Complex electromechanical system anomaly detection method and device based on domain-adaptive causal decoupling network

Abstract

The present disclosure relates to the technical field of anomaly detection in electromechanical systems, and in particular to a method and device for anomaly detection in complex electromechanical systems based on a domain-adaptive causal decoupling network. The method comprises: obtaining a training data set and causal information of the electromechanical system, the training data set comprising system monitoring data and corresponding system fault label data, the causal information being used to indicate the functional relationship between multiple components of the electromechanical system; and training an anomaly detection model based on a domain-adaptive causal decoupling network based on the training data set and the causal information, the domain-adaptive causal decoupling network being used to indicate the corresponding relationship between system monitoring data and component-level health status representation, and the anomaly detection model being used to perform anomaly detection on the electromechanical system. The embodiment of the present disclosure combines domain adaptation and causal decoupling technologies to train a domain-adaptive anomaly detection model, so that the model can adapt to a changing system operating environment.

Description

Technical Field

[0001] The present disclosure relates to the technical field of abnormality detection of electromechanical systems, and in particular to a method and device for abnormality detection of complex electromechanical systems based on a domain-adaptive causal decoupling network. Background Art

[0002] Anomaly detection, a key technology designed to identify potential system failures, plays a vital role in enhancing the reliability and safety of industrial systems. With the deepening digital transformation of the industrial sector, massive amounts of condition monitoring data provide a rich source of material for data-driven anomaly detection methods.

[0003] Although traditional supervised learning techniques, such as Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Convolutional Neural Networks (CNN), have demonstrated excellent performance in multiple fields, most of these methods are based on a key assumption: that training data and real-time monitoring data should follow the principle of independent and identically distributed (IID). However, in the ever-changing real-world industrial environment, this idealized assumption is often difficult to achieve. Therefore, ensuring that anomaly detection models can maintain their high performance under dynamically changing operating conditions has become a major challenge in the field of anomaly detection in complex systems. Summary of the Invention

[0004] In view of this, the present disclosure proposes a complex electromechanical system anomaly detection method and device based on a domain-adaptive causal decoupling network.

[0005] According to one aspect of the present disclosure, a method for detecting anomalies in complex electromechanical systems based on a domain-adaptive causal decoupling network is provided, the method comprising:

[0006] Obtaining a training data set and causal information of a mechatronic system, wherein the training data set includes system monitoring data and corresponding system fault label data, wherein the system monitoring data includes multiple variables related to the health status of the mechatronic system, the system fault label data is used to indicate whether the mechatronic system is in a fault state, and the causal information is used to indicate the functional relationship between multiple components of the mechatronic system;

[0007] According to the training data set and the causal information, an anomaly detection model based on a domain-adaptive causal decoupling network is trained. The domain-adaptive causal decoupling network is used to indicate the correspondence between the system monitoring data and the component-level health status representation. The component-level health status representation is used to indicate the health status of multiple components of the electromechanical system. The anomaly detection model is used to perform anomaly detection on the electromechanical system.

[0008] In one possible implementation, the types of the multiple variables include at least one of system monitoring parameters, system operating condition variables and system operating modes, the system monitoring parameters include variables collected by a data collection device set at a specified position of the electromechanical system, the system operating condition variables include external condition variables and / or environmental variables of the electromechanical system, and the system operating mode is the operating mode adopted by the electromechanical system when realizing its functions.

[0009] In another possible implementation, the training of an anomaly detection model based on a domain-adaptive causal decoupling network according to the training dataset and the causal information includes:

[0010] Training the domain-adaptive causal decoupling network based on the training data set and the causal information;

[0011] Decoupling the component-level health status representation from the system monitoring data through the domain-adaptive causal decoupling network;

[0012] The anomaly detection model is determined according to the component-level health status representation.

[0013] In another possible implementation, the training of the domain-adaptive causal decoupling network based on the training dataset and the causal information includes:

[0014] constructing a domain-adaptive causal decoupling network based on the training data set and the causal information, the domain-adaptive causal decoupling network including a causal adjustment matrix and a coordinate rotation matrix, the causal adjustment matrix being used to indicate the influence of the system operating condition variables and / or system operating mode on the causal information, and the coordinate rotation matrix being used to match the coordinate system of the system monitoring parameters to the coordinate system of the component-level health status representation;

[0015] By minimizing the loss function, the causal adjustment matrix and the coordinate rotation matrix are optimized, and the domain adaptive causal decoupling network is obtained through training.

[0016] In another possible implementation, determining the anomaly detection model according to the component-level health status representation includes:

[0017] determining a domain-stable anomaly detection model based on the component-level health status representation, the domain-stable anomaly detection model being used to indicate a correspondence between the component-level health status representation and an anomaly detection result, the anomaly detection result being used to indicate whether the electromechanical system is in a fault state;

[0018] The anomaly detection model is determined according to the domain adaptive causal decoupling network and the domain stable anomaly detection model.

[0019] In another possible implementation, the anomaly detection model includes: the domain-adaptive causal decoupling network and the domain-stable anomaly detection model, the input parameter of the domain-adaptive causal decoupling network is the system monitoring data, and the output parameter is the component-level health status representation; the input parameter of the domain-stable anomaly detection model is the component-level health status representation, and the output parameter is the anomaly detection result.

[0020] In another possible implementation, the method further includes:

[0021] Acquiring real-time monitoring data of the electromechanical system during operation;

[0022] According to the real-time monitoring data, the pre-trained anomaly detection model is called to output an anomaly detection result, where the anomaly detection result is used to indicate whether the electromechanical system is in a fault state.

[0023] According to another aspect of the present disclosure, a complex electromechanical system anomaly detection device based on a domain-adaptive causal decoupling network is provided, the device comprising:

[0024] an acquisition module, configured to acquire a training data set and causal information of the electromechanical system, wherein the training data set includes system monitoring data and corresponding system fault label data, wherein the system monitoring data includes multiple variables related to the health status of the electromechanical system, the system fault label data is used to indicate whether the electromechanical system is in a fault state, and the causal information is used to indicate the functional relationship between multiple components of the electromechanical system;

[0025] A training module is used to train an anomaly detection model based on a domain-adaptive causal decoupling network based on the training data set and the causal information, wherein the domain-adaptive causal decoupling network is used to indicate the correspondence between the system monitoring data and the component-level health status representation, the component-level health status representation is used to indicate the health status of multiple components of the electromechanical system, and the anomaly detection model is used to detect anomalies in the electromechanical system.

[0026] In one possible implementation, the types of the multiple variables include at least one of system monitoring parameters, system operating condition variables and system operating modes, the system monitoring parameters include variables collected by a data collection device set at a specified position of the electromechanical system, the system operating condition variables include external condition variables and / or environmental variables of the electromechanical system, and the system operating mode is the operating mode adopted by the electromechanical system when realizing its functions.

[0027] In another possible implementation, the training module is further configured to:

[0028] Training the domain-adaptive causal decoupling network based on the training data set and the causal information;

[0029] Decoupling the component-level health status representation from the system monitoring data through the domain-adaptive causal decoupling network;

[0030] The anomaly detection model is determined according to the component-level health status representation.

