Method and system for predicting service life of bearing of electric energy meter production line, equipment and medium

Abstract

The application discloses a kind of life prediction method and system, equipment, medium of bearing of electric energy meter production line, the life prediction method of the present application can improve the weight of important feature and reduce the weight of unimportant feature by the extraction of time domain feature and frequency domain feature of improved TPA algorithm, can be according to the importance of feature at each time node Reallocates weight proportion, realizes the dynamic adjustment of time-frequency domain feature weight, so as to retain the feature that is more influential to remaining useful life prediction, and exclude redundant features, so that the feature after fusion is more refined and representative, reduce subsequent calculation amount, and, reduce the interference of unimportant feature to model final decision, focus attention on prediction target, so as to improve the prediction accuracy and robustness of prediction model.

Description

Technical Field

[0001] This invention relates to the field of service life prediction technology, and in particular, to a method and system for predicting the service life of bearings in an electricity meter production line, an electronic device, and a computer-readable storage medium. Background Technology

[0002] In the mass production of smart meters, the bearings (generally rolling bearings) on the meter production line are the core power transmission components of the production line's drive system, and their health directly determines the production line's operational efficiency. Given the current industry demand for over ten million smart meters annually, these production line bearings typically need to operate continuously at high speeds (generally 2000-3000 rpm) for 24 hours. A single bearing failure can cause a 4-8 hour downtime for the entire line, resulting in economic losses exceeding one million yuan. More seriously, micron-level vibration deviations (<50μm) caused by early bearing wear can be transmitted to precision assembly stages, leading to exceedances of key parameters (such as metering accuracy and communication module positioning), causing batch quality incidents. Therefore, improving the prediction accuracy of the remaining useful life (RUL) of bearings on the meter production line can not only reduce the risk of production line downtime but also prevent batch quality incidents such as metering accuracy deviations caused by early bearing wear.

[0003] The Remaining Life (RUL) prediction problem for bearings in electricity meter production lines is a typical time-series problem. Current research often employs attention mechanisms to capture key features of time-series data for lifetime prediction. For example, the paper "An Improved CNN-LSTM Model for Rolling Bearing Remaining Life Prediction" uses Temporal Pattern Attention Mechanism (TPA) to focus on key information and then performs data prediction within an LSTM model. Compared to other models, this model demonstrates higher accuracy and better generalization ability. Classical TPA effectively highlights information from key time nodes and is more suitable for time-series model prediction. However, because classical TPA does not consider the contribution rate and relevance of data nodes when assigning weights to all time-series information, it leads to the TPA model consuming significant computational resources for irrelevant or unrepresentative data nodes, resulting in wasted resources. More importantly, it fails to eliminate the interference of unimportant and noisy features on the model's final decision, affecting the model's prediction accuracy. Summary of the Invention

[0004] This invention provides a method and system for predicting the lifespan of bearings in an electricity meter production line, an electronic device, and a computer-readable storage medium, which can improve the prediction accuracy and robustness of the prediction model.

[0005] According to one aspect of the present invention, a method for predicting the lifespan of bearings in an electricity meter production line is provided, comprising the following:

[0006] Vibration signal data of bearings on the electricity meter production line are collected, and the time-domain and frequency-domain signals of the vibration signal data are extracted.

[0007] An improved TPA algorithm is used to extract time-domain feature signals and frequency-domain feature signals from the time-domain signal and the frequency-domain signal respectively, and then fuse them to obtain a time-frequency fused signal. The improved TPA algorithm increases the weight of important features and decreases the weight of unimportant features.

[0008] The time-frequency domain fusion signal is input into the prediction model to predict the remaining service life of the bearings in the electricity meter production line.

[0009] Furthermore, the process of extracting time-domain feature signals from time-domain signals using the improved TPA algorithm includes the following:

[0010] The past information is convolved by a convolutional neural network to obtain a feature matrix, and the score of each row in the feature matrix is ​​calculated using a scoring function to obtain a score matrix.

