Rotating Machinery Life Prediction Method and System Based on Interactive Learning Convolutional Network

Abstract

The present invention discloses a method and system for predicting the remaining useful life of rotating machinery based on an interactive learning convolutional network, which relates to the field of predicting the remaining useful life of rotating machinery. The method includes obtaining vibration signals of rotating machinery and extracting shallow features in the time domain and time-frequency domain; constructing an ILCANet model framework; dividing the shallow features into a training sample set and a test sample set, and pasting labels to the training sample set to form a training label set; inputting the training sample set and the training label set into the ILCANet model for training, and after the training is completed, inputting the test sample set into the ILCANet model to obtain a prediction result; comparing the difference between the prediction result and the true value to verify the accuracy and adaptability of the model. The present invention extracts different eigenvalue of vibration signals and assigns weights through a feature attention mechanism, improving the prediction accuracy. Through odd-even segmentation and interactive learning, the problem of insufficient receptive field of the convolutional network for long sequence data is improved.

Description

Technical Field

[0001] The present invention relates to the field of rotating machinery life prediction, in particular to a rotating machinery life prediction method and system based on an interactive learning convolutional network. Background Art

[0002] With the development of industry towards modernization, complex and sophisticated mechanical equipment is increasing day by day. The development of deep learning technology has improved the prediction performance of the remaining useful life (RUL) of mechanical rotating parts. Traditional methods for predicting the remaining useful life of rolling bearings generally assume that a single vibration feature data is used to predict the life through a recurrent neural network. However, in many practical applications, this assumption does not hold. For example, the complexity of the working environment results in the fact that a single feature cannot reflect the complex degradation process of the bearing, and the recurrent neural network cannot perform parallel operations, making it difficult to predict the remaining useful life of the bearing in a timely manner.

[0003] Existing methods either adopt a recurrent neural network or use its variant network, and these methods do not consider the time cost of prediction. The present invention intends to study a method for predicting the remaining useful life of bearings relying on a convolutional neural network to solve the problems of long training time, easy occurrence of gradient explosion and gradient disappearance of the recurrent neural network, thereby improving the remaining useful life prediction system.

[0004] In view of the fact that the original sampling data of the rotating machinery vibration signal is too single, this paper attempts to extract the shallow features in the time domain and time-frequency domain of the original vibration signal to solve this problem, and this method can fully express the degradation information of the bearing. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the problems to be solved by the present invention are as follows: The existing technical solutions do not consider the time cost of prediction, and the original sampling data is too single.

[0007] To solve the above technical problems, the present invention provides the following technical solution: A rotating machinery life prediction method based on an interactive learning convolutional network, which includes obtaining a rotating machinery vibration signal, and extracting the shallow features in the time domain and time-frequency domain of the signal; constructing an ILCANet model framework; dividing the shallow features into a training sample set and a test sample set, pasting labels to the training sample set to form a training label set, and the label pasting rule is that a group of vibration feature data corresponds to a type of life label; inputting the training sample set and the training label set into the ILCANet model for training, and after the training is completed, inputting the test sample set into the ILCANet model to obtain a prediction result; comparing the difference between the prediction result and the true value to verify the model accuracy and adaptability.

[0008] As a preferred embodiment of the method for predicting the remaining useful life of rotating machinery based on the interactive learning convolutional network according to the present invention, wherein: the ILCANet model framework includes a one-dimensional convolutional module, an interactive learning module, a feature attention module, and a fully connected layer decoding module; the one-dimensional convolutional module consists of a replication padding layer, two one-dimensional convolutional layers, a LeakyRelu activation layer, a Tanh activation layer, and a Dropout layer, and its specific function is to map the input samples into representations with the same dimension as the input.

[0009] As a preferred embodiment of the method for predicting the remaining useful life of rotating machinery based on the interactive learning convolutional network according to the present invention, wherein: the interactive learning module includes performing dot product addition and subtraction operations on the data after odd-even splitting and the data processed by the one-dimensional convolutional module, which is expressed as

[0010]

[0011] wherein, I odd represents the odd data, I even represents the even data, ⊙ represents the Hadamard product, exp represents the exponential function, c() represents the data processed by the one-dimensional convolutional module, I o ′ dd represents the final output odd data sequence, I e ′ ven represents the final output even data sequence.

