Fault detection method and system based on double-hidden layer feature adversarial auto-encoding network

Abstract

The application discloses a kind of based on double hidden layer feature counteracting auto-encoding network's fault detection method and system, belongs to process industry process fault detection technical field, the method includes: obtaining historical production data in industrial production process;Double hidden layer feature counteracting auto-encoding network model is constructed, and historical production data is used to train double hidden layer feature counteracting auto-encoding network model, and the trained model is obtained;Real-time production data in industrial production process is obtained;Real-time production data is input into the trained double hidden layer feature counteracting auto-encoding network model, based on the output result of double hidden layer feature counteracting auto-encoding network model, the fault detection of industrial production process is realized.The reliability of the present application scheme can improve fault detection and the accuracy of fault detection.

Description

Technical Field

[0001] This invention relates to the field of process fault detection technology in the process industry, and in particular to a fault detection method and system based on a dual-hidden-layer feature adversarial autoencoder network. Background Technology

[0002] With the continuous expansion of modern industrial production scale, the high interconnectivity between production units makes process failures more destructive than ever before. In actual production, changes in process parameters, sensor problems, and interference from the external environment can all lead to abnormalities in system components and functions, thus causing failures. Utilizing effective process monitoring methods can ensure the safe and stable operation of the system and continuous production under normal operating conditions, while also having significant practical implications for ensuring product quality, reducing consumption losses, and lowering costs. The main research tasks of process monitoring include process failure detection, failure magnitude identification, failure time estimation, failure mechanism diagnosis, and failure recovery. Failure detection is a fundamental step in process monitoring and one of its core tasks. By monitoring and analyzing various parameters and variables in the production process, it can help to promptly identify abnormalities and potential failures in industrial production processes, allowing for timely measures to prevent accidents and ensure the safety of production equipment and personnel.

[0003] Fault detection methods include model-based methods, knowledge-based methods, and data-driven methods. Model-based methods monitor systems by establishing analytical models of industrial processes; however, the complex and uncertain structures of actual systems make it difficult to establish accurate mechanistic models. Knowledge-based methods heavily rely on process knowledge and production experience, making them difficult to apply to large-scale, tightly coupled modern industrial production processes. In recent years, the widespread application of fieldbus technology and distributed control systems in industrial processes has enabled the storage of large amounts of industrial process data through mature data acquisition and storage systems, leading to the rapid development of data-driven methods.

[0004] Due to their unique advantages in handling high-dimensional coupled data, multivariate statistical methods have rapidly developed and been widely applied to various industrial processes in recent decades. However, it is worth noting that most data in actual industrial processes are time-series dynamic data. There is temporal correlation between different stages of a process industry process; the process state at one moment affects the state at another. Furthermore, production processes are affected by various external disturbances and operational changes, such as fluctuations in raw material composition, changes in environmental conditions, equipment aging, and adjustments to operating parameters. These factors cause dynamic changes in process variables, resulting in time-varying data. Therefore, incorporating temporal information into the model building process is crucial. Dynamic extensions of multivariate statistical process monitoring methods, such as dynamic principal component analysis, dynamic independent component analysis, and dynamic partial least squares, use multiple historical samples to expand the process data matrix. However, the autocorrelation of fractional components may negatively impact the monitoring statistics. Therefore, the aforementioned dynamic methods based on augmented matrix strategies cannot fundamentally overcome the dynamic characteristics.

[0005] In recent years, deep neural networks capable of handling temporal dependencies have been extensively studied. For example, by introducing gating mechanisms and memory units into Long Short-Term Memory (LSTM) networks, long-term dependencies in time-series data can be effectively captured. However, these methods often have limitations in capturing the complex and uneven dynamic characteristics of industrial data. They focus too much on extracting time-series information while ignoring the overall trend of process changes, and the networks are too simple in structure to extract multi-layered information. This results in insufficient temporal information and affects the accuracy of fault detection. Summary of the Invention

[0006] This invention provides a fault detection method and system based on a dual-hidden-layer feature adversarial autoencoder network, which solves the technical problem that existing fault detection methods are unable to extract multi-layer information, resulting in low detection accuracy.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] On one hand, the present invention provides a fault detection method based on a dual-hidden-layer feature-adversarial autoencoder network, the fault detection method based on a dual-hidden-layer feature-adversarial autoencoder network comprising:

[0009] Acquire historical production data during the industrial production process;

[0010] A dual-hidden-layer feature adversarial autoencoder network model is constructed, and historical production data is used to train the dual-hidden-layer feature adversarial autoencoder network model to obtain a trained dual-hidden-layer feature adversarial autoencoder network model.