[0011] Compare all the ratings in the rating matrix with the preset first threshold and second threshold. If a rating is greater than the first threshold, increase the rating. If a rating is less than the second threshold, decrease the rating. If a rating is greater than or equal to the second threshold and less than or equal to the first threshold, keep the rating unchanged and generate a new rating matrix.

[0012] The new scoring matrix and the feature matrix are weighted and summed to obtain the feature vector. This feature vector is then concatenated with the current information to obtain the final feature vector.

[0013] Furthermore, the scoring is adjusted based on the following formula:

[0014]

[0015] in, Represents the first in the rating matrix x Each rating express Adjusted rating β This indicates that the parameters are being adjusted. X Indicates the first threshold. Y This represents the second threshold.

[0016] Furthermore, all scores in the rating matrix are sorted from smallest to largest, and the first quartile is taken as the second threshold, and the third quartile is taken as the first threshold.

[0017] Furthermore, after obtaining the time-frequency domain fused signal, the following is also included:

[0018] A temporal convolutional network is used to perform multiple dilation operations on the time-frequency domain fused signal to extract deep features.

[0019] Furthermore, the process of performing multiple dilation operations on the time-frequency domain fused signal using a temporal convolutional network includes the following:

[0020] The temporal convolutional network is used to first perform a dilation operation on the time-frequency domain fused signal by signal overlap, and then deep features are extracted through two dilation causal convolution operations.

[0021] Furthermore, the temporal convolutional network includes three convolutional kernels of different sizes, distributed in three parallel temporal convolutional branches, to extract short, medium, and long-scale features respectively.

[0022] In addition, the present invention also provides a life prediction system for bearings in an electricity meter production line, comprising:

[0023] The time-frequency domain signal processing module is used to collect vibration signal data of the bearings in the electricity meter production line and extract the time-domain and frequency-domain signals of the vibration signal data.

[0024] The time-frequency domain signal fusion module is used to extract time-domain feature signals and frequency-domain feature signals from the time-domain signal and the frequency-domain signal respectively using an improved TPA algorithm, and then fuse them to obtain a time-frequency domain fused signal. The improved TPA algorithm increases the weight of important features and decreases the weight of unimportant features.

[0025] The remaining service life prediction module is used to input the time-frequency domain fusion signal into the prediction model to predict the remaining service life of the bearings in the electricity meter production line.

[0026] In addition, the present invention also provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the method described above by calling the computer program stored in the memory.

[0027] In addition, the present invention provides a computer-readable storage medium for storing a computer program for predicting the bearing life of an electricity meter production line, wherein the computer program executes the steps of the method described above when run on a computer.

[0028] The present invention has the following beneficial effects:

[0029] The life prediction method for bearings in an electricity meter production line of the present invention extracts time-domain and frequency-domain features through an improved TPA algorithm. This method increases the weight of important features and decreases the weight of unimportant features. The weight ratio can be reallocated according to the importance of features at each time node, realizing dynamic adjustment of the weight of time-frequency domain features. This retains features that have a greater impact on the prediction of remaining lifespan and eliminates redundant features, making the fused features more refined and representative. This reduces the subsequent computational load and minimizes the interference of unimportant features on the final decision of the model, allowing all attention to be focused on the prediction target, thereby improving the prediction accuracy and robustness of the prediction model.

[0030] In addition, the life prediction system for the bearings of the electricity meter production line of the present invention also has the above-mentioned advantages.

[0031] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0032] The accompanying drawings, which form part of this application, 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 undue limitation of the invention. In the drawings:

[0033] Figure 1 This is a flowchart illustrating a preferred embodiment of the life prediction method for bearings in an electricity meter production line.