[0012] As a preferred embodiment of the method for predicting the remaining useful life of rotating machinery based on the interactive learning convolutional network according to the present invention, wherein: the feature attention module includes receiving the data sequence output by the interactive learning module as the input x of the feature attention module, and using global average pooling GAP to focus on the global view, which is expressed as

[0013]

[0014] wherein, represents the GAP output of the c-th channel, j represents the j-th group of data sequences, α represents the total number of sequences, Mean represents the average operation; through the 1×16 convolutional mapping function F1, the dependence relationship between x GAP and the input x is encoded to obtain the feature matrix f, which is expressed as

[0015] f = Tanh(F1[cat(x, x GAP )])

[0016] Among them, cat represents the merged data; f is decomposed into x' and others, and only x' is retained. The feature matrix x' contains the original input feature information and the information of global average pooling. x' is transformed into a variable with the same number of features as x through another convolutional mapping function F2, which is expressed as

[0017] g = Sigmoid[F2(f x′ )]

[0018] The fully connected layer decoding module is a decoder that maps the sequence to an RUL label.

[0019] As a preferred solution of the rotary machinery life prediction method based on the interactive learning convolutional network of the present invention, among them: the training includes, when training the ILCANet model, aiming to minimize the loss function MSELoss, and using the adaptive moment estimation method to adaptively adjust the learning rate of each parameter;

[0020] The loss function MSELoss is expressed as

[0021]

[0022] Among them, y i represents the data of the i-th group of input sample sets, represents the average value of the sample set data, and N represents the total number of groups of data in the sample set.

[0023] As a preferred solution of the rotary machinery life prediction method based on the interactive learning convolutional network of the present invention, among them: the adjustment of the learning rate of each parameter is expressed as

[0024]

[0025] m t =β1m t-1 +(1 - β1)g t

[0026]

[0027] Among them, g t represents the gradient, represents the loss function, m t represents the first moment estimation of the gradient in the momentum form at time t, v t represents the second moment estimation of the gradient in the momentum form at time t, and respectively represent the first and second moment estimations after bias correction, β1 and β2 represent the smoothing constants, θ t represents the parameter at time t, and η represents the learning rate.

[0028] As a preferred solution of the method for predicting the remaining useful life of a rotating machine based on an interactive learning convolutional network according to the present invention, wherein: verifying the model accuracy includes evaluating the prediction results and the true values using the mean absolute error (MAE), root mean square error (RMSE), and scoring function (SF), which is expressed as

[0029]

[0030]

[0031] wherein, er t represents the difference between the predicted value and the true value at time t, n represents the total number of time steps, m represents the number of early time steps, m = n / 2, and ω1 and ω2 respectively represent the weight ratios for early and late bearing predictions.

[0032] Another object of the present invention is to provide a remaining useful life prediction system for a rotating machine based on an interactive learning convolutional network, which can evaluate the remaining useful life of the rotating machine.

[0033] To solve the above technical problems, the present invention provides the following technical solution: A system for a method for predicting the remaining useful life of a rotating machine based on an interactive learning convolutional network, including: an acquisition module, a model construction module, a data classification module, a training module, and a verification module; the acquisition module is used to obtain the vibration signals of the rotating machine and extract the shallow features in the time domain and time-frequency domain of the signals; the model construction module is used to construct an ILCANet model framework, which also includes a one-dimensional convolutional module, an interactive learning module, a feature attention module, and a fully connected layer decoding module; the data classification module is used to divide the shallow features into a training sample set and a test sample set, paste labels on the training sample set to form a training label set, and the label pasting rule is that a set of vibration feature data corresponds to a type of remaining useful life label; the training module is used to input the training sample set and the training label set into the ILCANet model for training, and after training, input the test sample set into the ILCANet model to obtain the prediction results; the verification module is used to compare the differences between the prediction results and the true values to verify the model accuracy and adaptability.

