Method and device for predicting service life of rolling bearing and computer readable storage medium
Through the Seq2Seq model combined with BIGRU and LSTM, a convolutional neural network and attention mechanism were introduced to extract time domain features to predict rolling bearing life, solving the problem of poor prediction effect in the existing technology and achieving more efficient life prediction.
Patent Information
- Application Number
- CN202280096745.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2022-07-29
- Publication Date
- 2025-01-17
AI Technical Summary
The prior art has poor long-term dependence, weak core feature extraction ability and weak generalization ability when predicting the service life of rolling bearings, and is unable to effectively process non-standardized mechanical signals and redundant data, resulting in poor prediction results.
The Seq2Seq model is adopted to combine bidirectional gating recursive unit (BIGRU) and long-term short-term memory model (LSTM), and introduce convolutional neural network and attention mechanism to predict the service life of rolling bearings by extracting time domain features. The model input and output lengths are flexible , enhances the interpretability and predictive performance of features.
Improves the accuracy and model performance of rolling bearing life prediction, reduces errors, and is more effective in processing non-standardized vibration signals and redundant data.
Smart Images
Figure CN119325556A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of artificial intelligence (AI), and in particular to a method and a device for predicting the service life of a rolling bearing and a computer-readable storage medium. Background Art
[0002] Rolling bearings are important transmission components of mechanical equipment, and their working conditions have a great impact on the equipment. Rolling bearing failures often reduce the reliability and accuracy of the equipment, which not only affects production and reduces the service life of the equipment, but also causes accidents. Therefore, predicting the service life of rolling bearings is of great significance.
[0003] In recent years, many scholars have devoted themselves to the study of service life prediction of rolling bearings. Existing methods include model-based physical methods, statistical methods, and data-based artificial intelligence methods.
[0004] However, the service status of rolling bearings changes over time, and data at previous times have a specific impact on the predicted results at later times. General model-based physical methods, statistical methods, and data-based artificial intelligence methods are not ideal in the processing of the timing characteristics of rolling bearings. Summary of the invention
[0005] Embodiments of the present invention provide a method and an apparatus for predicting the service life of a rolling bearing and a computer-readable storage medium.
[0006] In a first aspect, a method for predicting the service life of a rolling bearing is provided. The method comprises:
[0007] Acquire a first vibration signal of the rolling bearing;
[0008] Extracting a time domain feature of the first vibration signal, wherein the time domain feature indicates a degradation state of the rolling bearing;
[0009] Inputting the time-domain features into a trained Seq2Seq model, the Seq2Seq model comprising an encoder and a decoder, the encoder comprising a bidirectional gated recurrent unit (BIGRU), the decoder comprising a long short-term memory model (LSTM), the BIGRU being adapted to output a hidden state based on the time-domain features, the LSTM being adapted to predict the service life of the rolling bearing based on the hidden state; and
[0010] The service life is output.
[0011] Preferably, the encoder further comprises a convolutional neural network, the output of the convolutional neural network is related to the input of the BIGRU, and the convolutional neural network is suitable for performing feature compression on time domain features.
[0012] Preferably, outputting a hidden state based on the time domain feature comprises:
[0013] Obtain the forward hidden state through the forward time recurrent layer;
[0014] Obtain the reverse hidden state through the reverse time recurrent layer;
[0015] The forward hidden state and the reverse hidden state are concatenated to obtain the hidden state.
[0016] Preferably, the decoder further comprises an attention mechanism, the output of the attention mechanism is related to the input of the LSTM, and the attention mechanism is adapted to perform a weighted summation of the hidden states.
[0017] Preferably, it further includes: acquiring a second vibration signal of the rolling bearing; dividing the second vibration signal into a training set and a test set; training a Seq2Seq model based on the training set; and using the test set to test the Seq2Seq model trained based on the training set.
[0018] Preferably, training a Seq2Seq model based on a training set comprises:
[0019] The root mean square error (RMSE) is taken as the loss function of the training, where
[0020]
[0021] Where n is the number of training samples in the training set; i is the index of the training sample; is the predicted value of service life; y i is the actual value of the service life.
[0022] Preferably, using the test set to test the Seq2Seq model trained based on the training set includes:
[0023] The mean absolute error (MAE) is taken as the evaluation function of the test, where
[0024]
[0025] Where n is the number of test samples in the test set; i is the index of the test sample; is the predicted value of service life; y i is the actual value of the service life.