[0031] In another possible implementation, the training module is further configured to:

[0032] constructing a domain-adaptive causal decoupling network based on the training data set and the causal information, the domain-adaptive causal decoupling network including a causal adjustment matrix and a coordinate rotation matrix, the causal adjustment matrix being used to indicate the influence of the system operating condition variables and / or system operating mode on the causal information, and the coordinate rotation matrix being used to match the coordinate system of the system monitoring parameters to the coordinate system of the component-level health status representation;

[0033] By minimizing the loss function, the causal adjustment matrix and the coordinate rotation matrix are optimized, and the domain adaptive causal decoupling network is obtained through training.

[0034] In another possible implementation, the training module is further configured to:

[0035] determining a domain-stable anomaly detection model based on the component-level health status representation, the domain-stable anomaly detection model being used to indicate a correspondence between the component-level health status representation and an anomaly detection result, the anomaly detection result being used to indicate whether the electromechanical system is in a fault state;

[0036] The anomaly detection model is determined according to the domain adaptive causal decoupling network and the domain stable anomaly detection model.

[0037] In another possible implementation, the anomaly detection model includes: the domain-adaptive causal decoupling network and the domain-stable anomaly detection model, the input parameter of the domain-adaptive causal decoupling network is the system monitoring data, and the output parameter is the component-level health status representation; the input parameter of the domain-stable anomaly detection model is the component-level health status representation, and the output parameter is the anomaly detection result.

[0038] In another possible implementation, the apparatus further includes: a calling module configured to:

[0039] Acquiring real-time monitoring data of the electromechanical system during operation;

[0040] According to the real-time monitoring data, the pre-trained anomaly detection model is called to output an anomaly detection result, where the anomaly detection result is used to indicate whether the electromechanical system is in a fault state.

[0041] According to another aspect of the present disclosure, a computing device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0042] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the above method is implemented.

[0043] According to another aspect of the present disclosure, a computer program product is provided, comprising a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of a computing device, the processor in the computing device executes the above method.

[0044] An embodiment of the present disclosure provides a method for detecting anomalies in complex electromechanical systems based on a domain-adaptive causal decoupling network. The method obtains a training data set and causal information of the electromechanical system, wherein the training data set includes system monitoring data and corresponding system fault label data. The system monitoring data includes multiple variables related to the health status of the electromechanical system. The system fault label data is used to indicate whether the electromechanical system is in a fault state, and the causal information is used to indicate the interaction relationship between multiple components of the electromechanical system. According to the training data set and the causal information, an anomaly detection model based on a domain-adaptive causal decoupling network is trained. The domain-adaptive causal decoupling network is used to indicate the correspondence between the system monitoring data and the component-level health status representation. The component-level health status representation is used to indicate the respective health status of multiple components of the electromechanical system. The anomaly detection model is used to detect anomalies in the electromechanical system. That is, by combining domain adaptation and causal decoupling technology, a domain-adaptive anomaly detection model can be trained so that the model can adapt to the changing system operating environment. The method can more accurately identify anomalies in the electromechanical system and maintain high accuracy even when the data distribution changes.

[0045] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0047] Figure 1 A flowchart of a complex electromechanical system anomaly detection method based on a domain-adaptive causal decoupling network provided by an exemplary embodiment of the present disclosure is shown.

[0048] Figure 2 A schematic diagram illustrating the cause-effect relationship of different types of variables in a mechatronic system provided by an exemplary embodiment of the present disclosure is shown.

[0049] Figure 3 A schematic diagram illustrating the impact of different system operation modes on the causal relationship structure provided by an exemplary embodiment of the present disclosure is shown.

[0050] Figure 4 A schematic diagram illustrating the influence of system operating condition variables on causal relationship strength under the same system operating mode provided by an exemplary embodiment of the present disclosure is shown.

[0051] Figure 5 A schematic diagram of the structure of a domain-adaptive causal decoupling network provided by an exemplary embodiment of the present disclosure is shown.

[0052] Figure 6A schematic diagram of a causal network between system monitoring parameters provided by an exemplary embodiment of the present disclosure is shown.

[0053] Figure 7 A schematic diagram of two-dimensional t-SNE projection of system monitoring parameters and YM provided by an exemplary embodiment of the present disclosure is shown.

[0054] Figure 8 A schematic diagram of a component-level health status representation and a two-dimensional t-SNE projection of YM provided by an exemplary embodiment of the present disclosure is shown.

[0055] Figure 9 A schematic structural diagram of a complex electromechanical system anomaly detection device based on a domain-adaptive causal decoupling network provided by an exemplary embodiment of the present disclosure is shown.

[0056] Figure 10 The present invention is a block diagram of an apparatus for executing a complex electromechanical system anomaly detection method based on a domain-adaptive causal decoupling network according to an exemplary embodiment. DETAILED DESCRIPTION

[0057] Various exemplary embodiments, features, and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0058] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0059] In addition, numerous specific details are provided in the following detailed description to better illustrate the present disclosure. Those skilled in the art will appreciate that the present disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main points of the present disclosure.

[0060] Ensuring that anomaly detection models maintain their performance under changing operating environments has become one of the key challenges facing anomaly detection in complex systems. To address this issue, transfer learning techniques have been introduced to adapt well-trained models to new domains, that is, new operating conditions. However, for complex systems, due to the lack of prior knowledge and the complex interactions between monitoring features, finding the most effective fine-tuning strategy is often a time-consuming and difficult task. In this context, domain adaptation methods based on feature mapping have become a widely considered solution. This method achieves model transfer by encoding data from the source and target domains into a unified latent feature space, making data from different domains difficult to distinguish in this space.

[0061] While traditional domain adaptation methods have achieved some success, they often overlook the diversity of system monitoring variables and the mechanisms of system failure (i.e., causal information). Current research suggests that directly using and encoding all raw monitoring variables can lead to latent feature spaces containing information that is difficult to generalize, resulting in suboptimal predictions in the target domain. Therefore, understanding the practical significance of different monitoring variables—that is, analyzing their specific impact on system state—becomes crucial.

[0062] Causal relationships can reveal the most fundamental causal relationships in observational data. Data-based causal discovery techniques have been used to explore the causal information of a system to improve the stability and interpretability of domain adaptation. These methods, through constructing joint causal inference frameworks and mechanism transfer, identify invariant conditional distributions across domains to accommodate different distributions caused by nonparametric shifts.

[0063] In response to the limitations of current research, the embodiments of the present disclosure propose a method for anomaly detection in complex electromechanical systems based on a domain-adaptive causal decoupling network, so that the anomaly detection model can adapt to the changing system operating environment. First, by constructing a general causal framework for complex systems, the different categories of system monitoring variables and system causal information are analyzed. Secondly, a domain-adaptive causal strength matrix is ​​proposed to characterize the impact of operating modes and conditions on the system causal information (structure and strength). Then, a domain-adaptive causal decoupling network is designed to learn a domain-adaptive representation of component-level health status representation. Finally, based on the obtained component-level health status representation, a domain-adaptive anomaly detection model is constructed. The embodiments of the present disclosure aim to decouple the complex interactions within complex electromechanical systems, reveal potential, domain-adaptive causal information, and thus improve the accuracy and interpretability of anomaly detection.

