Process fault detection method, device and medium based on introspective orthogonal autoencoder

Abstract

The present invention relates to a process fault detection method, device, and medium based on an introspective orthogonal autoencoder. Data samples to be tested are collected from a target manufacturing process test set, and random noise is generated. This data is then input into a fault detection network. Within the fault detection network, an encoder uses the data samples and random noise to obtain latent features of a hidden layer. A decoder reconstructs data based on the latent features. The squared prediction error between the data samples and the reconstructed data is calculated, and the squared prediction error is compared with a fault warning threshold to obtain a fault detection result. In the introspective orthogonal training module of the fault detection network, the encoder and decoder modules are introspectively trained to update network parameters, and negative sample data is added as training values. Compared with existing technologies, the present invention offers advantages such as high accuracy, strong adaptability, and robustness.

Description

Technical Field

[0001] The present invention relates to the field of multivariable fault detection in manufacturing processes, and in particular to a process fault detection method, device and medium based on an introspective orthogonal autoencoder. Background Art

[0002] Modern industrial manufacturing systems, such as chemical, steel, and semiconductor manufacturing, are experiencing a trend toward increasing complexity and informatization. This places higher demands on the reliability and safety of the manufacturing process, necessitating timely and accurate fault detection to improve production efficiency and product quality, and to prevent machine damage and casualties. With advances in sensing and data storage technologies, existing fault detection methods are trained using large amounts of collected historical multivariate fault data. In real-world manufacturing processes, the collected data exhibits nonlinearity, high redundancy, and noise, while also presenting significant challenges in collecting large amounts of fault type data. Therefore, in the absence of fault data, effective training and accurate detection of fault detection models present challenges.

[0003] The autoencoder model is an unsupervised deep model consisting primarily of an encoder, a hidden layer, and a decoder. The training objective is to transform the input into latent features in the hidden layer and then reconstruct the original input from the latent features. By feeding only normal manufacturing process data, the autoencoder model can be trained to recover normal data but not faulty data. During fault detection, if the fault detection metric of the autoencoder model exceeds the fault warning threshold, a fault is detected. While autoencoders can detect faults without faulty process data, real-world manufacturing scenarios often present issues such as noise interference and differences between normal and faulty data only in individual variables. This makes it difficult for models lacking negative samples to detect subtle faulty data. Therefore, improving the model's sensitivity to normal data and improving fault detection accuracy in the absence of faulty samples remains a technical challenge. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a process fault detection method, equipment and medium based on introspective orthogonal autoencoder. By introspectively training the encoder module and decoder module in the introspective orthogonal training module of the fault detection network to update the network parameters, negative sample data is added to improve the sensitivity to abnormal data in the absence of fault samples, thereby achieving a higher fault detection accuracy.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] According to one aspect of the present invention, a process fault detection method based on an introspective orthogonal autoencoder is provided, comprising the following steps:

[0007] S1. Collect data samples to be tested in the target manufacturing process test set, generate random noise, and input it into the fault detection network;

[0008] S2. In the fault detection network, the encoder obtains the latent features of the hidden layer through data samples and random noise;

[0009] S3, the decoder obtains the reconstructed data based on the latent features;

[0010] S4. Calculate the square prediction error of the data sample and the reconstructed data, compare the square prediction error with the fault warning threshold, and obtain a fault detection result.

[0011] Furthermore, the encoder in S2 is composed of a convolutional layer, and the decoder in S3 is composed of a transposed convolutional layer.

[0012] Furthermore, the reconstructed data includes reconstructed positive samples and reconstructed negative samples, wherein the reconstructed negative samples are obtained according to random noise.

[0013] Furthermore, in the introspective orthogonal training module of the fault detection network, introspective training is performed on the encoder module and the decoder module to update the network parameters.

[0014] Furthermore, the encoder updates the network parameters through introspective orthogonal training, and the encoder loss function Loss E The expression is:

[0015]

[0016] Among them, L R is the basic training function of positive samples, α is the weight parameter, L F is the basic training function for negative samples.

[0017] Furthermore, the decoder updates the network function through introspective orthogonal training, and the decoder loss function Loss D The expression is:

[0018] Loss D =L R +L F ,

[0019] Among them, L R is the basic training function of positive samples, L F is the basic training function for negative samples.

[0020] Furthermore, the positive sample basic training function L R The expression is:

[0021]

[0022] Among them, K is the number of positive samples, x k is the kth positive sample, y k represents the kth positive sample reconstruction data, λ represents the orthogonal regularization term weight coefficient, I is the identity matrix, and Z is the potential feature of the hidden layer corresponding to the positive sample;

[0023] Negative sample basic training function L F The expression is:

[0024]

[0025] Among them, K' is the number of negative samples, x k ' is the kth negative sample, y k ' represents the kth negative sample reconstruction data, λ represents the orthogonal regularization term weight coefficient, I is the identity matrix, and Z' is the potential feature of the hidden layer corresponding to the negative sample.

