A mechanical equipment fault diagnosis method, device, equipment and readable storage medium

Abstract

The application discloses a mechanical equipment fault diagnosis method, device and equipment and a readable storage medium. The method comprises the following steps: obtaining a one-dimensional vibration signal of a mechanical equipment to be diagnosed; inputting the one-dimensional vibration signal into a Morlet kernel convolution capsule network algorithm model trained to obtain a probability vector corresponding to a fault category; the Morlet kernel convolution capsule network algorithm model is trained by using historical one-dimensional vibration signals with fault category labels; the Morlet kernel convolution capsule network algorithm model can comprise a complex Morlet kernel convolution layer, a main capsule layer, a digital capsule layer and a classification output layer; and the fault category corresponding to the maximum probability in the probability vector is determined as the fault category of the mechanical equipment to be diagnosed. According to the technical scheme disclosed by the application, the Morlet kernel convolution capsule network algorithm model trained is used to perform fault diagnosis on the mechanical equipment to be diagnosed, so that automatic fault diagnosis is realized, and the fault diagnosis efficiency and accuracy are improved.

Description

Technical Field

[0001] This application relates to the field of mechanical equipment fault diagnosis technology, and more specifically, to a mechanical equipment fault diagnosis method, apparatus, equipment, and readable storage medium. Background Technology

[0002] Rolling bearings, cutting tools, and other devices that output one-dimensional mechanical vibration signals are common components in mechanical equipment. They can be damaged due to various factors during actual processing or application. Timely and effective monitoring of their status is crucial for ensuring equipment safety and stability and reducing maintenance costs.

[0003] Currently, for fault diagnosis of mechanical equipment that outputs one-dimensional mechanical vibration signals, accelerometer vibration sensors are generally used to collect the vibration signals. Then, Fourier transform is performed for time-frequency conversion, and the frequency distribution on the spectrum is manually analyzed to determine the operating status of the mechanical equipment. Specifically, experts construct feature engineering based on experience, and then identify the state of the mechanical equipment based on the constructed feature engineering and classifier.

[0004] In the aforementioned implementation methods, manual feature extraction relies heavily on expert experience, resulting in high time costs. Furthermore, experience-based manual feature selection can miss valuable information from the original signal. Additionally, feature extraction from a single domain cannot comprehensively reflect the state of the mechanical equipment, leading to relatively low accuracy in fault diagnosis. Moreover, in existing fault diagnosis methods, the scalar transmission of data within neurons results in insufficient detailed features, further contributing to low accuracy.

[0005] In conclusion, improving the accuracy of fault diagnosis for mechanical equipment is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the purpose of this application is to provide a method, apparatus, device and readable storage medium for diagnosing mechanical equipment faults, so as to improve the accuracy of fault diagnosis of mechanical equipment.

[0007] To achieve the above objectives, this application provides the following technical solution:

[0008] A method for diagnosing mechanical equipment faults, comprising:

[0009] Acquire one-dimensional vibration signals of the mechanical equipment to be diagnosed;

[0010] The one-dimensional vibration signal is input into the Morlet kernel convolutional capsule network algorithm model obtained by training to obtain the probability vector corresponding to the fault category; the Morlet kernel convolutional capsule network algorithm model is trained using historical one-dimensional vibration signals with fault category labels; the Morlet kernel convolutional capsule network algorithm model includes a complex Morlet kernel convolutional layer, a main capsule layer, a digital capsule layer and a classification output layer.

[0011] The fault category corresponding to the highest probability in the probability vector is determined as the fault category of the mechanical equipment to be diagnosed.

[0012] Preferably, the Morlet kernel convolutional capsule network algorithm model is an adaptive Morlet kernel convolutional capsule network algorithm model, the complex Morlet kernel convolutional layer is an adaptive complex Morlet kernel convolutional layer, and the translation factor and scaling factor in the adaptive complex Morlet kernel convolutional layer are updated through backpropagation during training of the adaptive Morlet kernel convolutional capsule network algorithm model.

[0013] Preferably, the adaptive complex Morlet kernel convolutional layer uses complex Morlet wavelet basis functions as convolution kernels;

[0014] The complex Morlet wavelet basis function is: Where f(t) is the one-dimensional vibration signal, t is time, a is the scaling factor, b is the translation factor, ω0 is the angular frequency, and i is the imaginary unit;

[0015] The output of the adaptive complex Morlet kernel convolutional layer is represented as follows:

[0016]

[0017] in, This represents the i-th complex Morlet kernel in the l-th layer. Let represent the i-th convolutional kernel of the l-th layer, W be the size of the complex Morlet kernel, j' represent the size of the receptive field, and r be the size of the receptive field. j This represents the receptive field of the j-th complex Morlet kernel. This represents the signal input to the j-th receptive field;

[0018] The process of updating the translation and scaling factors in the adaptive complex Morlet kernel convolutional layer during backpropagation of the adaptive Morlet kernel convolutional capsule network algorithm model training includes:

[0019]

[0020]

[0021] Where L is the loss function of the adaptive Morlet kernel convolutional capsule network algorithm model, and the loss function is the sum of the interval loss and the reconstruction loss. This represents the gradient derivative of the intermediate layer between the adaptive complex Morlet kernel convolutional layer and the output layer in the adaptive Morlet kernel convolutional capsule network algorithm model. For the k-th complex Morlet convolution kernel, a k and b k Here, η is the scaling factor and translation factor, and η is the learning rate. This represents the deviation corresponding to the scaling factor. The deviation corresponding to the translation factor, z k It is the general term for the differentiation of the gradient of the capsule network layer during backpropagation.

[0022] Preferably, the operation in the main capsule layer is as follows:

[0023]

[0024] Among them, u l(i,j) Indicates the main capsule, f represents the feature data output after the output of the adaptive complex Morlet kernel convolutional layer has passed through the activation function. f Let f represent the activation function. s This represents the squeezing function.

[0025] Preferred options also include:

[0026] Obtain the fault category label corresponding to the one-dimensional vibration signal of the mechanical equipment to be diagnosed;

[0027] The one-dimensional vibration signal of the mechanical equipment to be diagnosed and the corresponding fault category label are used as a new training dataset.