[0034] A computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the method for predicting the remaining useful life of a rotating machine based on an interactive learning convolutional network as described above are implemented.

[0035] A computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the method for predicting the remaining useful life of a rotating machine based on an interactive learning convolutional network as described above are implemented.

[0036] The beneficial effects of the present invention are as follows: Different characteristic values of vibration signals are extracted by the present invention, and weights are assigned through a characteristic attention mechanism, improving the prediction accuracy.

[0037] Through odd-even segmentation and interactive learning, the problem of insufficient receptive fields of convolutional networks for long-sequence data is improved, significantly enhancing the degree of feature extraction and the lifespan prediction performance during the lifespan prediction process. At the same time, the robustness of lifespan prediction is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. Among them:

[0039] Figure 1 It is a flowchart of the rotational machinery lifespan prediction method based on the interactive learning convolutional network in Embodiment 1.

[0040] Figure 2 It is a structural diagram of the ILCANet of the rotational machinery lifespan prediction method based on the interactive learning convolutional network in Embodiment 1.

[0041] Figure 3 It is a diagram of the one-dimensional convolutional module of the rotational machinery lifespan prediction method based on the interactive learning convolutional network in Embodiment 1.

[0042] Figure 4 It is a diagram of the interactive learning module of the rotational machinery lifespan prediction method based on the interactive learning convolutional network in Embodiment 1.

[0043] Figure 5 It is a diagram of the characteristic attention mechanism module of the rotational machinery lifespan prediction method based on the interactive learning convolutional network in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following will provide a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification.

[0045] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0046] Embodiment 1

[0047] Refer to Figures 1 - 5, which is the first embodiment of the present invention. The method for predicting the remaining useful life of a rotating machine based on an interactive learning convolutional network provided by this embodiment includes, as Figure 1 shown:

[0048] Step 1: Obtain the vibration signal of the rotating machine, and extract the shallow features in the time domain and time-frequency domain of the signal.

[0049] Step 2: As Figure 2 shown, construct the ILCANet model framework. The ILCANet model framework includes a one-dimensional convolutional module, an interactive learning module, a feature attention module, and a fully connected layer decoding module.

[0050] As Figure 3 shown: The one-dimensional convolutional module consists of a replication padding layer, two one-dimensional convolutional layers, a LeakyRelu activation layer, a Tanh activation layer, and a Dropout layer. Its specific function is to map the input samples into representations with the same dimension as the input, and the parameter is θ c .

[0051] As Figure 4 shown, the interactive learning module performs dot product addition and subtraction operations on the data after odd-even segmentation and the data processed by the one-dimensional convolutional module, expressed as,

[0052]

[0053] where, I odd represents the odd data, I even represents the even data, ⊙ represents the Hadamard product, exp represents the exponential function, c() represents the data processed by the one-dimensional convolutional module, I o ′ dd represents the final output odd data sequence, and I e ′ ven represents the final output even data sequence.

[0054] As Figure 5 shown: The parameter of the feature attention module is θ A . The input is the output recombination sequence of the multi-layer interactive learning module, and the output is the weighted sequence data. Since there are too many data features to determine the contribution degree of each feature to the prediction of RUL, the attention mechanism is used to capture the dependence relationship between data, pay more attention to important features, and prevent interference from useless information.

[0055] In it, global average pooling (GAP) is adopted to focus on the global view. Because the remaining useful life prediction has a global receptive field and GAP does not lose position information, GAP is more suitable than global max pooling (GMP). The GAP calculation formula for the c-th channel is expressed as,

[0056]

[0057] Among them, represents the GAP output of the c-th channel, j represents the j-th group of data sequences, α represents the total number of sequences, and Mean represents the average operation.

