A fault diagnosis method and system based on digital-analog linkage

Abstract

The application discloses a kind of based on digital-analog linkage's fault diagnosis method and system, belong to fault diagnosis technical field, method includes: the bond graph model of constructing multiple redundancy electric static pressure servo mechanism;Deduce analytical redundancy relationship;The operation data of multiple redundancy electric static pressure servo mechanism is brought into analytical redundancy relationship, generates redundancy formula residual error, and is divided based on redundancy formula residual error redundancy formula residual error threshold interval;Based on analytical redundancy relationship and the system parameter of multiple redundancy electric static pressure servo mechanism, establish the fault characteristic matrix of multiple redundancy electric static pressure servo mechanism;With fault characteristic matrix and redundancy formula residual error threshold interval, the contribution degree of each system parameter is calculated, and low contribution degree system parameter is isolated;The long short-term memory neural network model is constructed;Long short-term memory neural network model is trained;Fault data is classified using trained long short-term memory neural network model, and the fault reason of fault data is determined.The range of fault diagnosis and diagnostic accuracy are improved.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, specifically relating to a fault diagnosis method and system based on digital-analog linkage. Background Technology

[0002] Redundant electrostatic servo mechanisms are advanced motion control systems used to achieve precise position and motion control, particularly in industrial and aerospace applications requiring high precision, high speed, high reliability, and high load capacity. These mechanisms typically include multiple electrostatic actuators, sensors, control units, and feedback loops that work together to achieve complex motion control tasks, such as high-precision motion control and high-speed response in flight control. Correspondingly, in applications requiring high precision and reliability, mechanism failures can lead to inaccurate motion control or even jeopardize system safety. Early fault detection and resolution are crucial for ensuring system reliability and safety. Redundant electrostatic servo mechanisms often contain complex electronic and mechanical components; repairs can be costly and time-consuming if a failure occurs. Early fault diagnosis allows for better planning of maintenance and repair operations. Failures can lead to performance degradation, such as reduced accuracy and slower speed; accurate fault diagnosis helps maintain normal system operation and expected performance.

[0003] Currently, fault diagnosis of redundant electrostatic servo mechanisms is often limited to the diagnosis of some system parameters, resulting in low fault diagnosis parameter coverage. For low-contribution system parameters that are unlikely to cause changes in redundancy when a fault occurs, the cause of the fault often cannot be located. Summary of the Invention

[0004] To address the technical problems of existing technologies that often limit fault diagnosis of redundant electrostatic servo mechanisms to the diagnosis of only some system parameters, resulting in low fault diagnosis parameter coverage and the inability to locate the cause of faults for low-contribution system parameters that are unlikely to cause changes in redundancy when faults occur, this invention provides a fault diagnosis method and system based on analog-digital linkage.

[0005] First aspect

[0006] This invention provides a fault diagnosis method based on digital-analog linkage, comprising:

[0007] S101: Construct the bond graph model of the redundant electrostatic servo mechanism;

[0008] S102: Derive analytical redundancy relationships based on the bond graph model;

[0009] S103: Input the operating data of the redundant electrostatic servo mechanism into the analytical redundancy relation to generate redundant residuals, and divide the redundant residual threshold range based on the redundant residuals.

[0010] S104: Establish the fault characteristic matrix of the redundant electrostatic servo mechanism based on the analytical redundancy relation and the system parameters of the redundant electrostatic servo mechanism;

[0011] S105: Calculate the contribution of each system parameter by combining the fault feature matrix and the redundant residual threshold interval, and isolate the system parameters with low contribution based on the contribution.

[0012] S106: Construct a long short-term memory neural network model including an input gate, a forget gate, and an output gate;

[0013] S107: Use the operating data and corresponding fault types of low-contribution system parameters as sample data to train the long short-term memory neural network model;

[0014] S108: Use the trained Long Short-Term Memory Neural Network model to classify fault data and determine the cause of the fault.

[0015] Second aspect

[0016] The present invention provides a fault diagnosis system based on digital-analog linkage, used to execute the fault diagnosis method based on digital-analog linkage in the first aspect.