[0028] The newly added training dataset is merged with the existing training dataset to obtain a merged training dataset; the existing training dataset contains historical one-dimensional vibration signals with fault category labels;

[0029] The model is trained using the merged training dataset to obtain a newly trained Morlet kernel convolutional capsule network algorithm model.

[0030] Replace the existing Morlet kernel convolutional capsule network algorithm model with the newly trained Morlet kernel convolutional capsule network algorithm model.

[0031] Preferred options also include:

[0032] Output the fault category of the mechanical equipment to be diagnosed.

[0033] Preferably, the fault category of the mechanical equipment to be diagnosed is output, including:

[0034] The fault category of the mechanical equipment to be diagnosed is sent to the terminal via email and / or SMS.

[0035] A mechanical equipment fault diagnosis device, comprising:

[0036] The first acquisition module is used to acquire the one-dimensional vibration signal of the mechanical equipment to be diagnosed.

[0037] The input module is used to input the one-dimensional vibration signal into the Morlet kernel convolutional capsule network algorithm model obtained by training, and to obtain the probability vector corresponding to the fault category. The Morlet kernel convolutional capsule network algorithm model is trained using historical one-dimensional vibration signals with fault category labels. The Morlet kernel convolutional capsule network algorithm model includes a complex Morlet kernel convolutional layer, a main capsule layer, a digital capsule layer, and a classification output layer.

[0038] The determination module is used to determine the fault category corresponding to the highest probability in the probability vector as the fault category of the mechanical equipment to be diagnosed.

[0039] A mechanical equipment fault diagnosis device, comprising:

[0040] Memory, used to store computer programs;

[0041] A processor, configured to execute the computer program to implement the steps of the mechanical equipment fault diagnosis method as described in any of the preceding claims.

[0042] A readable storage medium storing a computer program that, when executed by a processor, implements the steps of the mechanical equipment fault diagnosis method as described in any of the preceding claims.

[0043] This application provides a method, apparatus, device, and readable storage medium for diagnosing mechanical equipment faults. The method includes: acquiring a one-dimensional vibration signal of the mechanical equipment to be diagnosed; inputting the one-dimensional vibration signal into a trained Morlet kernel convolutional capsule network algorithm model to obtain a probability vector corresponding to the fault category; the Morlet kernel convolutional capsule network algorithm model is trained using historical one-dimensional vibration signals with fault category labels, and the Morlet kernel convolutional capsule network algorithm model may include a complex Morlet kernel convolutional layer, a main capsule layer, a digital capsule layer, and a classification output layer; and determining the fault category corresponding to the highest probability in the probability vector as the fault category of the mechanical equipment to be diagnosed.

[0044] The technical solution disclosed in this application utilizes a trained Morlet kernel convolutional capsule network algorithm model for fault diagnosis of mechanical equipment. Specifically, the trained Morlet kernel convolutional capsule network algorithm model directly uses the one-dimensional vibration signal of the mechanical equipment as model input. This end-to-end method enables automatic fault diagnosis of the mechanical equipment, eliminating the need for prior feature engineering by experts, reducing reliance on expert experience, and improving the efficiency and accuracy of fault diagnosis. Furthermore, this application utilizes the Morlet kernel convolutional capsule network algorithm model to extract features from the one-dimensional vibration signal across multiple domains, comprehensively reflecting the state of the mechanical equipment and thus improving the accuracy of fault diagnosis. Additionally, the Morlet kernel convolutional capsule network algorithm model uses vectors for transmission, which can represent the interrelationships and positional relationships between features, enhancing the ability to extract spatial features and improving the accuracy of fault diagnosis. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0046] Figure 1 A flowchart of a mechanical equipment fault diagnosis method provided in this application embodiment;

[0047] Figure 2 An architecture diagram of the Morlet kernel convolutional capsule network algorithm model provided in the embodiments of this application;

[0048] Figure 3 A schematic diagram illustrating the online model update process provided in this application embodiment;

[0049] Figure 4 A schematic diagram of the confusion matrix obtained by fault identification using the adaptive Morlet kernel convolutional capsule network algorithm model provided in an embodiment of this application;

[0050] Figure 5 This is a schematic diagram of the structure of a mechanical equipment fault diagnosis device provided in an embodiment of this application;

[0051] Figure 6 This is a schematic diagram of the structure of a mechanical equipment fault diagnosis device provided in an embodiment of this application. Detailed Implementation

[0052] The core of this application is to provide a method, apparatus, equipment, and readable storage medium for diagnosing mechanical equipment faults, so as to improve the accuracy of fault diagnosis of mechanical equipment.

[0053] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] See Figure 1 and Figure 2 ,in, Figure 1 A flowchart of a mechanical equipment fault diagnosis method provided in an embodiment of this application is shown. Figure 2 This diagram illustrates the architecture of the Morlet kernel convolutional capsule network algorithm model provided in this application. An embodiment of this application provides a method for diagnosing mechanical equipment faults, which may include:

[0055] S11: Obtain the one-dimensional vibration signal of the mechanical equipment to be diagnosed.

[0056] In this application, an acceleration vibration sensor can be used to detect a one-dimensional vibration signal (specifically, an acceleration vibration signal) of the mechanical equipment to be diagnosed, and the detected one-dimensional vibration signal of the mechanical equipment to be diagnosed can be sent to a mechanical equipment fault diagnosis device, so that the mechanical equipment fault diagnosis device can automatically diagnose the fault of the mechanical equipment to be diagnosed using the one-dimensional vibration signal of the mechanical equipment to be diagnosed.

[0057] The mechanical equipment to be diagnosed can be either a rolling bearing or a cutting tool.

[0058] S12: Input the one-dimensional vibration signal into the Morlet kernel convolutional capsule network algorithm model obtained by training to obtain the probability vector corresponding to the fault category; the Morlet kernel convolutional capsule network algorithm model is trained using historical one-dimensional vibration signals with fault category labels. The Morlet kernel convolutional capsule network algorithm model includes complex Morlet kernel convolutional layers, main capsule layers, digital capsule layers and classification output layers.

