Bearing fault detection method and system based on determination learning and convolutional neural network

Abstract

This invention belongs to the field of bearing fault diagnosis and provides a bearing fault detection method and system based on deterministic learning and convolutional neural networks. The method includes acquiring a one-dimensional bearing vibration signal, expanding it into a two-dimensional signal using a high-gain observer, standardizing the two-dimensional signal to obtain a normalized signal, modeling the internal dynamics of the bearing vibration signal based on a deterministic learning mechanism, generating a dynamic information graph of the bearing vibration signal, and performing fault detection using a pre-trained convolutional neural network model based on the dynamic information graph to obtain the bearing fault detection result. This invention, for the first time from a dynamic system perspective, visualizes the dynamics of bearing vibration signals as a dynamic information graph and detects fault occurrence through changes in the dynamics.

Description

Technical Field

[0001] This invention belongs to the field of bearing fault diagnosis technology, specifically relating to a bearing fault detection method and system based on deterministic learning and convolutional neural networks. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Fault diagnosis, by monitoring the operating status of mechanical equipment and identifying the types and timing of faults, is crucial for improving the reliability and safety of mechanical systems. Bearings are critical components of rotating equipment and have a high failure rate (e.g., bearing-related failures account for approximately 41% of all motor failures). Detecting bearing failures is essential, as such failures not only compromise equipment reliability and safety but can also lead to severe economic losses and personal injury.

[0004] Feature extraction is a crucial step in fault diagnosis methods. Commonly used manual feature extraction methods include statistical analysis, short-time Fourier transform, wavelet transform, and empirical mode decomposition. However, the effectiveness of manual feature extraction is highly dependent on prior knowledge. In recent years, deep learning-based intelligent fault diagnosis has enabled automatic feature extraction for end-to-end fault detection, with convolutional neural networks (CNNs) being particularly suitable for deep feature extraction. Currently, some methods convert vibration time-series signals into vibration images, enabling fault diagnosis through CNN feature learning on these images. Existing research indicates that system failures often result from a change in dynamics from healthy to faulty. However, there is currently no CNN-based fault diagnosis method that converts dynamic information into images from the perspective of the dynamic system. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a bearing fault detection method and system based on deterministic learning and convolutional neural networks. This invention, for the first time from a dynamic system perspective, visualizes the dynamic information of bearing vibration signals as a dynamic information graph, and detects the occurrence of faults by changing the dynamic information.

[0006] According to some embodiments, the first aspect of the present invention provides a bearing fault detection method based on deterministic learning and convolutional neural networks, employing the following technical solution:

[0007] Bearing fault detection methods based on deterministic learning and convolutional neural networks include:

[0008] A one-dimensional bearing vibration signal of bearing failure is obtained, and a high-gain observer is used to extend the one-dimensional bearing vibration signal into a two-dimensional bearing vibration signal.

[0009] The two-dimensional bearing vibration signal is normalized by standardizing the maximum value to obtain the normalized bearing vibration signal.

[0010] The internal dynamic information of bearing vibration signals is modeled based on a deterministic learning mechanism, and a dynamic information graph of bearing vibration signals is generated based on the internal dynamic information of bearing vibration signals.

[0011] Based on the dynamic information graph of the bearing vibration signal, fault detection is performed using a pre-trained convolutional neural network model to obtain bearing fault detection results.

[0012] Furthermore, the step of standardizing the two-dimensional bearing vibration signal to obtain a normalized bearing vibration signal involves the following steps:

[0013] The maximum value is normalized for each dimension of the two-dimensional bearing vibration signal;

[0014] The normalization function is:

[0015]

[0016] Where, x i This represents a one-dimensional time series of the i-th sample. This indicates that the vibration data is concentrated at the maximum value of all samples in dimension j. This represents the normalized sample.

[0017] Furthermore, the step of modeling the internal dynamics information of the bearing vibration signal based on the deterministic learning mechanism, and generating a dynamics information graph of the bearing vibration signal based on the internal dynamics information of the bearing vibration signal, specifically involves:

[0018] A deterministic learning mechanism is used to model the bearing vibration signal, resulting in a dynamic information model of the bearing vibration signal.