[0011] To acquire real-time production data during the industrial production process;

[0012] Real-time production data is input into a trained dual-hidden-layer feature adversarial autoencoder network model. Based on the output of the dual-hidden-layer feature adversarial autoencoder network model, fault detection in the industrial production process is achieved.

[0013] Furthermore, the dual-hidden-layer feature adversarial autoencoder model includes an encoder, a decoder, and a discriminator;

[0014] The encoder's output layer includes a first hidden layer structure and a second hidden layer structure; wherein, the first hidden layer structure is used to output the latent variables corresponding to the encoder input data; and the second hidden layer structure is used to output the slow features corresponding to the encoder input data as feature data based on slow feature analysis.

[0015] The decoder is used to obtain the reconstructed data corresponding to the encoder input data based on the latent variables output by the first hidden layer structure in the encoder.

[0016] The discriminator is used during training to identify whether its input data originates from slow features output by the second hidden layer structure in the encoder or from real data that follows a Gaussian distribution.

[0017] Furthermore, the loss function of the dual-hidden-layer feature adversarial autoencoder network model includes the reconstruction loss L. r Slow Feature Loss L s And the losses of the generator and discriminator; among which,

[0018] Reconstruction loss L r Defined as minimizing the difference between the reconstructed data output by the decoder and the input data by the encoder;

[0019] Slow Feature Loss L s Defined as:

[0020]

[0021] Among them, z si 1 =[z si,2 ,z si,3 ,z si,4 ,…,z si,n ], z si 2 =[z si,1 ,z si,2 ,z si,3 ,…,z si,n-1 ], z si,j Let z represent the i-th slow feature vector. si The j-th sample; j = 1, 2, ..., n; n represents the number of training samples; q represents the number of features;

[0022] The generator's loss aims to make the slow features output by the encoder follow a Gaussian distribution, so that the discriminator is confused and has difficulty distinguishing whether its input data is the slow features output by the encoder or the real data that follows a Gaussian distribution. The discriminator's loss aims to improve the discriminator's classification ability so that the discriminator can distinguish whether its input data is the slow features output by the encoder or the real data that follows a Gaussian distribution.

[0023] Furthermore, the dual-hidden-layer feature adversarial autoencoder network model is trained using historical production data, including:

[0024] Historical production data is normalized to obtain training samples. These training samples are then fed into a dual-hidden-layer feature adversarial autoencoder network model for training according to a set batch and period. During training, the encoder is used as the generator, forming a generative adversarial network with the discriminator. The slow feature z output from the second hidden layer of the encoder is used as the generator. s And randomly generated real data x that follows a Gaussian distribution D The data is fed into a discriminator, which determines that the input data is z. s or x D The generator is designed to make z s The data follows a Gaussian distribution to confuse the discriminator, making it difficult for it to distinguish whether the input data is z. s or x D The discriminator aims to improve its classification ability to distinguish z. s and x D The generator and discriminator are trained adversarially until the loss reaches a Nash equilibrium.

[0025] The encoder and decoder are trained iteratively using reconstruction loss and slow feature loss, and the discriminator and encoder are trained iteratively using adversarial and generative losses, thus obtaining a trained model.

[0026] Furthermore, after training the dual-hidden-layer feature adversarial autoencoder network model using historical production data, the fault detection method based on the dual-hidden-layer feature adversarial autoencoder network also includes:

[0027] The reconstructed data and feature data of the training data are obtained by using a dual-hidden-layer feature adversarial autoencoder network model.

[0028] The residual space of the training data is constructed using the reconstructed data and the training data. The feature space of the training data is constructed using the feature data. The SPE statistic of the training data is established in the residual space, and the T-statistic of the training data is established in the feature space. 2 Statistic;

[0029] The control limits and T of the SPE statistic were calculated based on the training data. 2 Control limits for statistics.