[0059] In this application, historical one-dimensional vibration signals of the same mechanical equipment as the equipment to be diagnosed can be acquired in advance, and / or historical one-dimensional vibration signals of the equipment to be diagnosed can be acquired. Then, engineers can label the acquired historical one-dimensional vibration signals with fault category labels to obtain historical one-dimensional vibration signals with fault category labels, and these labeled historical one-dimensional vibration signals can be used as a training set. Subsequently, the initial Morlet kernel convolutional capsule network algorithm model can be trained using the training set to obtain the Morlet kernel convolutional capsule network algorithm model.

[0060] The initial Morlet kernel convolutional capsule network algorithm model and the trained Morlet kernel convolutional capsule network algorithm model have the same architecture, both including complex Morlet kernel convolutional layers, main capsule layers, digital capsule layers, and classification output layers. The complex Morlet kernel convolutional layers use complex Morlet wavelet basis functions as convolution kernels. In other words, this application uses capsule networks as the basic structure combined with complex Morlet wavelet transform as a fault diagnosis model for mechanical equipment, thereby diagnosing the health status of the mechanical equipment.

[0061] Wavelet analysis is a novel time-frequency analysis method, particularly suitable for analyzing non-stationary signals. It features multi-resolution power analysis, capable of simultaneously characterizing local signal features in both the time and frequency domains. It can depict the spectral information of a signal within a specific time period and describe the corresponding time-domain information. This application incorporates complex Morlet wavelet analysis into the fault diagnosis of mechanical equipment, effectively adapting to one-dimensional vibration signals under complex operating conditions and thus providing a superior method for fault diagnosis.

[0062] After acquiring the one-dimensional vibration signal of the mechanical equipment to be diagnosed, the acquired one-dimensional vibration signal can be input into the trained Morlet kernel convolutional capsule network algorithm model to obtain the probability vector corresponding to the fault category.

[0063] In the trained Morlet kernel convolutional capsule network algorithm model, the complex Morlet kernel convolutional layer is used to extract feature data of the one-dimensional vibration signal. The feature data output by the complex Morlet kernel convolutional layer is capsuled in the main capsule layer. The capsules are passed in vector form in the digital capsule layer, and the values ​​in the capsules are compressed to between 0 and 1 by a squeezing function and updated through a dynamic routing mechanism. The feature data output by the digital capsule layer is used in the classification output layer to calculate the loss function value and the probability vector corresponding to the fault category.

[0064] The probability vector contains multiple probability values, each of which corresponds to a fault category. In other words, there is a one-to-one correspondence between fault categories and probability values ​​in the probability vector.

[0065] S13: Determine the fault category corresponding to the highest probability in the probability vector as the fault category of the mechanical equipment to be diagnosed.

[0066] Based on step S12, the maximum probability in the probability vector can be obtained, and the fault category corresponding to the maximum probability can be determined as the fault category of the mechanical equipment to be diagnosed.

[0067] The above method enables the processing of mechanical equipment fault data using a Morlet kernel convolutional capsule network algorithm model based on training. It directly uses the one-dimensional raw signal as model input. This end-to-end approach reduces the reliance on expert experience in the early stages of traditional mechanical equipment fault diagnosis methods, enhances model robustness, and improves the efficiency and accuracy of fault diagnosis. Furthermore, this application uses capsule networks as the basic structure combined with complex Morlet wavelet transform to diagnose the health status of mechanical equipment. Compared with traditional mechanical equipment fault diagnosis methods, data in capsule networks is transmitted in a vector format, which can represent the interrelationships and positional relationships between features, enhancing the ability to extract spatial features. The dynamic routing update mechanism between capsules also improves the model's learning ability. This vector representation method can retain the precise position, orientation, and magnitude parameters of features in the input data. Even subtle changes in the input data are intuitively reflected in the model's output, ensuring the model's accuracy.

[0068] The technical solution disclosed in this application utilizes a trained Morlet kernel convolutional capsule network algorithm model for fault diagnosis of mechanical equipment. Specifically, the trained Morlet kernel convolutional capsule network algorithm model directly uses the one-dimensional vibration signal of the mechanical equipment as model input. This end-to-end method enables automatic fault diagnosis of the mechanical equipment, eliminating the need for prior feature engineering by experts, reducing reliance on expert experience, and improving the efficiency and accuracy of fault diagnosis. Furthermore, this application utilizes the Morlet kernel convolutional capsule network algorithm model to extract features from the one-dimensional vibration signal across multiple domains, comprehensively reflecting the state of the mechanical equipment and thus improving the accuracy of fault diagnosis. Additionally, the Morlet kernel convolutional capsule network algorithm model uses vectors for transmission, which can represent the interrelationships and positional relationships between features, enhancing the ability to extract spatial features and improving the accuracy of fault diagnosis.

[0069] This application provides a method for diagnosing mechanical equipment faults. The Morlet kernel convolutional capsule network algorithm model is an adaptive Morlet kernel convolutional capsule network algorithm model, and the complex Morlet kernel convolutional layer is an adaptive complex Morlet kernel convolutional layer. The translation factor and scaling factor in the adaptive complex Morlet kernel convolutional layer are updated through backpropagation during the training of the adaptive Morlet kernel convolutional capsule network algorithm model.

[0070] In this application, the Morlet kernel convolutional capsule network algorithm model used for fault diagnosis can specifically be an adaptive Morlet kernel convolutional capsule network algorithm model, meaning that the parameters in the model can be adaptively changed. Specifically, the complex Morlet kernel convolutional layer in the adaptive Morlet kernel convolutional capsule network algorithm model is an adaptive complex Morlet kernel convolutional layer, and the translation factor and scaling factor in the adaptive complex Morlet kernel convolutional layer can be updated iteratively through backpropagation as the adaptive Morlet kernel convolutional capsule network algorithm model is trained.

[0071] Traditional convolutional neural networks (CNNs) use a floating-point matrix as the kernel function, initialized with weights. The dimension of this matrix depends on the kernel width, and the scale of kernel translation depends on the stride. In traditional CNNs, the kernel width and stride cannot be updated based on model iterations. This leads to the loss of more state information when processing low-noise-ratio vibration signals. However, the adaptive complex Morlet kernel convolutional layer design in this application can adaptively adjust the translation and scaling factors based on the information contained in the input signal, making it more suitable for low signal-to-noise ratio vibration signals and thus improving the accuracy of mechanical equipment fault diagnosis.