[0026] Preferably, the method further comprises: denoising the second vibration signal by discrete wavelet transform; and performing rolling segmentation on the denoised second vibration signal based on a window length of a preset sliding window.
[0027] Preferably, the time domain features include at least one of the following: mean value; variance; root mean square amplitude; root mean square value; maximum value; minimum value; waveform index; peak index; pulse index; marginal index; skewness; kurtosis.
[0028] In a second aspect, a device for predicting the service life of a rolling bearing is provided. The device comprises:
[0029] An acquisition module, which is configured to acquire a first vibration signal of a rolling bearing; an extraction module, which is configured to extract time domain features of the first vibration signal, wherein the time domain features represent a degradation state of the rolling bearing; an input module, which is configured to input the time domain features into a trained Seq2Seq model, wherein the Seq2Seq model includes an encoder and a decoder, the encoder includes a BIGRU, the decoder includes an LSTM, the BIGRU is suitable for outputting a hidden state based on the time domain features, and the LSTM is suitable for predicting the service life of the rolling bearing based on the hidden state; and an output module, which is configured to output the service life.
[0030] In a third aspect, an electronic device is provided, comprising a processor and a memory, wherein an application executable by the processor is stored in the memory for causing the processor to execute the method for predicting the service life of a rolling bearing according to any one of the above texts.
[0031] In a fourth aspect, a computer-readable medium is provided, comprising computer-readable instructions stored thereon, wherein the computer-readable instructions are used to execute the method for predicting the service life of a rolling bearing according to any one of the above.
[0032] In a fifth aspect, a computer program product comprises a computer program for executing the method for predicting the service life of a rolling bearing according to any one of the above when the computer program is executed by a processor.
[0033] The embodiment of the present invention proposes a novel Seq2Seq structure for bearing life prediction. The Seq2Seq network integrating artificial and convolutional feature extraction increases the interpretability of features. The length definition of the input and output of the model is flexible, and the model performance is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to make the technical solutions of the examples of the present disclosure clearer, only the drawings used to describe the examples will be introduced below. Obviously, the drawings to be described below are only some examples of the present disclosure. Those skilled in the art can obtain other drawings based on these drawings without creative work.
[0035] Figure 1is a flow chart of a method for predicting the service life of a rolling bearing according to an embodiment of the present invention.
[0036] Figure 2 4 is a schematic diagram of a process for predicting the service life of a rolling bearing according to an embodiment of the present invention.
[0037] Figure 3 It is a schematic diagram of the training process and the testing process of the service life prediction model of the rolling bearing according to an embodiment of the present invention.
[0038] Figure 4 is an exemplary line graph of loss values according to an embodiment of the present invention.
[0039] Figure 5 is an exemplary structural diagram of an apparatus for predicting the service life of a rolling bearing according to an embodiment of the present invention.
[0040] Figure 6 is a diagram showing an exemplary configuration of an electronic device according to an embodiment of the present invention.
[0041] List of reference numerals:
[0042]
[0043] DETAILED DESCRIPTION
[0044] In order to make the purpose, technical solutions and advantages of the present invention clearer, the following examples are given to further explain the present invention in detail.
[0045] In order to be concise and intuitive in the description, the scheme of the present invention is described below by describing several representative embodiments. Many details in the embodiments are only used to help understand the scheme of the present invention. However, it is obvious that the technical scheme of the present invention can be implemented without being limited to these details. In order to avoid unnecessarily obscuring the scheme of the present invention, some embodiments are not described in detail, but only a framework is given. In the following, "including" refers to "including but not limited to", and "according to..." refers to "at least according to..., but not limited to...". Due to the language habits of Chinese, when the number of elements is not specifically indicated below, it means that the elements can be one or more, or can be understood as at least one.
[0046] The applicant found through research that traditional feature extraction is designed to solve the problem of sequence regression. In the process of learning and training, there are many problems, such as poor long-term dependencies, weak core feature extraction capabilities, weak generalization capabilities, gradient disappearance, etc., which result in the inability of the residual life prediction model to achieve the expected diagnostic effect. Knowledge-based prediction methods can be based on expert system combination situations and rule reasoning. However, it is only suitable for high-demand qualitative analysis that requires the establishment of a knowledge base, and the maintenance cost of updating the knowledge base is high, which is difficult to popularize in large areas. Data-based prediction methods use mathematical theoretical knowledge to limit and improve the computing power of computers and data-based analysis capabilities, so that the management and maintenance of equipment become information-based. However, due to the low standardization of bearing vibration signals and excessive redundant data compared to core features, a single recurrent neural network cannot be well solved.
