Fault intelligent monitoring and fine diagnosis method and device, and storage medium

Abstract

The application discloses a kind of fault intelligent monitoring and fine diagnosis method, device and storage medium, the method includes: according to the uncertain generalized system of fault to be detected, the actuator fault augmented system is additional specific state of uncertain generalized system, constructs generalized system model;Based on neural network, the uncertain model caused by part of unmodeled dynamic and parameter disturbance is obtained with the boundary information of dynamic characteristics using boundary identification algorithm;According to generalized system model, the upper and lower bound dynamic system state and output are coupled using Min-Max function, and a dynamic-output multi-coupling Luenberger interval observer is constructed, and then an observer error variable system is constructed, to realize the fault diagnosis of machine in a large class;According to the fine diagnosis of large class fault.The method provided by the application can timely and effectively find faults, and finely judge fault types when faults occur, to realize intelligent monitoring and fault diagnosis of the system.

Description

Technical Field

[0001] This invention relates to the field of safety and reliability technology, and in particular to a method, apparatus, electronic device and storage medium for intelligent fault monitoring and refined diagnosis. Background Technology

[0002] With the continuous development of economy and technology, the functions of various equipment are becoming increasingly sophisticated, the scale of industry is constantly expanding, and the complexity of various control systems is gradually increasing. Ensuring the reliable, stable, and safe operation of these systems is a huge challenge we will face. In recent years, significant research results have been achieved in the field of system fault diagnosis based on interval observers, which have the advantages of relatively low detection cost and fast response. However, fault diagnosis methods based on interval state observation are still immature. For example, fault monitoring based on interval observation can only determine whether a major category of fault has occurred, but cannot accurately identify the fault type. Therefore, there are still many areas that need improvement and theoretical problems that urgently need to be solved. On the other hand, data-driven machine learning or deep learning-based motor fault classification and diagnosis methods have also been a research hotspot in recent years. They have the characteristics of not requiring specific mechanistic models and being able to perform refined fault type diagnosis. However, fault diagnosis methods based on artificial intelligence algorithms often require a large number of sensors to detect the operating status of the device in real time, thereby obtaining a large number of data samples that occur during normal or faulty conditions. Due to the complexity of actual operating scenarios, when there are many types of faults, the computational cost of building the detection model will also increase, thus greatly increasing the operation and maintenance cost of fault diagnosis equipment.

[0003] In summary, the fault diagnosis method based on interval observers is a soft detection method that can detect faults in sensorless scenarios. However, this method can only identify whether a general category of fault has occurred, and cannot accurately identify the precise type of fault. On the other hand, data-driven fault diagnosis methods often require a large amount of sample data to build a network model, and in actual operation, they need to use a large number of sensors to detect the machine's status in real time. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method, apparatus, electronic device, and storage medium for intelligent fault monitoring and refined diagnosis. This method effectively solves the boundary construction problem of nonlinear perturbation modes by constructing a dynamic-output multi-coupled Luenberger interval observer. Furthermore, it effectively solves the cooperative construction problem of the interval observer error dynamic system without using any linear transformation operations, thus exhibiting lower conservatism. In addition, the dynamic-output multi-coupled Luenberger interval observer, combined with deep learning, can achieve higher accuracy in fault diagnosis and identify more fault types without requiring a large amount of sample data. The method provided by this invention enables timely and effective fault detection, and allows for refined fault type determination upon occurrence, achieving intelligent monitoring and fault diagnosis of the system.

[0005] The first objective of this invention is to provide a method for intelligent fault monitoring and refined diagnosis.

[0006] The second objective of this invention is to provide a fault intelligent monitoring and refined diagnosis system.

[0007] A third objective of this invention is to provide an electronic device.

[0008] A fourth objective of this invention is to provide a storage medium.

[0009] The first objective of this invention can be achieved by adopting the following technical solution:

[0010] A method for intelligent fault monitoring and refined diagnosis, the method comprising:

[0011] Based on the uncertain generalized system for which faults need to be detected, the actuator fault augmented system is treated as an additional specific state of the uncertain generalized system, and a generalized system model is constructed.