[0072] This application provides a mechanical equipment fault diagnosis method, in which an adaptive complex Morlet kernel convolutional layer uses complex Morlet wavelet basis functions as convolutional kernels;

[0073] The complex Morlet wavelet basis functions are: Where f(t) is a one-dimensional vibration signal, t is time, a is a scaling factor, b is a translation factor, ω0 is the angular frequency, and i is the imaginary unit;

[0074] The output of the adaptive complex Morlet kernel convolutional layer is represented as:

[0075]

[0076] in, This represents the i-th complex Morlet kernel in the l-th layer. Let represent the i-th convolutional kernel in the l-th layer, W be the size of the complex Morlet kernel, j' represent the size of the receptive field, and r be the size of the receptive field. j This represents the receptive field of the j-th complex Morlet kernel. This represents the signal input to the j-th receptive field;

[0077] The process of updating the translation and scaling factors in the adaptive complex Morlet kernel convolutional layer through backpropagation during training of the adaptive Morlet kernel convolutional capsule network algorithm model can include:

[0078]

[0079]

[0080] Where L is the loss function of the adaptive Morlet kernel convolutional capsule network algorithm model, which is the sum of the margin loss and the reconstruction loss. This represents the gradient derivative of the intermediate layer between the adaptive complex Morlet kernel convolutional layer and the output layer in the adaptive Morlet kernel convolutional capsule network algorithm model. For the k-th complex Morlet convolution kernel, a k and b k Here, η is the scaling factor and translation factor, and η is the learning rate. This represents the deviation corresponding to the scaling factor. The deviation corresponding to the translation factor, z k It is the general term for the differentiation of the gradient of the capsule network layer during backpropagation.

[0081] In this application, the input to the adaptive complex Morlet kernel convolutional layer is a one-dimensional vibration signal. In this application, the signal is convolved using a complex Morlet wavelet basis function. The complex Morlet wavelet is a complex cosine Gaussian wave, which can be expressed as:

[0082]

[0083] Among them, f c f is the wavelet center frequency. b The wavelet bandwidth parameter, after Fourier transform, takes the following form:

[0084]

[0085] Where f is the frequency component after Fourier transform.

[0086] After increasing the scaling factor 'a' and the translation factor 'b', and then adding a one-dimensional vibration signal, it can be expressed as:

[0087]

[0088] Where f(t) is a one-dimensional vibration signal, t is time, a is a scaling factor, b is a translation factor, ω0 is the angular frequency, and i is the imaginary unit.

[0089] In an adaptive complex Morlet kernel convolutional layer, a complex Morlet wavelet basis function is used as the convolution kernel. The scaling factor 'a' and the translation factor 'b' in the convolution kernel are updated with the model training gradient and are therefore updatable parameters. The adaptive complex Morlet kernel convolutional layer can be represented as follows:

[0090]

[0091] in, This represents the i-th complex Morlet kernel in the l-th layer. Let represent the i-th convolutional kernel in the l-th layer, W be the size of the complex Morlet kernel, j' represent the size of the receptive field, and r be the size of the receptive field. j This represents the receptive field of the j-th complex Morlet kernel. This represents the signal input to the j-th receptive field.

[0092] For backpropagation parameter updates: During the forward propagation of the adaptive complex Morlet kernel convolutional layer, the scaling factor a and the translation factor b can be updated. The update process is expressed as follows:

[0093]

[0094]

[0095] in, This represents the derivative of the gradient. This represents the gradient derivative of the intermediate layers between the adaptive complex Morlet kernel convolutional layer and the classification output layer in the adaptive Morlet kernel convolutional capsule network algorithm model (the intermediate layers refer to the layers contained in the capsule network, i.e., the gradient derivatives of the digit capsule layer and the main capsule layer). For the k-th complex Morlet convolution kernel, a k and b k Let a be the scaling factor and translation factor. k b k , In this context, k represents the k-th update process, and η is the learning rate. This represents the deviation corresponding to the scaling factor. Let L be the deviation corresponding to the translation factor, and let L be the loss function of the adaptive Morlet kernel convolutional capsule network algorithm model. The loss function uses the sum of the interval loss and reconstruction loss as the total loss function of the adaptive Morlet kernel convolutional capsule network algorithm model. k It can be viewed as the general term for the gradient differentiation of capsule network layers during backpropagation.

[0096] In the two sets of formulas mentioned above, the first set of formulas is the expression for the backpropagation update of the entire network (i.e., the adaptive Morlet kernel convolutional capsule network algorithm model); the second set of formulas is the gradient update process of the scaling factor and translation factor in the adaptive complex Morlet kernel convolutional layer, which is to update the data of the next time step (i.e., k) with the data of the previous time step (i.e., k-1).

[0097] The above process enables adaptive updating of the translation and scaling factors in the adaptive complex Morlet kernel convolutional layer, thereby improving the accuracy of fault diagnosis for mechanical equipment.

[0098] This application provides a method for diagnosing mechanical equipment faults, wherein the calculation in the main capsule layer is as follows:

[0099]

[0100] Among them, u l(i,j) Indicates the main capsule, The output f represents the feature data f output after the output of the adaptive complex Morlet kernel convolutional layer is processed by the activation function. a f represents the activation function. s This represents the squeezing function.