[0026] Furthermore, in S3, the expression for reconstructing data according to the potential features is:

[0027]

[0028] Among them, Y is the reconstructed data, a j represents the jth transposed convolution kernel, represents the transposed convolution operation, Z is the potential feature, and the expression is:

[0029] Z=σ(ΣX*w i +b),

[0030] Among them, σ is the hyperbolic tangent activation function, X is the input data sample to be detected, and w i is the i-th convolution kernel, * is the convolution operation, and b is the bias value.

[0031] According to a second aspect of the present invention, an electronic device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the method when executing the program.

[0032] According to a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method described above is implemented.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] (1) Improve the accuracy of fault detection: By introducing introspective orthogonal training, the encoder module and decoder module are introspectively trained and the network parameters are updated. In the introspective training process, the input noise is used as a negative sample through the decoder output to participate in the training, which makes up for the problem that the model is not sensitive enough to non-obvious abnormal data. At the same time, an orthogonal regularization term is added to the training function to improve the orthogonality of each potential feature and reduce the redundancy of the model feature learning information, so that the model can better identify potential fault data, improve the model feature learning ability, and improve the accuracy of fault detection.

[0035] (2) Adaptation to the lack of fault samples: The process fault detection method based on the introspective orthogonal autoencoder can be trained using only normal manufacturing process data. By introducing introspective training and orthogonal regularization terms, the model's sensitivity to abnormal data is improved, solving the problem of model training under the condition of lack of fault samples.

[0036] (3) Improve the efficiency and robustness of fault detection: The autoencoder network model based on convolutional layers and transposed convolutional layers is adopted to improve the efficiency and robustness of fault detection by optimizing the network structure and training process. The convolutional layer can extract the local features of the input data, while the transposed convolutional layer restores these features to an approximate representation of the original input data. This network structure enables the model to process large amounts of data more quickly and accurately identify fault data. At the same time, the introduction of introspective orthogonal training also enhances the robustness of the model, enabling it to maintain stable performance in the face of noise interference and data changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Flowchart of the process fault detection method based on introspective orthogonal autoencoder;

[0038] Figure 2 Schematic diagram of fault detection network training. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0040] like Figure 1 FIG. 1 is a flow chart of a process fault detection method based on an introspective orthogonal autoencoder, comprising the following steps:

[0041] S1. Collect data samples to be tested in the target manufacturing process test set, generate random noise, and input it into the fault detection network;

[0042] S2. In the fault detection network, the encoder obtains the latent features of the hidden layer through data samples and random noise;

[0043] S3, the decoder obtains the reconstructed data based on the latent features;

[0044] S4. Calculate the square prediction error of the data sample and the reconstructed data, compare the square prediction error with the fault warning threshold, and obtain a fault detection result.

[0045] like Figure 2 As shown in Figure 2, the training process of the fault detection network is shown in Figure 2. Figure 2 As can be seen, the encoder is composed of convolutional layers and the decoder is composed of transposed convolutional layers. The reconstructed data includes reconstructed positive samples and reconstructed negative samples, where the reconstructed negative samples are obtained based on random noise.

[0046] During the fault detection network training process, historical data from the target manufacturing process is first used as a training set. This data includes 52 items of pressure, temperature, flow rate, and composition collected at each moment. These items are then standardized to achieve a uniform order of magnitude and dimension. Based on this training set, a fault network model is constructed.

[0047] In the introspection orthogonal training module of the fault detection network, the encoder module and the decoder module are trained introspectively to update the network parameters. The encoder updates the network parameters through introspection orthogonal training, and the encoder loss function Loss E The expression is:

[0048]

[0049] Among them, L R is the basic training function of positive samples, α is the weight parameter, L F is the basic training function for negative samples. The decoder updates the network function through introspection orthogonal training, and the decoder loss function Loss D The expression is:

[0050] Loss D =L R +L F ,

[0051] Among them, L R is the basic training function of positive samples, L F is the basic training function for negative samples.

[0052] Positive sample basic training function L R The expression is:

[0053]

[0054] Among them, K is the number of positive samples, x k is the kth positive sample, y k represents the kth positive sample reconstruction data, λ represents the orthogonal regularization term weight coefficient, I is the identity matrix, and Z is the potential feature of the hidden layer corresponding to the positive sample;

[0055] Negative sample basic training function L F The expression is:

[0056]

[0057] Among them, K' is the number of negative samples, x k ' is the kth negative sample, y k ' represents the kth negative sample reconstruction data, λ represents the orthogonal regularization term weight coefficient, I is the identity matrix, and Z' is the potential feature of the hidden layer corresponding to the negative sample.