[0034] Figure 2 yes Figure 1 A schematic diagram of the sub-process of step S2;

[0035] Figure 3 This is a schematic diagram illustrating the principle of the improved TPA algorithm for accurate feature memorization in a preferred embodiment of this application;

[0036] Figure 4 This is another schematic diagram of the life prediction method for bearings in an electricity meter production line according to a preferred embodiment of this application;

[0037] Figure 5 This is a schematic diagram showing the comparison results of the prediction method of this application with existing prediction methods in a preferred embodiment of this application;

[0038] Figure 6 This is a schematic diagram of the prediction results before and after adding 5dB of noise during the noise robustness test in a preferred embodiment of this application;

[0039] Figure 7 This is a schematic diagram of the prediction results before and after adding 10dB of noise during a noise robustness test in a preferred embodiment of this application.

[0040] Figure 8 This is a schematic diagram of the module structure of a life prediction system for bearings in an electricity meter production line, according to another embodiment of this application. Detailed Implementation

[0041] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0042] Reference Figure 1 A preferred embodiment of this application provides a method for predicting the lifespan of bearings in an electricity meter production line, comprising the following:

[0043] Step S1: Collect vibration signal data of the bearings on the electricity meter production line, and extract the time domain signal and frequency domain signal of the vibration signal data;

[0044] Step S2: The improved TPA algorithm is used to extract time-domain feature signals and frequency-domain feature signals from the time-domain signal and the frequency-domain signal respectively, and then fused to obtain the time-frequency domain fused signal. The improved TPA algorithm increases the weight of important features and decreases the weight of unimportant features.

[0045] Step S3: Input the time-frequency domain fusion signal into the prediction model to predict the remaining service life of the bearings in the electricity meter production line.

[0046] It is understood that the life prediction method for bearings on an electricity meter production line in this embodiment extracts time-domain and frequency-domain features through an improved TPA algorithm. This increases the weight of important features and decreases the weight of unimportant features. The weight ratio can be reallocated according to the importance of features at each time node, realizing dynamic adjustment of the weight of time-frequency domain features. This retains features that have a greater impact on the prediction of remaining lifespan and eliminates redundant features, making the fused features more refined and representative, reducing the subsequent computational load, and reducing the interference of unimportant features on the model's final decision. This allows all attention to be focused on the prediction target, thereby improving the prediction accuracy and robustness of the prediction model.

[0047] In step S1, vibration signal data of the bearings on the electricity meter production line is first collected using vibration sensors, including horizontal and vertical vibration signal data. Then, the time-domain signal is extracted from the vibration signal data using the EEMD algorithm, and the frequency-domain signal is extracted using the FFT algorithm. Compared to the time-domain signal, the frequency-domain signal contains more bearing degradation information and can more accurately reflect the periodic characteristics of the bearing degradation process. Both the EEMD and FFT algorithms are existing technologies, and their specific principles will not be elaborated here. Optionally, after collecting the vibration signal data, noise reduction processing can be performed to filter out some noise interference.

[0048] Furthermore, due to the complex composition and large amount of information in bearing vibration signals, traditional TPA algorithms do not consider the importance of current information when processing data, indiscriminately assigning weights to all current information, resulting in a significant waste of computational resources by consuming computational power for unimportant features. Therefore, in step S2, this application utilizes an improved TPA algorithm to extract time-domain feature signals and frequency-domain feature signals from the time-domain signal and the frequency-domain signal respectively, increasing the weight of important features and decreasing the weight of unimportant features, and fusing the extracted time-domain feature signals and frequency-domain feature signals to obtain a time-frequency fused signal. For example, Figure 2 As shown, the process of extracting time-domain feature signals from time-domain signals using the improved TPA algorithm includes the following:

[0049] Step S21: Perform convolution operations on past information using a convolutional neural network to obtain a feature matrix, and use a scoring function to calculate the score of each row in the feature matrix to obtain a scoring matrix;

[0050] Step S22: Compare all the ratings in the rating matrix with the preset first threshold and second threshold. If a rating is greater than the first threshold, increase the rating. If a rating is less than the second threshold, decrease the rating. If a rating is greater than or equal to the second threshold and less than or equal to the first threshold, keep the rating unchanged and generate a new rating matrix.