[0017] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0018] In this invention, the energy transfer relationship inside the redundant electrostatic servo mechanism is characterized by a bond graph model. Then, an analytical redundancy relation including the constraint relationship of the servo mechanism is derived to monitor the state of the servo mechanism. After that, the contribution of the redundancy is calculated, and low contribution system parameters that cannot be observed through the constraint relationship are isolated. A long short-term memory neural network model including an input gate, a forget gate, and an output gate is constructed to effectively locate the faults of the low contribution system parameters. Information is selectively retained and forgotten, and long-term dependencies are modeled to avoid premature or excessive forgetting or updating of information, thereby reducing the risk of gradient vanishing and gradient explosion. This combination of numerical and modeling improves the parameter coverage, fault diagnosis efficiency, and diagnostic accuracy of fault diagnosis. Attached Figure Description

[0019] The preferred embodiments will now be described in a clear and easy-to-understand manner, in conjunction with the accompanying drawings, to further explain the above-mentioned characteristics, technical features, advantages, and implementation methods of the present invention.

[0020] Figure 1 This is a flowchart illustrating a fault diagnosis method based on digital-analog linkage provided by the present invention;

[0021] Figure 2This is a schematic diagram of the structure of a bond graph model provided by the present invention;

[0022] Figure 3 This is a schematic diagram of the structure of a long short-term memory neural network model provided by the present invention. Detailed Implementation

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.

[0024] To keep the drawings concise, each figure only schematically shows the parts relevant to the invention, and these do not represent the actual structure of the product. Furthermore, to facilitate understanding, in some figures, only one of components with the same structure or function is schematically depicted, or only one is labeled. In this document, "one" not only means "only one," but can also mean "more than one."

[0025] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0026] In this document, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0027] Furthermore, in the description of this invention, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0028] Example 1

[0029] In one embodiment, refer to the appendix to the specification. Figure 1 This diagram illustrates a flowchart of the fault diagnosis method based on analog-digital linkage provided by the present invention. (See attached specification.) Figure 2 The diagram shows a structural schematic of a bond graph model provided by the present invention.

[0030] This invention provides a fault diagnosis method based on analog-digital linkage, applied to a redundant electrostatic servo mechanism. The redundant electrostatic servo mechanism includes a servo motor, a hydraulic system, and a load system. The method includes:

[0031] S101: Construct the bond graph model of the redundant electrostatic servo mechanism.

[0032] Among them, the bond graph model is an abstract diagram used to represent the connection relationship between internal components of a redundant electrostatic servo mechanism. In the field of mechanical and electrical system control, the bond graph model is a commonly used method to represent the relationship between components, such as connection, energy flow, and signal transmission.

[0033] In one possible implementation, the bond graph model includes a servo motor bond graph model, a hydraulic system bond graph model, and a load system bond graph model, and S101 specifically includes:

[0034] S1011: The three-phase coupling circuit in the servo motor is converted into a two-phase circuit using the synchronous coordinate system. The servo motor bonding model includes the data acquisition interface se, a-axis current, q-axis current, q-axis resistance, q-axis inductance, q-axis current and variable modulus gyroscope MGY. The d-axis current in the synchronous coordinate system is simplified to 0. The data acquisition interface se is connected to the common current junction I1 representing the q-axis current. The q-axis resistance and q-axis inductance in the synchronous coordinate system are connected to the common current junction I1. The common current junction I1 is connected to the variable modulus gyroscope MGY to complete the construction of the servo motor bonding graph model.

[0035] S1012: The bond graph model of the hydraulic system includes data entry S1, friction coefficient fp between the shaft and bearing of the servo motor and pump, rotational inertia Jp of the shaft between the servo motor and pump, transducer TF1 representing the piston pump in the hydraulic system, transducer coefficient Dp representing the displacement of the piston pump, piston pump leakage coefficient ep, pipeline friction loss coefficient epipe, hydraulic cylinder leakage coefficient eh, transducer TF2 representing the hydraulic cylinder, transducer coefficient Spis representing the effective area of ​​the hydraulic cylinder piston, common flow junction I2, common potential junction O1, and common flow junction. I3 and common potential junction O2, combined with the energy interaction relationship in the hydraulic system, the friction coefficient fp, the moment of inertia Jp are connected to common potential junction I2, common potential junction I2 is connected to common potential junction O1 through converter TF2 and converter coefficient Dp, the piston pump leakage coefficient ep is connected to common potential junction O1, common potential junction O1 is connected to common potential junction O2 through common flow junction I3, the pipeline friction loss coefficient epipe is connected to common flow junction I3, the hydraulic cylinder leakage coefficient eh is connected to common potential junction O2, and common potential junction O2 is connected to converter TF2 and converter coefficient Spis;

[0036] S1013: The load system bond graph model includes a common junction I4 and input interfaces S2, all of which are connected to the common junction I4, load mass m, load comprehensive friction coefficient Ch, and load elasticity coefficient k;

[0037] S1014: Connect the variable modulus gyroscope MGY to the data input S1 and the converter TF2 to the input interface S2 to connect the servo motor bond graph model, the hydraulic system bond graph model and the load system bond graph model to obtain the bond graph model.