[0101] In this application, the output of the adaptive complex Morlet kernel convolutional layer is the input to the main capsule. The main capsule layer vectorizes the feature data output from the adaptive complex Morlet kernel convolutional layer into the main capsule (specifically, the feature data output from the adaptive complex Morlet kernel convolutional layer is first processed by the activation function f). a Then, vectorized into capsule vectors, the capsule size is compressed by the compression function, where the operation in the main capsule layer can be represented as:

[0102]

[0103] Among them, u l(i,j) Indicates the main capsule, This represents the feature data output after the output of the adaptive complex Morlet kernel convolutional layer passes through an activation function. There is more than one kernel function (i.e., the complex Morlet wavelet kernel mentioned above, also known as the complex Morlet wavelet basis function). Therefore, there is also more than one intermediate feature map calculated by the kernel function, a total of m, where m represents the number of kernel functions and also the number of intermediate feature maps. l(i,j) and In this context, i represents the i-th core, j represents the core size, l represents the l-th layer, and f a f represents the activation function. s The extrusion function can be represented as:

[0104]

[0105] After being squeezed by the squeezing function, the capsule vector is compressed to between 0 and 1 without changing its direction. This allows the relationships between features to be properly transferred to the digital capsule layer, while also accelerating the convergence speed of the model.

[0106] For the digital capsule layer: During the transfer of the main capsule from the main capsule layer to the digital capsule layer, the data output by the squeezing function in the main capsule layer is first multiplied by a weight matrix to obtain the prediction vector, which can be specifically represented as:

[0107]

[0108] Among them, u i Let W represent the i-th input vector. ij Let represent the weight matrix, where i represents the number of rows in the weight matrix and j represents the number of columns in the weight matrix. Let represent the prediction vector, specifically the i-th prediction vector subcapsule in the current propagation layer, where j is the size of the subcapsule vector (i.e., the prediction vector subcapsule).

[0109] Through the coupling coefficient c ij Predicted vector The result is passed to the digital capsule layer, where the coupling coefficient is updated by the dynamic routing mechanism. The dynamic routing update mechanism can be represented as:

[0110]

[0111]

[0112]

[0113]

[0114] The dynamic route update mechanism consists of several steps, each introducing s j and v j To represent the intermediate quantity between each step, i.e., s j and v j These are all intermediate quantities introduced in the dynamic routing update mechanism, c ij Denotes the coupling coefficients and satisfies ∑ i c ij =1, b ij The initial value of b represents the prior probability of capsule i and capsule j being coupled. ij This is the initial value, typically 0. The optimal coupling coefficient, s, is obtained after several iterations. j Indicates capsule After weight matrix transformation, the j-th parent capsule in the next layer, v j The parent capsule after compression, with a value between 0 and 1, c ij Let b be the weight of the i-th child capsule relative to the j-th parent capsule. ij Let be the bias of the i-th child capsule relative to the j-th parent capsule.

[0115] The loss function uses the sum of the margin loss and the reconstruction loss as the overall loss function of the adaptive Morlet kernel convolutional capsule network algorithm model, which can be specifically expressed as:

[0116] L jx =T jx max(0,m + -||v j ||) 2 +λ(1-T jx max(0,||v) j ||-m - ) 2

[0117] Among them, ||v j || represents the output layer capsule (data is transmitted in capsules in the network, so the final output is also in capsule form, v) j Let L be the L2 norm of the j-th compressed parent capsule vector. jx and T jx The subscript jx in T indicates the bearing fault condition category. jx Indicates whether the fault category jx exists, m + The upper boundary for false positives is set at 0.9, m. - The lower boundary for false negatives is 0.1, and λ is the proportionality coefficient with a value of 0.5.

[0118] For the classification output layer: the probability vector of the model for each fault category is calculated using the loss function. The category with the largest probability vector is output as the model's recognition result, thus mapping the probability vector to the fault category of the mechanical equipment to be diagnosed. The classification output layer can be represented as:

[0119]

[0120] Where, [y1,y2,…,y i ] represents the calculated probability vector, class represents the fault category, and i represents the number of fault types.

[0121] The structure allows for accurate and rapid identification of the fault type of the mechanical equipment to be diagnosed.

[0122] See Figure 3 This illustration shows a schematic diagram of the online model update process provided in an embodiment of this application. A mechanical equipment fault diagnosis method provided in an embodiment of this application may further include:

[0123] Obtain the fault category label corresponding to the one-dimensional vibration signal of the mechanical equipment to be diagnosed;

[0124] The one-dimensional vibration signal of the mechanical equipment to be diagnosed and the corresponding fault category label are used as a new training dataset.

[0125] The newly added training dataset is merged with the existing training dataset to obtain the merged training dataset; the existing training dataset contains historical one-dimensional vibration signals with fault category labels.

[0126] The model was trained using the merged training dataset to obtain a newly trained Morlet kernel convolutional capsule network algorithm model;

[0127] Replace the existing Morlet kernel convolutional capsule network algorithm model with the newly trained Morlet kernel convolutional capsule network algorithm model.

[0128] In this application, during system operation, one-dimensional vibration signals of the mechanical equipment to be diagnosed can be continuously collected. Then, engineers label the data to obtain fault category labels for the one-dimensional vibration signals. Correspondingly, the mechanical equipment fault diagnosis equipment can acquire the fault category labels corresponding to the one-dimensional vibration signals of the mechanical equipment to be diagnosed. Alternatively, one-dimensional vibration signals of the same mechanical equipment can be continuously collected and labeled by engineers to obtain fault category labels for the one-dimensional vibration signals. These fault category labels can then be acquired by the mechanical equipment fault diagnosis equipment and incorporated into subsequent new training datasets.

[0129] After obtaining the fault category labels corresponding to the one-dimensional vibration signals of the machinery to be diagnosed, the one-dimensional vibration signals and corresponding fault category labels can be used as a new training dataset. Then, the new training dataset can be merged with the existing training dataset to obtain a merged training dataset, thereby continuously enriching the training dataset. The existing training dataset refers to the dataset that already exists and has participated in model training; this existing training dataset contains historical one-dimensional vibration signals with fault category labels.

[0130] After obtaining the merged training dataset, the model can be trained using the merged training dataset to obtain a newly trained Morlet kernel convolutional capsule network algorithm model. Then, the newly trained Morlet kernel convolutional capsule network algorithm model can replace the existing Morlet kernel convolutional capsule network algorithm model and be used in the fault diagnosis of the mechanical equipment to be diagnosed.

[0131] It should be noted that, specifically, if the adaptive Morlet kernel convolutional capsule network algorithm model is used, then the following steps are taken: the model is trained using the merged training dataset to obtain a newly trained adaptive Morlet kernel convolutional capsule network algorithm model; the newly trained adaptive Morlet kernel convolutional capsule network algorithm model replaces the existing adaptive Morlet kernel convolutional capsule network algorithm model and participates in the fault diagnosis of the mechanical equipment to be diagnosed.