[0019] The trajectory region of the normalized bearing vibration signal is gridded to obtain gridded dynamic information input;

[0020] Based on the gridded dynamic information input, the dynamic information map of the bearing vibration signal is obtained by using the dynamic information model of the bearing signal.

[0021] Furthermore, the dynamic information model of the bearing vibration signal is specifically as follows:

[0022]

[0023] in, Represents the state trajectory The neighborhood, This indicates that the weights of the RBF neural network converge at Tk. a To Tkb The average weight over the time interval, i = 1, 2, where T is the sampling time. Let S(z) represent the weights of the neural network to be estimated, and let S(z) represent the regression vector composed of radial basis functions.

[0024] Furthermore, the dynamic information diagram of the bearing vibration signal is an m×m matrix, specifically as follows:

[0025]

[0026] Among them, z i (1),…,z i (m), i = 1, 2 means that the interval [-1, 1] is evenly divided into m-1 corresponding data points;

[0027] Converting the above matrix into a grayscale image creates a dynamic information map for each bearing vibration signal.

[0028] Furthermore, the training process of the convolutional neural network model is as follows:

[0029] For the generated vibration signal dynamic information graph, the training set and the test set are divided in a 7:3 ratio;

[0030] The dynamic information graph of bearing vibration signals from the training set is input into a convolutional neural network for bearing fault detection.

[0031] The loss function is cross-entropy loss, and the gradient descent algorithm is selected to optimize and update the network parameters to obtain a trained convolutional neural network model.

[0032] The dynamic information graph of bearing vibration signal from the test set is input into the trained convolutional neural network model to obtain the network's fault detection results.

[0033] Furthermore, the structure of the convolutional neural network model includes a convolutional layer, a pooling layer, a convolutional layer, a pooling layer, a convolutional layer, a pooling layer, a convolutional layer, a pooling layer, a flattening layer, a fully connected layer, and a fully connected layer connected in sequence.

[0034] The ELU activation function and batch normalization are used between the convolutional layer and the pooling layer. The extracted feature map is flattened and input into the fully connected layer. The output of the fully connected layer is used as the output of the entire convolutional neural network.

[0035] According to some embodiments, a second aspect of the present invention provides a bearing fault detection system based on deterministic learning and convolutional neural networks, employing the following technical solution:

[0036] A bearing fault detection system based on deterministic learning and convolutional neural networks includes:

[0037] The data acquisition module is configured to acquire one-dimensional bearing vibration signals of bearing faults and expand the one-dimensional bearing vibration signals into two-dimensional bearing vibration signals using a high-gain observer.

[0038] The standardization processing module is configured to perform maximum value standardization processing on the two-dimensional bearing vibration signal to obtain a normalized bearing vibration signal.

[0039] The dynamic information acquisition module is configured to model the internal dynamic information of the bearing vibration signal based on a deterministic learning mechanism, and generate a dynamic information diagram of the bearing vibration signal based on the internal dynamic information of the bearing vibration signal.

[0040] The bearing fault detection module is configured to perform fault detection based on the dynamic information graph of the bearing vibration signal and use a pre-trained convolutional neural network model to obtain the bearing fault detection result.

[0041] According to some embodiments, a third aspect of the present invention provides a computer-readable storage medium.

[0042] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the bearing fault detection method based on deterministic learning and convolutional neural networks as described in the first aspect above.

[0043] According to some embodiments, a fourth aspect of the present invention provides a computer device.

[0044] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the bearing fault detection method based on deterministic learning and convolutional neural networks as described in the first aspect above.