[0038] Specifically, in the context of fault diagnosis, this bond graph model describes the connection methods between various components in a redundant electrostatic servo mechanism, the paths for transmitting information, and the mutual influence relationships between them. Such a model helps to understand the structure of the mechanism and provides a foundation for subsequent fault analysis and diagnosis.

[0039] S102: Derive the analytical redundancy relation based on the bond graph model.

[0040] Analytical Redundancy Relation (ARR) is a fault detection and isolation method based on the system's physical structure. Its core principle is to establish constraint equations under normal system conditions. When a fault occurs, the equations related to the fault parameters generate residuals. By analyzing the residuals of multiple equations, the fault range can be significantly narrowed, and specific faults can be located. Analytical Redundancy Relation helps determine the occurrence and location of faults. By comparing actual residuals with predefined thresholds, it can identify which equations produced abnormal residuals, thereby narrowing the fault range and ultimately locating a specific fault type or location. Because it is based on the system's physical structure and does not rely on statistical models or large amounts of historical data, it is applicable to various systems and applications. By combining information from multiple equations, Analytical Redundancy Relation can provide more accurate and reliable fault diagnosis results, facilitating the rapid identification and resolution of faults in the system.

[0041] In one possible implementation, S102 specifically includes:

[0042] S1021: Characterize the common potential nodes and common current nodes in the bond graph model using potential variables and current variables respectively, and obtain the analytical redundancy relational expressions corresponding to each common potential node and common current node.

[0043] Among them, the analytical redundancy relation includes the first analytical redundancy relation to the sixth analytical redundancy relation.

[0044] In one possible implementation, S1021 specifically includes:

[0045] S1031A: The current flow of the common junction I1 represents the q-axis current i. q The potential variables of the cocurrent junction I1 satisfy the algebraic sum of 0, and the first analytical redundancy relation corresponding to the cocurrent junction I1 is:

[0046]

[0047] Among them, U q L represents the q-axis voltage. q Indicates q-axis inductance, i q R represents the q-axis current. q Represents the q-axis resistance, ω e Indicates the electric angular velocity of the motor. Indicates the magnetic flux linkage of the motor;

[0048] S1031B: The flow rate of the common junction I2 represents the mechanical angular velocity ω output by the servo motor. m The potential variables of the cocurrent junction I2 satisfy the algebraic sum of 0, and the second analytical redundancy relation corresponding to the cocurrent junction I2 is:

[0049]

[0050] Where P represents the number of pole pairs of the motor, J p D represents the moment of inertia of the piston pump rotor, fp represents the sum of the viscous rotational friction coefficient of the motor and the viscous rotational friction coefficient of the piston pump, and D... p P represents the displacement of a plunger pump. p This indicates the pressure difference across the plunger pump.

[0051] S1031C: The potential variable of the common potential junction O1 represents the pressure difference Pp between the two ports of the plunger pump. The flow variables of the common potential junction O1 satisfy the algebraic sum of 0. The third analytical redundancy relation corresponding to the common potential junction O1 is:

[0052] D p ω m -ε p P p -q pipe =0

[0053] Where, ε p q represents the leakage coefficient of a plunger pump. pipe Indicates hydraulic flow rate;

[0054] S1031D: The flow variable of the co-flow junction I3 represents the pipe flow rate q, and the potential variable of the co-potential junction I3 satisfies an algebraic sum of 0. The fourth analytical redundancy relation corresponding to the co-flow junction I3 is:

[0055] P p -ε pipe -q pipe=0

[0056] Where, ε pipe Indicates the leakage coefficient of hydraulic pipelines;

[0057] S1031E: The potential variable of the common potential junction O2 represents the pressure difference Ph between the two oil ports of the hydraulic cylinder. The flow variables of the common potential junction O2 satisfy the algebraic sum of 0. The fifth analytical redundancy relation corresponding to the common potential junction O2 is:

[0058] q pipe -ε h P h -V pis S pis =0

[0059] Where, ε h P represents the leakage coefficient of the hydraulic cylinder. h V represents the pressure difference between the two ends of the hydraulic cylinder. pis S represents the volume of the hydraulic pipeline. pis Indicates the effective area of ​​the hydraulic cylinder piston;

[0060] S1031F: The flow variable of the common flow junction I4 represents the load velocity v, the potential variable of the common flow junction I4 satisfies an algebraic sum of 0, and the sixth analytical redundancy relation corresponding to the common flow junction I4 is:

[0061]

[0062] Among them, P Δ The pressure difference in the pipeline is represented by m, the load mass by a, and the piston rod acceleration by C. h The coefficient of friction of the hydraulic cylinder is represented by v, the load speed (i.e., piston rod speed) is represented by k, the load elasticity coefficient is represented by x, and the piston rod displacement is represented by x.