[0030] Furthermore, the step of inputting real-time production data into a trained dual-hidden-layer feature adversarial autoencoder network model, and realizing fault detection in the industrial production process based on the output of the dual-hidden-layer feature adversarial autoencoder network model, includes:

[0031] The real-time production data is normalized, and the normalized real-time production data is input into the trained dual-hidden-layer feature adversarial autoencoder network model. The dual-hidden-layer feature adversarial autoencoder network model is used to obtain the reconstructed data and feature data of the real-time production data.

[0032] A residual space for real-time production data is constructed using reconstructed data and real-time production data. A feature space for real-time production data is constructed using its characteristic data. The SPE statistic for real-time production data is established within the residual space, and the T-statistic for real-time production data is established within its feature space. 2 Statistic;

[0033] SPE statistics and T based on real-time production data 2 Statistical measures are used to determine whether a malfunction has occurred in the current production process.

[0034] Furthermore, the SPE statistics and T based on real-time production data 2 Statistical measures are used to determine whether a fault has occurred in the current production process, including:

[0035] If the SPE statistic of real-time production data or the T of real-time production data 2 If the statistical value exceeds the corresponding control limit, the current production process is determined to have malfunctioned.

[0036] If the SPE statistic of real-time production data or the T of real-time production data 2 If none of the statistical values ​​exceed the corresponding control limits, it is determined that no malfunction has occurred in the current production process.

[0037] On the other hand, the present invention also provides a fault detection system based on a dual-hidden-layer feature adversarial autoencoder network, the fault detection system based on the dual-hidden-layer feature adversarial autoencoder network comprising:

[0038] The offline modeling module is used for:

[0039] Acquire historical production data during the industrial production process;

[0040] A dual-hidden-layer feature adversarial autoencoder network model is constructed, and historical production data is used to train the dual-hidden-layer feature adversarial autoencoder network model to obtain a trained dual-hidden-layer feature adversarial autoencoder network model.

[0041] The online monitoring module is used for:

[0042] To acquire real-time production data during the industrial production process;

[0043] Real-time production data is input into a trained dual-hidden-layer feature adversarial autoencoder network model. Based on the output of the dual-hidden-layer feature adversarial autoencoder network model, fault detection in the industrial production process is achieved.

[0044] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0045] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the above method.

[0046] The beneficial effects of the technical solution provided by this invention include at least the following:

[0047] The technical solution of this invention constructs a dual-hidden-layer network, enabling the network to learn multi-level information, including more abstract and diverse features, thus enhancing the model's expressive and generalization abilities. Furthermore, by constructing slowly varying feature loss constraints for the features, features containing system operational modal information can be extracted, thereby reflecting the system's long-term behavior and trends. Based on this, a generative adversarial model is used to make the features follow a prior Gaussian distribution, improving the reliability and accuracy of fault detection. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the execution flow of the fault detection method based on a dual-hidden-layer feature adversarial autoencoder network provided in an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the structure of the dual hidden layer feature adversarial autoencoder network model provided in an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the fault detection process based on a dual-hidden-layer feature adversarial autoencoder network provided in an embodiment of the present invention;

[0052] Figure 4 This is a system block diagram of the electronic device provided in the embodiments of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0054] First, it should be noted that in the embodiments of the present invention, the words "exemplarily," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplarily" is intended to present the concept in a specific manner. Furthermore, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either one or the other.

[0055] Furthermore, in the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0056] First Embodiment

[0057] This embodiment provides a fault detection method based on a dual-hidden-layer feature adversarial autoencoder network. Addressing the problem of existing basic networks having a single layer and difficulty in extracting multi-layer information, it proposes an autoencoder network with a dual-hidden-layer structure that simultaneously considers both temporal information and the overall trend of process changes. One set of hidden-layer features is used for input data reconstruction, while the other set focuses on extracting process trend information, learning the dynamic features and deep representation of variables in an unsupervised learning manner. To address the insufficient process information in the feature representation, a slow feature loss constraint is constructed for the features extracting dynamic information, ensuring they reflect the process change trend of the research object. This extracts Gaussian distribution features that reflect the process change trend of the research object, thus solving the problem of insufficient temporal information caused by the dynamism of complex large-scale industrial processes. Based on this network, online fault detection of dynamic industrial processes is achieved, improving detection accuracy. This method can be implemented by an electronic device, which can be a terminal or a server.