[0047] Considering that the signals in the mechanical field are non-standardized, there are a lot of redundant data, and there are some problems in feature extraction, such as low efficiency and inconvenient extraction. Traditional RNN or LSTM can only handle problems with fixed lengths of input and output, that is, one-to-one or many-to-many. The format of data input and output has certain restrictions, and the generalization is not strong enough.
[0048] The present disclosure provides a method for a Seq2Seq recurrent network that integrates artificial and convolutional feature extraction for life prediction. The Seq2seq recurrent network typically includes an encoder, a decoder, and a fixed-size state vector connecting the two. By learning the input, the encoder encodes the input into a fixed-size state vector, and then transmits the state vector to the decoder, which outputs the state vector by learning the state vector.
[0049] Figure 1 is a flow chart of a method for predicting the service life of a rolling bearing according to an embodiment of the present invention.
[0050] like Figure 1 As shown, the method includes:
[0051] Step 101: Acquire a first vibration signal of a rolling bearing.
[0052] Here, the vibration sensor may be used to obtain a first vibration signal of the rolling bearing. The first vibration signal is a signal used to predict the service life of the rolling bearing.
[0053] For example, a first vibration signal in analog format is collected by a piezoelectric acceleration sensor, and then a first vibration signal in digital format that can be recognized by a microcomputer is obtained by a charge amplifier and an A / D conversion circuit with a filter. Alternatively, the first vibration signal of the rolling bearing can be read from a control system of the motor, or the first vibration signal can be converted based on a current value or a torque value of a driving motor. The first vibration signal can be a time series signal.
[0054] Step 102: extracting time domain features of the first vibration signal, wherein the time domain features represent a degradation state of the rolling bearing.
[0055] Here, the time domain features are automatically or manually extracted. In the real-time dynamic detection process, the time domain features of the first vibration signal can directly reflect the degradation state of the bearing. In one embodiment, the time domain features include at least one of the following: mean value; variance; root mean square amplitude; root mean square value; maximum value; minimum value; waveform index; peak index; pulse index; marginal index; skewness; kurtosis.
[0056] Step 103: Input the time domain features into the trained Seq2Seq model, the Seq2Seq model includes an encoder and a decoder, the encoder includes a bidirectional gated recurrent unit (BIGRU), the decoder includes a long short-term memory model (LSTM), the BIGRU is suitable for outputting hidden states based on the time domain features, and the LSTM is suitable for predicting the service life of the rolling bearing based on the hidden states.
[0057] In one embodiment, outputting a hidden state based on time domain features includes: obtaining a forward hidden state through a forward time recurrent layer; obtaining a reverse hidden state through a reverse time recurrent layer; and concatenating the forward hidden state and the reverse hidden state to obtain a hidden state.
[0058] Gated recurrent unit (GRU) is an optimized type of RNN. It alleviates the vanishing gradient problem of RNN through a gating mechanism, so it can learn long-term dependencies in long sequences. However, unidirectional GRU only learns sequence information from front to back. In some problems, the output of the current time is not only related to the above sequence information, but also to the following sequence information. For example, when predicting the missing words in a sentence, we need to combine the above and the following at the same time. BIGRU is a neural network that can handle such problems. BIGRU consists of two Gru recurrent layers with opposite information transmission directions, where the first layer transmits information in chronological order (forward time recurrent layer) and the second layer transmits information in reverse chronological order (reverse time recurrent layer). The basic idea of BIGRU is: obtain the forward hidden state through the forward time recurrent layer, obtain the reverse hidden state through the reverse time recurrent layer, and then splice the forward hidden state and the reverse hidden state to obtain the final output hidden state of BIGRU.
[0059] In one embodiment, the encoder further includes a convolutional neural network, the output of which is related to the input of the BIGRU, and the convolutional neural network is suitable for performing feature compression on the time domain features. The compressed features are provided to the BIGRU. Therefore, an embodiment of the present invention proposes a novel Seq2Seq structure for bearing life prediction. The Seq2Seq network that integrates manual and convolutional feature extraction increases the interpretability of the features. The length definition of the input and output of the model is flexible, and the model performance is improved.