[0012] Based on neural networks, for uncertain models caused by some unmodeled dynamics and parameter perturbations, boundary identification algorithms are used to obtain boundary information with dynamic characteristics;

[0013] Based on the generalized system model, the Min-Max function is used to comprehensively couple the upper and lower bound dynamic system states and outputs to construct a dynamic-output multi-coupled Luenberger interval observer with richer degrees of freedom.

[0014] Based on the dynamic-output multi-coupled Luenberger interval observer, an observer error variable system is constructed to realize major fault diagnosis of sensors and actuators;

[0015] The multidimensional data collected by the dynamic-output multi-coupled Luenberger interval observer is fused with features. The fused data and the major fault types diagnosed are then input into a deep convolutional neural network model to output refined fault types.

[0016] Furthermore, the boundary information is as follows:

[0017]

[0018]

[0019] in, The output of the neural network is ΔA; ΔA represents the uncertainty structure of the system matrix parameters. x∈R n Let f be the state vector, f∈R n The problem is an actuator malfunction. For unmodeled nonlinear dynamics, t is the time variable; δ is the network approximation error.

[0020] Furthermore, based on the generalized system model, the Min-Max function is used to comprehensively couple the upper and lower bound dynamic system states and outputs to construct a dynamic-output multi-coupled Luenberger interval observer with richer degrees of freedom, including:

[0021] By coupling the boundary information at the dynamic system and the output equation, a dynamic-output multi-coupled Luenberger interval observer is constructed as follows:

[0022]

[0023] in, and q (t) represents the upper and lower bounds of the state estimation variables; and N (*) represents the comprehensive boundary information of uncertainty; X(t) = [i a ω] T i a Let ω be the armature current and ω be the rotational speed; Max and Min represent the extremum functions; U(t) represents the control state; Y(t) represents the output state; and the matrices N, R, G, M, L are the observer gains to be solved.

[0024] Furthermore, the step of constructing the observer error variable system based on the dynamic-output multi-coupled Luenberger interval observer includes:

[0025] The constructed error variables are:

[0026]

[0027] The error dynamic system matrix corresponding to the error variable is: The matrix is ​​in Metzler form;

[0028] The total error dynamic system is further constructed as follows:

[0029]

[0030] Wherein, the system matrix is ​​N q =Max(N,0)-Min(N,0); Based on the system matrix, the state estimation for the first-order augmented system is finally achieved, i.e., satisfying Max(N,0)-Min(N,0);

[0031] Furthermore, based on the observer error variable system, major categories of fault diagnosis for sensors and actuators are achieved, including:

[0032] The output of the dynamic-output multi-coupled Luenberger interval observer is:

[0033]

[0034]

[0035] Where, y∈R n This is the output vector;

[0036] When no faults occur, the actuator fault f is 0, and the corresponding actual output and the estimate of a specific state interval satisfy... When an actuator failure occurs, the following conditions are met.

[0037] Furthermore, the feature fusion includes data feature mining and data fusion;

[0038] The data feature mining involves characterizing multidimensional data based on short-time Fourier transform, including:

[0039] For the time-domain discrete signal of the multidimensional data, the spectral power spectrum signal x(n,w) is obtained using the discrete short-time Fourier transform, as shown in the following formula:

[0040]

[0041] Where x(m) is the discrete sequence form of the multidimensional signal, and ω(n) is the window function. When the sample size n takes different values, the window function ω(nm) slides along the x(m) sequence.

[0042] The formula for performing a Fast Fourier Transform on the power spectrum signal is as follows:

[0043]

[0044] When k is odd -1 when k is even =1;

[0045] Decompose X(k) into even and odd numbers. When k is odd, i.e., k = 2m + 1, m = 0, 1, ... hour:

[0046]

[0047] When k is even, i.e., k = 2m, m = 0, 1, ..., hour:

[0048]

[0049] The power spectrum matrix of the power spectrum signal is finally obtained through fast Fourier transform.

[0050] Furthermore, the data fusion includes:

[0051] The power spectrum matrix is ​​converted into a two-dimensional power spectrum with three RGB channels;

[0052] The features of multiple layers are fused through skip connections in the two-dimensional power spectrum.