[0051] Step S23: Perform a weighted summation of the new scoring matrix and the feature matrix to obtain the feature vector, and then concatenate the feature vector with the current information to obtain the final feature vector.

[0052] Specifically, such as Figure 3 As shown, the feature matrix is ​​obtained by first performing convolution operations on the past information nodes of the convolutional neural network (CNN). H cCNNs extract features from input data through convolution and pooling operations. Convolution uses kernels to extract local features, while pooling downsamples the features extracted by the convolutional layers, reducing feature dimensionality, parameter count, and preventing overfitting. Then, a scoring function is used to calculate the score for each row of the feature matrix, resulting in a scoring matrix. This scoring function assesses the influence of past information on the present; a higher influence results in a higher score. The specific scoring function and its underlying principles are existing technologies and will not be elaborated here. Next, all scores in the scoring matrix are compared to preset first and second thresholds. If a score is greater than the first threshold, its weight is increased, allowing the model to focus more on important features and capture key temporal patterns. If a score is less than the second threshold, its weight is decreased, reducing the influence of unimportant or noisy features and minimizing their interference with the final decision. If a score is greater than or equal to the second threshold and less than or equal to the first threshold, it remains unchanged, generating a new scoring matrix, the precise memory matrix. Finally, the new rating matrix will be... With characteristic matrix H c A weighted summation is performed to obtain a context vector that reflects the importance of past information to the current node's information. P t Then the context vector P t With current information h t The features are concatenated to obtain the final feature vector. The final feature vector not only contains information about the current node, but also includes a representation of the importance of past information to the current node information.

[0053] Optional, the score may be adjusted based on the following formula:

[0054]

[0055] in, Represents the first in the rating matrix x Each rating express Adjusted rating β This indicates that the parameter is being adjusted, typically to 0.1. X Indicates the first threshold. Y This represents the second threshold.

[0056] Optionally, all scores in the rating matrix are sorted from smallest to largest, and the first quartile is used as the second threshold, and the third quartile as the first threshold. It can be understood that by selecting quartiles as thresholds, the thresholds possess stable statistical properties, are insensitive to outliers, reflect the overall data distribution, avoid abnormal weights interfering with the model, and allow for automatic threshold adjustment by using quartiles to preserve key degradation patterns, ensuring the rationality and interpretability of the threshold settings.

[0057] It is understood that this invention improves upon the traditional TPA algorithm by introducing a threshold filtering mechanism after the scoring function. This adaptively adjusts the output of the scoring function, avoiding situations where the output of the scoring function is too concentrated or scattered. The algorithm focuses its attention on the most relevant and important parts, deepening the memory of important information while completely forgetting unimportant information, thus achieving accurate memory. This not only improves the accuracy of the algorithm and saves computing resources, but also helps to capture data degradation characteristics and enhances the robustness of the algorithm.

[0058] Furthermore, the process of extracting frequency domain feature signals from frequency domain signals using the improved TPA algorithm is the same as described above and will not be repeated here. Additionally, after extracting the time-domain and frequency-domain feature signals using the improved TPA algorithm, weighted fusion or splicing methods can be used to fuse the time-domain and frequency-domain feature signals to obtain a time-frequency fused signal. The specific fusion process is existing technology and will not be elaborated here.

[0059] Furthermore, in step S3, the prediction model employs a BiLSTM model. The BiLSTM model consists of a bidirectional LSTM layer and a fully connected layer. The forward LSTM inputs data into the model in the forward direction, while the backward LSTM inputs data in the reverse direction. The fully connected layer is responsible for concatenating the hidden states of the forward and backward LSTM at each time step. The training process and prediction principle of the BiLSTM model are existing technologies and will not be elaborated upon here.

[0060] Optional, such as Figure 4 As shown, the life prediction method for the bearings of the electricity meter production line, after obtaining the time-frequency domain fusion signal, also includes the following:

[0061] Step S20: Use a temporal convolutional network to perform multiple dilation operations on the time-frequency domain fused signal to extract deep-level features.

