Multi-working-condition fault identification method of multi-scale cascade deep belief network
Through the end-to-end fault diagnosis architecture of a multi-scale cascading deep confidence network, advanced features are automatically learned from the original vibration data and identified fault patterns, solving the problem of artificial intervention and poor universality of shallow learning methods in industrial equipment fault detection. Achieve more efficient and accurate troubleshooting.
Patent Information
- Application Number
- CN202410968400.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The data-driven shallow learning fault diagnosis method in the prior art requires a lot of human intervention in industrial equipment fault detection, relying on the experience of technicians, feature extraction and classification are carried out separately, with poor versatility, and is only suitable for specific diagnostic occasions.
Adopting the end-to-end fault diagnosis architecture of a multi-scale cascading deep confidence network, the multi-scale coarse-grained layer, multi-scale feature learning layer and classification layer automatically learns advanced features from the original vibration data and recognizes the fault mode to achieve end-to-end Troubleshooting.
It improves the adaptability and versatility of fault diagnosis, reduces human intervention, can accurately identify mechanical faults under different working conditions, provides more complete and richer fault characteristic information, and improves diagnostic capabilities.
Smart Images

Figure CN119004251A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of industrial fault identification, and in particular to a multi-operating condition fault identification method of a multi-scale cascaded deep belief network. Background Art
[0002] In actual engineering applications, due to various influencing factors (such as fatigue, wear, overload and corrosion), industrial equipment is prone to local damage, which may lead to mechanical system shutdown or even casualties. Therefore, in order to ensure the safety and stability of mechanical systems, improve the reliability of equipment and reduce maintenance costs, the condition monitoring and fault diagnosis of mechanical systems have received more and more attention. In addition, efficient and practical intelligent diagnosis methods have always been the goal of scholars. In other words, it is of great significance to develop a new technology with excellent automation and high precision to identify mechanical faults under different working conditions. In recent years, data-driven fault diagnosis has been a hot research direction in the field of mechanical intelligent fault diagnosis. Traditionally, data-driven fault diagnosis methods mainly focus on two stages (i.e., feature extraction and fault identification). At present, data-driven related intelligent fault diagnosis methods have been studied. A large number of studies have been conducted, including back propagation neural network (BPNN), extreme learning machine (ELM), support vector machine (SVM), k-nearest neighbor (KNN), hidden Markov model (HMM) and cluster analysis; however, the above fault diagnosis methods belong to shallow learning processes, which are essentially manual feature extraction and then fault classification, which is relatively complex and cumbersome; in addition, the above data-driven diagnostic procedures require a lot of prior knowledge and expert experience about signal processing, thus limiting their adaptability and diagnostic capabilities; when fault detection is performed on industrial equipment, a data-driven shallow learning fault diagnosis scheme is used for detection. The shallow learning scheme requires a lot of human intervention and relies on the experience of technicians, and feature extraction and classification are performed separately and cannot be optimized at the same time. At the same time, the shallow learning scheme is only applicable to specific diagnostic occasions and has poor versatility. Summary of the invention
[0003] In order to overcome the problem of using data-driven shallow learning fault diagnosis schemes for fault detection of industrial equipment, shallow learning schemes require a lot of human intervention, rely on the experience of technicians, and feature extraction and classification are performed separately and cannot be optimized at the same time. At the same time, shallow learning schemes are only applicable to specific diagnostic occasions and have poor versatility.
[0004] The technical solution of the present invention is: a multi-condition fault identification method of a multi-scale cascaded deep belief network, comprising the following steps:
[0005] S11: Collect fault data of industrial equipment that needs to be monitored for faults;
[0006] S12: inputting the collected fault data into a multi-scale cascaded deep belief network, and training the end-to-end fault diagnosis architecture based on the multi-scale cascaded deep belief network;
[0007] S13: At the same time, an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network is used to identify fault results.