[0053] Furthermore, the uncertain generalized system is:

[0054]

[0055] Where, X∈R n Let Y be the state vector; Y∈R n The output vector; U∈R n The control vector; f∈R n For actuator failure; w∈R n and v∈R n These represent system interference and output interference, respectively; E, A∈R n×n , B∈R n×m , F∈R n×q , C∈R p×n And rank(E)≤n.

[0056] The second objective of this invention can be achieved by adopting the following technical solution:

[0057] A fault intelligent monitoring and refined diagnosis device, the device comprising:

[0058] The interval observer category diagnosis module is used to determine whether a machine has malfunctioned and to classify the malfunction types into major categories. It includes a construction unit for the dynamic-output multi-coupled Luenberger interval observer and a malfunction type identification unit, wherein:

[0059] The construction unit is used to construct a generalized system model by taking the actuator fault augmented system as an additional specific state of the uncertain generalized system to be detected, based on the uncertain generalized system. It also uses a boundary identification algorithm to obtain boundary information with dynamic characteristics based on a neural network for uncertain models caused by some unmodeled dynamics and parameter perturbations. Finally, based on the generalized system model, it uses the Min-Max function to comprehensively couple the upper and lower bound dynamic system states and outputs to construct a dynamic-output multi-coupled Luenberger interval observer with richer degrees of freedom.

[0060] The identification unit is used to construct an observer error variable system based on the dynamic-output multi-coupled Luenberger interval observer to realize major fault diagnosis of sensors and actuators.

[0061] The deep network refined diagnosis module is used to perform refined diagnosis based on the fault types obtained by the interval observer. This includes: fusing the multidimensional data collected by the dynamic-output multi-coupled Luenberger interval observer, inputting the fused data and the major fault types diagnosed into the deep convolutional neural network model, and outputting the refined fault types.

[0062] The third objective of this invention can be achieved by adopting the following technical solution:

[0063] An electronic device includes a processor and a memory for storing a processor-executable program, wherein when the processor executes the program stored in the memory, it implements the above-described intelligent fault monitoring and refined diagnosis method.

[0064] The fourth objective of this invention can be achieved by adopting the following technical solution:

[0065] A storage medium storing a program, which, when executed by a processor, implements the aforementioned intelligent fault monitoring and refined diagnosis method.

[0066] The present invention has the following advantages over the prior art:

[0067] 1. This invention effectively solves the boundary construction problem of nonlinear perturbation modes by constructing a multi-coupled Luenberger-type interval observer. Furthermore, it effectively solves the cooperative construction problem of the interval observer error dynamic system without using any linear transformation operations, thus exhibiting lower conservatism. In addition, by constructing a total error dynamic system, the design difficulty of stability analysis for coupled interval structures is further reduced.

[0068] 2. This invention improves detection and segmentation performance by employing multidimensional data acquisition and feature mining methods, using time-frequency analysis to mine highly discriminative feature representations; and by studying the model construction, training and optimization of deep convolutional neural networks based on multidimensional feature fusion data and classification conditions, and building a refined fault classification network system for normal and abnormal machines based on deep convolution, so as to realize fault early warning and intelligent diagnosis.

[0069] 3. This invention combines an interval observer and a deep convolutional network to achieve intelligent monitoring and refined diagnosis of faults in motor units, thereby improving the accuracy of fault diagnosis and identifying more types of faults. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0071] Figure 1 This is a flowchart of the fault intelligent monitoring and refined diagnosis method of Embodiment 1 of the present invention.

[0072] Figure 2 This is a schematic diagram of the fault intelligent monitoring and refined diagnosis method of Embodiment 1 of the present invention.

[0073] Figure 3 This is a flowchart of the short-time Fourier transform of Embodiment 1 of the present invention.

[0074] Figure 4 This is a schematic diagram of the training process of the deep convolutional neural network model in Embodiment 1 of the present invention.

[0075] Figure 5 This is a structural block diagram of the fault intelligent monitoring and refined diagnosis device of Embodiment 2 of the present invention.