[0060] In one embodiment, the decoder includes an attention mechanism, the output of the attention mechanism is related to the input of the LSTM, and the attention mechanism is adapted to perform a weighted summation of hidden states. The weighted summation result of the hidden states is provided to the LSTM.
[0061] Based on the weighted sum of the hidden states, LSTM predicts the service life of the rolling bearing, preferably the residual service life. LSTM is a time recurrent neural network that is specifically designed to solve the long-term dependency problem of general RNN. LSTM is a neural network containing LSTM blocks or other blocks. In the literature or other materials, LSTM blocks can be described as intelligent network units because they can remember values of variable time lengths. The gates in the block can determine whether the input is important enough to be remembered and whether the output can be output.
[0062] Step 104: Output the service life.
[0063] In one embodiment, the method further comprises: obtaining a second vibration signal of the rolling bearing; dividing the second vibration signal into a training set and a test set; training a Seq2Seq model based on the training set; using the test set to test the Seq2Seq model trained based on the training set, wherein training the Seq2Seq model based on the training set comprises: taking RMSE as a loss function for training, wherein Where n is the number of training samples in the training set; i is the index of the training sample; is the predicted value of service life; y i Here, the second vibration signal is used to train and test the Seq2Seq model.
[0064] In one embodiment, using the test set to test the Seq2Seq model trained based on the training set includes: taking MAE as the evaluation function of the test, where Where n is the number of test samples in the test set; i is the index of the test sample; is the predicted value of service life; y i is the actual value of the service life.
[0065] In one embodiment, the method further comprises: denoising the second vibration signal by discrete wavelet transform; and performing rolling segmentation on the denoised second vibration signal based on a window length of a preset sliding window.
[0066] Figure 2 4 is a schematic diagram of a process for predicting the service life of a rolling bearing according to an embodiment of the present invention.
[0067] like Figure 2 As shown, a Seq2Seq recurrent network integrating artificial and convolutional feature extraction and attention mechanism is shown. The Seq2Seq model includes an encoder 22 and a decoder 29. The residual life prediction algorithm of the Seq2Seq recurrent network integrating artificial and convolutional feature extraction and attention mechanism specifically uses the following steps, and Figure 2 The implementation process is shown in , where the parameter 'h' represents the hidden layer features of BIGRU. For example, Figure 2 , as h1, h2, ha3, .... The parameter 'a' represents the attention weight of the attention mechanism. For example, Figure 2 Shown as a1, a2, a3, ... in the figure.
[0068] First, a vibration signal 20 of a rolling bearing is collected. Then, time domain features of the vibration signal 20 are extracted. For example, the following 12 time domain features 21 of the vibration signal 20 are extracted: mean value; variance; root mean square amplitude; root mean square value; maximum value; minimum value; waveform index; peak index; pulse index; marginal index; skewness; kurtosis.
[0069] Next, the corresponding time series data of the time domain features 21 is sent to the encoder 22 of the Seq2Seq model. The encoder 22 contains one or more convolutional neural networks 23. One or more convolutional neural networks 23 perform feature compression on the time domain features 21 to reduce the total length of the time series to an appropriate size to reduce the impact of gradient disappearance or explosion on the recurrent neural network. The output signal of the convolutional neural network 23 is transmitted to the BIGRU 24. BIGRU 24 outputs the hidden layer state based on the time domain features 21. BIGRU 24 sends the hidden layer state to the decoder 29. The attention mechanism 25 in the decoder 29 performs a weighted sum of the hidden layer states. The weighted sum of the hidden states is provided to the LSTM 27. LSTM 27 predicts the service life of the rolling bearing based on the weighted sum 26 of the hidden states. The output layer 28 outputs the service life.
[0070] Figure 3 Schematic diagram of the training process and testing process of the service life prediction model of the rolling bearing according to an embodiment of the present invention. The training process and the testing process include:
[0071] Step 40: Obtain the vibration signal of the bearing.
[0072] Step 41: Perform wavelet analysis and denoising on the vibration signal.
[0073] Step 42: Perform equal-interval sampling on the de-noised vibration signal.
[0074] Step 43: According to the preset window length of the sliding window, the vibration signals sampled at equal intervals are subjected to rolling segmentation to form a training set 44 and a test set 50 .