[0008] Preferably, the scheme is mainly divided into three steps: fault data collection, end-to-end fault diagnosis architecture training and recognition result output based on multi-scale cascaded deep belief network; decomposing the collected original fault data into a series of sub-training sets of the same size; then calculating the improved multi-scale coarse-grained time series and Fourier spectra corresponding to different scales of each sub-signal to capture multi-scale features, thereby realizing preliminary feature learning; finally, inputting the multi-scale features into the multi-scale cascaded deep belief network, directly learning high-level feature representations from the multi-scale features in a parallel manner, and automatically identifying fault modes through a cascade method and a softmax classifier; being able to capture the complex input-output projection relationship between the original data and the label, thereby providing an end-to-end diagnosis scheme.
[0009] Preferably, when collecting fault data for industrial equipment that needs to be monitored for faults, a data collection unit and monitoring software are used to collect fault data for the industrial equipment that needs to be monitored for faults, wherein the fault data is vibration data of the corresponding industrial equipment.
[0010] Preferably, when the collected fault data is input into a multi-scale cascaded deep belief network and trained using an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network, the multi-scale cascaded deep belief network includes a multi-scale coarse-grained layer, a multi-scale feature learning layer and a classification layer; the multi-scale coarse-grained layer is used to perform a multi-scale coarse-grained process on the input original vibration data to obtain coarse-grained time series at different scales; the multi-scale feature learning layer is used to automatically learn high-level and efficient feature representations from improved multiple coarse-grained time series in a parallel manner; the classification layer is used to output the probability distribution of each fault mode, solve multi-class problems, and automatically provide classification results.
[0011] Preferably, the classification layer uses a softmax classifier to output the probability distribution of each fault mode, solve multi-class problems, and automatically provide classification results.
[0012] Preferably, first, in the improved multi-scale coarse-grained layer, the original signal is subjected to an improved multi-scale coarse-grained process to obtain coarse-grained time series at different scales; secondly, in the multi-scale feature learning layer, high-level and efficient feature representations are automatically learned from the improved multiple coarse-grained time series in parallel; these features are learned from the multi-scales of the original signal, and therefore, the feature learning ability can be enhanced and the distinction between categories can be improved; finally, in the classification layer, a softmax classifier is used to output the probability distribution of each fault mode, solve multi-class problems, and automatically provide classification results; the improved multi-scale coarse-grained process is incorporated into the deep belief network in parallel to demonstrate multi-scale learning capabilities, thereby obtaining more complete and richer fault feature information, and achieving better diagnostic capabilities compared to the traditional deep belief network architecture.
[0013] Preferably, when the multi-scale coarse-grained layer performs a multi-scale coarse-grained process on the input original vibration data to obtain coarse-grained time series at different scales, the multi-scale coarse-grained layer includes the following steps:
[0014] S21: using a sliding window with data overlap to decompose the collected original fault vibration signal into a group of sub-signals of equal size;
[0015] S22: An improved multi-scale coarse-grained process is performed on each sub-signal to obtain coarse-grained time series at different scales, and Fourier spectra at different scales are calculated to capture multi-scale features.
[0016] Preferably, when the collected fault data is input into a multi-scale cascaded deep belief network, and an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network is used for training, the multi-scale cascaded deep belief network is constructed and trained, and the following steps are included:
[0017] S31: using a data acquisition unit and monitoring software to collect vibration data from a mechanical fault simulator to form training data samples and test data samples;
[0018] S32: Construct an improved multi-scale coarse-grained process to characterize the original signal at a range of scales;
[0019] S33: Integrate the constructed improved multi-scale coarse-grained process with the traditional deep belief network in parallel to obtain a multi-scale cascaded deep belief network, and use the established training data samples to train the multi-scale cascaded deep belief network;
[0020] S34: Input the test data into the trained multi-scale cascaded deep belief network architecture to automatically learn multi-scale features.
[0021] Preferably, when using the data acquisition unit and the monitoring software to collect vibration data from the mechanical fault simulator to form training data samples and test data samples, the following steps are included:
[0022] S41: Collect vibration data from multiple failure modes of a mechanical failure simulator using a data acquisition unit and monitoring software;
[0023] S42: using a sliding window with data overlap to divide the collected vibration data into a plurality of groups of sub-signals of equal size, and treating each sub-signal as a sample;
[0024] S43: Randomly distribute the data samples of each fault mode to the training data samples and the test data samples in a one-to-one ratio.