[0076] Figure 6 This is a structural block diagram of the electronic device according to Embodiment 3 of the present invention. Detailed Implementation

[0077] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be understood that the specific embodiments described are merely used to explain this application and are not intended to limit this application.

[0078] Example 1:

[0079] like Figure 1 As shown in the figure, the intelligent fault monitoring and refined diagnosis method provided in this embodiment will be explained using motor set fault diagnosis as an example. Its principle diagram is shown in the figure. Figure 2 As shown, the method includes the following steps:

[0080] S1. Construct a system model.

[0081] First, establish a comprehensive model of the interval observer motor and clarify the fault types under the sensors and actuators.

[0082] The electromotive force equation of the motor can be expressed as follows:

[0083]

[0084] Among them, U a i is the armature voltage; a R is the armature current; a L is the armature circuit resistance. a For armature circuit self-inductance; E a It is the back electromotive force.

[0085] Torque balance equation of the motor:

[0086]

[0087] Where ω is the rotational speed; J is the total moment of inertia of the rotating parts of the unit; T e T represents the electromagnetic torque of the motor. L T is the load braking torque of the motor; f B is the static friction torque of the motor; ω It is a friction system.

[0088] Combining the above two types of component models and fault classifications, let X(t) = [i a ω] T And introduce sensor and actuator fault signals f a (t) and f b(t) then the overall model of the generator system can be obtained:

[0089]

[0090] Y(t)=CX(t)+n(t)+f b (t)

[0091] in ΔA represents the uncertainty structure of the system matrix parameters. The generator set is not modeled for nonlinear dynamics, d(t) and n(t) are external disturbances to the system, and f a (t) and f b (t) represents actuator and sensor faults, respectively. To effectively separate sensor and actuator faults, an augmented system is first constructed, treating actuator faults as a specific additional state of the motor, then:

[0092]

[0093]

[0094] in:

[0095]

[0096] S2. Based on the system model, construct the Luenberger coupled interval observer.

[0097] Compared to existing structures, this structure offers greater design freedom.

[0098] First, select and design a suitable neural network, and obtain dynamic boundary information based on the boundary identification algorithm:

[0099]

[0100]

[0101] in, The output of the neural network is ΔA; ΔA represents the uncertainty structure of the system matrix parameters. δ represents the unmodeled nonlinear dynamics; δ is the network approximation error.

[0102] This embodiment uses a boundary identification algorithm to obtain boundary information with dynamic characteristics.

[0103] Secondly, based on the matrix Min-Max function, and considering the upper and lower bounds of the state estimation variables, a coupling structure similar to Luenberger's is proposed at the dynamic system and output equation ends:

[0104]

[0105] in, and q (t) represents the upper and lower bounds of the state estimation variables; t represents the time variable. and N (*) represents the comprehensive boundary information of uncertainty; Max and Min represent the extremum function; the matrices N, R, G, M, L are the observer gains to be solved.

[0106] This embodiment employs a matrix-based Min-Max function to comprehensively couple the upper and lower bounds of the dynamic system state and output, resulting in a Luenberger-like structure with richer degrees of freedom. Compared to existing structures, this structure offers significantly greater design freedom.

[0107] Based on the dynamic-output multi-coupled Luenberger interval observer, construct the observer error variable system, including:

[0108] The constructed error variables are:

[0109]

[0110] The error dynamic system matrix corresponding to the error variable is: The matrix is ​​in Metzler form, requiring no linear transformation, which fundamentally and effectively relaxes the strong constraints on synergy.

[0111] Furthermore, a dynamic system for the total error is constructed:

[0112]

[0113] Wherein, the system matrix is ​​N q =Max(N,0)-Min(N,0); Based on the system matrix, the state estimation for the first-order augmented system is finally achieved, i.e., satisfying Max(N,0)-Min(N,0);

[0114] Furthermore, the output of the above-mentioned coupled structure interval observer is:

[0115]

[0116]

[0117] When no fault occurs, f a =0,f b =0, the corresponding true output and the estimate of the specific state interval satisfy When only actuator failure occurs, there are When the sensor malfunctions When simultaneous failures occur Therefore, the motor fault logic rules based on the interval observer effectively realize the monitoring of motor faults and the diagnosis of major types of faults in sensors and actuators.