[0058] Specifically, the execution flow of this method is as follows: Figure 1 As shown, it includes the following steps:

[0059] S1, acquire historical production data during the industrial production process;

[0060] The application scenario of this method can be fault detection in chemical production processes, such as fault detection in the vinyl acetate production process. When applied to fault detection in the vinyl acetate production process, the production data obtained in this embodiment can include: temperature, concentration, liquid level, pressure, raw material supply rate, etc.

[0061] S2, construct a dual-hidden-layer feature adversarial autoencoder network model, and train the dual-hidden-layer feature adversarial autoencoder network model using historical production data to obtain a trained dual-hidden-layer feature adversarial autoencoder network model.

[0062] In this embodiment, the dual-hidden-layer feature adversarial autoencoder network model is constructed based on an autoencoder network. The dual-hidden-layer feature adversarial autoencoder network model is constructed by modifying the output of the encoder in the autoencoder to a dual-hidden-layer structure and adding a discriminator to the autoencoder network. The autoencoder includes an encoder and a decoder, and aims to learn the low-dimensional feature representation of the data while reconstructing the input data through unsupervised learning.

[0063] Specifically, such as Figure 2 As shown, the dual-hidden-layer feature adversarial autoencoder network model constructed in this embodiment includes an encoder, a decoder, and a discriminator. The encoder and decoder are used to reconstruct data and generate low-dimensional features. By minimizing the reconstruction loss, reconstructed data similar to the training data and features containing information from the training data can be obtained. The encoder's output is a dual-hidden-layer model. One layer of features is input to the decoder for data reconstruction, while the other layer, through the addition of slow feature constraints, extracts slow features containing process temporal information and global change trends, enabling the network to extract more abstract and multi-layered features.

[0064] During training, normal samples are collected as training samples and normalized. The training data is fed into the network in predetermined batches and periods. Slow features and data following a Gaussian distribution are fed into the discriminator, while the encoder simultaneously acts as the generator of the generative adversarial network. The discriminator determines whether the data originates from slow features or the original Gaussian data. The generator produces slow features following a Gaussian distribution to confuse the discriminator and make it difficult for it to distinguish. The encoder and decoder are iteratively trained using reconstruction loss and slow feature loss, and the discriminator and generator (i.e., the encoder) are iteratively trained using adversarial and generative losses. This allows the encoder and decoder to obtain reconstructed data similar to the training data, as well as features containing information from the training data. Simultaneously, the encoder can output slow features following a Gaussian distribution, thus obtaining a well-trained model. The generative adversarial network (GAN) is a network that includes a generative model and a discriminative model, aiming to estimate the distribution of the sample data using a binary minimax game problem.

[0065] Furthermore, the loss function of the dual-hidden-layer feature adversarial autoencoder network model includes the reconstruction loss L. r Slow Feature Loss L s In addition to adversarial and generative losses, the last two fully connected layers of the encoder are defined as f(·) and g(·), where f(·) and g(·) are nonlinear functions. The other modules of the encoder besides these two layers are G... e (·;θ e The decoder is represented as G. d (·;θ d The discriminator is G. y (·;θ y ), θ e θ d and θ y These are the parameters of the encoder, decoder, and discriminator networks, respectively. For a given normalized input x, f(G) e (·;θ e The decoder G maps the input x to the latent variable z. d (·;θ d Map z to the reconstructed data Reconstruction loss L r Defined as minimizing reconstructed data The difference between the input x and the latent variable z and the reconstructed data. It is expressed as follows:

[0066] z = f(G) e (x;θ e ))

[0067]

[0068] Slow feature z s It is calculated by the following formula:

[0069] z s =g(G e (x;θ e ))

[0070] Furthermore, this embodiment uses a slow feature analysis method to analyze z. s =[z si,1 ,z si,2 ,z si,3 ,…,z si,j ,…,z si,n A slow feature loss L was constructed. s , z si,j Represents the i-th eigenvector z siThe j-th sample, where 1 < j < n and 1 ≤ i << q. Here, n represents the number of training samples and q represents the number of features. Among them, Slow Feature Analysis (SFA) is an unsupervised dimensionality reduction method. For a given multi-dimensional input, it seeks an input-output function mapping that makes the output signal change as slowly as possible.