[0058] In addition, in order to find the position information, through the 1×16 convolutional mapping function F1, for x GAP and the dependency of the input x are encoded to obtain the feature matrix f, which is expressed as

[0059] f = Tanh(F1[cat(x, x GAP )])

[0060] Among them, cat represents merging data; then, f is decomposed into x' and others, and only x' is retained. The feature matrix x' contains the original input feature information and the information of global average pooling. x' is transformed into a variable with the same number of features as x through another convolutional mapping function F2, which is expressed as

[0061] g = Sigmoid[F2(f x′ )]

[0062] The parameters of the fully connected layer decoding module are θ F . The fully connected layer is the decoder, which maps the sequence to the RUL label.

[0063] Step 3: Divide the shallow features into a training sample set and a test sample set, and paste labels on the training sample set to form a training label set. The label pasting rule is that a group of vibration feature data corresponds to a type of life label.

[0064] Step 4: Input the training sample set and the training label set into the ILCANet model for training. After training, input the test sample set into the ILCANet model to obtain the prediction result.

[0065] When training the ILCANet model, the initialization strategy of the network model training is very important for constructing the deep architecture. To ensure the quality of the learned network parameters, the goal is to minimize the loss function MSELoss, and an adaptive moment estimation method is used to adaptively adjust the learning rate of each parameter.

[0066] The loss function MSELoss is expressed as

[0067]

[0068] Among them, y i represents the data of the i-th group of input sample sets, It is denoted as the average value of the sample set data, and N is denoted as the total number of data groups in the sample set.

[0069] The learning rate for adjusting each parameter is denoted as

[0070]

[0071] m t =β1m t-1 +(1 - β1)g t

[0072]

[0073] where g t is denoted as the gradient, is denoted as the loss function expression which is consistent with the above-mentioned mean squared error loss function MSELoss, m t is denoted as the first moment estimate of the gradient in momentum form at time t, v t is denoted as the second moment estimate of the gradient in momentum form at time t, and respectively denote the first and second moment estimates after bias correction, θ t is denoted as the parameter at time t, η is denoted as the learning rate, β1 and β2 are denoted as smoothing constants also known as decay coefficients, and in machine learning, the default values are usually β1 = 0.9 and β2 = 0.999, and the value range is (0, 1). Their values mainly depend on the performance of the time series data in the dataset. If the time series is relatively stable, the smoothing coefficient should be taken smaller to reduce the correction amplitude and make the prediction model contain information of a longer time series; if the time series has a rapid and obvious tendency to change, the smoothing coefficient should be taken larger to improve the sensitivity of the prediction model. It has been proved by experiments that in most cases, β1 = 0.9 and β2 = 0.999 have the best optimization effect.

[0074] Step 5: Compare the difference between the predicted result and the true value to verify the model accuracy and adaptability.

[0075] The mean absolute error MAE, root mean square error RMSE, and scoring function SF are used to evaluate the predicted result and the true value, denoted as

[0076]

[0077] where er t is denoted as the difference between the predicted value and the true value at time t, n is denoted as the total number of time steps, m is denoted as the number of early time steps, m = n / 2, and ω1 and ω2 respectively denote the weight ratios for early and late bearing predictions.

[0078] For bearing degradation, late prediction is more important than early prediction. Therefore, the analysis of a large amount of experimental data is set with ω1 = 0.35 and ω2 = 0.65.

[0079] Example 2

[0080] In the second embodiment of the present invention, which is different from the first embodiment: the rotating machinery life prediction method based on the interactive learning convolutional network further includes the bearing remaining service life prediction method based on the multi-feature attention mechanism proposed in this example, and experiments are carried out on the phm2012 data set provided by the FEMTO-ST Institute. Through the method of the present invention: the experimental results comparison and analysis of the bearing remaining service life prediction method based on the multi-feature attention mechanism of the interactive learning convolutional network with TCN-SA, TCN-RSA, and TCN-RSCB show that the present invention has a great improvement in MAE, RMSE, and SF. It should be noted that the above life prediction method is only illustrated by experiments using the above benchmark data set. In actual applications, the above method can be applied to different scenarios for experimental analysis according to needs.