[0064] The disclosed embodiments are not only applicable to abnormality detection in complex electromechanical systems, but can also be extended to fault diagnosis of other complex systems such as electronic systems and software systems, as well as a wider range of complex classification task scenarios, and have good application prospects.

[0065] Below, several exemplary embodiments are used to introduce the complex electromechanical system anomaly detection method based on a domain-adaptive causal decoupling network provided by the embodiments of the present disclosure.

[0066] Please refer to Figure 1 , which shows a flowchart of a complex electromechanical system anomaly detection method based on a domain-adaptive causal decoupling network provided by an exemplary embodiment of the present disclosure. This embodiment uses the method applied to a computing device as an example. The method includes the following steps.

[0067] Step 101: Obtain a training data set and causal information of the electromechanical system. The training data set includes system monitoring data and corresponding system fault label data. The system monitoring data includes multiple variables related to the health status of the electromechanical system. The system fault label data is used to indicate whether the electromechanical system is in a fault state. The causal information is used to indicate the functional relationship between multiple components of the electromechanical system.

[0068] The training dataset includes system monitoring data of the electromechanical system in both normal and faulty states, along with corresponding system fault label data. System monitoring data refers to data collected from the electromechanical system, such as temperature, pressure, and vibration, which reflects the health of the system. System monitoring data can be collected in real time by data acquisition devices (such as sensors) and processed using data preprocessing techniques such as wavelet analysis, median filtering, and Kalman filtering to improve data quality and reduce the impact of noise and interference.

[0069] System fault label data is a label associated with system monitoring data, used to indicate whether the electromechanical system corresponding to the system monitoring data is normal or in a faulty state. These labels can be binary (faulty / no faulty) or multi-category (different types of faults). System fault label data can be derived from maintenance logs, fault records, or expert assessments, and is used to indicate whether each set of data corresponds to a system fault.

[0070] Causal information describes how components in a mechatronic system interact and influence each other. This information helps understand the causal relationships within the system. Causal information indicates the interactions between multiple components in a mechatronic system, also known as causal relationships. In other words, causal information indicates the structure and strength of the causal relationships between multiple components in a mechatronic system. Causal relationships between components in a mechatronic system can be determined through expert knowledge, historical data analysis, or causal inference techniques.

[0071] In some embodiments, multiple sets of original monitoring data and corresponding system fault label data are collected, and the multiple sets of original monitoring data and corresponding system fault label data are preprocessed to obtain a training data set; based on the training data set, the causal information of the electromechanical system is determined by a preset algorithm.

[0072] Step 102: Based on the training data set and causal information, an anomaly detection model based on a domain-adaptive causal decoupling network is trained. The domain-adaptive causal decoupling network is used to indicate the correspondence between system monitoring data and component-level health status representations. The component-level health status representations are used to indicate the health status of multiple components of the electromechanical system. The anomaly detection model is used to detect anomalies in the electromechanical system.

[0073] Using the collected training dataset and causal information, a computing device can train a domain-adaptive causal decoupling network. This network learns the correspondence between system monitoring data and component-level health status representations. Domain adaptation refers to the ability of a model to migrate from a source domain (training dataset) to a target domain (actual application environment), maintaining or improving its performance in the target domain. In this method, domain adaptation is achieved through data from different operating conditions included in the training dataset. This means that the model learns system behavior under different operating conditions during training, allowing it to adapt to the various operating conditions likely to be encountered in real-world applications. Causal decoupling decouples the complex causal relationships between system monitoring data and component-level health status to identify and represent the health status of individual components. By constructing a causal framework and a causal strength matrix, the model understands the interactions and influences between different components, enabling more accurate predictions of the health status of each component. This decoupling capability enables the model to adaptively adjust to changing operating conditions to maintain its prediction accuracy.

[0074] Component-level health status representation means that the model can generate a health status representation for each component of the mechatronic system. This representation can capture the health changes of the component under different operating conditions, enabling the model to perform more refined anomaly detection for specific components and specific conditions.

[0075] In practical applications, domain-adaptive causal decoupling networks can dynamically adjust their parameters and structure based on real-time monitoring data to adapt to current operating conditions. This dynamic adjustment capability is achieved through continuous learning or online learning mechanisms, allowing the model to continuously optimize its performance to adapt to changing environments.

[0076] Based on a domain-adaptive causal decoupling network, an anomaly detection model can be constructed. The purpose of this model is to monitor electromechanical systems and promptly detect abnormal behavior, such as whether the system is in a faulty state. During model training, various machine learning algorithms, such as random forests, support vector machines, and neural networks, as well as deep learning methods such as autoencoders and generative adversarial networks (GANs), can be used. Through domain adaptation and causal decoupling training, the model not only performs well on the training dataset but also generalizes well to unseen operating conditions. This generalization capability enables the model to accurately detect anomalies even when faced with new or changing operating conditions.

[0077] After model training is complete, it can be evaluated to ensure its effectiveness in real-world applications. Model performance can be assessed using metrics such as cross-validation, precision, and recall. Based on the evaluation results, the model may need to be fine-tuned to optimize its effectiveness in real-world mechatronic systems.

[0078] It should be noted that the implementation details of the model training process can be referred to the relevant description in the following embodiments and will not be introduced here.

[0079] In summary, the embodiments of the present disclosure provide a method for detecting anomalies in complex electromechanical systems based on a domain-adaptive causal decoupling network, by obtaining a training data set and causal information of the electromechanical system, the training data set includes system monitoring data and corresponding system fault label data, the system monitoring data includes multiple variables related to the health status of the electromechanical system, the system fault label data is used to indicate whether the electromechanical system is in a fault state, and the causal information is used to indicate the interaction relationship between multiple components of the electromechanical system; according to the training data set and causal information, an anomaly detection model based on a domain-adaptive causal decoupling network is trained, the domain-adaptive causal decoupling network is used to indicate the correspondence between the system monitoring data and the component-level health status representation, the component-level health status representation is used to indicate the respective health status of multiple components of the electromechanical system, and the anomaly detection model is used to perform anomaly detection on the electromechanical system; that is, by combining domain adaptation and causal decoupling technology, a domain-adaptive anomaly detection model can be trained, so that the model can adapt to the changing system operating environment, and the method can more accurately identify anomalies of the electromechanical system and maintain high accuracy even when the data distribution changes.

[0080] In some embodiments, the system monitoring data includes multiple variables related to the health status of the electromechanical system, the types of the multiple variables include at least one of system monitoring parameters, system operating condition variables, and system operating modes, the system monitoring parameters include variables collected by a data collection device set at a specified location on the electromechanical system, the system operating condition variables include external condition variables and / or environmental variables of the electromechanical system, and the system operating mode is the operating mode adopted by the electromechanical system when implementing its functions. Based on the training data set and causal information, an anomaly detection model based on a domain-adaptive causal decoupling network is trained, including: based on the training data set and causal information, a domain-adaptive causal decoupling network is trained; component-level health status representations are decoupled from the system monitoring data through the domain-adaptive causal decoupling network; and an anomaly detection model is determined based on the component-level health status representations.