[0132] The above methods can continuously enrich the training dataset to enhance the robustness of the trained model, thereby improving the model's generalization accuracy and fault diagnosis accuracy.

[0133] The mechanical equipment fault diagnosis method provided in this application embodiment may further include:

[0134] Output the fault category of the mechanical equipment to be diagnosed.

[0135] In this application, after determining the fault category corresponding to the highest probability in the probability vector as the fault category of the mechanical equipment to be diagnosed, the fault category of the mechanical equipment to be diagnosed can also be output so that relevant personnel can obtain the fault category of the mechanical equipment to be diagnosed.

[0136] This application provides a method for diagnosing mechanical equipment faults, which outputs the fault category of the mechanical equipment to be diagnosed, and may include:

[0137] The fault category of the mechanical equipment to be diagnosed is sent to the terminal via email and / or SMS.

[0138] In this application, when outputting the fault category of the mechanical equipment to be diagnosed, the fault category can be sent to the terminal via email and / or SMS, so that relevant personnel can obtain the fault category of the mechanical equipment to be diagnosed through the terminal. Of course, the fault category of the mechanical equipment to be diagnosed can also be output via voice broadcast or other methods.

[0139] Specifically, the effectiveness of the above method has been verified through a completed project's rolling bearing failure data example.

[0140] 1. The object to be monitored in the experiment was a bearing of a certain model connected to the motor drive end, with an inner diameter of 25mm, an outer diameter of 52mm, and a rolling element diameter of 7.94mm. Vibration signals under different fault types were collected using a vibration sensor at a sampling frequency of 12kHz. The fault types included: a) inner ring fault; b) outer ring fault; c) rolling element fault. The damage at the fault location had three different damage sizes: 0.007inch, 0.014inch, and 0.021inch. Including the normal bearing condition dataset, there were a total of ten category labels. Signal segments with a sequence length of 784 were extracted sequentially from a single data file as samples, for a total of 3000 sets of samples, with 300 samples for each type of fault.

[0141] 2. Data Labeling and Segmentation: The bearing fault signals include a total of 10 fault types, each of which is one-hot encoded. The dataset contains 300 samples, which are manually divided into a training set (1800 samples) and a test set (1200 samples).

[0142] 3. Directly use the one-dimensional original signal sample data as the input to the adaptive Morlet kernel convolutional capsule network algorithm model, denoted as... It is a tensor with a shape of 1×784, and the feature data output by the adaptive complex Morlet kernel convolutional layer has a shape of 20×20×256.

[0143] 4. The feature data output by the adaptive complex Morlet kernel convolutional layer is capsuled in the main capsule layer. The capsule size is set to caps = 8, and the shape of the output data of the main capsule layer is 6×6×8×32.

[0144] 5. Capsules are passed in the digital capsule layer in vector form. The values ​​in the capsules are compressed to between 0 and 1 by the squeezing function and updated through a dynamic routing mechanism. The shape of the output data of the digital capsule layer is 16×10.

[0145] 6. The feature data output from the digital capsule layer is used in the classification output layer to calculate the loss function value and the prediction vector, realizing the mapping between the prediction vector and the bearing fault category. The mapping between the prediction vector value and the bearing fault category can be expressed as:

[0146]

[0147] When i = n, the algorithm model identifies the bearing fault as the nth type.

[0148] Regarding the effects achieved by the mechanical equipment fault diagnosis method of this application:

[0149] 1. As can be seen from the example application, this end-to-end method based on the Morlet kernel convolutional capsule network algorithm model can reduce the dependence of traditional bearing fault diagnosis methods on expert experience in the early stage. While ensuring diagnostic accuracy, it avoids complex feature engineering in the early stage, making it more suitable for practical industrial applications.

[0150] 2. This application uses a capsule network structure combined with complex Morlet wavelet transform to extract the time-frequency domain features of the fault signal, providing a more comprehensive characterization of bearing condition changes. Simultaneously, features in the network model are transmitted in vector form, preserving the fault feature information of the original data while also visually reflecting subtle changes in the bearing condition in the model's classification results, thus ensuring the model's accuracy.

[0151] 3. The adaptive Morlet kernel convolutional capsule network algorithm model achieved a diagnostic accuracy of 98.9167% on 10 types of fault samples. The confusion matrix of the classification results is as follows: Figure 4 As shown, it illustrates a schematic diagram of the confusion matrix obtained by fault identification using the adaptive Morlet kernel convolutional capsule network algorithm model provided in an embodiment of this application.

[0152] This application also provides a mechanical equipment fault diagnosis device, see [link to relevant documentation]. Figure 5 It shows a structural schematic diagram of a mechanical equipment fault diagnosis device provided in an embodiment of this application, which may include:

[0153] The first acquisition module 51 is used to acquire the one-dimensional vibration signal of the mechanical equipment to be diagnosed.

[0154] Input module 52 is used to input one-dimensional vibration signals into the Morlet kernel convolutional capsule network algorithm model obtained by training, and obtain the probability vector corresponding to the fault category. The Morlet kernel convolutional capsule network algorithm model is trained using historical one-dimensional vibration signals with fault category labels. The Morlet kernel convolutional capsule network algorithm model may include complex Morlet kernel convolutional layers, main capsule layers, digital capsule layers and classification output layers.

[0155] The determination module 53 is used to determine the fault category corresponding to the highest probability in the probability vector as the fault category of the mechanical equipment to be diagnosed.

[0156] This application provides a mechanical equipment fault diagnosis device. The Morlet kernel convolutional capsule network algorithm model is an adaptive Morlet kernel convolutional capsule network algorithm model, and the complex Morlet kernel convolutional layer is an adaptive complex Morlet kernel convolutional layer. The translation factor and scaling factor in the adaptive complex Morlet kernel convolutional layer are updated through backpropagation during the training of the adaptive Morlet kernel convolutional capsule network algorithm model.