[0081] In this example, the database used in the experiment adopts the phm2012 rolling bearing full life test database collected by the FEMTO-ST Research through the PRONOSTIA test bench. The bearing degradation data mainly consists of two parts, namely vibration data and temperature data. The vibration sensor consists of two micro accelerometers positioned at 90° to each other, one placed on the vertical axis and the other placed on the horizontal axis. The two accelerometers are placed radially on the outer ring of the bearing, and the sampling frequency is 25.6KHz. The temperature sensor is a resistance temperature detector placed in a hole near the outer bearing ring, and the sampling frequency is 0.1Hz. The data set contains the full life data of different bearings under three working conditions, as shown in Table 1. Bearing is the bearing.

[0082] Table 1 Details of the data set

[0083]

[0084] Since the vertical signal is less sensitive to bearing degradation, the horizontal vibration signals under the first two working conditions are selected for experimental research in the experiment of the present invention. The detailed description of the data set is shown in Table 1. Each bearing data in Table 1 is collected once every 10 seconds at a sampling frequency of 25.6KHz for 0.1 second, that is, each sample contains 2560 data points.

[0085] To test the life prediction ability of the present invention, the training set and the test set are divided by the leave-one-out method for cross-validation. The model proposed by the present invention is compared with TCN-SA, TCN-RSA, and TCN-RSCB in a comparative experiment. The experimental results of the three evaluation index values finally obtained are shown in Tables 2, 3, and 4.

[0086] Table 2 MAE of the prediction results of different models

[0087]

[0088]

[0089] Table 3 RMSE of the prediction results of different models

[0090]

[0091] Table 4 SF of the prediction results of different models

[0092]

[0093]

[0094] It can be seen from the prediction results that our method is superior to other methods for most bearings.

[0095] Embodiment 3

[0096] In the third embodiment of the present invention, which is different from the previous two embodiments, a system for a rotary machinery life prediction method based on an interactive learning convolutional network includes an acquisition module, a model construction module, a data classification module, a training module, and a verification module; the acquisition module is used to obtain the vibration signals of the rotary machinery and extract the shallow features in the time domain and the time-frequency domain of the signals; the model construction module is used to construct the ILCANet model framework, which also includes a one-dimensional convolutional module, an interactive learning module, a feature attention module, and a fully connected layer decoding module; the data classification module is used to divide the shallow features into a training sample set and a test sample set, paste labels on the training sample set to form a training label set, and the label pasting rule is that a group of vibration feature data corresponds to a type of life label; the training module is used to input the training sample set and the training label set into the ILCANet model for training, and after training, input the test sample set into the ILCANet model to obtain the prediction results; the verification module is used to compare the difference between the prediction results and the true values to verify the model accuracy and adaptability.

[0097] If the described functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, etc., which can store program codes of various kinds.

[0098] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0099] More specific examples (nonexhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), optical fiber devices, and portable compact disc read-only memories (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as necessary, and then storing it in a computer memory.

[0100] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting the remaining useful life of rotating machinery based on an interactive learning convolutional network, characterized in that: including Obtain the vibration signals of rotating machinery, and extract the shallow features in the time domain and time-frequency domain of the signals; Construct an ILCANet model framework, which includes a one-dimensional convolutional module, an interactive learning module, a feature attention module, and a fully connected layer decoding module; The one-dimensional convolutional module consists of a replication padding layer, two one-dimensional convolutional layers, a LeakyRelu activation layer, a Tanh activation layer, and a Dropout layer. Its specific function is to map the input samples into representations with the same dimension as the input; The feature attention module includes receiving the data sequence output by the interactive learning module as the input x of the feature attention module, and using global average pooling GAP to focus on the global view, which is expressed as Among them, is expressed as the GAP output of the c-th channel, j is expressed as the j-th group of data sequences, α is expressed as the total number of sequences, and Mean is expressed as the mean operation; Encode the dependency relationship between x and the input x through the 1×16 convolutional mapping function F1 to obtain the feature matrix f, expressed as GAP ​ f = Tanh(F1[cat(x, x GAP )]) where cat represents merging data; Decompose f into x' and others, only keep x'. The feature matrix x' contains the original input feature information and the information of global average pooling. Transform x' into a variable with the same number of features as x through another convolutional mapping function F2, which is expressed as g = Sigmoid[F2(f x′ )] The fully connected layer decoding module is a decoder that maps the sequence to an RUL label; Divide the shallow features into a training sample set and a test sample set, and paste labels on the training sample set to form a training label set. The label pasting rule is that a set of vibration feature data corresponds to a type of life label; Input the training sample set and the training label set into the ILCANet model for training. After training, input the test sample set into the ILCANet model to obtain the prediction results; Compare the differences between the prediction results and the true values to verify the model accuracy and adaptability.