[0025] Preferably, when constructing an improved multi-scale coarse-grained process to characterize the original signal at a series of scales, the improved multi-scale coarse-grained process is essentially equivalent to a compact moving average process with overlapping adjacent data; in the improved multi-scale coarse-grained process, for a given time series {x(j):1≤j≤n}, the improved multi-scale coarse-grained time series with a scaling factor of κ can be obtained by the following equation:
[0026]
[0027] in, Improved multi-scale coarse-grained layer output; κ represents the scale factor; n is the length of the original time series; x(j) represents the jth sample.
[0028] Preferably, an improved multi-scale coarse-grained process is constructed to characterize the original signal at a series of scales, and the improved multi-scale coarse-grained process is essentially equivalent to a compact moving average process with adjacent data overlap, so that multi-scale features can be automatically learned from the original vibration signal. Compared with traditional shallow learning schemes, no manual feature extraction and experience are required, thereby obtaining better adaptability.
[0029] As a preferred method, when the constructed improved multi-scale coarse-grained process is integrated in parallel with the traditional deep belief network to obtain a multi-scale cascaded deep belief network, and the established training data samples are used to train the multi-scale cascaded deep belief network, the principle formula of the multi-scale cascaded deep belief network is:
[0030]
[0031] in, is the output of the multi-scale feature learning layer; q(α) i Classification layer output; s represents the sigmoid activation function; Hidden is the connection weight between units; is the hidden unit bias; α is the network parameter.
[0032] Preferably, when using an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network to identify fault results, the fault mode is automatically identified through a cascade method and a softmax classifier.
[0033] Beneficial effects of the present invention:
[0034] 1. Compared with the existing technology, when detecting faults in industrial equipment, a data-driven shallow learning fault diagnosis scheme is used for detection. The shallow learning scheme requires a lot of human intervention and relies on the experience of technicians. Feature extraction and classification are performed separately and cannot be optimized at the same time. At the same time, the shallow learning scheme is only applicable to specific diagnostic occasions and has poor versatility. This fault identification method incorporates the improved multi-scale coarse-grained process into the deep belief network to demonstrate multi-scale learning capabilities, thereby obtaining more complete and richer fault feature information. Compared with the traditional deep belief network architecture, it achieves better diagnostic capabilities;
[0035] 2. In the improved multi-scale coarse-grained layer, the original signal is subjected to an improved multi-scale coarse-grained process to obtain coarse-grained time series at different scales; secondly, in the multi-scale feature learning layer, high-level and efficient feature representations are automatically learned from the improved multiple coarse-grained time series in parallel; these features are learned from the multi-scales of the original signal, so that the feature learning ability can be enhanced and the distinction between categories can be improved; finally, in the classification layer, a softmax classifier is used to output the probability distribution of each fault mode, solve multi-class problems, and automatically provide classification results; the improved multi-scale coarse-grained process is incorporated into the deep belief network in parallel to demonstrate multi-scale learning capabilities, thereby obtaining more complete and richer fault feature information, and achieving better diagnostic capabilities compared to the traditional deep belief network architecture;
[0036] 3. By constructing an improved multi-scale coarse-grained process to characterize the original signal at a series of scales, and the improved multi-scale coarse-grained process is essentially equivalent to a compact moving average process with overlapping adjacent data, it is possible to automatically learn multi-scale features from the original vibration signal. Compared with the traditional shallow learning scheme, no manual feature extraction and experience are required, thus achieving better adaptability;
[0037] 4. The scheme is mainly divided into three steps: fault data collection, end-to-end fault diagnosis architecture training and recognition result output based on multi-scale cascaded deep belief network; decomposing the collected original fault data into a series of sub-training sets of the same size; then calculating the improved multi-scale coarse-grained time series and Fourier spectra corresponding to different scales of each sub-signal to capture multi-scale features, thereby realizing preliminary feature learning; finally, inputting the multi-scale features into the multi-scale cascaded deep belief network, directly learning high-level feature representations from multi-scale features in a parallel manner, and automatically identifying fault modes through a cascade method and a softmax classifier; it is able to capture the complex input-output projection relationship between the original data and the label, thereby providing an end-to-end diagnosis scheme. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 What is shown is a flow chart of a multi-condition fault identification method of a multi-scale cascaded deep belief network of the present invention;
[0039] Figure 2 What is shown is a specific flow chart of the multi-condition fault identification method of the multi-scale cascaded deep belief network of the present invention;
[0040] Figure 3 What is shown is a structural schematic diagram of a multi-scale cascaded deep belief network in the multi-condition fault identification method of the multi-scale cascaded deep belief network of the present invention. DETAILED DESCRIPTION
[0041] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0042] See also Figure 1-2 The present invention provides an embodiment: a multi-condition fault identification method of a multi-scale cascaded deep belief network, comprising the following steps:
[0043] S11: Collect fault data of industrial equipment that needs to be monitored for faults;
[0044] S12: inputting the collected fault data into a multi-scale cascaded deep belief network, and training the end-to-end fault diagnosis architecture based on the multi-scale cascaded deep belief network;
[0045] S13: At the same time, an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network is used to identify fault results.
