to the preset fault phenomenon, ensuring that the causal direction is consistent with the physical transmission order.

[0069] Step S1062: Connect the nodes in the root cause node sequence according to the search order, causal relationship weight and causal direction of the downstream detection nodes to generate a fault propagation path.

[0070] In step S1062, the root cause node sequence obtained in S1061 is connected according to traversal order, causal relationship weights, and causal direction to generate a weighted directed fault propagation path. The corresponding causal relationship weights can be directly mapped through visualization, such as path thickness and color intensity, to show the contribution of each intermediate node in the fault evolution process. This allows maintenance personnel to identify key propagation nodes and weak links at a glance, enabling rapid and interpretable troubleshooting guidance.

[0071] Please refer to Figure 7This is a second flowchart of the industrial equipment fault root cause diagnosis method provided in the embodiments of this application. After generating the fault propagation path corresponding to the fault root cause based on the cause-effect graph, the industrial equipment fault root cause diagnosis method further includes steps S201-S202.

[0072] Step S201: The root cause of the fault, its corresponding probability decrease value, and the fault propagation path are used as update samples to update the causal relationship weights between the corresponding detection nodes in the causal graph.

[0073] In step S201, the root cause nodes, probability decrease values, and fault propagation paths output from this round of diagnosis are integrated into an update sample to update the causal graph. Specifically, a preset attenuation coefficient is applied to the corresponding causal relationship weights of all causal relationships corresponding to the fault propagation paths in the causal graph. Then, the probability decrease value is accumulated to the corresponding causal relationship according to the node contribution ratio to form the causal relationship weights corresponding to the updated causal relationships. This allows the causal graph to continuously evolve with the aging of industrial equipment, load changes, and other operating conditions, avoiding misjudgments caused by the solidification of the fault probability inference model.

[0074] Step S202: Use the updated causal graph to train the causal perception fusion network to update the fault probability inference model, so that the updated fault probability inference model can be used for the next round of fault probability inference.

[0075] In step S202, the updated causal graph is used to regenerate the attention prior to train the current causal perception fusion network, resulting in a new fault probability inference model. After training, the new fault probability inference model is written to the field memory via hot replacement, seamlessly replacing the old fault probability inference model for the next cycle of fault probability inference and intervention re-inference, achieving online self-learning and performance self-improvement.

[0076] Please refer to Figure 8 This is a structural block diagram of the industrial equipment fault root cause diagnosis system provided in the embodiments of this application. This application also provides an industrial equipment fault root cause diagnosis system 10. The industrial equipment 1000 acquires detection data of different physical quantities through multiple sensors 200. The system 10 includes a cause-effect graph generation module 1, a priori training module 2, a fault probability acquisition module 3, an intervention and adjustment module 4, a root cause acquisition module 5, and a path generation module 6.

[0077] The causal graph generation module 1 is used to process historical data of different physical quantities using a preset causal discovery algorithm to obtain a causal graph. The causal graph is used to characterize the causal relationship weight and causal direction between each detection node. The detection node corresponds one-to-one with the detection data, and each detection node corresponds to the physical quantity collected by the corresponding sensor 200.

[0078] Prior training module 2 is used to train the causal perception fusion network with the causal graph as a prior to obtain the fault probability inference model.

[0079] When the current detection data of industrial equipment 1000 is detected, the fault probability acquisition module 3 is used to input the current detection data into the fault probability inference model to perform fault probability inference and obtain the current fault probability.

[0080] When the current failure probability reaches or exceeds the preset threshold, the intervention adjustment module 4 is used to adjust each detection node in sequence to obtain intervention data.

[0081] The root cause acquisition module 5 is used to record the probability decrease value corresponding to each detection node based on the intervention data, and to mark the detection node with the largest probability decrease value as the root cause of the fault.

[0082] The path generation module 6 is used to generate the fault transmission path corresponding to the fault root cause according to the cause-effect graph. The fault transmission path is used to connect the fault root cause and the preset fault phenomenon in the order of the cause-effect graph to form an explanatory transmission chain from the fault root cause to the preset fault phenomenon.