2. The method for predicting the service life of a rotating machine based on an interactive learning convolutional network according to claim 1, wherein: The interactive learning module includes performing dot product addition and subtraction operations on the data after odd-even segmentation and the data processed by the one-dimensional convolutional module, which is expressed as Among them, I odd represents odd data, I even represents even data, ⊙ represents the Hadamard product, exp represents the exponential function, c() represents the data after being processed by the one-dimensional convolution module, I o ′ dd represents the odd data sequence of the final output, I e ′ ven represents the even data sequence of the final output.

3. The method for predicting the service life of a rotating machine based on an interactive learning convolutional network according to claim 2, characterized in that: The training includes, when training the ILCANet model, aiming to minimize the loss function MSELoss, and using the adaptive moment estimation method to adaptively adjust the learning rate of each parameter; The loss function MSELoss is expressed as Among them, y i represents the data of the i-th group of input sample sets, represents the average value of the sample set data, and N represents the total number of groups of sample set data.

4. The method for predicting the service life of a rotating machine based on an interactive learning convolutional network according to claim 3, characterized in that: The adjustment of the learning rate of each parameter is expressed as m t = β1m t-1 + (1 - β1)g t v t = β2v t-1 + (1 - β2)g t 2 Among them, g t is represented as the gradient, is represented as the loss function, m t is represented as the first moment estimate of the gradient in the momentum form at time t, v t is represented as the second moment estimate of the gradient in the momentum form at time t, and are respectively represented as the first and second moment estimates after bias correction, β1 and β2 are represented as smoothing constants, θ t is represented as the parameter at time t, and η is represented as the learning rate.

5. The method for predicting the service life of a rotating machine based on an interactive learning convolutional network according to claim 4, characterized in that: The verification of the model accuracy includes using the mean absolute error MAE, root mean square error RMSE, and scoring function SF to evaluate the prediction results and the true values, which is expressed as where er t represents the difference between the predicted value and the true value at time t, n represents the total number of time steps, m represents the number of early time steps, m = n / 2, and ω1 and ω2 respectively represent the weight ratios for early and late bearing predictions.

6. A system adopting the rotating machinery life prediction method based on an interactive learning convolutional network according to any one of claims 1 to 5, characterized in that: including an acquisition module, a model construction module, a data classification module, a training module, and a verification module; The acquisition module is used to obtain the vibration signals of rotating machinery and extract the shallow features in the time domain and time-frequency domain of the signals; The model construction module is used to construct an ILCANet model framework, which also includes a one-dimensional convolutional module, an interactive learning module, a feature attention module, and a fully connected layer decoding module; The data classification module is used to divide the shallow features into a training sample set and a test sample set, and paste labels on the training sample set to form a training label set. The label pasting rule is that a set of vibration feature data corresponds to a type of life label; The training module is used to input the training sample set and the training label set into the ILCANet model for training. After training, input the test sample set into the ILCANet model to obtain the prediction results; The verification module is used to compare the differences between the prediction results and the true values to verify the model accuracy and adaptability.

7. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for predicting the remaining useful life of a rotating machine based on an interactive learning convolutional network according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for predicting the remaining useful life of a rotating machine based on an interactive learning convolutional network according to any one of claims 1 to 5.

Images