[0118] This embodiment achieves a reduction in the order of observer stability analysis based on comprehensive error interval analysis, which is a simpler stability criterion.

[0119] S3. Construct a deep learning-based intelligent fault monitoring and refined diagnosis method.

[0120] The multidimensional data collected by the interval observer is fused into features. The multidimensional feature fusion data and major fault types are used as inputs to the deep learning, and the refined fault types are used as the network output to train a deep convolutional neural network model.

[0121] Furthermore, step S3 includes:

[0122] (1) Multidimensional fusion data includes data feature mining and data fusion.

[0123] like Figure 3 As shown, the steps for multidimensional data fusion include:

[0124] S301. Feature characterization of multidimensional signal data is performed based on Short-Time Fourier Transform (STFT). For the time-domain discrete signal of the acquired multidimensional data, the discrete STFT is used to obtain the spectral power spectrum signal x(n,w), as shown in the following formula:

[0125]

[0126] Here, x(m) is the discrete sequence form of the multidimensional signal, and ω(n) is the window function. When the sample size n takes different values, the window function ω(nm) slides along the x(m) sequence.

[0127] In this example, a Hamming window is used as the window function. Hamming windows have a wide main lobe and large side lobe attenuation, meaning there is less spectral leakage at other frequency components caused by convolution. The formula is as follows:

[0128]

[0129] To balance frequency and time resolution, a window function size of 256 sampling points was chosen. To counteract the attenuation effect at both ends caused by the window function, overlap between blocks is necessary; here, a repeating interval of 128 sampling points was selected.

[0130] S302. Perform a Fast Fourier Transform (FFT) on the power spectrum signal, where the sequence x(n) has a length n = 256 = 2. 8 Divide it in half lengthwise to obtain the following subsequence:

[0131]

[0132] in, The value is -1 when k is odd and 1 when k is even.

[0133] Decompose X(k) into even and odd subarrays, where k is odd. hour,

[0134]

[0135] When k is even hour,

[0136]

[0137] In summary, X(k) can be decomposed into odd-even groups by calculating the results of N / 2 Discrete Fourier Transforms of the discrete sequence x(n) within the curly braces in the formula. Similarly, the same odd-even grouping can be performed on the N / 2 discrete points of the discrete sequence x(n) within the curly braces. After 7 decompositions, it can be decomposed into 128 groups of two discrete points' Discrete Fourier Transforms. This algorithm has a much lower computational cost than directly using the DFT algorithm. Finally, a 92×129 power spectrum matrix can be obtained by using the STFT feature extraction method on the multidimensional signal, where 92 represents the time dimension and 129 represents the frequency dimension. The proposed method is to extract features from the multidimensional signal using the STFT method.

[0138] S303. The matrix-form features are converted into a two-dimensional power spectrum with RGB three channels. The principle is to select different colors based on the different energy spectral density values, with the colors changing from warm to cool as the energy spectral density values ​​decrease. Multiple layers of features are fused using a skip connection method on the two-dimensional power spectrum, and then a prediction period is trained on the fused features. The skip connection method includes concatenation and add operations.

[0139] The concatenation operation directly joins two features. If the dimensions of the two input features x and y are p and q, the dimension of the output feature z is p+q.

[0140] The add operation combines the two feature vectors into a complex vector. For input features x and y, z = x + iy, where i is the imaginary unit.

[0141] (2) Construct a deep convolutional network.

[0142] like Figure 4 As shown, the construction of a deep convolutional network includes the following steps:

[0143] The basic parameters of the deep convolutional network are set. In this example, the deep convolutional network includes 3 convolutional layers, 2 SE modules, and 3 fully connected layers. The first convolutional layer uses an 11×11 kernel with a stride of 4, the second convolutional layer uses a 5×5 kernel with a stride of 1, and the third convolutional layer uses a 3×3 kernel with a stride of 2. All three convolutional layers include a 3×3 max pooling layer and a ReLU layer. Each SE module consists of a Squeeze operation and an Excitation operation. The outputs of the two SE modules are fused with the feature maps obtained from the first two convolutional blocks. The last fully connected layer uses n nodes as output, where n is the number of refined fault types. The output layer uses a Softmax classifier.