[0075] For the test set 44, a position 45 close to the endpoint is randomly selected and input to a Figure 2 In the Seq2Seq model 47 of the architecture shown in . Forward propagation 46 and prediction of the residual life 48 are performed in the Seq2Seq model 47. In addition, backward propagation 49 is performed in the Seq2Seq model 47, a loss function value is determined based on the difference between the predicted residual life 48 and the actual residual life, and the model parameters of the Seq2Seq model 47 are adjusted by the loss function value.
[0076] During the testing process, the test set 50 is input into the trained Seq2Seq model 47 and the predicted residual lifetime is output.
[0077] Generally, embodiments of the present invention have the following features:
[0078] (1) Calculation of characteristics of time series data based on the common time domain index of vibration signals.
[0079] Preferably, 12 items such as mean value, variance, square root amplitude, root mean square value, maximum value, minimum value, waveform index, peak index, impulse index, marginal index, skewness and kurtosis are selected as the time series data for calculation because the source data contains vibration signals divided into horizontal and vertical directions.
[0080] (2) Encoder and decoder of Seq2Seq recurrent network with manual and convolutional feature extraction and attention mechanism.
[0081] The main body of the network Seq2Seq model is based on an encoding-decoding structure. First, the encoding process is that the convolutional layer acts as a feature compressor to reduce the total length of the sequence to an appropriate size to reduce the impact of gradient vanishing or exploding in the recurrent neural network. Subsequently, the feature sequence is inspected by using a bidirectional gated recurrent unit network and the hidden state (HT) is output. Finally, the decoding process is that the LSTM network acts as a decoder to gradually predict the hidden layer feature values representing the health indicator (HI). In each step, the attention score obtained based on the presentation of the overall information provided by the encoder helps to find the most important information. The structural parameters of the Seq2Seq network that integrates manual and convolutional feature extraction and attention mechanism are shown in Table 1.
[0082] Table 1 Network structure parameters
[0083]
[0084] (3) Overall flow chart of residual life prediction algorithm
[0085] A Seq2Seq recurrent network that integrates manual and convolutional feature extraction and attention mechanism is proposed. The training and testing process is as follows Figure 3 shown.
[0086] (4) Model optimization and training
[0087] In the input layer, the feature data is standardized to accelerate the convergence of the model. In the encoding process, first, the convolution layer is used as a feature compressor to reduce the total length of the sequence to an appropriate size to reduce the impact of gradient vanishing or exploding in the recurrent neural network, and the weights and biases are adjusted through backpropagation to minimize the error between the output data and the expected data. Finally, the decoding process is that the LSTM network acts as a decoder to gradually predict the hidden layer feature values representing the health indicator (HI).
[0088] The purpose of training is to reduce the value of the constructed loss function. The training process takes the root mean square error (RMSE) as the loss function, that is, the mean square error (MSE) with a square root. The larger the error, the larger its value. Finally, the mean absolute error (MAE) is used as an index to measure the model. When the predicted value is exactly the same as the actual value, it is equal to 0, which is a perfect model. Similarly, the larger the error, the larger the value.
[0089] The two formulas are as follows:
[0090]
[0091] (5) Analysis of experimental results
[0092] In order to verify the superiority of the Seq2Seq recurrent network model (hereinafter referred to as TCA-Seq2Seq) of the present invention that integrates manual and convolutional feature extraction and attention mechanism, several comparative experiments were designed and tested on the phm2012 dataset. The results are shown in Table 2. In Table 2, A-GRU represents a gated recurrent unit model based on an attention mechanism. T-LSTM represents a model that uses a combination of temporal features and long-term and short-term memory units. Seq2seq represents the structure of the encoding and decoding structure, with BIGRU as the encoding part and LSTM as the decoding part. TC-Seq2Seq represents a feature extraction part based on a combination of the time domain features of the vibration signal and the convolutional compression features of the above-mentioned Seq2Seq, and TCA-Seq2Seq represents the attention calculation in the encoding and decoding process based on the above-mentioned TC-Seq2Seq.
[0093] The experimental results show that the error of the Seq2Seq model with LSTM decoding is reduced by 0.04 compared with the BIGRU model. The TC-Seq2Seq model, which adds the feature extraction part of the combination of the time domain features of the vibration signal and the convolution compression features, reduces the error by 0.05 compared with the Seq2Seq model. Compared with the TC-Seq2Seq model, the error of TCA-Seq2Seq with the attention mechanism is reduced by 0.04. Compared with A-GRU, the error of TCA-Seq2Seq is reduced by 0.07. Compared with T-LSTM, the error of TCA-Seq2Seq is reduced by 0.13. The comparison shows that the TCA-Seq2Seq algorithm designed in this subject presents good results in remaining life prediction, which reflects the advantages of the TCA-Seq2Seq algorithm designed in this subject in remaining life prediction.