[0062] Specifically, while retaining the powerful feature extraction capabilities of CNNs, Temporal Convolutional Networks (TCNs) are better suited for processing time-series data, extracting deeper features and improving prediction accuracy. However, traditional TCNs only perform one dilation convolution on the time-series data, which can easily lead to the loss of important information. Therefore, this application utilizes a temporal convolutional network to perform multiple dilation operations on the time-frequency domain fused signal, enabling the comprehensive and thorough learning of the important features of the vibration signal without any omissions, and ensuring that no important information is lost.

[0063] The process of performing multiple dilation operations on the time-frequency domain fused signal using a temporal convolutional network includes the following:

[0064] The temporal convolutional network is used to first perform a dilation operation on the time-frequency domain fused signal by signal overlap, and then deep features are extracted through two dilation causal convolution operations.

[0065] Specifically, assuming the size of the time-frequency domain fusion signal is j×3, the TCN first divides the time-frequency domain fusion signal into three segments, each segment being j / 3×3 in size, and then performs overlapping processing so that each segment has signal stacking. Finally, the three stacked signals are spliced ​​together to form a stacked signal of size 3j / 2×3, where j represents the length of the time-frequency domain fusion signal sequence. Then, the stacked signals undergo two dilated causal convolution operations. Dilated causal convolution is a fundamental operation in traditional TCN. Dilated convolution amplifies the receptive field by filling in holes "0" without increasing model computational cost. Causal convolution appropriately pads the input data to ensure that the convolution output value depends only on past and present inputs, not future information. Specifically, the input of the first dilated convolution is 3j / 2×3, and the output is 3j×3; the input of the first causal convolution is 3j×3, and the output is (3j+2)×3; the input of the second dilated convolution is (3j+2)×3, and the output is (6j+4)×3; the input of the second causal convolution is (6j+4)×3, and the output is (6j+6)×3. Furthermore, the TCN's input / output matrices both have 3 columns, representing time information, horizontal vibration signals, and vertical vibration signals, respectively.

[0066] As can be understood, this invention first segments and overlaps the signal, ensuring that each segment overlaps with other segments. These overlapping portions provide more contextual information, ensuring that important signal features are repeatedly captured across multiple segments. This increases the likelihood of retaining crucial information, enabling the model to better predict signal trends and further improving prediction accuracy. Furthermore, the two dilated causal convolution operations extract deeper features, further enhancing prediction accuracy.

[0067] Furthermore, traditional TCNs have fixed kernel sizes, resulting in fixed receptive field sizes. This limits their ability to extract features at a single scale, restricting their capacity to capture features at different scales. Therefore, optionally, the temporal convolutional network of this application includes three convolutional kernels of different sizes, distributed across three parallel temporal convolutional branches, to extract short, medium, and long-scale features respectively. Specifically, the TCN of this application includes three parallel TCN branches, using large, medium, and small convolutional kernels to extract long, medium, and short-scale features of the time series, enabling the model to more comprehensively capture data dependencies.

[0068] Preferably, the sizes of the three convolutional kernels are set to 2, 3, and 5, respectively. The formula for calculating the data length that the convolutional kernels can cover is: l = k +( k -1)×( d -1), l This represents the receptive field length of the convolution kernel, that is, the length of input data that the convolution kernel can cover. k This is the original size of the convolution kernel, that is, the width or height of the convolution kernel without dilation. d The dilation rate, or expansion rate, defines the spacing between elements in the convolution kernel. d The size is typically set to 4. Therefore, the receptive field can be calculated. l The parameters 5, 9, and 17 respectively fully consider the multi-scale characteristics of bearing vibration signals. Smaller kernel sizes (k=2, k=3) can effectively capture high-frequency transient features in the signal, while larger kernel sizes (k=5) are more suitable for extracting low-frequency trend information. By using dilated convolutions with an expansion rate of 4, while ensuring sufficient receptive fields (5, 9, 17) to cover complete signal features, the computational redundancy problem caused by traditional large convolution kernels is avoided. This parameter setting not only achieves a good balance between computational efficiency and feature extraction capability, but also significantly improves the predictive performance of the model.