[0045] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0046] This invention proposes a deterministic learning convolutional neural network framework for bearing fault detection by combining the accurate modeling of the intrinsic dynamics of data through deterministic learning with the automatic feature extraction capabilities of convolutional neural networks. Specifically, firstly, a deterministic learning mechanism is used to accurately model the intrinsic dynamic information of the original vibration time series. Secondly, the dynamic information learned by the deterministic learning mechanism is visualized as a dynamic information graph. Finally, a convolutional neural network is used to extract features from the dynamic information graph, and based on the extracted discriminative features, different bearing fault types are effectively classified and detected. This invention provides an efficient and accurate method for bearing fault detection, which can effectively improve the reliability and safety of rotating machinery systems. Attached Figure Description

[0047] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0048] Figure 1 This is a flowchart of the bearing fault detection method based on deterministic learning and convolutional neural networks in an embodiment of the present invention;

[0049] Figure 2 This is a flowchart of the training process of the dynamic radial basis function neural network model in an embodiment of the present invention;

[0050] Figure 3 This is a diagram of the convolutional neural network structure in an embodiment of the present invention. Detailed Implementation

[0051] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0052] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0053] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0054] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0055] Example 1

[0056] like Figure 1As shown, this embodiment provides a bearing fault detection method based on deterministic learning and convolutional neural networks. This embodiment uses the application of this method to a server as an example for illustration. It is understood that this method can also be applied to terminals, and can also be applied to systems including terminals, servers, and other components, and can be implemented through interaction between the terminal and the server. The server can be an independent physical server, a server cluster composed of multiple physical servers, or a distributed system. It can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communication, middleware services, domain name services, CDN security services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. In this embodiment, the method includes the following steps:

[0057] A one-dimensional bearing vibration signal of bearing failure is obtained, and a high-gain observer is used to extend the one-dimensional bearing vibration signal into a two-dimensional bearing vibration signal.

[0058] The two-dimensional bearing vibration signal is normalized by standardizing the maximum value to obtain the normalized bearing vibration signal.

[0059] The internal dynamic information of bearing vibration signals is modeled based on a deterministic learning mechanism, and a dynamic information graph of bearing vibration signals is generated based on the internal dynamic information of bearing vibration signals.

[0060] Based on the dynamic information graph of the bearing vibration signal, fault detection is performed using a pre-trained convolutional neural network model to obtain bearing fault detection results.

[0061] This embodiment discloses a bearing fault detection method based on a combination of deterministic learning and convolutional neural networks. Combining the precise modeling capability of deterministic learning mechanisms for the intrinsic dynamics of data with the automatic feature extraction capability of convolutional neural networks, a deterministic learning convolutional neural network framework for bearing fault detection is proposed, and bearing fault detection is performed based on this framework. Specifically, firstly, the intrinsic dynamics of the original vibration time series are learned, i.e., modeled, through a deterministic learning mechanism. The dynamic information learned by the deterministic learning mechanism is visualized as a dynamic information graph, and then input into the convolutional neural network designed in this invention for automatic feature extraction. The extracted discriminative features are used to effectively classify and detect different bearing fault types. This invention provides an efficient and accurate method for bearing fault detection, which can effectively improve the reliability and safety of rotating machinery systems.

[0062] The dynamic information extraction method used in this embodiment comes from a deterministic learning mechanism, and the further fault diagnosis method uses a convolutional neural network.

[0063] Deterministic learning theory offers an effective approach to addressing the challenges of knowledge acquisition, representation, and utilization in dynamic systems. Under the condition of continuous excitation (PE), deterministic learning mechanisms provide a solution for the accurate local identification or modeling of nonlinear dynamic systems. The modeled dynamics can be represented and stored using constant RBF neural networks. Furthermore, this stored knowledge can be effectively used for tasks in dynamic environments, such as dynamic pattern recognition and intelligent control.

[0064] (1) A radial basis function (RBF) neural network is used;

[0065] (2) The periodic (or regression) trajectory satisfies some of the continuous incentive conditions;

[0066] (3) Achieve accurate local modeling of the dynamics of nonlinear systems along the trajectory;

[0067] (4) Stored in a constant neural network manner.

[0068] As a type of deep learning neural network, Convolutional Neural Networks (CNNs) are specifically designed to process structured array data like those shown in the image. Their superior feature extraction capabilities make them a primary choice for many vision applications. A CNN consists of convolutional layers, pooling layers, and fully connected layers.

