Centrifugal pump fault diagnosis method and device, electronic equipment and storage medium

Abstract

The invention provides a centrifugal pump fault diagnosis method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the feature extraction of a vibration signal of a to-be-diagnosed centrifugal pump, and obtaining a vibration feature; the vibration characteristics comprise time domain characteristics, frequency domain characteristics and time-frequency domain characteristics; inputting the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model; wherein the fault diagnosis model is obtained through training based on sample vibration characteristics and fault labels corresponding to the sample vibration characteristics, the fault diagnosis model is obtained through construction based on a bidirectional long short-term memory network BiLSTM, and weight vectors in a forward propagation layer and a reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalar quantities. According to the centrifugal pump fault diagnosis method and device, the electronic equipment and the storage medium, the prediction precision of the fault diagnosis model can be improved, and then the fault diagnosis result can be accurately obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment fault diagnosis, and in particular to a centrifugal pump fault diagnosis method, device, electronic equipment and storage medium. Background Art

[0002] Centrifugal pumps are widely used in petrochemical, metallurgical, mechanical, electric power and other industries to transport liquids or lift hydraulic pressure. Due to the complex structure of centrifugal pumps, the relatively harsh working environment, and the influence of various random factors, the working system composed of motors, speed change gear boxes, centrifugal pump bodies, etc. is prone to various faults. When the fault is serious, it will cause the production system to be paralyzed and safety accidents to occur. Therefore, conducting fault diagnosis research on centrifugal pumps is of great significance to ensure the safe, efficient and reliable operation of mechanical equipment.

[0003] At present, common centrifugal pump fault diagnosis methods include mechanism diagnosis methods (such as noise detection diagnosis method, vibration detection method, etc.), which mainly focus on extracting effective feature parameters through numerical calculation or signal processing technology, mainly including time-frequency domain analysis methods. However, due to the harsh working conditions, the collected fault vibration signals often have strong noise, nonlinearity and non-stationarity. At this time, traditional diagnostic methods cannot effectively extract fault features and classify them, which leads to low fault diagnosis accuracy and difficulty in meeting diagnostic needs under different working conditions. Summary of the invention

[0004] The present invention provides a centrifugal pump fault diagnosis method, device, electronic equipment and storage medium, which are used to solve the defect of low fault diagnosis accuracy in the prior art.

[0005] The present invention provides a centrifugal pump fault diagnosis method, comprising:

[0006] Extracting features of the vibration signal of the centrifugal pump to be diagnosed to obtain vibration features; the vibration features include time domain features, frequency domain features and time-frequency domain features;

[0007] Inputting the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model;

[0008] Among them, the fault diagnosis model is trained based on sample vibration characteristics and fault labels corresponding to the sample vibration characteristics, the fault diagnosis model is constructed based on a bidirectional long short-term memory network BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

[0009] According to a centrifugal pump fault diagnosis method provided by the present invention, the inputting of the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model comprises:

[0010] Inputting the vibration feature into the forward propagation layer to obtain a first hidden state feature output by the forward propagation layer;

[0011] Inputting the vibration feature into the back propagation layer to obtain a second hidden state feature output by the back propagation layer;

[0012] The first hidden state feature and the second hidden state feature are used to form a hidden layer of the fault diagnosis model, thereby obtaining a concatenated feature output by the hidden layer;

[0013] The splicing features are input into the classification layer of the fault diagnosis model to obtain the fault diagnosis result output by the classification layer.

[0014] According to a centrifugal pump fault diagnosis method provided by the present invention, the loss value of the fault diagnosis model is:

[0015] FL=-α t (1-p t ) γ log(p t );

[0016]

[0017] Among them, FL represents the loss value of the fault diagnosis model, α t represents the shared weight used to control the loss value of the positive sample vibration feature and the negative sample vibration feature, γ represents the modulation coefficient, y represents the accuracy of the fault prediction result obtained based on the sample vibration feature, and p represents the probability corresponding to the fault prediction result.