[0094] Taking the mean absolute error (MAE) as the loss value index, the loss values of a specific training process, evaluation process, and test process can reach a good convergence level, such as Figure 4 shown. Figure 4 is an exemplary line graph of loss values according to an embodiment of the present invention. The horizontal axis represents time, and the vertical axis represents loss values. Figure 4 The training loss 60, validation loss 61 and test loss 62 are shown in FIG.
[0095] Table 2 Experimental comparison results
[0096] Model MAE TCA-Seq2Seq 0.0251 TC-Seq2Seq 0.0608 Seq2Seq 0.1149 BIGRU 0.1527 A-GRU 0.0938 T-LSTM 0.1579
[0097] Taking the phm2012 bearing data set as a validation, several different residual life prediction models of BIGRU, Seq2Seq, TCseq2seq and TCASeq2Seq are compared according to the mean absolute error index. The results show that the TCASeq2Seq model in the present invention improves the ability of feature extraction and regression fitting, and reflects the usability and superiority of this model.
[0098] Traditional RNN or LSTM can only handle problems where the input and output have fixed lengths, i.e., one-to-one or many-to-many, while Seq2Seq can handle one-to-many. It is also the most important variant of RNN:n vs m (the input and output sequence lengths are different). An attention mechanism is added to give a higher weight to the attention part to obtain more effective information.
[0099] By using the teacher forcing mechanism, the monitoring parameters are set: at each time during the training process, there is a certain probability of using the output of the previous time as input and a certain probability of using the correct target value as input, which can improve the scalability of the model. At the same time, if the probability is used to determine the input value during the iteration process during the training process, it can make the model uncertain during the training process and reduce the risk of poor performance during the evaluation process of the model.
[0100] Figure 5 is an exemplary structural diagram of a device for predicting the service life of a rolling bearing according to an embodiment of the present invention. The device 500 for predicting the service life of a rolling bearing comprises:
[0101] An acquisition module 501 is configured to acquire a first vibration signal of a rolling bearing;
[0102] An extraction module 502 is configured to extract a time domain feature of the first vibration signal, wherein the time domain feature indicates a degradation state of the rolling bearing;
[0103] An input module 503, which is configured to input the time domain features into a trained Seq2Seq model, wherein the Seq2Seq model includes an encoder and a decoder, the encoder includes a BIGRU, the decoder includes an LSTM, the BIGRU is suitable for outputting a hidden state based on the time domain features, and the LSTM is suitable for predicting the service life of a rolling bearing based on the hidden state; and an output module 504, which is configured to output the service life; and an output module 504, which is configured to output the service life.
[0104] An embodiment of the present invention further provides an electronic device having a processor-memory architecture and suitable for performing a prediction of the service life of a rolling bearing. Figure 6 is a structural diagram of an electronic device having a processor-memory architecture according to an embodiment of the present invention. Figure 6As shown, the electronic device 600 includes a processor 601, a memory 602, and a computer program stored in the memory 602 and running on the processor 601. When the computer program is executed by the processor 601, it implements the operation of any one of the methods for predicting the service life of a rolling bearing based on a Seq2Seq model. Among them, the memory 602 can be specifically implemented as various storage media, such as an electrically erasable programmable read-only memory (EEPROM), a flash memory (Flash memory), and a programmable program read-only memory (PROM). The processor 601 can be implemented to include one or more central processing units or one or more field programmable gate arrays, wherein the field programmable gate array integrates one or more central processing unit cores. Specifically, the central processing unit or the central processing unit core can be implemented as a CPU, an MCU, or a DSP, etc.
[0105] It should be noted that not all steps and modules in the above-mentioned process and structure diagram are necessary, and some steps or modules can be omitted according to actual needs. The execution order of each step is not fixed, and can be adjusted as needed. The division of each module is only for the convenience of describing the functional division. In the actual implementation scheme, the module can be implemented by multiple modules, and the functions of multiple modules can also be implemented by the same module. These modules can be located in the same device. It can also be in different devices.