[0071] In model training, by minimizing L s makes z s able to capture the trend of process changes and extract the temporal dynamic information of the process; specifically, the slow feature loss L s is defined as:

[0072]

[0073] where z si 1 = [z si,2 , z si,3 , z si,4 , …, z si,n , that is, the second to the last sample of the i-th feature z s of z si . z si 2 = [z si,1 , z si,2 , z si,3 , …, z si,n-1 , that is, the first to the second-to-last sample of the i-th feature z s of z si ; z si,j represents the j-th sample of the i-th slow feature vector z si ; j = 1, 2, …, n; n represents the number of training samples; q represents the number of features;

[0074] Furthermore, based on adversarial training, this embodiment constructs the losses of the generator and the discriminator respectively:

[0075] y fake = G y (z s ; θ y ) = G y (g(G e (x; θ e )); θ y )

[0076] y real = G y (x D ; θ y )

[0077]

[0078] Where, x D G represents randomly generated real data that follows a Gaussian distribution; y Represents the discriminator; θ y Indicates the parameters of the discriminator; z s G represents the slow feature output by the second hidden layer structure. e The encoder is represented by x; the encoder input data is represented by θ. e represents the encoder parameters; g(·) represents the nonlinear function.

[0079] The generator aims to generate slow features z s The data follows a Gaussian distribution to confuse the discriminator and make it difficult to distinguish the source of the data. s or x D The discriminator aims to improve its classification ability to distinguish z. s and x D The generator and discriminator are trained adversarially until the loss reaches a Nash equilibrium, ultimately achieving the goal of adversarial training, i.e., z. s It follows a Gaussian distribution but retains its essential characteristic information.

[0080] Using the above scheme, this embodiment constructs a dual-hidden-layer autoencoder network structure, adding a slow feature loss constraint to the features, enabling the model features to capture the dynamic temporal information and global change trends of the process industry production process. The features contain richer information, thereby improving the accuracy of the network model. Furthermore, by performing generative adversarial training on the slow features, the distribution of the slow features is made to follow a Gaussian distribution during training, providing a reasonable basis for the subsequent calculation of the control limits of fault detection statistics, thereby improving detection accuracy.

[0081] S3 acquires real-time production data during the industrial production process;

[0082] S4 inputs real-time production data into a trained dual-hidden-layer feature adversarial autoencoder network model, and uses the output of the dual-hidden-layer feature adversarial autoencoder network model to achieve fault detection in the industrial production process.

[0083] Specifically, the process of fault detection based on the above model is as follows: Figure 3 As shown, it includes the following steps:

[0084] S41. Obtain normal process data in the industrial production process, and normalize the obtained normal process data to obtain training data. Input the training data into the dual hidden layer feature adversarial autoencoder network model to obtain the reconstructed data and feature data of the training data obtained by the dual hidden layer feature adversarial autoencoder network model.

[0085] S42. Construct the residual space of the training data using the reconstructed data and the training data; construct the feature space of the training data using the feature data of the training data; establish the SPE statistic of the training data in the residual space of the training data; and establish the T value of the training data in the feature space of the training data. 2 Statistic;

[0086] S43. Calculate SPE and T based on the reconstructed data and slow features from the training data. 2 Control limits for statistics;

[0087] S44. Obtain real-time production data, normalize the obtained real-time production data, input the normalized real-time production data into the trained dual-hidden-layer feature adversarial autoencoder network model, and use the dual-hidden-layer feature adversarial autoencoder network model to obtain the reconstructed data and feature data of the real-time production data.

[0088] S45. Construct the residual space of real-time production data using reconstructed data and real-time production data. Construct the feature space of real-time production data using characteristic data. Establish the SPE statistic of real-time production data in the residual space of real-time production data. Establish the T statistic of real-time production data in the feature space of real-time production data. 2 Statistics; thus, using reconstructed data from real-time production data and slow characteristics, the values ​​of two statistics are calculated at each sampling time;

[0089] S46. SPE statistics and T based on real-time production data 2 Statistics determine whether a fault has occurred in the current production process; if the SPE statistic of real-time production data or the T statistic of real-time production data... 2 If the statistic exceeds the corresponding control limit, a fault is determined to have occurred in the current production process; if the SPE statistic of the real-time production data or the T statistic of the real-time production data... 2 If none of the statistical values ​​exceed the corresponding control limits, it is determined that no malfunction has occurred in the current production process.