[0144] In this example, the activation function formula used in the ReLU layer is as follows:

[0145] f(x) = max(0,x)

[0146] The calculation formula for the Squeeze operation of the SE module is as follows:

[0147]

[0148] Where x represents the multidimensional data fusion feature map, c represents the number of channels, and x c This represents the feature map of the c-th channel, z. c This represents the compressed value of the feature map, where W and H represent the width and height of the feature map, respectively.

[0149] The calculation formula for the Excitation operation is:

[0150] s = F ex (z,W)=σ(g(z,W))=σ(W2δ(W1z))

[0151] x' c =F scale (x c ,s c )=x c ×s c

[0152] Where z represents the result after the Squeeze operation, and s represents the weights of each channel after the Excitation operation. Represents the dimensionality reduction parameter. The parameters for increasing dimensionality are denoted by δ, ReLU function, and Sigmoid function; x' c F represents the final output of the SE module. scale (x c ,s c ) represents the feature map x of the c-th channel. c and corresponding weights s c The product;

[0153] The calculation formula for the Softmax classifier is as follows:

[0154]

[0155] Among them, a i Let y represent the i-th dimension of the input vector. i This represents the i-th value of the output result y.

[0156] In this embodiment, by constructing a multi-coupled Luenberger-type interval observer, the boundary construction problem of nonlinear perturbation modes is effectively solved. Furthermore, without using any linear transformation operations, the cooperative construction problem of the interval observer error dynamic system is effectively addressed, resulting in lower conservatism. In addition, by studying multidimensional data acquisition and feature mining methods, time-frequency analysis is used to mine highly discriminative feature representations, improving detection and segmentation performance. Based on multidimensional feature fusion data and classification conditions, the model construction, training, and optimization of deep convolutional neural networks are studied, establishing a refined fault classification network system for normal and abnormal machines based on deep convolution, enabling fault early warning and intelligent diagnosis.

[0157] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.

[0158] It should be noted that although the method operations of the above embodiments are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the described steps may be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0159] Example 2:

[0160] like Figure 5 As shown, this embodiment provides a fault intelligent monitoring and refined diagnosis device, which includes an interval observer category diagnosis module 501 and a deep network refined diagnosis module 502, wherein:

[0161] The interval observer category diagnosis module 501 is used to determine whether a machine has malfunctioned and to classify its malfunction types into major categories. It includes a construction unit for a dynamic-output multi-coupled Luenberger interval observer and a malfunction type identification unit, wherein:

[0162] The construction unit is used to construct a generalized system model by taking the actuator fault augmented system as an additional specific state of the uncertain generalized system to be detected, based on the uncertain generalized system. It also uses a boundary identification algorithm to obtain boundary information with dynamic characteristics based on a neural network for uncertain models caused by some unmodeled dynamics and parameter perturbations. Finally, based on the generalized system model, it uses the Min-Max function to comprehensively couple the upper and lower bound dynamic system states and outputs to construct a dynamic-output multi-coupled Luenberger interval observer with richer degrees of freedom.

[0163] The identification unit is used to construct an observer error variable system based on the dynamic-output multi-coupled Luenberger interval observer to realize major fault diagnosis of sensors and actuators.

[0164] The deep network refined diagnosis module 502 is used to perform refined diagnosis based on the fault types obtained by the interval observer, including: fusing the multidimensional data collected by the dynamic-output multi-coupled Luenberger interval observer, inputting the fused data and the major fault types diagnosed into the deep convolutional neural network model, and outputting the refined fault types.

[0165] The specific implementation of each module in this embodiment can be found in Embodiment 1 above, and will not be repeated here. It should be noted that the device provided in this embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure can be divided into different functional modules to complete all or part of the functions described above.

[0166] Example 3:

[0167] This embodiment provides an electronic device, which can be a computer, such as... Figure 6As shown, the processor 602, memory, input device 603, display 604, and network interface 605 are connected via system bus 601. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium 606 and internal memory 607. The non-volatile storage medium 606 stores the operating system, computer programs, and database. The internal memory 607 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the processor 602 executes the computer programs stored in the memory, it implements the fault intelligent monitoring and refined diagnosis method of Embodiment 1 described above, as follows:

[0168] Based on the uncertain generalized system for which faults need to be detected, the actuator fault augmented system is treated as an additional specific state of the uncertain generalized system, and a generalized system model is constructed.