[0069] In addition, to verify the effectiveness of the prediction method in this application, it was compared and verified with some existing prediction methods, using MAE, MSE, and RMSE as evaluation indicators. The comparison results are as follows: Figure 5 As shown. From Figure 5It can be seen that the prediction method of this application outperforms the existing eight algorithms in all three evaluation metrics. This is because this application adopts a more advanced architecture design, which improves the TPA algorithm to allocate proportional weights to the time-frequency signal, retains features that have a greater impact on the prediction of remaining lifetime, and eliminates redundant features. It also combines the TCN with multiple expansion and multi-core mechanism to deeply mine and extract features, which can effectively capture key information in the data and extract features from it. Therefore, the prediction method of this application has a smaller error value.

[0070] In addition, to verify the robustness of the prediction method in this application, noise robustness tests were also conducted, and the test results are as follows: Figure 6 and Figure 7 As shown, where, Figure 6 A comparison of prediction performance before and after adding 5dB of noise to the model. Figure 7 A comparison of the prediction performance before and after adding 10dB of noise to the model shows that the model's prediction results are very close to the actual data without noise, indicating that the model has high prediction accuracy in a noise-free environment. Moreover, the model's prediction results still maintain a high degree of consistency with the actual data even with 5dB and 10dB of noise added, indicating that the model has strong noise resistance.

[0071] In addition, such as Figure 8 As shown, another embodiment of the present invention also provides a life prediction system for bearings in an electricity meter production line, preferably employing the life prediction method for bearings in an electricity meter production line as described above, including:

[0072] The time-frequency domain signal processing module is used to collect vibration signal data of the bearings in the electricity meter production line and extract the time-domain and frequency-domain signals of the vibration signal data.

[0073] The time-frequency domain signal fusion module is used to extract time-domain feature signals and frequency-domain feature signals from the time-domain signal and the frequency-domain signal respectively using an improved TPA algorithm, and then fuse them to obtain a time-frequency domain fused signal. The improved TPA algorithm increases the weight of important features and decreases the weight of unimportant features.

[0074] The remaining service life prediction module is used to input the time-frequency domain fusion signal into the prediction model to predict the remaining service life of the bearings in the electricity meter production line.

[0075] It is understood that the life prediction system for bearings on the electricity meter production line in this embodiment extracts time-domain and frequency-domain features through an improved TPA algorithm. This increases the weight of important features and decreases the weight of unimportant features. The weight ratio can be reallocated according to the importance of features at each time node, realizing dynamic adjustment of the weight of time-frequency domain features. This retains features that have a greater impact on the prediction of remaining lifespan and eliminates redundant features, making the fused features more refined and representative. This reduces the subsequent computational load and minimizes the interference of unimportant features on the model's final decision. It focuses all attention on the prediction target, thereby improving the prediction accuracy and robustness of the prediction model.

[0076] In addition, the life prediction system for the bearings of the electricity meter production line also includes:

[0077] The multiple dilation operation module is used to perform multiple dilation operations on the time-frequency domain fused signal using a temporal convolutional network in order to extract deep-level features.

[0078] It is understood that each module of this system embodiment corresponds to each step of the above method embodiment, so the working principle of each module will not be repeated here, and you can refer to the steps of the above method embodiment.

[0079] In addition, another embodiment of the present invention provides an electronic device including a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the method described above by calling the computer program stored in the memory.

[0080] In addition, another embodiment of the present invention provides a computer-readable storage medium for storing a computer program for predicting the bearing life of an electricity meter production line, wherein the computer program executes the steps of the method described above when run on a computer.