[0069] Convolutional layers are two-dimensional convolutional neural networks. They use multiple convolutional kernels to repeatedly act on the entire receptive field, traversing all input data to extract features. Pooling layers, also known as sampling layers, are typically used to reduce the spatial resolution of feature maps while retaining the most important feature information to prevent overfitting during training. This invention employs the max pooling method.

[0070] A fully connected layer is the output layer of a CNN. Each node is connected to all nodes in the previous layer, responsible for connecting all neurons from the feature map output of the previous layer to each node, ultimately resulting in the output of an interconnected neuron. Fully connected layers are used as classifiers.

[0071] A bearing rolling system is a stable, dynamic system whose vibration signals can be collected by sensors. When damage occurs to the rolling elements, inner ring, or outer ring of the bearing due to overload, friction, corrosion, or other reasons, the vibration signal of this stable system will change. Both the vibration signals generated by a normal bearing rolling system and those generated after a fault can be called time-series signals. The essence of the fault detection method in this invention is how to effectively preserve and differentiate the dynamic information of the bearing rolling system, and how to quickly identify and classify the type of fault to be tested, ultimately obtaining a fault diagnosis result. The flowchart of this embodiment is as follows: Figure 1 As shown.

[0072] The main technical problem to be solved in this embodiment is: how to extract hidden features from rolling bearing signals from the perspective of dynamic feature extraction, and how to use this dynamic feature information in conjunction with a convolutional neural network to complete the fault detection task. Specifically, the method of this embodiment first converts the one-dimensional bearing vibration time series signal into a two-dimensional vibration time series signal through a high-gain observer. Then, it uses a deterministic learning mechanism to model the inherent dynamic information of the time series signal, visualizes the obtained dynamic information as a dynamic information graph, and finally uses a convolutional neural network to perform deep feature extraction. The extracted features are then used to determine whether the rolling bearing is in a normal state or a certain fault state.

[0073] The specific technical solution of this embodiment is achieved through the following steps:

[0074] (1) Collect fault vibration signals: The rolling bearing system’s one-dimensional vibration signal in the time domain is collected by an accelerated vibration sensor. The frequency of the sensor’s collected signal is determined, and the collected vibration signal is cut into equal lengths to construct a fault dataset.

[0075] Vibration signals of rolling bearings with different fault types are collected by sensors, and bearing fault labels are recorded. The collected vibration signals are cut into equal lengths, and each piece is a sample to construct a dataset of rolling bearing vibration signals.

[0076] The vibration signal collected is at the drive end of the motor, with a sampling frequency of 12kHz. Other sampling frequencies can also be used. The shorter the sampling time, the higher the modeling accuracy of the learning mechanism.

[0077] (2) Use a high-gain observer to expand one-dimensional samples into two-dimensional data samples.

[0078] Two-dimensional data are obtained by observing the underlying internal state of the vibration system from discrete one-dimensional vibration data acquired by digital sensors using a high-gain observer (HGO).

[0079] The purpose of using a high-gain observer is to observe the hidden internal dynamics from the acquired one-dimensional vibration signal and take this observed signal as the second state of the bearing vibration system. Different high-gain parameters will affect the observed results, thereby further affecting the accuracy of the modeled dynamic information. Therefore, it is necessary to select appropriate parameters.

[0080] (3) Data standardization: Standardize the data by standardizing the maximum value of each dimension of the sample.

[0081] Perform maximum value standardization on each dimension of the data in step (2). The normalization function is:

[0082]

[0083] Where, x i Let x represent a one-dimensional time series of the i-th sample. max This indicates that the vibration data is concentrated at the maximum value of all samples in dimension j. This represents the normalized sample.

[0084] (4) Establishing a dynamic RBF neural network model for rolling bearing vibration signals: A deterministic learning mechanism is used to model the intrinsic dynamics of rolling bearing vibration signals. The deterministic learning mechanism can extract the dynamic characteristics of the vibration signals as they change over time by modeling the bearing vibration time series signals.

[0085] A simplified flowchart of the RBF neural network training process is shown in Figure 2. Here, x1 and x2 represent two states of the bearing vibration signal, respectively. This represents the estimated state of the RBF neural network, i.e., the output of the neural network.