[0063] S103: Input the operating data of the redundant electrostatic servo mechanism into the analytical redundancy relation to generate redundant residuals, and divide the redundant residual threshold range based on the redundant residuals.

[0064] It's important to note that the data collected during the actual operation of the redundant electrostatic servo mechanism is input into the previously derived analytical redundancy equations. These equations, constructed under normal operating conditions, describe the constraint relationships between various components. By substituting the actual data into these equations, a series of residuals are obtained—the differences between the actual data and the expected values. These generated residuals are known as "redundant residuals." They arise from the difference between the system's actual operating state and its normal operating state. Redundant residuals reflect, to some extent, abnormal or faulty conditions in the system. Then, based on these redundant residuals, different threshold intervals are defined. A threshold is a predefined numerical range used to determine the magnitude and degree of the residuals. By analyzing the distribution of redundant residuals, we can determine different thresholds to divide the residuals into different intervals. Each interval may represent different situations, such as normal operation, minor faults, or severe faults. Combining the actual operating data with the analytical redundancy equations generates redundant residuals, and different threshold intervals are defined based on the magnitude of the residuals. These intervals provide a basis for subsequent fault diagnosis, helping us determine the system's health status based on the degree and distribution of the residuals.

[0065] In one possible implementation, S103 specifically includes:

[0066] S1031: Calculate the probability contribution value of each data point in the running data using the kernel density function, where the kernel density function is a Gaussian function:

[0067]

[0068] h = 1.06σn -1 / 5

[0069] in, denoted by , where K represents the kernel density function, h represents the Scott estimation bandwidth, n represents the sample size, and σ represents the standard deviation of the sample data.

[0070] S1032: Determine the redundancy residual threshold range based on the probability contribution value and the 3σ principle.

[0071] S104: Establish the fault characteristic matrix of the redundant electrostatic servo mechanism based on the analytical redundancy relation and the system parameters of the redundant electrostatic servo mechanism.

[0072] In one possible implementation, the system parameters include the q-axis resistance, q-axis inductance, motor flux linkage, moment of inertia, coefficient of friction, piston pump displacement, piston pump leakage coefficient, pipeline friction coefficient, hydraulic cylinder leakage coefficient, piston area, load mass, load damping coefficient, and load elasticity coefficient of the redundant electrostatic servo mechanism. The fault characteristic matrix A is specifically as follows:

[0073]

[0074] In this fault feature matrix, the first and second columns represent system characterization parameters, the third to eighth columns represent the first to sixth analytical redundancy relations, the ninth and tenth columns represent the detectability and isolability of the redundant static voltage servo mechanism, the numbers 0 and 1 represent the fault feature vector of the system characterization parameter in the row, and the isolability number is 1 when it is isolable.

[0075] Among them, the analytical redundancy relation is a constraint equation used to describe the normal operating state of the system, while the system parameters of the redundant electrostatic servo mechanism cover various physical, electrical and control parameters of the mechanism. By combining these two to establish a matrix, the parts of the analytical redundancy relation related to the system parameters are filled into the corresponding positions to help us better understand the characteristics and redundancy relationships of the system, so as to conduct analysis and judgment in subsequent fault diagnosis.

[0076] S105: Calculate the contribution of each system parameter by combining the fault feature matrix and the redundant residual threshold interval, and isolate the system parameters with low contribution based on the contribution.

[0077] In one possible implementation, S105 specifically includes:

[0078] S1051: Correspond each system parameter to the analytical redundancy relation to generate a fault feature vector;

[0079] S1052: Unify the rate of change and direction of change of each system parameter;

[0080] S1053: Obtain the system residual values ​​of each system parameter when a fault occurs individually using AMESim simulation software;

[0081] S1054: Calculate the contribution using system residuals:

[0082]

[0083] Where T represents the signal period, Th u and Th d β represents the upper and lower limits of the redundant residual threshold interval. A Indicates contribution, δ A This represents the system residual value corresponding to a system parameter failure.