[0090] Among them, SPE and T 2 The calculation process of the statistic is as follows:

[0091] For training data X = [x1, x2, ..., x...] n ]∈R n×m Where n represents the number of samples, m represents the number of process variables, and the standardized X is fed into the network to obtain the slow feature Z. s and reconstructing data For test data X te =[x te,1 ,x te,2 ,…,x te,n ]∈Rp×m Where p represents the number of test data samples, m represents the number of process variables, and X is the standardized value. te It is fed into the network to obtain the slow feature Z. te,s and reconstructing data

[0092] T 2 The SPE statistic is used to monitor the feature space composed of slow features, and the SPE statistic is used to monitor the residual space composed of the original data and the reconstructed data. The calculation process is as follows;

[0093] T 2 =z s z s T

[0094]

[0095] Furthermore, the control limits for the two statistics are calculated as follows: given a confidence level α, T of the training data... 2 The control limits of SPE are determined by the F-distribution and χ², respectively. 2 The distribution was calculated.

[0096]

[0097]

[0098] Among them, F α (k,nu) and The distributions are F-distribution and χ² distribution. 2 The upper critical point of the α% distribution. u,n and Represents degrees of freedom, u is T 2 The statistic is a measure of independent information in the sample, and it determines T. 2 The distribution of the statistic and the associated critical values. m and v are the mean and variance of the SPE. If the statistic exceeds the control limits, a fault is considered detected. Otherwise, the process is considered normal.

[0099] In summary, this embodiment provides a fault detection method based on a dual-hidden-layer feature adversarial autoencoder network. A dual-hidden-layer feature adversarial autoencoder network is constructed, enabling the network to learn multi-level information and extract more comprehensive features. By constructing a slow feature loss, dynamic information and overall change trends of the production process are extracted, thereby improving the accuracy and reliability of fault detection.

[0100] Second Embodiment

[0101] This embodiment provides a fault detection system based on a dual-hidden-layer feature adversarial autoencoder network, which includes the following modules:

[0102] The offline modeling module is used for:

[0103] Acquire historical production data during the industrial production process;

[0104] A dual-hidden-layer feature adversarial autoencoder network model is constructed, and historical production data is used to train the dual-hidden-layer feature adversarial autoencoder network model to obtain a trained dual-hidden-layer feature adversarial autoencoder network model.

[0105] The online monitoring module is used for:

[0106] To acquire real-time production data during the industrial production process;

[0107] Real-time production data is input into a trained dual-hidden-layer feature adversarial autoencoder network model. Based on the output of the dual-hidden-layer feature adversarial autoencoder network model, fault detection in the industrial production process is achieved.

[0108] It should be noted that the fault detection system based on dual-hidden-layer feature adversarial autoencoders in this embodiment corresponds to the fault detection method based on dual-hidden-layer feature adversarial autoencoders in the first embodiment described above. The functions implemented by each module in the fault detection system based on dual-hidden-layer feature adversarial autoencoders in this embodiment correspond one-to-one with the process steps in the fault detection method based on dual-hidden-layer feature adversarial autoencoders in the first embodiment described above; therefore, they will not be repeated here.

[0109] Third Embodiment

[0110] This embodiment provides an electronic device, such as... Figure 4 As shown, the electronic device includes a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. Furthermore, the electronic device may also include a transceiver, the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0111] Below, in conjunction with Figure 4 A detailed introduction to each component of this electronic device is provided below:

[0112] The processor is the control center of the electronic device. The electronic device may include multiple processors, each of which can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The term "processor" can refer to a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), other general-purpose processors, application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), one or more field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0113] In a specific implementation, as one example, the processor may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 shown are, of course, merely illustrative examples.