[0083] Please refer to Figure 9 This is a structural block diagram of the industrial equipment provided in the embodiments of this application. This application also provides an industrial device 1000. The industrial device 1000 includes multiple sensors 200 and a main control device 100. The multiple sensors 200 are respectively used to acquire detection data corresponding to different physical quantities. The main control device 100 can be a computer device.

[0084] Please refer to Figure 10 This is a schematic diagram of the internal structure of the main control device for the fault root cause diagnosis method of industrial equipment provided in the embodiments of this application. Figure 10 As shown, the main control device 100 includes a memory 901 and a processor 902. The processor 902 is used to run computer program instructions stored in the memory 901 to implement a root cause diagnosis method for industrial equipment faults.

[0085] The memory 901 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 901 can be an internal storage unit of a computer device, such as a hard disk. In other embodiments, the memory 901 can be an external storage device of a computer device, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., configured in the computer device. Furthermore, the memory 901 can include both internal and external storage units of a computer device. The memory 901 can be used not only to store application software and various types of data installed on the computer device, such as code for root cause diagnosis methods of industrial equipment, but also to temporarily store data that has been output or will be output.

[0086] Furthermore, the main control device 100 also includes a bus 903. The bus 903 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0087] Furthermore, the main control device 100 may also include a display component 904. The display component 904 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an organic light-emitting diode (OLED) touch screen, etc. The display component 904 may also be appropriately referred to as a display device or display unit, used to display information processed in the main control device 100 and to display a visual user interface.

[0088] Furthermore, the main control device 100 may also include a communication component 905. The communication component 905 may optionally include a wired communication component and / or a wireless communication component (such as a Wi-Fi communication component, a Bluetooth communication component, etc.), which is typically used to establish a communication connection between the main control device 100 and other computer devices.

[0089] Figure 10 Only a partial view of the main control device 100, which includes a method for diagnosing the root causes of industrial equipment failures, is shown. Those skilled in the art will understand that... Figure 10 The structure shown does not constitute a limitation on the main control device 100, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0090] In the above embodiments, the implementation can be achieved, in whole or in part, through software, hardware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, in the form of a computer program product.

[0091] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to embodiments of the present invention is generated. The computer device may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0092] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0094] The unit described as a separate component may or may not be physically separate. The component shown as a unit may or may not be a physical unit; that is, it may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0095] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist independently, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0096] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard disks, read-only storage media (ROM), random access storage media (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0097] In the above embodiments, a causal graph is constructed using the Transformer causal discovery algorithm. The causal perception fusion network is trained with the causal graph prior to fuse the output fault probabilities of each detection data. When a fault is determined, each node is intervened and reasoned in turn. The probability decrease value is recorded and the largest value is marked as the root cause. The fault propagation path is generated in reverse along the causal graph, so as to realize low false alarm, traceable real-time fault root cause localization and fault propagation path output under operating condition drift, thereby reducing downtime troubleshooting costs.