[0046] Preferably, the scheme is mainly divided into three steps: fault data collection, end-to-end fault diagnosis architecture training and recognition result output based on multi-scale cascaded deep belief network; decomposing the collected original fault data into a series of sub-training sets of the same size; then calculating the improved multi-scale coarse-grained time series and Fourier spectra corresponding to different scales of each sub-signal to capture multi-scale features, thereby realizing preliminary feature learning; finally, inputting the multi-scale features into the multi-scale cascaded deep belief network, directly learning high-level feature representations from the multi-scale features in a parallel manner, and automatically identifying fault modes through a cascade method and a softmax classifier; being able to capture the complex input-output projection relationship between the original data and the label, thereby providing an end-to-end diagnosis scheme.
[0047] Preferably, when collecting fault data for industrial equipment that needs to be monitored for faults, a data collection unit and monitoring software are used to collect fault data for the industrial equipment that needs to be monitored for faults, wherein the fault data is vibration data of the corresponding industrial equipment.
[0048] See also Figure 3 In this embodiment, when the collected fault data is input into a multi-scale cascaded deep belief network and trained using an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network, the multi-scale cascaded deep belief network includes a multi-scale coarse-grained layer, a multi-scale feature learning layer and a classification layer; the multi-scale coarse-grained layer is used to perform a multi-scale coarse-grained process on the input original vibration data to obtain coarse-grained time series at different scales; the multi-scale feature learning layer is used to automatically learn high-level and efficient feature representations from improved multiple coarse-grained time series in a parallel manner; the classification layer is used to output the probability distribution of each fault mode, solve multi-class problems, and automatically provide classification results.
[0049] Preferably, the classification layer uses a softmax classifier to output the probability distribution of each fault mode, solve multi-class problems, and automatically provide classification results.
[0050] Preferably, first, in the improved multi-scale coarse-grained layer, the original signal is subjected to an improved multi-scale coarse-grained process to obtain coarse-grained time series at different scales; secondly, in the multi-scale feature learning layer, high-level and efficient feature representations are automatically learned from the improved multiple coarse-grained time series in parallel; these features are learned from the multi-scales of the original signal, and therefore, the feature learning ability can be enhanced and the distinction between categories can be improved; finally, in the classification layer, a softmax classifier is used to output the probability distribution of each fault mode, solve multi-class problems, and automatically provide classification results; the improved multi-scale coarse-grained process is incorporated into the deep belief network in parallel to demonstrate multi-scale learning capabilities, thereby obtaining more complete and richer fault feature information, and achieving better diagnostic capabilities compared to the traditional deep belief network architecture.
[0051] Preferably, when the multi-scale coarse-grained layer performs a multi-scale coarse-grained process on the input original vibration data to obtain coarse-grained time series at different scales, the multi-scale coarse-grained layer includes the following steps:
[0052] S21: using a sliding window with data overlap to decompose the collected original fault vibration signal into a group of sub-signals of equal size;
[0053] S22: An improved multi-scale coarse-grained process is performed on each sub-signal to obtain coarse-grained time series at different scales, and Fourier spectra at different scales are calculated to capture multi-scale features.