[0084] S1055: Set the isolation contribution threshold to separate system parameters according to the isolation contribution threshold and filter system parameters with low contribution.

[0085] It should be noted that low-contribution system parameters selected through parameter contribution calculation cannot cause redundant residual changes when they undergo abnormal changes. This means that such parameter faults cannot be diagnosed and isolated using the symbolic bond graph method. This would isolate low-contribution system parameters that may be at fault, allowing for secondary classification and precise fault type identification.

[0086] Reference Figure 3 The diagram shows a structural schematic of a long short-term memory neural network model provided by the present invention.

[0087] Depend on Figure 3 As can be seen from this, the input to the input gate is the input feature (x) at the current time step. t ) and the hidden state of the previous time step (h t-1 ), x t and h t-1 The input gate's weights are connected to the input gate's weights and multiplied by their respective weights. After linear combination, the gate is activated by the sigmoid function, yielding a value between 0 and 1 representing the importance of the new information. The output of the input gate is multiplied by the candidate values ​​processed by the tanh function to obtain the new information to be added to the cell state. The forget gate's input is the input feature (x) at the current time step. t ) and the hidden state of the previous time step (h t-1 ), x t and h t-1 The values ​​are concatenated with the weights of the forget gate, multiplied by their respective weights, and then linearly combined before being activated by the sigmoid function. This yields a value between 0 and 1, representing the degree of information retention in each cell state. The previous cell state C is then used as the basis for further processing. t-1 Multiplying the result by the output of the forget gate yields the information to be forgotten. During the update process, the output of the input gate (candidate values ​​processed by the tanh function) and the output of the forget gate (the previous cell state C) are used. t-1 (Multiplied by the output of the forget gate), the two input parts are concatenated with their corresponding weights, and the two input parts are summed element-wise to obtain the new cell state C. t The input to the output gate is the input feature (x) at the current time step. t ) and the hidden state of the previous time step (h t-1 ), x t and h t-1 The weights of the output gate are connected, multiplied by their respective weights, and after linear combination, activated by the sigmoid function to obtain a value between 0 and 1, representing the degree of information output in the cell state. The cell state C processed by the tanh function is then... t Multiplying the output by the output gate yields the final hidden state h. tHidden state h t This information is used for model output or passed to the next time step. This method addresses the vanishing and exploding gradient problems, leading to better diagnosis of fault characteristics.

[0088] S106: Construct a long short-term memory neural network model including an input gate, a forget gate, and an output gate.

[0089] The Long Short-Term Memory (LSTM) neural network model comprises three important gating units: the input gate, the forget gate, and the output gate. LSTM is a special variant of Recurrent Neural Network (RNN) designed to process sequential data, such as time series and text. The input gate determines which information needs to be updated to the cell state. It calculates a weight based on the input features of the current time step and the hidden state of the previous time step to control the update of new information. The forget gate determines which information needs to be forgotten. Similar to the input gate, the forget gate calculates a weight based on the input features of the current time step and the hidden state of the previous time step to control which information in the cell state needs to be retained or forgotten. The output gate determines the information output from the cell state. It calculates a weight based on the input features of the current time step and the hidden state of the previous time step to control the output information. These gating mechanisms enable better capture of long-term dependencies when processing sequential data, while effectively avoiding problems such as vanishing and exploding gradients. The LSTM model is built for subsequent fault diagnosis tasks. By using input data and the system state, the model can learn and analyze different fault modes, thereby achieving fault detection and diagnosis. This method can make full use of the nonlinear characteristics of neural networks to model and predict complex fault modes, thereby improving fault diagnosis capabilities.

[0090] S107: Use the operating data and corresponding fault types of low-contribution system parameters as sample data to train the long short-term memory neural network model.

[0091] Specifically, during training, the operational data corresponding to the low-contribution system parameters are the actual observed values ​​of these parameters under different operating states of the redundant electrostatic servo mechanism. This data is input into the Long Short-Term Memory (LSTM) neural network model to help the model learn the system's state and characteristics. For each sample data point, we know its corresponding fault type—the specific fault that occurs in a certain state. This information is used as a supervisory signal to help the model learn the correct fault mode. By inputting these training samples into the LSM model, the model will gradually understand the relationship between low-contribution system parameters and different fault types by learning the patterns and correlations between the data. As training progresses, the model adjusts its weights and parameters to fit the training data to the greatest extent possible, thereby improving its generalization ability to unknown data.