[0081] That is, the complex electromechanical system anomaly detection method based on a domain-adaptive causal decoupling network can include four parts: causal information analysis of complex electromechanical systems, decoupled component-level health status representation, domain-adaptive causal decoupling network construction, and domain-adaptive anomaly detection model construction. First, the causal information of the complex electromechanical system is analyzed in detail, that is, the impact of the system operating mode and system operating condition variables on the causal relationship structure and strength between components. Considering that domain-adaptive causal information cannot usually be learned directly, the embodiment of the present disclosure proposes a domain-adaptive causal decoupling network. In the domain-adaptive causal decoupling network, the fault information transmitted between components is gradually stripped from the system monitoring parameters using a linear causal model to achieve the purpose of decoupling to obtain component-level health status representation. Based on the domain-adaptive causal decoupling network, a component-level health status representation that can adapt to the system operating conditions can be obtained. Based on the component-level health status representation, an adaptive and inherently interpretable anomaly detection model can be established using an interpretable machine learning model. These four parts are introduced in turn below.

[0082] Part I: Causal information analysis of complex electromechanical systems.

[0083] Assume that the training data set of the electromechanical system is D = [V, Y], V∈R N×K is the system monitoring data related to the health status of the electromechanical system, Y∈R N×1 is the system fault label data, that is, the record data of whether the electromechanical system is in a fault state. The training data set includes N samples, and the system monitoring data includes K variables, where N and K are both positive integers. In this embodiment, the system monitoring data includes multiple variables V = (V1, V2, ..., V K) is divided into three categories: system monitoring parameters η, system operating condition variables C and system operating mode M. Among them, system monitoring parameters refer to variables collected by data collection devices (such as sensors) set at designated locations of the electromechanical system based on expert experience. System operating condition variables refer to the collected system exogenous variables, that is, the external condition variables and / or environmental variables of the electromechanical system. The system operating mode refers to the operating mode (that is, operation or setting mode) when it realizes different functions, which is usually determined by relevant staff based on actual needs and experience. Multiple variables V can be arranged in the order of system monitoring parameters η, system operating condition variables C and system operating mode M, that is, η=(η1,η2,...,η H )=(V1,V2,...,V H ), C=(C1,C2,...,C Q )=(V H+1 ,V H+2 ,...,V H+Q ).

[0084] Schematically, based on the classification of variable categories, the causal relationship of different types of variables in the electromechanical system can be described as Figure 2 The general structure shown. The variables in the solid box (i.e., system monitoring parameters, system operating condition variables, and system operating mode) are usually observable and collectable, while the variables in the dotted box (i.e., the health status of each component) cannot be observed. Among them, the health status of each component directly determines the fault state of the complex electromechanical system and directly affects the system monitoring data. The system operating mode and system operating condition variables affect the system monitoring parameters by affecting the interaction relationship between components (i.e., the causal relationship structure and strength). Different system operating modes often correspond to different combinations and connection relationships of components within the system, so that the system can achieve different functions. Different system operating condition variables will affect the frequency and intensity of interactions between components. In addition, it is often impossible to determine that the system operating mode and system operating condition variables will necessarily affect all system monitoring parameters.

[0085] Taking the electromechanical system of a high-speed rail electromechanical control system as an example, the health status of each component is unobservable but directly affects the system's fault state. System operating condition variables, such as external temperature, humidity, required speed, and input voltage, represent the application environment monitored instantaneously by the electromechanical system. Therefore, they do not affect the health status of each component, but they do affect system monitoring parameters (such as line current and internal temperature). The system operating mode corresponds to the functions implemented by the electromechanical system. Train operators usually select different operating modes based on mileage, stations, and emergencies. Therefore, there is no causal relationship between them and the system application environment, and they do not affect the health status of each component. However, they do affect system monitoring parameters.

[0086] In some embodiments, the system monitoring parameters include multiple state monitoring values, and there is a one-to-one correspondence between the multiple state monitoring values ​​and the multiple components, that is, each state monitoring value is used to uniquely indicate the health status of a component, and the causal information of the electromechanical system is the causal relationship between the system monitoring parameters.

[0087] In a single system operation mode and system operating condition, it is assumed that the electromechanical system contains H' = H components and each component has only a unique state monitoring value, H is a positive integer greater than 1. The structural causal relationship between components is assumed to be a linear regression equation, and the component-level health status representation U' = [U1', U2', ..., U' H ] to simplify the causal mechanism, the causal information of the electromechanical system (i.e., the generation mechanism of the system monitoring parameter η) can be expressed as follows:

[0088]

[0089] Among them, A S is the system-wide causal strength matrix, which is used to indicate the interaction relationship between multiple components of the electromechanical system (i.e., causal relationship structure and causal relationship strength), and I is the unit matrix.

[0090] Under different system operation modes, the causal relationship structure between components usually changes. Figure 3 As shown in the figure, it shows the effect of different system operation modes on the causal relationship structure. The M1 system operation mode and the M2 system operation mode use different component combinations to achieve different system functions. Under the same system operation mode, different system operating condition variables will affect the causal relationship strength between components. Figure 4 As shown, it shows a schematic diagram of the influence of system operating condition variables on the causal relationship strength under the same system operating mode. (1) and system operating condition C (2) The strength of the interaction between the corresponding components is different. For example, different values ​​(such as 0.5, 0.6, 0.7, etc.) can be used to represent the strength of the causal relationship between the components. Figure 3 and Figure 4 The solid lines in the figure represent used components and structural relationships, and the dotted lines represent unused components and structural relationships.

[0091] Based on the above considerations, for a complex electromechanical system containing m system operation modes, the generation process of the system monitoring parameter η can be expressed as follows:

[0092]

[0093] in, It's M i The causal strength matrix between H components under the system operation mode is affected by the system operation condition variables. i is a positive integer ranging from 1 to m, and m is a positive integer greater than 1.

[0094] The second part is decoupling the component-level health status representation.

[0095] When the causal relationships between components are all linear, based on the causal information of the complex electromechanical system (i.e., the generation process of the system monitoring parameter η shown in the above formula), the process of decoupling the component-level health status representation U′ is as follows:

[0096]

[0097] In order to prevent dimensional changes from affecting the fault information in the data, it is generally desired that the representation learning process does not change the data dimension. Therefore, the embodiment of the present disclosure normalizes the above-mentioned formula for decoupling component-level health status representation as shown below:

[0098]

[0099] in, is a diagonal matrix whose j-th row and j-th column element is (Right now The sum of the elements in the jth row of ), and all other elements are 0. Called the Causal Intensity Matrix The degree matrix of .

[0100] The disclosed embodiments repeatedly use linear causal models to fit the nonlinear causal relationships that are prevalent in complex electromechanical systems, decoupling component-level health status representations from system monitoring parameters. Therefore, under nonlinear causal mechanisms, the component-level health status representation U′ can be gradually decoupled from the system monitoring parameter η using the following formula:

[0101]

[0102] Among them, U j ′ and U j ' +1 are the input and output parameters of the jth causal information decoupling process. The input parameter of the first causal information decoupling is the system monitoring parameter η, and the output parameter of the last causal information decoupling is the component-level health status representation U′.

[0103] It can be seen that the causal strength matrix of the system operating conditions under each system operating mode is the key basis for decoupling component-level health status representation. However, system monitoring data for complex electromechanical systems is often limited and biased, and the system operating conditions are complex and variable. Mining causal information for each system operating mode and system operating condition directly based on system monitoring data is unrealistic, difficult to achieve, and cannot guarantee the validity of the results. In addition, the causal information of the system under different system operating modes and system operating conditions also has correlations, and directly distinguishing and treating them separately also ignores these correlations. Therefore, directly decoupling component-level health status representation based on adaptive causal relationships is almost impossible to achieve.