[0086] A radial basis function neural network (RBF NN) was used to train vibration signals under normal conditions and different bearing faults, with Gaussian functions selected as the radial basis functions. The training employed a learning method based on a deterministic learning mechanism to achieve convergence of the weights of the dynamic RBF NN to their optimal values. There were two convergence scenarios: one where neurons along the expanded vibration signal trajectory met the continuous excitation condition, and their weights converged to the optimal value; and another where neurons far from the trajectory were not excited and their parameters were not updated, remaining at their initial weights, which were almost zero.

[0087] An RBF neural network approximates the internal dynamics of each bearing vibration time series signal sample. The approximation is performed on the intrinsic dynamics near the vibration signal trajectory; dynamics far from the trajectory are not approximated. The mean of the weights of the dynamic RBF neural network over a period of time after convergence is saved to generate a constant RBF neural network, which provides the necessary knowledge / model for generating the dynamic information graph.

[0088] The bearing vibration signal generated by the rolling system can be represented as a nonlinear dynamic system described by ordinary differential equations;

[0089]

[0090] Where x = [x1, x2] T ∈R 2 F(x;p) = [f1(x;p), f2(x;p)] represents the state of the nonlinear dynamic system, p represents the parameter vector of the nonlinear dynamic system, and F(x;p) = [f1(x;p), f2(x;p)] represents the unknown dynamics of the nonlinear dynamic system.

[0091] φ(X)=[x(t0),x(t0+T),…,x(t0+(N-1)T)] is the sampling time sequence obtained from the nonlinear dynamic system (2) over a sampling period T, which is the real vibration signal obtained by the sensor. By selecting an appropriate sampling time, the sampling data can be represented by the following Euler sampling model:

[0092] x i (k+1)=x i (k)+Tf i (x(k);p) (3)

[0093] Where x(k)=[x1(k),x2(k)]∈R 2 It is the system state of the Euler model, f i (x(k); p) represents the internal dynamics of the sampled data, modeled by an RBF neural network with precise and convergent network parameters, and T is the sampling time.

[0094] The RBF neural network is constructed as follows:

[0095]

[0096] Where W is the weight vector of the network, Z is the input vector of the network, i.e., the state of the sampling model (3), and M is the number of neurons. S(Z)=[s1(||Z-ξ1||),…,s M (||Z-ξ M ||)]w is the regression vector of RBF. It is a Gaussian radial basis function, η is the width of the receptive field, and ξ is the value of the function. i =1,…,M are the neuron centers of the RBF neural network.

[0097] The dynamic neural network identifier is constructed as follows:

[0098]

[0099]

[0100] in, These are the two state vectors of the identifier, corresponding to the two states of the vibration signal. 0 < α1, α2 < 1 represents the identifier gain. This represents the weight vector to be estimated in the 1RBF neural network representing the state of the vibration signal. The value represents the weight vector to be estimated by the 2RBF neural network for the vibration signal state, and T is the sampling time.

[0101] The weight update strategy of the RBF neural network updates the weights based on a weight update rate designed according to Lyapunov stability, as described below:

[0102]

[0103]

[0104] Where γ is the learning gain parameter, and e1 and e2 represent the tracking errors of the two states, respectively.

[0105] According to the deterministic learning mechanism, the regression vector S(x(k)) with the regression trajectory input satisfies a partial continuous excitation condition, and the deterministic learning mechanism can locally and accurately approximate / model the unknown dynamic function f. i (x(k); p), and stored in a constant RBF neural network.

[0106]

[0107]

[0108] in, i = 1, 2 indicates that the RBF neural network weights converge at Tk. a To Tk b The average weight over the time interval, ∈ i This indicates the approximation error.

[0109] The established constant RBF neural network model represents the dynamic information of the vibration signal. That is, in the formula above... The dynamic information diagram described in this embodiment is essentially a graphical representation of the modeled dynamics, representing the dynamic information of the vibration signal.