[0018] According to a centrifugal pump fault diagnosis method provided by the present invention, the fault diagnosis model solves the minimum value of the loss value based on the first-order exponential smoothing value of the historical gradient, the first-order exponential smoothing value of the historical gradient square and the updated value of the variable;

[0019] The first-order exponential smoothing value of the historical gradient is determined based on the following formula:

[0020] m t =beta 1 ×m t-1 +(1-beta 1 )×g;

[0021] The first-order exponential smoothing value of the historical gradient square is determined based on the following formula:

[0022] v t =beta 2 ×v t-1 +(1-beta 2)×g×g;

[0023] The updated value of the variable is determined based on the following formula:

[0024]

[0025] Among them, m t represents the first-order exponential smoothing value of the historical gradient, beta 1 represents the first-order moment attenuation coefficient, g represents the gradient, v t represents the first-order exponential smoothing value of the square of the historical gradient, beta 2 represents the second-order moment attenuation coefficient, x represents the updated value of the variable, and ε is a constant.

[0026] According to a centrifugal pump fault diagnosis method provided by the present invention, the parameter vector is updated based on the following formula:

[0027]

[0028] The parameter scalar is updated based on the following formula:

[0029]

[0030] Among them, v represents the parameter vector, g represents the parameter scalar, and w represents the weight vector.

[0031] According to a centrifugal pump fault diagnosis method provided by the present invention, the forward propagation layer and the reverse propagation layer are sequentially provided with an input gate, a first update gate, a forget gate, a second update gate and an output gate.

[0032] According to a centrifugal pump fault diagnosis method provided by the present invention, the forward propagation layer and the reverse propagation layer are sequentially provided with an input gate, a first update gate, a forget gate, a second update gate and an output gate.

[0033] The present invention also provides a centrifugal pump fault diagnosis device, comprising:

[0034] An extraction unit is used to extract features from the vibration signal of the centrifugal pump to be diagnosed, and obtain vibration features; the vibration features include time domain features, frequency domain features, and time-frequency domain features;

[0035] A diagnosis unit, used for inputting the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model;

[0036] Among them, the fault diagnosis model is trained based on sample vibration characteristics and fault labels corresponding to the sample vibration characteristics, the fault diagnosis model is constructed based on a bidirectional long short-term memory network BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

[0037] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any one of the centrifugal pump fault diagnosis methods described above is implemented.

[0038] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the centrifugal pump fault diagnosis method as described in any one of the above is implemented.

[0039] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the centrifugal pump fault diagnosis method as described above is implemented.

[0040] The centrifugal pump fault diagnosis method, device, electronic device and storage medium provided by the present invention have a fault diagnosis model based on a bidirectional long short-term memory network BiLSTM, so that the sample vibration characteristics can transmit information in positive order and reverse order respectively, so that the information between historical characteristics and future characteristics can be obtained at the same time at the current moment, effectively improving the accuracy of model prediction; in addition, the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model in the embodiment of the present invention are decoupled into parameter vectors and parameter scalars, and the node weight vector dimensions of the fault diagnosis model are normalized, realizing optimization based on matrix decomposition, enhancing the convergence speed of the model and the accuracy of model prediction, and having better effect in noise-sensitive environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is a flow chart of the centrifugal pump fault diagnosis method provided by the present invention;

[0043] Figure 2 It is a flow chart of the fault diagnosis model training method provided by the present invention;

[0044] Figure 3It is a structural schematic diagram of a centrifugal pump fault diagnosis device provided by the present invention;

[0045] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] At present, common centrifugal pump fault diagnosis methods include mechanism diagnosis methods (such as noise detection diagnosis method, vibration detection method, etc.), which mainly focus on extracting effective feature parameters through numerical calculation or signal processing technology, mainly including time-frequency domain analysis methods. However, due to the harsh working conditions, the collected fault vibration signals often have strong noise, nonlinearity and non-stationarity. At this time, traditional diagnostic methods cannot effectively extract fault features and classify them, which leads to low fault diagnosis accuracy and difficulty in meeting diagnostic needs under different working conditions.

[0048] To this end, the present invention provides a centrifugal pump fault diagnosis method. Figure 1 FIG. 1 is a flow chart of a centrifugal pump fault diagnosis method provided by the present invention, such as Figure 1 As shown, the method comprises the following steps:

[0049] Step 110 , extracting features from the vibration signal of the centrifugal pump to be diagnosed, and obtaining vibration features; the vibration features include time domain features, frequency domain features, and time-frequency domain features.

[0050] Here, the centrifugal pump to be diagnosed is the centrifugal pump that needs to be diagnosed for fault. The vibration signal refers to the vibration signal generated by the centrifugal pump during operation, and the vibration signal is generated due to the relative movement between the rotating parts (such as impellers, bearings, etc.) and the fixed parts of the centrifugal pump.