[0104] Part III, construction of domain-adaptive causal decoupling network.

[0105] To comprehensively decouple the causal mechanisms in complex electromechanical systems, embodiments of the present disclosure design a domain-adaptive causal decoupling network to indirectly learn the impact of system operating conditions on causal information between components. In fact, this domain-adaptive causal decoupling network is a decoupled causal representation learning model, so named to distinguish it from other decoupling models. The proposed domain-adaptive causal decoupling network decouples the system operating mode into component-level health state representations, making the decoupled health state representations interpretable (separable) to the system operating mode. Furthermore, system operating condition variables are used as adjustment parameters for the overall system causal strength matrix, thereby improving the reliability of the causal strength matrix obtained. Here, the overall system causal strength matrix refers to the causal strength matrix obtained based on all modes and operating conditions. Analysis of actual application cases reveals that in most cases, the impact of system operating conditions on component action strength is relatively simple. Therefore, embodiments of the present disclosure use a single-layer neural network to fit the correspondence between system operating conditions and causal information. In some embodiments, the causal adjustment matrix affected by system operating condition variables is determined based on a sample weight matrix, system operating condition variables, a coefficient weight matrix for the system operating condition variables, and an adjoint weight matrix for the system operating condition variables. That is, according to the sample weight matrix, the system operating condition variables, the coefficient weight matrix of the system operating condition variables, and the adjoint weight matrix of the system operating condition variables, the causal adjustment matrix is ​​determined by a preset algorithm (such as an activation function). Schematically, the causal adjustment matrix W affected by the system operating condition variables is set C ∈R H×H as follows:

[0106] W C =Sigmoid(αCβ+γ)

[0107] Among them, W C It is used to indirectly find the causal intensity matrix Yes, W CIt is a linear matrix with unknown parameters, which is related to the overall causal intensity matrix A of the system. S Multiplication can be expressed as Part of the information, that is, the causal information affected by the working conditions. K×N Is the sample weight matrix, which is set to a full 1 matrix to adjust the data dimension and ensure that the data has the correct shape and size before entering the neural network. Q×K is the coefficient weight matrix of the system operating condition variable C. The coefficient weight matrix represents the influence of the system operating condition variable C on the strength of the causal relationship. It is a parameter matrix that is optimized through the learning process to best reflect the association between the operating condition variable and the strength of the causal relationship between components. γ∈R K×K It is the adjoint weight matrix of the system operating condition variable C. The adjoint weight matrix is ​​used to further adjust and optimize the output of the causal adjustment matrix. It can be regarded as a correction or enhancement to the output of the coefficient weight matrix. It is also a parameter matrix that needs to be optimized through the training process. Sigmoid() is the Sigmoid function, also known as the S-type function. The Sigmoid function is one of the commonly used activation functions in neural networks. Its definition is shown in the following formula:

[0108]

[0109] The characteristic of this function is that it can map the input value x to the interval (0,1), so that the output value has nonlinear characteristics, which is very important for neural networks to process complex data and pattern recognition.

[0110] Furthermore, the preceding assumptions all assume that the mechatronic system consists of H components, each with a unique condition monitoring value. Therefore, during the decoupling process, a coordinate rotation matrix should be set to align the coordinate system of the system monitoring parameters with the coordinate system of the component-level health status representation. This matrix adjusts the orientation and scale of the system monitoring parameters to ensure that they accurately reflect the component's health status.

[0111] Based on the above considerations, the schematic diagram of the domain-adaptive causal decoupling network is as follows: Figure 5 As shown in Figure 1, the domain-adaptive causal decoupling network consists of multiple causal decoupling layers, such as layer 1, layer 2, and layer J, where J is a positive integer greater than 1. Each layer is responsible for converting the coordinate system of system monitoring parameters to the coordinate system of component-level health status representation. The operation process of the jth causal decoupling layer is as follows:

[0112]

[0113] Among them, j is a positive integer, and the value range of j is 1 to J. S is the overall causal intensity matrix of the system; U′ j and U′j+1 are the input parameters and output parameters of the j-th causal decoupling process respectively; yes The degree matrix, which contains the measurement of each component condition monitoring value, H is the H-order identity matrix; W j ∈R H×H is the coordinate rotation matrix of the j-th causal decoupling layer, which is used to adjust the coordinate system.

[0114] Through multiple layers of causal decoupling, this domain-adaptive causal decoupling network gradually decouples system monitoring parameters down to the component level, enabling independent analysis of the health status of each component. The domain-adaptive causal decoupling network dynamically adjusts the system's overall causal strength matrix and coordinate rotation matrix to adapt to varying system states and environments based on changes in system operating conditions. Ultimately, the network's output vector provides a clear representation of component-level health status, facilitating system maintenance and fault diagnosis.

[0115] The optimization goal of the domain adaptive causal decoupling network is to make the decoupled causal representation vector interpretable to the system operation mode and fault state. Interpretability means that the output of the model can be understood and explained by humans. For the domain adaptive causal decoupling network, this means that the component-level health status representation vector output by the network should be able to clearly reflect the state of the system to distinguish different system operation modes and fault states, so that system engineers or analysts can understand how the health status of each component is related to the overall system performance. Schematically, with reference to the silhouette coefficient used in the clustering task, the loss function of the proposed domain adaptive causal decoupling network is Loss DA It can be:

[0116] Loss DA =-silhouette_score(U′,YM)

[0117] Among them, YM is a variable obtained by label encoding the combination of system fault label data Y and system operation mode M, indicating how many forms of combinations of Y and M there are; silhouette_score() is a function that calculates the average silhouette coefficient.

[0118] In some embodiments, a domain-adaptive causal decoupling network can be constructed based on a training data set and causal information. The domain-adaptive causal decoupling network includes a causal adjustment matrix and a coordinate rotation matrix. The causal adjustment matrix is ​​used to indicate the impact of system operating condition variables and / or system operating modes on causal information, and the coordinate rotation matrix is ​​used to match the coordinate system of system monitoring parameters to the coordinate system of component-level health status representation. By minimizing the loss function, the causal adjustment matrix and the coordinate rotation matrix are optimized to obtain a domain-adaptive causal decoupling network through training.

[0119] That is, the embodiment of the present disclosure proposes a trainable parameter in the domain adaptive causal decoupling network, namely the causal adjustment matrix W in the above formula C and the coordinate rotation matrix W j is optimized in sequence. First, all W j It is set to the unit matrix (ie, no coordinate rotation is performed), and the causal condition matrix W is optimized by minimizing the loss function C (Actually, this is to obtain the coefficient weight matrix β and the accompanying weight matrix γ). Secondly, based on the trained W C , W j The optimization is also achieved by minimizing the loss function. In addition, since the adaptive causal decoupling network in the field of graph convolution operation also has strong feature extraction performance, that is, it can efficiently decouple causal information, the model usually contains a maximum of three layers. When optimizing the model structure, gradually increase the number of layers and train W in sequence. C and W j Schematically, when the training fit accuracy is less than a preset accuracy threshold, the model structure layer number is increased, and when the training fit accuracy is greater than or equal to the preset accuracy threshold, the training stops. The training goal is to minimize the loss function as much as possible to obtain a component-level health status representation U′ that is more explainable to system faults and operating modes and can adapt to system operating conditions.