[0110] The learning and training process described in step (4) is a knowledge acquisition process. The learned knowledge is stored as constant RBF neural network weight vectors. The weight vectors of each neural network represent the intrinsic dynamics of a sample, serving as an expression for generating the dynamic information graph corresponding to that vibration signal sample.

[0111] (5) Generating a dynamic information graph: After maximum value normalization and deterministic learning to learn the intrinsic dynamics of the vibration data, a constant RBF network that captures the inherent dynamics along the state trajectory is obtained, namely: Represents the state trajectory The neighborhood of . Due to the maximum value normalization, we have That is, the neighborhood of the state trajectory is restricted to a rectangular area between [-1, 1] × [-1, 1]. Therefore, even the trajectory region of each sample... It's different, we can be in The dynamic information modeled for each sample is compared above, i.e. Trajectory region of each sample The difference lies primarily in the varying amplitude ranges of vibration signals collected by sensors for different fault types.

[0112] At a given resolution, this dynamic information can form an m×m matrix.

[0113]

[0114] Among them, z i (1),…,z i (m), i = 1, 2 means uniformly dividing the interval [-1, 1] into m-1 corresponding data points. Further converting the above matrix into a grayscale image creates a dynamic information map for each sample. The higher the required resolution, the larger the value of m can be.

[0115] The process of generating the dynamic information map in step (5) is a process of reusing the knowledge learned by the deterministic learning mechanism. The generated dynamic information map is a numerical matrix of size m×m. The value of m cannot be too small. If the value of m is too small, the number of data points in the generated dynamic information map will be small, and the effect of reproducing dynamic information will not be achieved. Furthermore, the value of m is not better the larger it is. If the value of m is too large, the resolution of the generated dynamic information image will be very high, which will increase the demand on computing power. Since the constant RBF neural network for generating dynamic information is learned from discrete vibration signal samples through deterministic learning mechanism, it is recommended that the value of m should not be greater than the data length of a single sample. If the value of m is greater than the data length of the sample, it will not be of much significance to improve the accuracy of fault detection.

[0116] The specific process for generating the dynamic diagram is as follows:

[0117] Deterministic learning for modeling unknown dynamics: After using deterministic learning to model the intrinsic dynamic information of vibration data, a constant RBF neural network along the state trajectory is obtained.

[0118] Right now Represents the state trajectory The neighborhood of.

[0119] Gridded input: Due to maximum value normalization, we have That is, the neighborhood of the state trajectory is restricted to a rectangular area between [-1, 1] × [-1, 1]. Therefore, even the trajectory region of each sample... It's different, it can be The dynamic information modeled for each sample is compared above, i.e. Divide the two-dimensional interval [-1,1]×[-1,1] evenly into (m-1)×(m-1) parts, which corresponds to m×m data points. These data points will be used as input data for generating the dynamic information graph.

[0120] Gridded output: Input the gridded m×m data points into the established model. The output matrix is ​​the dynamic information graph.

[0121] (6) Convolutional Neural Network Construction: In the convolutional neural network model proposed in this invention, the convolutional layer and the pooling layer constitute a sub-module. The ELU activation function and the batch normalization layer are used between the convolutional layer and the pooling layer to improve training efficiency and enhance the generalization performance of the network. This sub-module is repeated four times, followed by a flattening layer, and then two fully connected layers are used to complete the classification task.

[0122] A convolutional neural network (CNN) is built using the PyTorch framework. The network structure is as follows: convolutional layer -> pooling layer -> convolutional layer -> pooling layer -> convolutional layer -> pooling layer -> convolutional layer -> pooling layer -> flattening layer -> fully connected layer -> fully connected layer. The ELU activation function and batch normalization are used between the convolutional and pooling layers. The flattened feature map is then input to the fully connected layer, and the output of the fully connected layer becomes the output of the entire CNN. The network structure diagram is shown below. Figure 3 As shown.

[0123] The kernel sizes in each convolutional layer are 9×9, 5×5, 3×3, and 3×3, respectively, and the kernel size in the pooling layer is 2×2.