[0106] The hardware modules in each embodiment can be implemented mechanically or electronically. For example, a hardware module may include a specially designed permanent circuit or logic device (e.g., a dedicated processor, such as an FPGA or ASIC) to perform a specific operation. A hardware module may also include a programmable logic device or circuit (e.g., including a general-purpose processor or other programmable processor) temporarily configured by software to perform a specific operation. The specific use of a mechanical method, or the use of a dedicated permanent circuit, or the use of a temporarily configured circuit (e.g., software configured) to implement a hardware module can be determined based on cost and time considerations.
[0107] The above is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for predicting the service life of a rolling bearing, comprising: Acquiring (101) a first vibration signal of a rolling bearing; extracting (102) a time domain feature of the first vibration signal, wherein the time domain feature indicates a degradation state of the rolling bearing; Inputting (103) the time-domain features into a trained Seq2Seq model, the Seq2Seq model comprising an encoder and a decoder, the encoder comprising a bidirectional gated recurrent unit (BIGRU), the decoder comprising a long short-term memory model (LSTM), the BIGRU being adapted to output a hidden state based on the time-domain features, the LSTM being adapted to predict the service life of the rolling bearing based on the hidden state; and The service life is output (104).
2. The method according to claim 1, wherein the encoder further comprises a convolutional neural network, the output of the convolutional neural network is related to the input of the BIGRU, and the convolutional neural network is suitable for performing feature compression on the time domain features.
3. The method according to claim 1, wherein outputting the hidden state based on the time domain feature comprises: Obtain the forward hidden state through the forward time recurrent layer; Obtain the reverse hidden state through the reverse time recurrent layer; The forward hidden state and the reverse hidden state are concatenated to obtain the hidden state.
4. The method of claim 1 , wherein the decoder further comprises an attention mechanism, the output of the attention mechanism being related to the input of the LSTM, and the attention mechanism being adapted to perform a weighted summation of the hidden states.
5. The method according to claim 1, further comprising: Acquiring a second vibration signal of the rolling bearing; Dividing the second vibration signal into a training set and a test set; Training the Seq2Seq model based on the training set; The test set is used to test the Seq2Seq model trained based on the training set.
6. The method of claim 5, wherein training the Seq2Seq model based on the training set comprises: Take RMSE as the loss function of the training, where Where n is the number of training samples in the training set; i is the index of the training sample; is the predicted value of service life; y i is the actual value of the service life.
7. The method of claim 5, wherein using the test set to test the Seq2Seq model trained based on the training set comprises: MAE is taken as the evaluation function of the test, where Where n is the number of test samples in the test set; i is the index of the test sample; is the predicted value of service life; y i is the actual value of the service life.
8. The method according to claim 5, further comprising: De-noising the second vibration signal using discrete wavelet transform; Rolling segmentation is performed on the de-noised second vibration signal based on a window length of a preset sliding window.
9. The method according to any one of claims 1 to 8, wherein the time domain features include at least one of the following: Mean value; variance; RMS amplitude; RMS value; maximum value; minimum value; waveform index; peak index; impulse index; marginal index; skewness; kurtosis.
10. A device for predicting the service life of a rolling bearing, comprising: An acquisition module (501), configured to acquire a first vibration signal of a rolling bearing; An extraction module (502) configured to extract a time domain feature of the first vibration signal, wherein the time domain feature indicates a degradation state of the rolling bearing; An input module (503) configured to input the time domain features into a trained Seq2Seq model, the Seq2Seq model comprising an encoder and a decoder, the encoder comprising a bidirectional gated recurrent unit (BIGRU), the decoder comprising a long short-term memory model (LSTM), the BIGRU being adapted to output a hidden state based on the time domain features, and the LSTM being adapted to predict the service life of the rolling bearing based on the hidden state; as well as An output module (504) is configured to output the service life.
11. An electronic device comprising a processor (601) and a memory (602), wherein an application program executable by the processor (601) is stored in the memory (602) for causing the processor (601) to execute a method for predicting the service life of a rolling bearing according to any one of claims 1 to 9. 12 . A computer-readable medium comprising computer-readable instructions stored thereon, wherein the computer-readable instructions are used to execute the method for predicting the service life of a rolling bearing according to any one of claims 1 to 9.
13. A computer program product comprising a computer program, wherein the computer program is executed by a processor for performing the method for predicting the service life of a rolling bearing according to any one of claims 1 to 9.