[0098] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

[0099] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0100] The above-listed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A method for diagnosing the root cause of faults in industrial equipment, characterized in that, The industrial equipment acquires detection data of different physical quantities through multiple sensors; the method includes: A causal graph is obtained by processing the historical data of the different physical quantities using a preset causal discovery algorithm. The historical data is the data of the corresponding physical quantities read within a certain period of time before the industrial equipment stops due to failure. The causal graph is used to characterize the causal relationship weight and causal direction between each detection node. The detection node corresponds one-to-one with the detection data, and each detection node corresponds to the physical quantity collected by the corresponding sensor. Using the causal graph as a priori, a fault probability inference model is obtained by training the causal perception fusion network. When the current detection data of industrial equipment is detected, the current detection data is input into the fault probability reasoning model to obtain the current fault probability. When the current failure probability reaches or exceeds a preset threshold, the detection nodes are adjusted sequentially to obtain intervention data. Record the probability decrease value corresponding to each detection node based on the intervention data, and mark the detection node with the largest probability decrease value as the root cause of the fault. Based on the cause-effect graph, a fault propagation path corresponding to the root cause of the fault is generated. The fault propagation path is used to connect the root cause of the fault with a preset fault phenomenon in the order of the cause-effect graph. The industrial equipment fault root cause diagnosis method is executed in an online or offline phase, except for the step of processing historical data of different physical quantities using a preset causal discovery algorithm to obtain a causal graph, and the step of using the causal graph as a priori to train a causal perception fusion network to obtain a fault probability inference model, which is executed in the offline phase. The preset causal discovery algorithm is a Transformer-based causal discovery algorithm; the causal graph is obtained by processing the historical data of the different physical quantities using the preset causal discovery algorithm, including: The historical data for each physical quantity is preprocessed, including denoising, standardization, and alignment. The corresponding historical data that changes with each preprocessed historical data is identified one by one to obtain the causal relationship between different detection data. Specifically, the temporal lag information between any two detection nodes is calculated through a self-attention mechanism. If the change of the historical data of one detection node at time (t-Δt) is significantly earlier than the change of the historical data of another detection node at time t, and time Δt is within the lag interval allowed by the device dynamics, then it is determined that there is a potential candidate causal relationship between one detection node and another detection node, and the causal direction between the one detection node and the other detection node is characterized. Each causal relationship is quantified to obtain the causal relationship weight of the corresponding causal relationship; The corresponding causal relationships with weights lower than the preset weights are removed, so as to obtain the causal graph based on the retained causal relationships. The retained causal relationships are then checked for directional consistency to eliminate bidirectional conflicts and ensure that each causal relationship points to a unique direction. The method further includes, after generating the fault propagation path corresponding to the root cause of the fault based on the cause-effect graph: The root cause of the fault and its corresponding probability decrease value, as well as the fault propagation path, are used as update samples to update the causal relationship weights between the corresponding detection nodes in the causal graph. The update specifically involves applying a preset attenuation coefficient to the corresponding causal relationship weights of all causal relationships corresponding to the fault propagation path, and then accumulating the probability decrease value to the corresponding causal relationship according to the node contribution ratio. The updated fault probability inference model is updated by training a causal perception fusion network using the updated causal graph, so that the updated fault probability inference model can be used for the next round of fault probability inference.

2. The fault root cause diagnosis method as described in claim 1, characterized in that, Historical data for each different physical quantity includes a first sample representing the normal state and a second sample representing the fault state; Using the causal graph as a priori, a fault probability inference model is obtained by training the causal perception fusion network, including: The causal relationship weights between each detection node are used as attention priors to initialize the causal perception fusion network, so that the causal perception fusion network is used to strengthen the detection nodes with causal relationship weights and suppress the detection nodes without causal relationship weights. The causal perception fusion network is trained using the first and second samples, so that the causal perception fusion network learns the feature distribution of each detection node in the normal state and the fault state, respectively, and the fault probability inference model is obtained.

3. The fault root cause diagnosis method as described in claim 2, characterized in that, The current detection data is input into the fault probability inference model to perform fault probability inference and obtain the current fault probability, including: The current detection data is mapped to the causal perception fusion network according to the corresponding sensors; The causal perception fusion network is used to strengthen detection nodes with causal relationship weights and suppress detection nodes without causal relationship weights to obtain fused data. The current fault probability is obtained by performing fault probability inference on the fused data.

4. The fault root cause diagnosis method as described in claim 1, characterized in that, Based on the intervention data, the probability decrease value corresponding to each detection node is recorded, including: Select any detection node as the target node, replace the current detection data corresponding to the target node with the preset reference data under normal conditions, and keep the other detection nodes unchanged to form the intervention data; The intervention data and the remaining unchanged current detection data are input together into the fault probability inference model to obtain the fault probability after intervention; Calculate the difference between the current failure probability and the failure probability after intervention to obtain the probability reduction value of the target node; The intervention data is restored to the corresponding current detection data before the replacement, and any detection node among the remaining detection nodes is selected as the target node until the probability decrease value corresponding to each detection node is obtained.