[0081] Common computer-readable storage media include: floppy disks, flexible disks, hard disks, magnetic tapes, any other magnetic media, CD-ROMs, any other optical media, punch cards, paper tape, any other physical media with perforated patterns, random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), flash-erasable programmable read-only memory (FLASH-EPROM), any other memory chips or cartridges, or any other media readable by a computer. Instructions may further be transmitted or received by a transmission medium. The term transmission medium can include any tangible or intangible medium used to store, encode, or carry instructions for execution by a machine, and includes digital or analog carrier communication signals or intangible media that facilitate communication of such instructions. Transmission media include coaxial cables, copper wires, and optical fibers, which contain conductors for transmitting a bus of computer data signals.

[0082] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0083] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] 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 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] 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.

[0086] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0087] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

[0088] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the lifespan of bearings in an electricity meter production line, characterized in that, Includes the following: Vibration signal data of bearings on the electricity meter production line are collected, and the time-domain and frequency-domain signals of the vibration signal data are extracted. An improved TPA algorithm is used to extract time-domain feature signals and frequency-domain feature signals from the time-domain signal and the frequency-domain signal respectively, and then fuse them to obtain a time-frequency fused signal. The improved TPA algorithm increases the weight of important features and decreases the weight of unimportant features. The time-frequency domain fusion signal is input into the prediction model to predict the remaining service life of the bearings in the electricity meter production line. The process of extracting time-domain feature signals from time-domain signals using the improved TPA algorithm includes the following: The past information is convolved by a convolutional neural network to obtain a feature matrix, and the score of each row in the feature matrix is ​​calculated using a scoring function to obtain a score matrix. Compare all scores in the scoring matrix with preset first and second thresholds. If a score is greater than the first threshold, increase that score; if a score is less than the second threshold, decrease that score; if a score is greater than or equal to the second threshold and less than or equal to the first threshold, keep that score unchanged, and generate a new scoring matrix. Adjust the scores based on the following formula: ； in, Represents the first in the rating matrix x Each rating express Adjusted rating β This indicates that the parameters are being adjusted. X Indicates the first threshold. Y The second threshold is defined by sorting all the scores in the rating matrix from smallest to largest, taking the first quartile as the second threshold, and taking the third quartile as the first threshold. The new scoring matrix and the feature matrix are weighted and summed to obtain the feature vector. This feature vector is then concatenated with the current information to obtain the final feature vector. After obtaining the time-frequency domain fused signal, the following content is also included: A temporal convolutional network is used to perform multiple dilation operations on the time-frequency domain fused signal to extract deep features; The process of performing multiple dilation operations on the time-frequency domain fused signal using a temporal convolutional network includes the following: The temporal convolutional network first performs a first dilation operation on the time-frequency domain fused signal by signal overlap, and then extracts deep features through two dilation causal convolution operations. The temporal convolutional network includes three convolutional kernels of different sizes, distributed in three parallel temporal convolutional branches, to extract short, medium and long scale features respectively.

2. A life prediction system for bearings in an electricity meter production line, employing the life prediction method for bearings in an electricity meter production line as described in claim 1, characterized in that, include: The time-frequency domain signal processing module is used to collect vibration signal data of the bearings in the electricity meter production line and extract the time-domain and frequency-domain signals of the vibration signal data. The time-frequency domain signal fusion module is used to extract time-domain feature signals and frequency-domain feature signals from the time-domain signal and the frequency-domain signal respectively using an improved TPA algorithm, and then fuse them to obtain a time-frequency domain fused signal. The improved TPA algorithm increases the weight of important features and decreases the weight of unimportant features. The remaining service life prediction module is used to input the time-frequency domain fusion signal into the prediction model to predict the remaining service life of the bearings in the electricity meter production line.

3. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program, the processor executing the steps of the method as described in claim 1 by calling the computer program stored in the memory.

4. A computer-readable storage medium for storing a computer program for predicting the lifespan of bearings in an electricity meter production line, characterized in that, The computer program performs the steps of the method as described in claim 1 when it is run on a computer.

Images