[0120] Part 4, construction of domain adaptive anomaly detection model.

[0121] In some embodiments, determining an anomaly detection model based on a component-level health status representation includes: determining a domain-stable anomaly detection model based on the component-level health status representation, the domain-stable anomaly detection model being used to indicate a correspondence between the component-level health status representation and anomaly detection results, the anomaly detection results being used to indicate whether the electromechanical system is in a fault state; and determining an anomaly detection model based on a domain-adaptive causal decoupling network and the domain-stable anomaly detection model. Illustratively, based on the domain-adaptive component-level health status representation U′, the domain-stable anomaly detection model can be defined as follows:

[0122]

[0123] Considering that the optimization goal of component-level health status representation is to cluster samples in the data [U′, YM] according to different YMs, the resulting component-level health status representation has high separability, eliminating the need for complex machine learning models to fit the mapping relationship between it and the system fault label data Y. To ensure that anomaly detection models for complex electromechanical systems are inherently interpretable, domain-stable anomaly detection models based on component-level health status representation, as shown in the above formula, can typically use simple, interpretable machine learning models such as logistic regression and decision trees. Furthermore, the relationship between component-level health status representation and system faults is generally stable. That is, when a component degrades to a certain threshold, the corresponding system fault holds true under different operating conditions. Therefore, domain-stable anomaly detection models based on component-level health status representation are domain-stable. Therefore, the adaptive anomaly detection model is derived by combining two models: a domain-adaptive causal decoupling network based on system monitoring data and a domain-stable anomaly detection model based on component-level health status representation. That is, the anomaly detection model includes: a domain-adaptive causal decoupling network and a domain-stable anomaly detection model. The input parameters of the domain-adaptive causal decoupling network are system monitoring data, and the output parameters are component-level health status representations; the input parameters of the domain-stable anomaly detection model are component-level health status representations, and the output parameters are anomaly detection results.

[0124] After training the anomaly detection model, the application process of the anomaly detection model may include: obtaining real-time monitoring data of the electromechanical system during operation; calling the pre-trained anomaly detection model based on the real-time monitoring data, and outputting an anomaly detection result, which is used to indicate whether the electromechanical system is in a fault state. Calling the pre-trained anomaly detection model based on the real-time monitoring data and outputting the anomaly detection result may further include: pre-processing the real-time monitoring data to obtain pre-processed real-time monitoring data; inputting the pre-processed real-time monitoring data into the pre-trained anomaly detection model; and outputting the anomaly detection result. The anomaly detection result is typically a probability value or classification label indicating whether the electromechanical system is in a normal or faulty state.

[0125] In some embodiments, the anomaly detection model can be used to interpret and make decisions based on the anomaly detection results. If an anomaly is detected, appropriate measures may need to be taken, such as issuing an alarm, automatically shutting down the system, or performing maintenance. In some embodiments, the anomaly detection model can be evaluated and iteratively improved based on actual system performance and maintenance records. This may include retraining the model to improve its accuracy and robustness. In some embodiments, the anomaly detection model is integrated into the monitoring system of the electromechanical system to achieve automated anomaly detection and response.

[0126] In an illustrative example, the effectiveness of the proposed adaptive anomaly detection model is analyzed and verified using a high-speed rail electromechanical control system. To verify the model's domain adaptation performance, the observed data from the high-speed rail electromechanical control system is first divided into three sets: a training set, test set 1, and test set 2. Test set 1 and test set 2 consist of the top 10% of train speed and traction samples, respectively, while the remaining samples serve as the training set.

[0127] The high-speed train brake control system comprises pneumatic and electrical equipment, achieving diverse functions through the integration of various devices using microelectronics technology. Numerous factors can influence the health of the system, including train operating conditions, brake control system operating conditions, system component status, and environmental factors. To comprehensively monitor these factors, sensors were installed at key locations based on expert experience to record operational and environmental variables. Specifically, monitoring tools were installed at key locations in the electrical system to collect information such as voltage, current, and temperature. Sensors were installed on key components of the mechanical system to collect vibration signals, temperature, and power. State variables displayed by the software system, such as braking mode, traction level, train speed, position, and operating time, as well as environmental variables such as external temperature, humidity, and weather, were also recorded. The collected raw monitoring data includes 43 variables potentially related to system status, such as train speed, wind speed, line voltage, line current, internal temperature, external temperature, braking mode, and braking level. For ease of illustration, these variables are simply denoted as X1, X2, ..., X43.

[0128] Before causal discovery and anomaly modeling, the raw monitoring data needs to be preprocessed. First, meaningless variables (such as unique value variables and ID information variables) are detected, and then character variables are converted to numerical variables using one-hot encoding. Data samples with missing values ​​are deleted, and finally all monitoring variables are normalized using z-score standardization. After data preprocessing, the resulting dataset contains 73,007 samples and 38 monitoring variables. These monitoring variables include two system operating condition variables, two operating mode variables, and four environmental factors. The remaining 30 monitoring variables are system monitoring parameters, denoted as V1, V2, ..., V30.

[0129] First, it is necessary to mine causal information about the brake control system. This involves exploring the causal relationships between the system's monitoring parameters based on a training dataset under different operating modes and conditions. Due to limited expert and physical knowledge, and the inability to conduct randomized controlled trials, data-based causal discovery is employed to reveal the causal relationships between the system's monitoring parameters.

[0130] Considering the potential confounding factors and selection bias that are common in complex system monitoring data, a fast causal inference (FCI) algorithm can be used. For example, the FCI function can be used to easily obtain the causal network between system monitoring parameters.

[0131] In order to improve the accuracy and stability of the causal discovery results, some prior knowledge is added to the FCI function through the corresponding parameters. Schematically, the causal network between the system monitoring parameters is shown in the figure below. Figure 6 This network intuitively demonstrates the causal relationship between multiple system monitoring parameters, providing an important basis for further analysis and optimization of the braking control system.

[0132] Based on the training data set, a domain-adaptive causal decoupling network is constructed, by gradually increasing the number of causal decoupling layers and training W in sequence. C and W j , minimizing the loss function and decoupling the component-level health status representation U′ from the system monitoring parameters. In this case, the resulting domain-adaptive causal decoupling network contains two causal decoupling layers.

[0133] The proposed domain-adaptive causal decoupling network can decouple component-level health status representations from system monitoring parameters. Based on the causal information of complex electromechanical systems, the fault information transmitted between the system monitoring parameters is sequentially extracted and decomposed to obtain the health status representation of each component. The decoupling process is interpretable and reliable. Furthermore, the decoupling process can learn the influence of system operating condition variables on the causal strength matrix between system monitoring parameters, and obtain the mapping function of system operating condition variables to the causal strength matrix, thereby explaining the influence of operating conditions on system monitoring parameters and improving the interpretability of system monitoring data.