[0058] The expression for reconstructing data based on potential features is:

[0059]

[0060] Among them, Y is the reconstructed data, a j represents the jth transposed convolution kernel, represents the transposed convolution operation, Z is the potential feature, and the expression is:

[0061] Z=σ(ΣX*w i +b),

[0062] Among them, σ is the hyperbolic tangent activation function, X is the input data sample to be detected, and w i is the i-th convolution kernel, * is the convolution operation, and b is the bias value.

[0063] The process fault detection method based on the introspective orthogonal autoencoder in this embodiment significantly improves the accuracy of fault detection by introducing an introspective orthogonal training module. The model is trained using normal manufacturing process data, and the input data is converted into potential features of the hidden layer by the encoder, and then the data is reconstructed by the decoder. During the training process, noise is input into the hidden layer to generate negative samples for training, which enhances the model's sensitivity to subtle abnormal data. At the same time, by adding orthogonal regularization terms, the orthogonality of each potential feature is improved, the redundancy of feature learning information is reduced, and the feature learning ability of the model is improved. The method of this embodiment can still achieve effective fault detection of manufacturing process data in the absence of fault samples, and has a high detection accuracy.

[0064] The comparison of the results of fault detection by the method in this embodiment and the traditional method is shown in Table 1. As can be seen from Table 1, among the fault detection results obtained by each model method, the detection rate / false alarm rate combination of the fault diagnosis model of the introspective autoencoder is 92.36% / 3.65%. The false alarm rate of the introspective autoencoder is lower than that of the introspective orthogonal autoencoder in this embodiment, but the detection rate is not ideal. The detection rate of the orthogonal autoencoder is close to that of the introspective orthogonal autoencoder, but the false alarm rate is too high. Among the various fault detection methods, the method of this embodiment has the best effect, has excellent detection performance, and has high application value.

[0065] Table 1 Comparison of fault detection results of various models

[0066] Model Detection rate / false alarm rate Introspective Orthogonal Autoencoder Model 92.36 / 3.65 Introspective Autoencoder Model 75.12 / 0.07 Orthogonal Autoencoder Model 92.09 / 22.71 Conventional Autoencoder Model 86.10 / 5.35 Principal Component Analysis Model 78.40 / 3.26

[0067] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0068] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0069] Multiple components in the device are connected to the I / O interface, including: input units, such as a keyboard, mouse, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as magnetic disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunications networks. The processing unit performs the various methods and processes described above, such as the method of the present invention. For example, in some embodiments, the method of the present invention can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device via ROM and / or the communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the method of the present invention described above can be performed. Alternatively, in other embodiments, the CPU can be configured to perform the method of the present invention by any other suitable means (e.g., by means of firmware).

[0070] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0071] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0072] In the context of the present invention, machine-readable medium can be a tangible medium that can contain or store a program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0073] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A process fault detection method based on introspective orthogonal autoencoder, characterized in that: The following steps are involved: S1. Collect data samples to be tested in the target manufacturing process test set, generate random noise, and input it into the fault detection network; S2. In the fault detection network, the encoder obtains the latent features of the hidden layer through data samples and random noise; S3, the decoder obtains the reconstructed data based on the latent features; S4. Calculate the square prediction error between the data sample and the reconstructed data, compare the square prediction error with the fault warning threshold, and obtain a fault detection result; The encoder in S2 is composed of convolutional layers, and the decoder in S3 is composed of transposed convolutional layers; The reconstructed data includes reconstructed positive samples and reconstructed negative samples, wherein the reconstructed negative samples are obtained according to random noise; In the introspection orthogonal training module of the fault detection network, the encoder module and the decoder module are introspectively trained to update the network parameters; the encoder updates the network parameters through introspection orthogonal training, and the encoder loss function Loss E The expression is: ， in, L R is the basic training function of positive samples, α is the weight parameter, L F is the basic training function of negative samples; The decoder updates the network function through introspective orthogonal training, and the decoder loss function Loss D The expression is: ， in, L R is the basic training function of positive samples, L F is the basic training function of negative samples; Positive sample basic training function The expression is: ， in, K is the number of positive samples, x k For the k positive samples, y k Indicates the k Positive samples reconstruct the data, λ represents the weight coefficient of the orthogonal regularization term, I is the identity matrix, Z is the potential feature of the hidden layer corresponding to the positive sample; Negative sample basic training function The expression is: ， in, K ’ is the number of negative samples, x k ’ For the k negative samples, y k ’ Indicates the k Negative samples reconstruct data, λ represents the weight coefficient of the orthogonal regularization term, I is the identity matrix, Z ’ is the potential feature of the hidden layer corresponding to the negative sample; In S3, the expression for reconstructing data according to the potential features is: ， in, Y To reconstruct the data, a j Indicates the j transposed convolution kernels, represents the transposed convolution operation, Z is the potential feature, and the expression is: ， in, σ is the hyperbolic tangent activation function, X is the input data sample to be tested, w i For the i convolution kernels, ∗ is the convolution operation, b is the bias value.

2. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to claim 1 is implemented.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method as claimed in claim 1 is implemented.

Images