[0092] In one possible implementation, S107 specifically includes:

[0093] S1071: Determine capacitive and inertial elements, wherein capacitive and inertial elements include inductive elements, capacitive elements, and load elements;

[0094] S1072: Considering the correlation between capacitive and inertial components, extract the operating data, including timing characteristics, of the redundant electrostatic servo mechanism under low contribution parameters.

[0095] S1073: Calculate the feature vector of the operating data using the fault feature matrix;

[0096] S1074: Divide the operational data, whose fault types are not unique, into training and test sets to train the long short-term memory neural network model.

[0097] In one possible implementation, the process after S107 includes:

[0098] S107A: The trained Long Short-Term Memory Neural Network Model is evaluated using evaluation metrics based on the confusion matrix, including accuracy, precision, and recall.

[0099] S108: Use the trained Long Short-Term Memory Neural Network model to classify fault data and determine the cause of the fault.

[0100] It's important to note that actual fault data is input into the pre-trained Long Short-Term Memory (LSTM) neural network model. This fault data, collected by the redundant electrostatic servo mechanism when a fault occurs, includes system status and parameter information. The LTM model utilizes the relationships and patterns learned during training to predict the category or cause of the fault data. By classifying the fault data, the LTM model can infer the cause of the fault based on the input data patterns and characteristics. The model's output may be a fault type or a specific value representing the system state. This output helps engineers or operators identify the fault, pinpoint its specific cause, and take appropriate maintenance measures.

[0101] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0102] In this invention, the energy transfer relationship within the redundant electrostatic servo mechanism is characterized using a bond graph model. Then, an analytical redundancy relation, including constraints on the servo mechanism, is derived for state monitoring. The contribution to the redundancy is calculated, and low-contribution system parameters that cannot be observed through constraints are isolated. A long short-term memory neural network model, including input, forget, and output gates, is constructed to effectively locate faults in these low-contribution system parameters. Information is selectively retained and forgotten, enabling the modeling of long-term dependencies and avoiding premature or excessive forgetting or updating of information. This reduces the risk of gradient vanishing and gradient explosion. This combination of numerical and model-based approaches improves the parameter coverage, efficiency, and accuracy of fault diagnosis.

[0103] Example 2

[0104] In one embodiment, the present invention provides a fault diagnosis system based on digital-analog linkage, used to execute the fault diagnosis method based on digital-analog linkage in Embodiment 1.

[0105] The fault diagnosis system based on digital-analog linkage provided by the present invention can realize the steps and effects of the fault diagnosis method based on digital-analog linkage in Embodiment 1 above. To avoid repetition, the present invention will not repeat them.

[0106] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0107] In this invention, the energy transfer relationship inside the redundant electrostatic servo mechanism is characterized by a bond graph model. Then, an analytical redundancy relation including the constraint relationship of the servo mechanism is derived to monitor the state of the servo mechanism. After that, the contribution of the redundancy is calculated, and low contribution system parameters that cannot be observed through the constraint relationship are isolated. A long short-term memory neural network model including an input gate, a forget gate, and an output gate is constructed to effectively locate the faults of the low contribution system parameters. Information is selectively retained and forgotten, and long-term dependencies are modeled to avoid premature or excessive forgetting or updating of information, thereby reducing the risk of gradient vanishing and gradient explosion. This combination of numerical and modeling improves the parameter coverage, fault diagnosis efficiency, and diagnostic accuracy of fault diagnosis.

[0108] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0109] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A fault diagnosis method based on analog-digital linkage, characterized in that, An application to a redundant electrostatic servo mechanism, the redundant electrostatic servo mechanism comprising a servo motor, a hydraulic system, and a load system, the method comprising: S101: Construct the bond graph model of the redundant electrostatic servo mechanism; S102: Derive the analytical redundancy relation based on the bond graph model; S103: Input the operating data of the redundant electrostatic servo mechanism into the analytical redundancy relation to generate redundant residuals, and divide the redundant residual threshold range based on the redundant residuals. S104: Based on the analytical redundancy relation and the system parameters of the redundant electrostatic servo mechanism, establish the fault characteristic matrix of the redundant electrostatic servo mechanism; S105: Calculate the contribution of each system parameter by combining the fault feature matrix and the redundant residual threshold interval, and isolate the system parameters with low contribution based on the contribution. S106: Construct a long short-term memory neural network model including an input gate, a forget gate, and an output gate; S107: Use the operating data and corresponding fault types corresponding to the low contribution system parameters as sample data to train the long short-term memory neural network model; S108: Use the trained Long Short-Term Memory Neural Network model to classify the fault data and determine the cause of the fault data.