[0134] Indicative, Figure 7 A schematic diagram showing the two-dimensional t-distributed Stochastic Neighbor Embedding (t-SNE) projection of system monitoring parameters and YM is shown. Figure 8A schematic diagram of the component-level health status representation and the two-dimensional t-SNE projection of YM is shown. Among them, YM is a variable obtained by label encoding the combination of system fault label data Y and system operation mode M. t-SNE is a nonlinear dimensionality reduction technique for high-dimensional data visualization. It displays data points in two-dimensional or three-dimensional space by maintaining the local structure in the high-dimensional space. As can be seen from the figure, the component-level health status representation is clustered under different system operation modes and system fault label data and has high separability, which means that the component-level health status representation can explain (distinguish) different system operation modes and fault states. Since the component-level health status representation has good separability, a classifier with inherent interpretability can be used to establish an anomaly detection model. Here, a single-layer perceptron with a Relu activation function is used to establish an anomaly detection model based on the component-level health status representation, which has strong interpretability.

[0135] The proposed domain-adaptive causal graph convolutional anomaly detection model, denoted as "Domain-Adaptive DWGCN," compares the training and testing performance of different models. The methods compared with the domain-adaptive DWGCN include commonly used data-driven anomaly detection models, namely support vector machines (SVMs), artificial neural networks (ANNs), and convolutional neural networks (CNNs), as well as the classic domain adaptation (DA) method based on multi-layer perceptrons (MLPs). The classic domain adaptation method based on multi-layer perceptrons is denoted as MLP-DA. Table 1 shows the fitting performance comparison results of different models on the training set. Tables 2 and 3 compare the generalization performance results of different models on the test set and 1-test 2, respectively. Fitting performance is evaluated using two metrics: precision and recall.

[0136] Table 1

[0137] Model Accuracy Recall Support Vector Machine 0.90860 0.57877 ANN 0.95454 0.86301 CNN 0.96124 0.84932 MLP-DA 0.95548 0.89711 Domain-adaptive DWGCN 0.96319 0.85136

[0138] Table 2

[0139]

[0140]

[0141] Table 3

[0142] Model Accuracy Recall Support Vector Machine 0.00000 0.00000 ANN 0.42708 0.24193 CNN 0.36767 0.26041 MLP-DA 0.42708 0.53125 Domain-adaptive DWGCN 0.86301 0.67568

[0143] The table above shows that commonly used data-driven models (SVM, ANN, and CNN) exhibit poor generalization performance on both Test Sets 1 and 2. This is because commonly used data-driven anomaly detection models are based on the independent and identically distributed assumption and are less adaptable to changing system operating conditions. Classic domain adaptation methods exhibit better adaptability to operating conditions than commonly used data-driven models, but are significantly inferior to the proposed method. Furthermore, the proposed adaptive anomaly detection model, based on an interpretable domain-adaptive causal decoupling network and anomaly detection model, offers greater credibility.

[0144] In summary, the solutions provided by the embodiments of the present disclosure offer the following beneficial effects: 1. Improved detection accuracy: By combining domain adaptation and causal decoupling techniques, this method can more accurately identify anomalies in electromechanical systems, maintaining high accuracy even when data distribution changes. 2. Enhanced model generalization: The domain-adaptive causal decoupling network can adapt to different operating environments and working conditions, enabling the model to perform well not only on training data but also effectively detect anomalies on real-time monitoring data. 3. Component-level health status characterization: Through component-level health status characterization, this method can provide detailed health assessments for each component in the electromechanical system, facilitating more refined maintenance and fault prediction. 4. Causal relationship analysis: Leveraging causal information, this method can analyze and understand the interactions and influences between system components, thereby providing more in-depth fault diagnosis information. 5. Data-driven decision support: Based on a large amount of system monitoring data, this method provides data-driven decision support for the operation and maintenance of electromechanical systems, helping to achieve predictive maintenance and reduce unplanned downtime. 6. Reduced maintenance costs: Through accurate anomaly detection and component-level health status assessment, this method helps reduce unnecessary maintenance work and lower maintenance costs. 7. Improve system reliability and safety: Timely detection and handling of potential faults will help improve the reliability and safety of electromechanical systems and prevent major accidents. 8. Strong adaptability: This method does not rely on specific equipment or systems, has good adaptability, and can be widely applied to different types of complex electromechanical systems. 9. User-friendly: Through intuitive fault labels and health status characterization, this method provides users with an easy-to-understand and easy-to-operate interface, improving user experience. 10. Promote intelligent development: The implementation of this method will help promote the development of electromechanical systems towards intelligence and automation, and provide technical support for industry and intelligent manufacturing.

[0145] The following is an apparatus embodiment of the present disclosure. For parts not described in detail in the apparatus embodiment, reference may be made to the technical details disclosed in the above method embodiment.

[0146] Please refer to Figure 9, which shows a schematic diagram of the structure of a complex electromechanical system anomaly detection device based on a domain-adaptive causal decoupling network, provided by an exemplary embodiment of the present disclosure. This device can be implemented in whole or in part as a computing device using software, hardware, or a combination of both. The device includes an acquisition module 92 and a training module 94.

[0147] an acquisition module 92 for acquiring a training data set and causal information of the electromechanical system, wherein the training data set includes system monitoring data and corresponding system fault label data, wherein the system monitoring data includes multiple variables related to the health status of the electromechanical system, the system fault label data indicates whether the electromechanical system is in a fault state, and the causal information indicates the functional relationship between multiple components of the electromechanical system;

[0148] A training module 94 is used to train an anomaly detection model based on a domain-adaptive causal decoupling network based on a training data set and causal information. The domain-adaptive causal decoupling network is used to indicate the correspondence between system monitoring data and component-level health status representations. The component-level health status representations are used to indicate the health status of multiple components of the electromechanical system. The anomaly detection model is used to detect anomalies in the electromechanical system.

[0149] In one possible implementation, the types of multiple variables include at least one of system monitoring parameters, system operating condition variables, and system operating modes. The system monitoring parameters include variables collected by a data collection device set at a specified position of the electromechanical system. The system operating condition variables include external condition variables and / or environmental variables of the electromechanical system. The system operating mode is the operating mode adopted by the electromechanical system when implementing its functions.

[0150] In another possible implementation, the training module 94 is further configured to:

[0151] Based on the training dataset and causal information, a domain-adaptive causal decoupling network is trained;

[0152] Through domain-adaptive causal decoupling networks, component-level health status representation is decoupled from system monitoring data.

[0153] Determine the anomaly detection model based on the component-level health status characterization.

[0154] In another possible implementation, the training module 94 is further configured to:

[0155] Based on the training data set and causal information, a domain-adaptive causal decoupling network is constructed. The domain-adaptive causal decoupling network includes a causal adjustment matrix and a coordinate rotation matrix. The causal adjustment matrix is ​​used to indicate the impact of system operating condition variables and / or system operating modes on causal information. The coordinate rotation matrix is ​​used to match the coordinate system of system monitoring parameters to the coordinate system of component-level health status representation.

[0156] By minimizing the loss function, the causal adjustment matrix and coordinate rotation matrix are optimized, and a domain-adaptive causal decoupling network is trained.

[0157] In another possible implementation, the training module 94 is further configured to:

[0158] Determine a domain-stable anomaly detection model based on the component-level health status representation. The domain-stable anomaly detection model is used to indicate the correspondence between the component-level health status representation and the anomaly detection result. The anomaly detection result is used to indicate whether the electromechanical system is in a fault state.