[0054] Preferably, when the collected fault data is input into a multi-scale cascaded deep belief network, and an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network is used for training, the multi-scale cascaded deep belief network is constructed and trained, and the following steps are included:
[0055] S31: using a data acquisition unit and monitoring software to collect vibration data from a mechanical fault simulator to form training data samples and test data samples;
[0056] S32: Construct an improved multi-scale coarse-grained process to characterize the original signal at a range of scales;
[0057] S33: Integrate the constructed improved multi-scale coarse-grained process with the traditional deep belief network in parallel to obtain a multi-scale cascaded deep belief network, and use the established training data samples to train the multi-scale cascaded deep belief network;
[0058] S34: Input the test data into the trained multi-scale cascaded deep belief network architecture to automatically learn multi-scale features.
[0059] Preferably, when using the data acquisition unit and the monitoring software to collect vibration data from the mechanical fault simulator to form training data samples and test data samples, the following steps are included:
[0060] S41: Collect vibration data from multiple failure modes of a mechanical failure simulator using a data acquisition unit and monitoring software;
[0061] S42: using a sliding window with data overlap to divide the collected vibration data into a plurality of groups of sub-signals of equal size, and treating each sub-signal as a sample;
[0062] S43: Randomly distribute the data samples of each fault mode to the training data samples and the test data samples in a one-to-one ratio.
[0063] Preferably, when constructing an improved multi-scale coarse-grained process to characterize the original signal at a series of scales, the improved multi-scale coarse-grained process is essentially equivalent to a compact moving average process with overlapping adjacent data; in the improved multi-scale coarse-grained process, for a given time series {x(j):1≤j≤n}, the improved multi-scale coarse-grained time series with a scaling factor of κ can be obtained by the following equation:
[0064]
[0065] in, Improved multi-scale coarse-grained layer output; κ represents the scale factor; n is the length of the original time series; x(j) represents the jth sample.
[0066] Preferably, an improved multi-scale coarse-grained process is constructed to characterize the original signal at a series of scales, and the improved multi-scale coarse-grained process is essentially equivalent to a compact moving average process with adjacent data overlap, so that multi-scale features can be automatically learned from the original vibration signal. Compared with traditional shallow learning schemes, no manual feature extraction and experience are required, thereby obtaining better adaptability.
[0067] As a preferred method, when the constructed improved multi-scale coarse-grained process is integrated in parallel with the traditional deep belief network to obtain a multi-scale cascaded deep belief network, and the established training data samples are used to train the multi-scale cascaded deep belief network, the principle formula of the multi-scale cascaded deep belief network is:
[0068]
[0069]
[0070] in, is the output of the multi-scale feature learning layer; q(α) iClassification layer output; s represents the sigmoid activation function; Hidden is the connection weight between units; is the hidden unit bias; α is the network parameter.
[0071] Preferably, when using an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network to identify fault results, the fault mode is automatically identified through a cascade method and a softmax classifier.
[0072] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of those skilled in the art without departing from the spirit of the present invention.
Claims
1. A multi-condition fault identification method based on a multi-scale cascaded deep belief network; characterized in that: The following steps are included: S11: Collect fault data of industrial equipment that needs to be monitored for faults; S12: inputting the collected fault data into a multi-scale cascaded deep belief network, and training the end-to-end fault diagnosis architecture based on the multi-scale cascaded deep belief network; S13: At the same time, an end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network is used to identify fault results.
2. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 1 is characterized in that: When collecting fault data for industrial equipment that needs to be monitored for faults, a data collection unit and monitoring software are used to collect fault data for the industrial equipment that needs to be monitored for faults, wherein the fault data is vibration data of the corresponding industrial equipment.
3. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 2 is characterized in that: When the collected fault data is input into the multi-scale cascaded deep belief network and the end-to-end fault diagnosis architecture based on the multi-scale cascaded deep belief network is trained, the multi-scale cascaded deep belief network includes a multi-scale coarse-grained layer, a multi-scale feature learning layer and a classification layer; the multi-scale coarse-grained layer is used to perform a multi-scale coarse-grained process on the input raw vibration data to obtain coarse-grained time series at different scales; The multi-scale feature learning layer is used to automatically learn high-level and efficient feature representations from multiple improved coarse-grained time series in a parallel manner; The classification layer is used to output the probability distribution of each failure mode, solve multi-class problems, and automatically provide classification results.
4. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 3 is characterized in that: The classification layer uses a softmax classifier to output the probability distribution of each fault mode, solve multi-class problems, and automatically provide classification results.
5. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 4 is characterized in that: The multi-scale coarse-grained layer performs a multi-scale coarse-grained process on the input raw vibration data to obtain coarse-grained time series at different scales, including the following steps: S21: using a sliding window with data overlap to decompose the collected original fault vibration signal into a group of sub-signals of equal size; S22: An improved multi-scale coarse-grained process is performed on each sub-signal to obtain coarse-grained time series at different scales, and Fourier spectra at different scales are calculated to capture multi-scale features.
6. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 5 is characterized in that: When the collected fault data is input into the multi-scale cascaded deep belief network, and the end-to-end fault diagnosis architecture based on the multi-scale cascaded deep belief network is used for training, the multi-scale cascaded deep belief network is constructed and trained, including the following steps: S31: using a data acquisition unit and monitoring software to collect vibration data from a mechanical fault simulator to form training data samples and test data samples; S32: Construct an improved multi-scale coarse-grained process to characterize the original signal at a range of scales; S33: Integrate the constructed improved multi-scale coarse-grained process with the traditional deep belief network in parallel to obtain a multi-scale cascaded deep belief network, and use the established training data samples to train the multi-scale cascaded deep belief network; S34: Input the test data into the trained multi-scale cascaded deep belief network architecture to automatically learn multi-scale features.
7. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 6 is characterized in that: When using a data acquisition unit and monitoring software to collect vibration data from a mechanical fault simulator to form training data samples and test data samples, the following steps are included: S41: Collect vibration data from multiple failure modes of a mechanical failure simulator using a data acquisition unit and monitoring software; S42: using a sliding window with data overlap to divide the collected vibration data into a plurality of groups of sub-signals of equal size, and treating each sub-signal as a sample; S43: Randomly distribute the data samples of each fault mode to the training data samples and the test data samples in a one-to-one ratio.
8. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 7 is characterized in that: When constructing an improved multi-scale coarse-grained process to characterize the original signal at a series of scales, the improved multi-scale coarse-grained process is essentially equivalent to a compact moving average process with adjacent data overlap; In the improved multi-scale coarse-grained process, for a given time series {x(j):1≤j≤n}, the improved multi-scale coarse-grained time series with a scaling factor of κ can be obtained by the following equation: in, Improved multi-scale coarse-grained layer output; κ represents the scale factor; n is the length of the original time series; x(j) represents the jth sample.
9. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 8, characterized in that: When the constructed improved multi-scale coarse-grained process is integrated in parallel with the traditional deep belief network to obtain a multi-scale cascaded deep belief network, and the established training data samples are used to train the multi-scale cascaded deep belief network, the principle formula of the multi-scale cascaded deep belief network is: in, is the output of the multi-scale feature learning layer; q(α) i Classification layer output; s represents the sigmoid activation function; Hidden is the connection weight between units; is the hidden unit bias; α is the network parameter.
10. The multi-condition fault identification method of a multi-scale cascaded deep belief network according to claim 9, characterized in that: When using the end-to-end fault diagnosis architecture based on a multi-scale cascaded deep belief network to identify fault results, the fault mode is automatically identified through the cascade method and softmax classifier.
Citation Information
Patent Citations
SAR image classification method based on SAR-SIFT and DBN
CN105894035A
Identification method of rolling bearing state under variable load of EEMD-Hilbert envelope spectrum in combination with DBN
CN106886660A
Diesel engine fault diagnosis method
CN110926818A
Rolling bearing fault diagnosis method in noisy environment based on G-MSCNN
CN114169377A
Bearing fault detection method based on deep learning
CN118312874A
Cited By
Gear fault diagnosis method based on multi-scale coarse-grained convolutional neural network
CN119961754A
A Gear Fault Diagnosis Method Based on Multi-Scale Coarse-Grained Convolutional Neural Network
CN119961754B
Method and device for training network diagnosis model, equipment and storage medium
CN120185999A