5. The fault root cause diagnosis method as described in claim 1, characterized in that, Generating the fault propagation path corresponding to the root cause of the fault based on the cause-effect graph includes: Starting from the detection node identified as the root cause of the fault, downstream detection nodes are searched sequentially in the causal graph along the direction of decreasing causal weight until the target node representing the preset fault phenomenon is reached, forming a root cause node sequence. According to the search order of the downstream detection nodes, the causal relationship weight, and the causal direction, the nodes in the root cause node sequence are connected to generate the fault propagation path.

6. A root cause diagnosis system for industrial equipment, characterized in that, The industrial equipment acquires detection data of different physical quantities through multiple sensors; the system includes: The causal graph generation module is used to process the historical data of the different physical quantities using a preset causal discovery algorithm to obtain a causal graph. The historical data is the data of the corresponding physical quantities read within a certain period of time before the industrial equipment stops due to failure. The causal graph is used to characterize the causal relationship weight and causal direction between each detection node. The detection node corresponds one-to-one with the detection data, and each detection node corresponds to the physical quantity collected by the corresponding sensor. The prior training module is used to train the causal perception fusion network with the causal graph as a prior to obtain a fault probability inference model. The fault probability acquisition module is used to input the current detection data of industrial equipment into the fault probability inference model to obtain the current fault probability when the current detection data is detected. The intervention adjustment module is used to adjust each detection node sequentially to obtain intervention data when the current fault probability reaches or exceeds a preset threshold. The root cause acquisition module is used to record the probability decrease value corresponding to each detection node based on the intervention data, and to mark the detection node with the largest probability decrease value as the root cause of the fault. The path generation module is used to generate a fault propagation path corresponding to the fault root cause based on the cause-effect graph. The fault propagation path is used to connect the fault root cause with the preset fault phenomenon in the order of the cause-effect graph. The industrial equipment fault root cause diagnosis method is executed in an online or offline phase, except for the step of processing historical data of different physical quantities using a preset causal discovery algorithm to obtain a causal graph, and the step of using the causal graph as a priori to train a causal perception fusion network to obtain a fault probability inference model, which is executed in the offline phase. The preset causal discovery algorithm is a Transformer-based causal discovery algorithm; the causal graph is obtained by processing the historical data of the different physical quantities using the preset causal discovery algorithm, including: The historical data for each physical quantity is preprocessed, including denoising, standardization, and alignment. The corresponding historical data that changes with each preprocessed historical data is identified one by one to obtain the causal relationship between different detection data. Specifically, the temporal lag information between any two detection nodes is calculated through a self-attention mechanism. If the change of the historical data of one detection node at time (t-Δt) is significantly earlier than the change of the historical data of another detection node at time t, and time Δt is within the lag interval allowed by the device dynamics, then it is determined that there is a potential candidate causal relationship between one detection node and another detection node, and the causal direction between the one detection node and the other detection node is characterized. Each causal relationship is quantified to obtain the causal relationship weight of the corresponding causal relationship; The corresponding causal relationships with weights lower than the preset weights are removed, so as to obtain the causal graph based on the retained causal relationships. The retained causal relationships are then checked for directional consistency to eliminate bidirectional conflicts and ensure that each causal relationship points to a unique direction. The method further includes, after generating the fault propagation path corresponding to the root cause of the fault based on the cause-effect graph: The root cause of the fault and its corresponding probability decrease value, as well as the fault propagation path, are used as update samples to update the causal relationship weights between the corresponding detection nodes in the causal graph. The update specifically involves applying a preset attenuation coefficient to the corresponding causal relationship weights of all causal relationships corresponding to the fault propagation path, and then accumulating the probability decrease value to the corresponding causal relationship according to the node contribution ratio. The updated fault probability inference model is updated by training a causal perception fusion network using the updated causal graph, so that the updated fault probability inference model can be used for the next round of fault probability inference.

7. An industrial device, characterized in that, The industrial equipment includes: Multiple sensors are used to acquire detection data for different physical quantities; and The main control device includes: Memory, used to store computer programs; and A processor for executing the computer program to implement the root cause diagnosis method for industrial equipment as described in any one of claims 1-5.

Images