[0169] Based on neural networks, for uncertain models caused by some unmodeled dynamics and parameter perturbations, boundary identification algorithms are used to obtain boundary information with dynamic characteristics;

[0170] Based on the generalized system model, the Min-Max function is used to comprehensively couple the upper and lower bound dynamic system states and outputs to construct a dynamic-output multi-coupled Luenberger interval observer with richer degrees of freedom.

[0171] Based on the dynamic-output multi-coupled Luenberger interval observer, an observer error variable system is constructed to realize major fault diagnosis of sensors and actuators;

[0172] The multidimensional data collected by the dynamic-output multi-coupled Luenberger interval observer is fused with features. The fused data and the major fault types diagnosed are then input into a deep convolutional neural network model to output refined fault types.

[0173] Example 4:

[0174] This embodiment provides a storage medium, which is a computer-readable storage medium, storing a computer program. When the computer program is executed by a processor, it implements the intelligent fault monitoring and refined diagnosis method of Embodiment 1 above, as follows:

[0175] Based on the uncertain generalized system for which faults need to be detected, the actuator fault augmented system is treated as an additional specific state of the uncertain generalized system, and a generalized system model is constructed.

[0176] Based on neural networks, for uncertain models caused by some unmodeled dynamics and parameter perturbations, boundary identification algorithms are used to obtain boundary information with dynamic characteristics;

[0177] Based on the generalized system model, the Min-Max function is used to comprehensively couple the upper and lower bound dynamic system states and outputs to construct a dynamic-output multi-coupled Luenberger interval observer with richer degrees of freedom.

[0178] Based on the dynamic-output multi-coupled Luenberger interval observer, an observer error variable system is constructed to realize major fault diagnosis of sensors and actuators;

[0179] The multidimensional data collected by the dynamic-output multi-coupled Luenberger interval observer is fused with features. The fused data and the major fault types diagnosed are then input into a deep convolutional neural network model to output refined fault types.

[0180] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0181] In summary, the intelligent fault monitoring and refined diagnosis method provided by this invention includes: First, based on the uncertain generalized system to be detected, the actuator fault augmented system is treated as an additional specific state of the uncertain generalized system, and a system model is constructed; Second, based on a neural network, boundary information with dynamic characteristics is obtained from the uncertain model caused by some unmodeled dynamics and parameter perturbations using a boundary identification algorithm; Third, based on the Min-Max function, the upper and lower bound dynamic system states and outputs are comprehensively coupled to construct a dynamic-output multi-coupled Luenberger interval observer structure with richer degrees of freedom; Then, based on the interval state information provided by the dynamic-output multi-coupled Luenberger interval observer, an observer error variable system is constructed to achieve major fault diagnosis of sensors and actuators; Finally, based on the major fault types and state estimates of the interval observer, a refined fault diagnosis model based on deep learning is constructed, and refined fault types are obtained from the refined fault diagnosis model. The method provided by this invention is applicable to most fault detection systems, can well meet the requirements of system fault detection, and achieve refined fault diagnosis.

[0182] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for intelligent fault monitoring and refined diagnosis, characterized in that, The method includes: Based on the uncertain generalized system for which faults need to be detected, the actuator fault augmented system is treated as an additional specific state of the uncertain generalized system, and a generalized system model is constructed. Based on neural networks, for uncertain models caused by some unmodeled dynamics and parameter perturbations, boundary identification algorithms are used to obtain boundary information with dynamic characteristics; Based on the generalized system model, the Min-Max function is used to comprehensively couple the upper and lower bound dynamic system states and outputs to construct a dynamic-output multi-coupled Luenberger interval observer with richer degrees of freedom. Based on the dynamic-output multi-coupled Luenberger interval observer, an observer error variable system is constructed to realize major fault diagnosis of sensors and actuators; The multidimensional data collected by the dynamic-output multi-coupled Luenberger interval observer is fused with features. The fused data and the major fault types diagnosed are then input into a deep convolutional neural network model to output refined fault types. The dynamic-output multi-coupled Luenberger interval observer is as follows: In the formula, and These are the upper and lower bounds for estimating state variables; and Comprehensive boundary information for uncertainty; , For armature current, For rotational speed; Max and Min represent extreme value functions; Indicates the control status; Indicates the output state; matrix The observer gain to be solved; The output of the dynamic-output multi-coupled Luenberger interval observer is: In the formula, This is the output vector; , ; When no fault occurs, the actuator malfunctions. A value of 0 corresponds to the actual output and the estimated value for a specific state interval satisfying the following conditions: When an actuator failure occurs, the following conditions must be met: .