[0114] The memory is used to store the software program that executes the solution of the present invention, and the processor controls its execution. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0115] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may be integrated with the processor or may exist independently, and may be accessed through the interface circuit of the electronic device (…). Figure 4 (Not shown in the image) is coupled to the processor; however, this embodiment of the invention does not impose specific limitations on this.

[0116] The transceiver may include a receiver and a transmitter. Figure 4 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver can be integrated with the processor or exist independently, and is connected through the interface circuit of the electronic device (…). Figure 4 (Not shown in the image) is coupled to the processor, and this embodiment of the invention does not specifically limit this.

[0117] In addition, it should be noted that, Figure 4 The structure of the electronic device shown is not intended to limit the device. Actual devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Furthermore, the technical effects achieved by this electronic device when performing the method of the first embodiment described above can be referenced to the technical effects described in the first embodiment; therefore, they will not be repeated here.

[0118] Fourth embodiment

[0119] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0120] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely or partially hardware embodiment, a completely or partially software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented in software, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any usable medium accessible to a computer or a data storage device such as a server or data center containing one or more sets of usable media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive (SSD).

[0121] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0122] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0123] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element. Furthermore, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone, where A and B can be singular or plural. Additionally, the character " / " in this text generally indicates an "or" relationship between the preceding and following objects, but it can also indicate an "AND / OR" relationship. Please refer to the context for specific interpretations. "At least one" refers to one or more items, while "more than" refers to two or more items. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can be represented as: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0124] Furthermore, it is understood that in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0125] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0126] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of functional modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, the functional units in the various embodiments of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0127] If the method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0128] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A fault detection method based on a dual-hidden-layer feature adversarial autoencoder network, used for fault detection in the vinyl acetate production process, characterized in that, include: Acquire historical production data from the industrial production process; this production data includes temperature, concentration, liquid level, pressure, and raw material supply rate during the vinyl acetate production process. A dual-hidden-layer feature adversarial autoencoder network model is constructed, and historical production data is used to train the dual-hidden-layer feature adversarial autoencoder network model to obtain a trained dual-hidden-layer feature adversarial autoencoder network model. To acquire real-time production data during the industrial production process; Real-time production data is input into a trained dual-hidden-layer feature adversarial autoencoder network model. Based on the output of the dual-hidden-layer feature adversarial autoencoder network model, fault detection in the industrial production process is achieved. The dual-hidden-layer feature adversarial autoencoder network model includes an encoder, a decoder, and a discriminator; The encoder's output layer includes a first hidden layer structure and a second hidden layer structure; wherein, the first hidden layer structure is used to output the latent variables corresponding to the encoder input data; and the second hidden layer structure is used to output the slow features corresponding to the encoder input data as feature data based on slow feature analysis. The decoder is used to obtain the reconstructed data corresponding to the encoder input data based on the latent variables output by the first hidden layer structure in the encoder. The discriminator is used during training to identify whether its input data originates from slow features output by the second hidden layer structure in the encoder or from real data that follows a Gaussian distribution.

2. The fault detection method based on a dual-hidden-layer feature adversarial autoencoder network as described in claim 1, characterized in that, The loss function of the dual-hidden-layer feature adversarial autoencoder model includes reconstruction loss. Slow feature loss And the losses of the generator and discriminator; among which, Reconstruction losses Defined as minimizing the difference between the reconstructed data output by the decoder and the input data by the encoder; Slow feature loss Defined as: ； in, , , Indicates the first slow feature vectors The One sample; n represents the number of training samples; Indicates the number of features; The generator's loss aims to make the slow features output by the encoder follow a Gaussian distribution, so that the discriminator is confused and has difficulty distinguishing whether its input data is the slow features output by the encoder or the real data that follows a Gaussian distribution. The discriminator's loss aims to improve the discriminator's classification ability so that the discriminator can distinguish whether its input data is the slow features output by the encoder or the real data that follows a Gaussian distribution.