[0159] An anomaly detection model is determined based on the domain-adaptive causal decoupling network and the domain-stable anomaly detection model.

[0160] In another possible implementation, the anomaly detection model includes: a domain-adaptive causal decoupling network and a domain-stable anomaly detection model. The input parameters of the domain-adaptive causal decoupling network are system monitoring data, and the output parameters are component-level health status representations; the input parameters of the domain-stable anomaly detection model are component-level health status representations, and the output parameters are anomaly detection results.

[0161] In another possible implementation, the apparatus further includes: a calling module configured to:

[0162] Obtain real-time monitoring data of electromechanical systems during operation;

[0163] Based on real-time monitoring data, the pre-trained anomaly detection model is called to output anomaly detection results, which are used to indicate whether the electromechanical system is in a fault state.

[0164] It should be noted that, when the device provided in the above embodiment realizes its function, it only uses the division of the above-mentioned functional modules as an example. In actual application, the above-mentioned functions can be assigned to different functional modules according to actual needs, that is, the content structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0165] Regarding the apparatus in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0166] An embodiment of the present disclosure further provides a computing device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0167] The embodiment of the present disclosure further provides a non-volatile computer-readable storage medium having computer program instructions stored thereon, which implement the above method when the computer program instructions are executed by a processor.

[0168] An embodiment of the present disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of a computing device, the processor in the computing device executes the above method.

[0169] Figure 10 1 is a block diagram of an apparatus 1900 for executing a complex electromechanical system anomaly detection method based on a domain-adaptive causal decoupling network according to an exemplary embodiment. For example, the apparatus 1900 can be provided as a server or a terminal device. Figure 10 The apparatus 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions, such as an application, that can be executed by the processing component 1922. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute the instructions to perform the above-described method.

[0170] The device 1900 may also include a power supply component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input / output interface 1958 (I / O interface). The device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server™, MacOS X™, Unix™, Linux™, FreeBSD™, or the like.

[0171] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by the processing component 1922 of the apparatus 1900 to perform the above-described method.

[0172] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0173] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0174] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0175] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, and conventional procedural programming languages ​​such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., utilizing an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions. The electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0176] Various aspects of the present disclosure are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0177] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0178] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0179] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and the part of the module, program segment or instruction contains one or more executable instructions for realizing the prescribed logical function. In some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the prescribed function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0180] While various embodiments of the present disclosure have been described above, the foregoing description is intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A complex electromechanical system anomaly detection method based on a domain-adaptive causal decoupling network, characterized in that: The method comprises: Obtaining a training data set and causal information of a mechatronic system, wherein the training data set includes system monitoring data and corresponding system fault label data, wherein the system monitoring data includes multiple variables related to the health status of the mechatronic system, the system fault label data is used to indicate whether the mechatronic system is in a fault state, and the causal information is used to indicate the functional relationship between multiple components of the mechatronic system; training an anomaly detection model based on a domain-adaptive causal decoupling network based on the training data set and the causal information, wherein the domain-adaptive causal decoupling network is used to indicate a correspondence between the system monitoring data and a component-level health status representation, wherein the component-level health status representation is used to indicate the health status of each of a plurality of components of the electromechanical system, and the anomaly detection model is used to perform anomaly detection on the electromechanical system; The process of determining the anomaly detection model includes: determining a domain-stable anomaly detection model based on the component-level health status representation, the domain-stable anomaly detection model being used to indicate a correspondence between the component-level health status representation and an anomaly detection result, the anomaly detection result being used to indicate whether the electromechanical system is in a fault state; The anomaly detection model is determined according to the domain adaptive causal decoupling network and the domain stable anomaly detection model.

2. The method according to claim 1, characterized in that The types of the multiple variables include at least one of system monitoring parameters, system operating condition variables and system operating modes. The system monitoring parameters include variables collected by a data collection device set at a specified position of the electromechanical system. The system operating condition variables include external condition variables and / or environmental variables of the electromechanical system. The system operating mode is the operating mode adopted by the electromechanical system when realizing its functions.

3. The method according to claim 1 or 2, characterized in that The method of training an anomaly detection model based on a domain-adaptive causal decoupling network according to the training data set and the causal information includes: Training the domain-adaptive causal decoupling network based on the training data set and the causal information; The component-level health status representation is obtained by decoupling from the system monitoring data through the domain-adaptive causal decoupling network.

4. The method according to claim 3, characterized in that The training of the domain-adaptive causal decoupling network based on the training data set and the causal information includes: constructing a domain-adaptive causal decoupling network based on the training data set and the causal information, the domain-adaptive causal decoupling network including a causal adjustment matrix and a coordinate rotation matrix, the causal adjustment matrix being used to indicate the influence of the system operating condition variables and / or system operating mode on the causal information, and the coordinate rotation matrix being used to match the coordinate system of the system monitoring parameters to the coordinate system of the component-level health status representation; By minimizing the loss function, the causal adjustment matrix and the coordinate rotation matrix are optimized, and the domain adaptive causal decoupling network is obtained through training.

5. The method according to claim 4, characterized in that The anomaly detection model includes: the domain-adaptive causal decoupling network and the domain-stable anomaly detection model, the input parameter of the domain-adaptive causal decoupling network is the system monitoring data, and the output parameter is the component-level health status representation; the input parameter of the domain-stable anomaly detection model is the component-level health status representation, and the output parameter is the anomaly detection result.

6. The method according to any one of claims 1 to 5, characterized in that The method further comprises: Acquiring real-time monitoring data of the electromechanical system during operation; According to the real-time monitoring data, the pre-trained anomaly detection model is called to output an anomaly detection result, where the anomaly detection result is used to indicate whether the electromechanical system is in a fault state.

7. A complex electromechanical system anomaly detection device based on a domain-adaptive causal decoupling network, characterized in that: The device comprises: an acquisition module, configured to acquire a training data set and causal information of the electromechanical system, wherein the training data set includes system monitoring data and corresponding system fault label data, wherein the system monitoring data includes multiple variables related to the health status of the electromechanical system, the system fault label data is used to indicate whether the electromechanical system is in a fault state, and the causal information is used to indicate the functional relationship between multiple components of the electromechanical system; a training module for training, based on the training data set and the causal information, an anomaly detection model based on a domain-adaptive causal decoupling network, wherein the domain-adaptive causal decoupling network is used to indicate a correspondence between the system monitoring data and a component-level health status representation, wherein the component-level health status representation is used to indicate the health status of each of multiple components of the electromechanical system, and the anomaly detection model is used to perform anomaly detection on the electromechanical system; The process of determining the anomaly detection model includes: determining a domain-stable anomaly detection model based on the component-level health status representation, the domain-stable anomaly detection model being used to indicate a correspondence between the component-level health status representation and an anomaly detection result, the anomaly detection result being used to indicate whether the electromechanical system is in a fault state; The anomaly detection model is determined according to the domain adaptive causal decoupling network and the domain stable anomaly detection model.

8. A computing device, characterized in that The device comprises: processor; a memory for storing processor-executable instructions; The processor is configured to implement the method according to any one of claims 1 to 6 when executing the instructions stored in the memory.

9. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images