2. The fault intelligent monitoring and refined diagnosis method according to claim 1, characterized in that, The boundary information is: in, This is the network output of the neural network; The system matrix parameters are uncertain in structure; , For state vectors, The problem is an actuator malfunction. For unmodeled nonlinear dynamics, It is a time variable; This represents the network approximation error.

3. The intelligent fault monitoring and refined diagnosis method according to claim 1, characterized in that, The construction of the observer error variable system based on the dynamic-output multi-coupled Luenberger interval observer includes: The constructed error variables are: The error dynamic system matrix corresponding to the error variable is: The matrix is ​​in Metzler form; The total error dynamic system is further constructed as follows: Wherein, the system matrix is Based on the system matrix, the state estimation for the first-order augmented system is finally achieved, i.e., it satisfies... .

4. The intelligent fault monitoring and refined diagnosis method according to claim 1, characterized in that, The feature fusion includes data feature mining and data fusion; The data feature mining involves characterizing multidimensional data based on short-time Fourier transform, including: The discrete-time signal of the multidimensional data is used to obtain the spectral power spectrum signal using the discrete short-time Fourier transform. Perform a Fast Fourier Transform on the power spectrum signal; The power spectrum matrix of the power spectrum signal is finally obtained through fast Fourier transform.

5. The intelligent fault monitoring and refined diagnosis method according to claim 4, characterized in that, The data fusion includes: The power spectrum matrix is ​​converted into a two-dimensional power spectrum with three RGB channels; The features of multiple layers are fused through skip connections in the two-dimensional power spectrum.

6. The fault intelligent monitoring and refined diagnosis method according to any one of claims 1 to 5, characterized in that, The uncertain generalized system is: in, It is a state vector; This is the output vector; For control vectors; The problem is an actuator malfunction. and These are system interference and output interference, respectively. , , , , and have .

7. A fault intelligent monitoring and refined diagnosis device, used to implement the fault intelligent monitoring and refined diagnosis method according to any one of claims 1 to 6, characterized in that, The device includes: The interval observer category diagnosis module is used to determine whether a machine has malfunctioned and to classify the malfunction types into major categories. It includes a construction unit for the dynamic-output multi-coupled Luenberger interval observer and a malfunction type identification unit, wherein: The construction unit is used to construct a generalized system model by taking the actuator fault augmented system as an additional specific state of the uncertain generalized system to be detected, based on the uncertain generalized system. It also uses a boundary identification algorithm to obtain boundary information with dynamic characteristics based on a neural network for uncertain models caused by some unmodeled dynamics and parameter perturbations. Finally, based on the generalized system model, it uses the Min-Max function to comprehensively couple the upper and lower bound dynamic system states and outputs to construct a dynamic-output multi-coupled Luenberger interval observer with richer degrees of freedom. The identification unit is used to construct an observer error variable system based on the dynamic-output multi-coupled Luenberger interval observer to realize major fault diagnosis of sensors and actuators. The deep network refined diagnosis module is used to perform refined diagnosis based on the fault types obtained by the interval observer. This includes: fusing the multidimensional data collected by the dynamic-output multi-coupled Luenberger interval observer, inputting the fused data and the major fault types diagnosed into the deep convolutional neural network model, and outputting the refined fault types.

8. A storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the fault intelligent monitoring and refined diagnosis method according to any one of claims 1-6.

Images