[0124] To achieve a feature map output of a specified size, a padding operation is required, i.e., padding the edges of the feature map with zeros. Pooling layers maintain feature invariance, downsample the feature map, expand the receptive field, and prevent overfitting to some extent. The ELU activation function is chosen to introduce non-linearity into the network.

[0125] (7) Training the convolutional neural network:

[0126] Dataset partitioning: For the generated vibration signal dynamic information graph, the training set and the test set are partitioned in a 7:3 ratio.

[0127] First, a random seed is set for the training program to ensure that the results of each training iteration are reproducible. An appropriate batch size is selected, and the dynamic information graph of the bearing vibration signals in the training set is input into the convolutional neural network. Cross-entropy loss is used as the loss function, and gradient descent is selected to optimize and update the network parameters. The dynamic learning rate is initially set to 0.001, and the learning rate scheduler ReduceLROnPlateau is used. The learning rate is gradually reduced as training progresses for better convergence and a higher-performing model, until the loss function converges. The network parameters with the minimum cross-entropy loss during training are selected and saved, resulting in the trained convolutional neural network.

[0128] The learning rate is one of the most important hyperparameters affecting model performance. The role of the learning scheduler is to conditionally reduce the learning rate during training to improve model performance. After each batch of training is completed, the learning scheduler checks whether the model performance has improved. If the model performance does not improve when the set batch is reached, the learning rate will be reduced. After a cooling-off period, this process will continue until the last batch of training is completed.

[0129] (8) Classification of faulty bearing signals:

[0130] The dynamic information graph of the bearing vibration signal to be tested is input into the trained convolutional neural network to obtain the fault detection result of the network. By comparing the result with the fault type label, it can be determined whether the vibration signal is normal or in a certain fault state.

[0131] Example 2

[0132] This embodiment provides a bearing fault detection system based on deterministic learning and convolutional neural networks, including:

[0133] The data acquisition module is configured to acquire one-dimensional bearing vibration signals of bearing faults and expand the one-dimensional bearing vibration signals into two-dimensional bearing vibration signals using a high-gain observer.

[0134] The standardization processing module is configured to perform maximum value standardization processing on the two-dimensional bearing vibration signal to obtain a normalized bearing vibration signal.

[0135] The dynamic information acquisition module is configured to capture the dynamic information of the normalized bearing vibration signal based on a deterministic learning mechanism using a pre-trained dynamic radial basis function neural network model, and generate a dynamic information graph of the bearing vibration signal.

[0136] The bearing fault detection module is configured to perform fault detection based on the dynamic information graph of the bearing vibration signal and a pre-trained convolutional neural network model to obtain the bearing fault detection result.

[0137] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1 above. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0138] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0139] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0140] Example 3

[0141] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the bearing fault detection method based on deterministic learning and convolutional neural networks as described in Embodiment 1 above.

[0142] Example 4

[0143] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the bearing fault detection method based on deterministic learning and convolutional neural networks as described in Embodiment 1 above.

[0144] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0145] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0146] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function 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 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0147] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable 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.

[0148] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0149] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A bearing fault detection method based on deterministic learning and convolutional neural networks, characterized in that, include: A one-dimensional bearing vibration signal of bearing failure is obtained, and a high-gain observer is used to extend the one-dimensional bearing vibration signal into a two-dimensional bearing vibration signal. The two-dimensional bearing vibration signal is normalized by standardizing the maximum value to obtain the normalized bearing vibration signal. Based on a deterministic learning mechanism, the internal dynamic information of bearing vibration signals is modeled, and a dynamic information graph of the bearing vibration signals is generated according to the internal dynamic information of the bearing vibration signals, specifically: A deterministic learning mechanism is used to model the bearing vibration signal, resulting in a dynamic information model of the bearing vibration signal, specifically: ； in, , Represents the state trajectory The neighborhood, This indicates that after the weights of the RBF neural network converge, arrive Average weight over the time interval T is the sampling time. This represents the neural network weights to be estimated. This represents the regression vector composed of radial basis functions; The trajectory region of the normalized bearing vibration signal is gridded to obtain gridded dynamic information input. Based on the gridded dynamic information input, a dynamic information graph of the bearing vibration signal is obtained using the dynamic information model of the bearing vibration signal. The dynamic information graph of the bearing vibration signal is an m×m matrix, specifically: ； in, This refers to dividing the interval Evenly divided into The corresponding data points, ; Convert the above matrix into a grayscale image to form a dynamic information map for each bearing vibration signal; Based on the dynamic information graph of the bearing vibration signal, a pre-trained convolutional neural network model is used for fault detection to obtain bearing fault detection results.