2. The fault diagnosis method based on analog-digital linkage according to claim 1, characterized in that, The bond graph model includes a servo motor bond graph model, a hydraulic system bond graph model, and a load system bond graph model. S101 specifically includes: S1011: The three-phase coupling circuit in the servo motor is converted into a two-phase circuit using a synchronous coordinate system. The servo motor bonding model includes a data acquisition interface se, a-axis current, q-axis current, q-axis resistance, q-axis inductance, q-axis current, and a variable modulus gyroscope MGY. The d-axis current in the synchronous coordinate system is simplified to 0. The data acquisition interface se is connected to the common current junction I1 representing the q-axis current. The q-axis resistance and q-axis inductance in the synchronous coordinate system are connected to the common current junction I1. The common current junction I1 is connected to the variable modulus gyroscope MGY to complete the construction of the servo motor bonding graph model. S1012: The hydraulic system bond graph model includes data entry S1, friction coefficient fp between the shaft and bearing of the servo motor and pump, rotational inertia Jp of the shaft between the servo motor and pump, transducer TF1 representing the piston pump in the hydraulic system, transducer coefficient Dp representing the piston pump displacement, piston pump leakage coefficient ep, pipeline friction loss coefficient epipe, hydraulic cylinder leakage coefficient eh, transducer TF2 representing the hydraulic cylinder, transducer coefficient Spis representing the effective piston area of ​​the hydraulic cylinder, common flow junction I2, common potential junction O1, common flow junction I3, and common potential junction O2, combined with the energy in the hydraulic system. The interaction relationships are as follows: the friction coefficient fp, the moment of inertia Jp, and the common potential junction I2 are connected; the common potential junction I2 is connected to the common potential junction O1 through the converter TF2 and the converter coefficient Dp; the piston pump leakage coefficient ep is connected to the common potential junction O1; the common potential junction O1 is connected to the common flow junction I3 and the common potential junction O2; the pipeline friction loss coefficient epipe is connected to the common flow junction I3; the hydraulic cylinder leakage coefficient eh is connected to the common potential junction O2; and the common potential junction O2 is connected to the converter TF2 and the converter coefficient Spis. S1013: The load system bond graph model includes a common flow junction I4 and input interfaces S2, load mass m, load comprehensive friction coefficient Ch, and load elasticity coefficient k, all of which are connected to the common flow junction I4. S1014: Connect the variable modulus gyroscope MGY to the data input S1, and connect the converter TF2 to the input interface S2 to connect the servo motor bond graph model, the hydraulic system bond graph model, and the load system bond graph model to obtain the bond graph model.

3. The fault diagnosis method based on analog-digital linkage according to claim 1, characterized in that, Specifically, S102 is as follows: S1021: The common potential junctions and common current junctions in the bond graph model are characterized by potential variables and current variables, respectively, to obtain the analytical redundancy relational expressions corresponding to each common potential junction and common current junction, wherein the analytical redundancy relational expressions include the first analytical redundancy relational expression to the sixth analytical redundancy relational expression.