3. The fault detection method based on a dual-hidden-layer feature adversarial autoencoder network as described in claim 2, characterized in that, Training a dual-hidden-layer feature adversarial autoencoder network model using historical production data includes: Historical production data is normalized to obtain training samples. These training samples are then fed into a dual-hidden-layer feature adversarial autoencoder network model for training according to a set batch and period. During training, the encoder is used as the generator, forming a generative adversarial network with the discriminator. The slow features output from the second hidden layer of the encoder are used as the training samples. and randomly generated real data that follows a Gaussian distribution The data is fed into a discriminator, which determines whether the input data is... still The generator is designed to make The data follows a Gaussian distribution to confuse the discriminator, making it difficult to distinguish whether the input data is... still The discriminator aims to improve its classification ability to distinguish between different categories. and The generator and discriminator are trained adversarially to bring the loss to a Nash equilibrium. The encoder and decoder are trained iteratively using reconstruction loss and slow feature loss, and the discriminator and encoder are trained iteratively using adversarial and generative losses, thus obtaining a trained model.

4. The fault detection method based on a dual-hidden-layer feature adversarial autoencoder network as described in claim 1, characterized in that, After training the dual-hidden-layer feature adversarial autoencoder network model using historical production data, the fault detection method based on the dual-hidden-layer feature adversarial autoencoder network further includes: The reconstructed data and feature data of the training data are obtained by using a dual-hidden-layer feature adversarial autoencoder network model. The residual space of the training data is constructed using the reconstructed data and the training data. The feature space of the training data is constructed using the feature data of the training data. Finally, the training data is established within the residual space. Statistics are used to construct training data in the feature space of the training data. Statistic; Calculated based on training data Control limits of statistics and Control limits for statistics.

5. The fault detection method based on a dual-hidden-layer feature adversarial autoencoder network as described in claim 4, characterized in that, The process of inputting real-time production data into a trained dual-hidden-layer feature adversarial autoencoder (DAU) network model and then using the output of the DAU model to achieve fault detection in the industrial production process includes: The real-time production data is normalized, and the normalized real-time production data is input into the trained dual-hidden-layer feature adversarial autoencoder network model. The dual-hidden-layer feature adversarial autoencoder network model is used to obtain the reconstructed data and feature data of the real-time production data. The residual space of real-time production data is constructed using reconstructed data and real-time production data. The feature space of real-time production data is then constructed using its characteristic data. Finally, the residual space of real-time production data is established within the residual space of the real-time production data. Statistics are used to establish real-time production data within the feature space of real-time production data. Statistic; Based on real-time production data Statistics and Statistical measures are used to determine whether a malfunction has occurred in the current production process.

6. The fault detection method based on a dual-hidden-layer feature adversarial autoencoder network as described in claim 5, characterized in that, The basis of real-time production data Statistics and Statistical measures are used to determine whether a fault has occurred in the current production process, including: If real-time production data Statistical quantities or real-time production data If the statistical value exceeds the corresponding control limit, the current production process is determined to have malfunctioned. If real-time production data Statistical quantities or real-time production data If none of the statistical values ​​exceed the corresponding control limits, it is determined that no malfunction has occurred in the current production process.

7. A fault detection system based on a dual-hidden-layer feature adversarial autoencoder network, used for fault detection in the vinyl acetate production process, characterized in that, include: The offline modeling module is used for: Acquire historical production data from the industrial production process; this production data includes temperature, concentration, liquid level, pressure, and raw material supply rate during the vinyl acetate production process. A dual-hidden-layer feature adversarial autoencoder network model is constructed, and historical production data is used to train the dual-hidden-layer feature adversarial autoencoder network model to obtain a trained dual-hidden-layer feature adversarial autoencoder network model. The online monitoring module is used for: To acquire real-time production data during the industrial production process; Real-time production data is input into a trained dual-hidden-layer feature adversarial autoencoder network model. Based on the output of the dual-hidden-layer feature adversarial autoencoder network model, fault detection in the industrial production process is achieved. The dual-hidden-layer feature adversarial autoencoder network model includes an encoder, a decoder, and a discriminator; The encoder's output layer includes a first hidden layer structure and a second hidden layer structure; wherein, the first hidden layer structure is used to output the latent variables corresponding to the encoder input data; and the second hidden layer structure is used to output the slow features corresponding to the encoder input data as feature data based on slow feature analysis. The decoder is used to obtain the reconstructed data corresponding to the encoder input data based on the latent variables output by the first hidden layer structure in the encoder. The discriminator is used during training to identify whether its input data originates from slow features output by the second hidden layer structure in the encoder or from real data that follows a Gaussian distribution.

Images