2. The bearing fault detection method based on deterministic learning and convolutional neural networks as described in claim 1, characterized in that, The process of standardizing the maximum value of the two-dimensional bearing vibration signal to obtain a normalized bearing vibration signal is as follows: The maximum value is normalized for each dimension of the two-dimensional bearing vibration signal; The normalization function is: ； in, Indicates the first One-dimensional time series of the sample, This indicates that the vibration data is concentrated in The maximum value of all samples in dimension . This represents the normalized sample.

3. The bearing fault detection method based on deterministic learning and convolutional neural networks as described in claim 1, characterized in that, The training process of a convolutional neural network model is as follows: For the generated dynamic information graph of bearing vibration signal, the training set and the test set are divided in a 7:3 ratio; The dynamic information graph of bearing vibration signals from the training set is input into a convolutional neural network for bearing fault detection. The loss function is cross-entropy loss, and the gradient descent algorithm is selected to optimize and update the network parameters to obtain a trained convolutional neural network model. The dynamic information graph of bearing vibration signal from the test set is input into the trained convolutional neural network model to obtain the network's fault detection results.

4. The bearing fault detection method based on deterministic learning and convolutional neural networks as described in claim 3, characterized in that, The structure of the convolutional neural network model includes a convolutional layer, a pooling layer, a convolutional layer, a pooling layer, a convolutional layer, a pooling layer, a convolutional layer, a pooling layer, a flattening layer, a fully connected layer, and a fully connected layer connected in sequence. The ELU activation function and batch normalization are used between the convolutional layer and the pooling layer. The extracted feature map is flattened and input into the fully connected layer. The output of the fully connected layer is used as the output of the entire convolutional neural network.

5. A bearing fault detection system based on deterministic learning and convolutional neural networks, characterized in that, include: The data acquisition module is configured to acquire one-dimensional bearing vibration signals of bearing faults and expand the one-dimensional bearing vibration signals into two-dimensional bearing vibration signals using a high-gain observer. The standardization processing module is configured to perform maximum value standardization processing on the two-dimensional bearing vibration signal to obtain a normalized bearing vibration signal. The dynamics information acquisition module is configured to model the internal dynamics information of the bearing vibration signal based on a deterministic learning mechanism, and generate a dynamics information graph of the bearing vibration signal based on the internal dynamics information of the bearing vibration signal, specifically: A deterministic learning mechanism is used to model the bearing vibration signal, resulting in a dynamic information model of the bearing vibration signal, specifically: ； in, , Represents the state trajectory The neighborhood, This indicates that after the weights of the RBF neural network converge, arrive Average weight over the time interval T is the sampling time. This represents the neural network weights to be estimated. This represents the regression vector composed of radial basis functions; The trajectory region of the normalized bearing vibration signal is gridded to obtain gridded dynamic information input. Based on the gridded dynamic information input, a dynamic information graph of the bearing vibration signal is obtained using the dynamic information model of the bearing vibration signal. The dynamic information graph of the bearing vibration signal is an m×m matrix, specifically: ； in, This refers to dividing the interval Evenly divided into The corresponding data points, ; Convert the above matrix into a grayscale image to form a dynamic information map for each bearing vibration signal; The bearing fault detection module is configured to perform fault detection based on the dynamic information graph of the bearing vibration signal and use a pre-trained convolutional neural network model to obtain the bearing fault detection result.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the bearing fault detection method based on deterministic learning and convolutional neural networks as described in any one of claims 1-4.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the bearing fault detection method based on deterministic learning and convolutional neural networks as described in any one of claims 1-4.

Images