[0157] This application provides a mechanical equipment fault diagnosis device, in which the adaptive complex Morlet kernel convolutional layer uses complex Morlet wavelet basis functions as convolutional kernels;

[0158] The complex Morlet wavelet basis functions are: Where f(t) is the one-dimensional vibration signal, t is time, a is the scaling factor, b is the translation factor, ω0 is the angular frequency, and i is the imaginary unit;

[0159] The output of the adaptive complex Morlet kernel convolutional layer is represented as:

[0160]

[0161] in, This represents the i-th complex Morlet kernel in the l-th layer. Let represent the i-th convolutional kernel in the l-th layer, W be the size of the complex Morlet kernel, j' represent the size of the receptive field, and r be the size of the receptive field. j This represents the receptive field of the j-th complex Morlet kernel. This represents the signal input to the j-th receptive field;

[0162] The process of updating the translation and scaling factors in the adaptive complex Morlet kernel convolutional layer through backpropagation during training of the adaptive Morlet kernel convolutional capsule network algorithm model can include:

[0163]

[0164]

[0165] Where L is the loss function of the adaptive Morlet kernel convolutional capsule network algorithm model, which is the sum of the margin loss and the reconstruction loss. This represents the gradient derivative of the intermediate layer between the adaptive complex Morlet kernel convolutional layer and the output layer in the adaptive Morlet kernel convolutional capsule network algorithm model. For the k-th complex Morlet convolution kernel, a k and b k Here, η is the scaling factor and translation factor, and η is the learning rate. This represents the deviation corresponding to the scaling factor. The deviation corresponding to the translation factor, z k It is the general term for the differentiation of the gradient of the capsule network layer during backpropagation.

[0166] This application provides a mechanical equipment fault diagnosis device, wherein the operation in the main capsule layer is:

[0167]

[0168] Among them, u l(i,j) Indicates the main capsule, f represents the feature data output after the output of the adaptive complex Morlet kernel convolutional layer is processed by the activation function. a f represents the activation function. s This represents the squeezing function.

[0169] The mechanical equipment fault diagnosis device provided in this application embodiment may further include:

[0170] The second acquisition module is used to acquire the fault category label corresponding to the one-dimensional vibration signal of the mechanical equipment to be diagnosed.

[0171] As a module, it is used to add a new training dataset by taking the one-dimensional vibration signal of the mechanical equipment to be diagnosed and the corresponding fault category label;

[0172] The merging module is used to merge the newly added training dataset with the existing training dataset to obtain the merged training dataset; the existing training dataset contains historical one-dimensional vibration signals with fault category labels.

[0173] The training module is used to train the model using the merged training dataset to obtain a newly trained Morlet kernel convolutional capsule network algorithm model.

[0174] The replacement module is used to replace the existing Morlet kernel convolutional capsule network algorithm model with a newly trained Morlet kernel convolutional capsule network algorithm model.

[0175] The mechanical equipment fault diagnosis device provided in this application embodiment may further include:

[0176] The output module is used to output the fault category of the mechanical equipment to be diagnosed.

[0177] This application provides a mechanical equipment fault diagnosis device, the output module of which may include:

[0178] The output unit is used to send the fault category of the mechanical equipment to be diagnosed to the terminal via email and / or SMS.

[0179] This application also provides a mechanical equipment fault diagnosis device, see [link to relevant documentation]. Figure 6 It shows a structural schematic diagram of a mechanical equipment fault diagnosis device provided in an embodiment of this application, which may include:

[0180] Memory 61 is used to store computer programs;

[0181] When processor 62 executes a computer program stored in memory 61, it can perform the following steps:

[0182] A one-dimensional vibration signal of the mechanical equipment to be diagnosed is acquired; the one-dimensional vibration signal is input into the Morlet kernel convolutional capsule network algorithm model trained to obtain the probability vector corresponding to the fault category; the Morlet kernel convolutional capsule network algorithm model is trained using historical one-dimensional vibration signals with fault category labels, and the Morlet kernel convolutional capsule network algorithm model may include complex Morlet kernel convolutional layers, main capsule layers, digital capsule layers and classification output layers; the fault category corresponding to the highest probability in the probability vector is determined as the fault category of the mechanical equipment to be diagnosed.

[0183] This application embodiment also provides a readable storage medium storing a computer program, which, when executed by a processor, can perform the following steps:

[0184] A one-dimensional vibration signal of the mechanical equipment to be diagnosed is acquired; the one-dimensional vibration signal is input into the Morlet kernel convolutional capsule network algorithm model trained to obtain the probability vector corresponding to the fault category; the Morlet kernel convolutional capsule network algorithm model is trained using historical one-dimensional vibration signals with fault category labels, and the Morlet kernel convolutional capsule network algorithm model may include complex Morlet kernel convolutional layers, main capsule layers, digital capsule layers and classification output layers; the fault category corresponding to the highest probability in the probability vector is determined as the fault category of the mechanical equipment to be diagnosed.

[0185] The readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0186] For a description of the relevant parts of the mechanical equipment fault diagnosis device, equipment and readable storage medium provided in the embodiments of this application, please refer to the detailed description of the relevant parts of the mechanical equipment fault diagnosis method provided in the embodiments of this application, and will not be repeated here.