4. The fault diagnosis method based on analog-digital linkage according to claim 3, characterized in that, S1021 specifically includes: S1021A: The current-carrying capacity of the common junction I1 represents the q-axis current. i q The potential variables of the cocurrent junction I1 satisfy the algebraic sum of 0, and the first analytical redundancy relation corresponding to the cocurrent junction I1 is: in, Represents the q-axis voltage. Indicates q-axis inductance. Represents the q-axis current. This represents the q-axis resistance. Indicates the electric angular velocity of the motor. Indicates the magnetic flux linkage of the motor; S1021B: The flow rate of the common junction I2 represents the mechanical angular velocity output by the servo motor. The potential variables of the cocurrent junction I2 satisfy the algebraic sum of 0, and the second analytical redundancy relation corresponding to the cocurrent junction I2 is: in, P Indicates the number of pole pairs of the motor. This represents the moment of inertia of the piston pump rotor. This represents the sum of the viscous rotational friction coefficient of the electric motor and the viscous rotational friction coefficient of the piston pump. Indicates the displacement of the plunger pump. This indicates the pressure difference across the plunger pump. S1021C: Co-junction The potential variable represents the pressure difference between the two ports of the plunger pump. The current variables of the common potential junction O1 satisfy the algebraic sum of 0, and the third analytical redundancy relation corresponding to the common potential junction O1 is: in, Indicates the leakage coefficient of a plunger pump. Indicates hydraulic flow rate; S1021D: The flow rate of the common flow junction I3 represents the pipe flow rate. q The potential variables of the common-potential junction I3 satisfy the algebraic sum of 0, and the fourth analytical redundancy relation corresponding to the common-current junction I3 is: in, Indicates the leakage coefficient of hydraulic pipelines; S1021E: The potential change of the common junction O2 represents the pressure difference between the two ports of the hydraulic cylinder. Ph The current variables of the common potential junction O2 satisfy the algebraic sum of 0, and the fifth analytical redundancy relation corresponding to the common potential junction O2 is: in, Indicates the leakage coefficient of the hydraulic cylinder. This indicates the pressure difference between the two ends of the hydraulic cylinder. Indicates the volume of the hydraulic pipeline. Indicates the effective area of ​​the hydraulic cylinder piston; S1021F: The flow rate of the common junction I4 represents the load speed. v The potential variables of the cocurrent junction I4 satisfy the algebraic sum of 0, and the sixth analytical redundancy relation corresponding to the cocurrent junction I4 is: in, Indicates the pipeline pressure difference. m Indicates load quality. Indicates the piston rod acceleration. This represents the viscous friction coefficient of the hydraulic cylinder. v This indicates that the load speed is the piston rod speed. k Indicates the load resilience coefficient. x This indicates the displacement of the piston rod.

5. The fault diagnosis method based on analog-digital linkage according to claim 1, characterized in that, S103 specifically includes: S1031: Calculate the probability contribution value of each data point in the running data using a kernel density function, wherein the kernel density function is a Gaussian function: in, This represents the probability contribution value. K Represents the kernel density function, h This indicates that Scott estimated the bandwidth. n Indicates the number of samples. Indicates the standard deviation of the sample data; S1032: Based on the aforementioned probability contribution value, according to The principle determines the redundant residual threshold range.

6. The fault diagnosis method based on analog-digital linkage according to claim 1, characterized in that, The system parameters include the q-axis resistance, q-axis inductance, motor flux linkage, moment of inertia, friction coefficient, plunger pump displacement, plunger pump leakage coefficient, pipeline friction coefficient, hydraulic cylinder leakage coefficient, piston area, load mass, load damping coefficient, and load elasticity coefficient of the redundant electrostatic servo mechanism. The fault characteristic matrix A specifically comprises: In the fault feature matrix, the first and second columns represent system characterization parameters, the third to eighth columns represent the first to sixth analytical redundancy relations, the ninth and tenth columns represent the detectability and isolability of the redundant static voltage servo mechanism, the numbers 0 and 1 represent the fault feature vector of the system characterization parameters in the row, and the isolability value of 1 indicates isolability.

7. The fault diagnosis method based on analog-digital linkage according to claim 1, characterized in that, Specifically, S105 includes: S1051: Correspond each of the system parameters to the analytical redundancy relation to generate a fault feature vector; S1052: Unify the rate of change and direction of change of each of the system parameters; S1053: Obtain the system residual values ​​of each of the system parameters when a fault occurs individually using AMESim simulation software; S1054: Calculate the contribution using the system residual values: Where T represents the signal period. This represents the upper and lower limits of the redundant residual threshold interval. This indicates the degree of contribution. This represents the system residual value corresponding to the occurrence of a fault in the system parameter; S1055: Set an isolation contribution threshold, separate the system parameters according to the isolation contribution threshold, and filter the low contribution system parameters.

8. The fault diagnosis method based on analog-digital linkage according to claim 1, characterized in that, S107 specifically includes: S1071: Determine capacitive and inertial elements, wherein the capacitive and inertial elements include inductive elements, capacitor elements, and load elements; S1072: Considering the correlation between the capacitive element and the inertial element, extract the operating data, including timing characteristics, of the redundant electrostatic servo mechanism under the low contribution parameter; S1073: Calculate the feature vector of the operating data using the fault feature matrix; S1074: Divide the operating data corresponding to the feature vectors, which have non-unique fault types, into a training set and a test set, and train the long short-term memory neural network model.

9. The fault diagnosis method based on analog-digital linkage according to claim 1, characterized in that, Following S107, the following is also included: S107A: The trained Long Short-Term Memory Neural Network Model is evaluated using evaluation metrics based on the confusion matrix, wherein the evaluation metrics include accuracy, precision, and recall.

Images