[0187] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that the elements inherent in a process, method, article, or apparatus that includes a list of elements are included. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Additionally, portions of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0188] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A mechanical equipment failure diagnosis method characterized by, The method comprises the following steps: obtaining a one-dimensional vibration signal of a mechanical equipment to be diagnosed; inputting the one-dimensional vibration signal into a trained Morlet kernel convolution capsule network algorithm model to obtain a probability vector corresponding to a fault category; the Morlet kernel convolution capsule network algorithm model is trained by using historical one-dimensional vibration signals with fault category labels; the Morlet kernel convolution capsule network algorithm model comprises a complex Morlet kernel convolution layer, a primary capsule layer, a digital capsule layer and a classification output layer; determining a fault category corresponding to a maximum probability in the probability vector as the fault category of the mechanical equipment to be diagnosed; the Morlet kernel convolution capsule network algorithm model is an adaptive Morlet kernel convolution capsule network algorithm model; the complex Morlet kernel convolution layer is an adaptive complex Morlet kernel convolution layer; translation factors and scaling factors in the adaptive complex Morlet kernel convolution layer are updated through back propagation during training of the adaptive Morlet kernel convolution capsule network algorithm model; wherein the adaptive complex Morlet kernel convolution layer uses a complex Morlet wavelet basis function as a convolution kernel; the complex Morlet wavelet basis function is: ； wherein, is the one-dimensional vibration signal, t is time, is a scaling factor, b is a translation factor, is an angular frequency, i is the imaginary unit; the output of the adaptive complex Morlet kernel convolution layer is represented as: ; wherein, represents the i-th layer, represents the i-th layer, represents the i-th is the size of the complex Morlet kernel, and represents the size of the receptive field, represents the receptive field of the j-th complex Morlet kernel, represents the input signal on the j-th receptive field; the process of updating the translation factors and the scaling factors in the adaptive complex Morlet kernel convolution layer through back propagation during training of the adaptive Morlet kernel convolution capsule network algorithm model comprises: ; wherein, is a loss function of the adaptive Morlet kernel convolution capsule network algorithm model, the loss function being a sum of a margin loss and a reconstruction loss, denotes a gradient derivation of an intermediate layer between the adaptive complex Morlet kernel convolution layer and an output layer in the adaptive Morlet kernel convolution capsule network algorithm model, is a k-th complex Morlet kernel, and is a scaling factor and a translation factor, is a learning rate, is a scaling factor a corresponding bias, is a translation factor a corresponding bias, is a generalization of the gradient derivation of the capsule network layer in the backpropagation process.

2. The mechanical equipment failure diagnosis method according to claim 1, characterized by, the operation in the primary capsule layer is: ; wherein, denotes the main capsule, denotes the feature data outputted after the output of the adaptive complex Morlet kernel convolution layer passes through the activation function, denotes the activation function, denotes the squeezing function.

3. The mechanical equipment failure diagnosis method according to claim 1, characterized by, The method further comprises the following steps: obtaining a fault category label corresponding to the one-dimensional vibration signal of the mechanical equipment to be diagnosed; using the one-dimensional vibration signal of the mechanical equipment to be diagnosed and the corresponding fault category label as new training data sets; merging the new training data sets with existing training data sets to obtain merged training data sets; the existing training data sets comprise historical one-dimensional vibration signals with fault category labels; training a model by using the merged training data sets to obtain a newly trained Morlet kernel convolution capsule network algorithm model; replacing the existing Morlet kernel convolution capsule network algorithm model with the newly trained Morlet kernel convolution capsule network algorithm model.

4. The mechanical equipment failure diagnosis method according to claim 1, characterized by, The method further comprises the following steps: outputting the fault category of the mechanical equipment to be diagnosed.

5. The mechanical equipment failure diagnosis method according to claim 4, characterized by, The outputting the fault category of the mechanical equipment to be diagnosed comprises the following steps: sending the fault category of the mechanical equipment to be diagnosed to a terminal through an email and / or a short message.

6. A mechanical equipment failure diagnosis device characterized by comprising: The method comprises the following steps: a first obtaining module is configured to obtain a one-dimensional vibration signal of a mechanical equipment to be diagnosed; an inputting module is configured to input the one-dimensional vibration signal into a trained Morlet kernel convolution capsule network algorithm model to obtain a probability vector corresponding to a fault category; the Morlet kernel convolution capsule network algorithm model is trained by using historical one-dimensional vibration signals with fault category labels; the Morlet kernel convolution capsule network algorithm model comprises a complex Morlet kernel convolution layer, a primary capsule layer, a digital capsule layer and a classification output layer; a determining module is configured to determine a fault category corresponding to a maximum probability in the probability vector as the fault category of the mechanical equipment to be diagnosed; The Morlet kernel convolution capsule network algorithm model is an adaptive Morlet kernel convolution capsule network algorithm model, the complex Morlet kernel convolution layer is an adaptive complex Morlet kernel convolution layer, and the translation factor and the scaling factor in the adaptive complex Morlet kernel convolution layer are updated with back propagation during training of the adaptive Morlet kernel convolution capsule network algorithm model; The adaptive complex Morlet kernel convolution layer uses a complex Morlet wavelet basis function as a convolution kernel. The complex Morlet wavelet basis function is as follows: ； wherein is the one-dimensional vibration signal, t is time, is the scaling factor, b is the translation factor, is the angular frequency, i is the imaginary unit; The output of the adaptive complex Morlet kernel convolution layer is represented as follows: ; wherein, represents the i-th complex Morlet kernel of the j-th layer, represents the i-th complex Morlet kernel of the j-th layer, represents the i-th complex Morlet kernel of the j-th layer, represents the i-th complex Morlet kernel of the j-th layer, represents the i-th complex Morlet kernel of the j-th layer, is the size of the complex Morlet kernel, and represents the size of the receptive field, represents the receptive field of the j-th complex Morlet kernel, represents the input signal on the j-th receptive field; The process in which the translation factor and the scaling factor in the adaptive complex Morlet kernel convolution layer are updated with back propagation during training of the adaptive Morlet kernel convolution capsule network algorithm model includes: ; wherein, is a loss function of the adaptive Morlet kernel convolution capsule network algorithm model, the loss function being a sum of a margin loss and a reconstruction loss, denotes a gradient derivation of an intermediate layer between the adaptive complex Morlet kernel convolution layer and an output layer in the adaptive Morlet kernel convolution capsule network algorithm model, is a kth complex Morlet convolution kernel, and is a scaling factor and a translation factor, is a learning rate, is a scaling factor a corresponding bias, is a translation factor a corresponding bias, is a generalization of the gradient derivation of the capsule network layer in the backpropagation process.

7. A mechanical equipment failure diagnosing apparatus characterized by comprising: including: a memory for storing a computer program; a processor for implementing the steps of the mechanical equipment fault diagnosis method according to any one of claims 1 to 5 when the computer program is executed.

8. A readable storage medium, characterized by, The computer program is stored in the readable storage medium and is executed by the processor to implement the steps of the mechanical equipment fault diagnosis method according to any one of claims 1 to 5.

Images