[0051] Optionally, statistical knowledge can be used to calculate statistical parameters such as the mean, peak, standard deviation, variance, root amplitude, skewness, kurtosis, etc. of the vibration signal samples to obtain the time domain characteristics, frequency domain characteristics, and time-frequency domain characteristics of the data samples.

[0052] Compared with the prior art, the embodiments of the present invention achieve sufficient feature extraction of vibration signals in the time domain, frequency domain, and time-frequency domain.

[0053] Step 120, inputting the vibration characteristics into the fault diagnosis model to obtain the fault diagnosis result output by the fault diagnosis model;

[0054] Among them, the fault diagnosis model is trained based on the sample vibration characteristics and the fault labels corresponding to the sample vibration characteristics. The fault diagnosis model is constructed based on the bidirectional long short-term memory network BiLSTM. The weight vectors in the forward propagation layer and the backward propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

[0055] Specifically, the sample vibration characteristics can be extracted from the sample vibration signal. For example, a centrifugal pump fault experimental platform can be built, and a vibration sensor can be used to collect the sample vibration signal of the continuous centrifugal pump, which is recorded as x(i), i=1, 2, 3, ... N, N is the total number of sampling points, and the vibration sensor can be arranged in the vertical direction of the centrifugal pump. Then, the sample vibration signal x(i) is subjected to sliding window slicing processing, and the sliding overlap is a. The vibration signal sequence F(t) after slicing processing can be expressed as:

[0056]

[0057] Among them, X[t:t+n] is the slice processing operation, n is the sliding window slice width, an is the sliding window step length, Represents the largest integer not exceeding N / an.

[0058] At the same time, the fault labels corresponding to the sample vibration characteristics can be one-hot encoded, that is, an N-bit status register is used to encode N types of data samples. Each type of sample has its own independent register bit, and at any time, only one of them is valid.

[0059] In addition, after slicing the sample vibration signal, the sample time domain characteristics, sample frequency domain characteristics and sample time-frequency domain characteristics of the sample vibration signal can be calculated using signal processing techniques such as Hilbert transform, wavelet transform, and empirical mode decomposition, and the sample feature matrix can be formed in sequence.

[0060] The extracted sample feature matrix can be normalized using the min-max normalization method:

[0061]

[0062] Where: x i_trans is the normalized vibration feature of the ith sample, x i is the vibration characteristic of the ith sample, min(x j ) is the minimum sample vibration feature, max(x j ) is the maximum sample vibration feature.

[0063] In addition, the fault diagnosis model is constructed based on the bidirectional long short-term memory network BiLSTM (Bi-directional LongShort-Term Memory), in which the BiLSTM is composed of a forward LSTM and a reverse LSTM to transmit the forward and reverse sequence information of the sample vibration characteristics. In addition, the weight vector w in the forward propagation layer and the reverse propagation layer of the fault diagnosis model is decoupled into a parameter vector v and a parameter scalar g in its Euclidean norm and its direction.

[0064]

[0065] Where: ||v|| represents the Euclidean norm of v. The updated values ​​of v and g can be calculated by the optimizer:

[0066]

[0067]

[0068] Among them: FL is the loss value, is the gradient value of w under FL.

[0069] It can be seen that the centrifugal pump fault diagnosis method provided by the embodiment of the present invention, the fault diagnosis model is based on the bidirectional long short-term memory network BiLSTM, so that the sample vibration characteristics can be transmitted in positive order and reverse order respectively, so that the information between historical characteristics and future characteristics can be obtained at the same time at the current moment, which effectively improves the accuracy of model prediction; in addition, the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model in the embodiment of the present invention are decoupled into parameter vectors and parameter scalars, and the node weight vector dimensions of the fault diagnosis model are normalized, which realizes optimization based on matrix decomposition, enhances the convergence speed of the model and the accuracy of model prediction, and has better effects in noise-sensitive environments.

[0070] Based on the above embodiment, the vibration characteristics are input into the fault diagnosis model to obtain the fault diagnosis result output by the fault diagnosis model, including:

[0071] Input the vibration feature into the forward propagation layer to obtain the first hidden state feature output by the forward propagation layer;

[0072] Input the vibration feature into the back propagation layer to obtain the second hidden state feature output by the back propagation layer;

[0073] The hidden layer of the fault diagnosis model of the first hidden state feature and the second hidden state feature is used to obtain a concatenated feature outputted by the hidden layer;

[0074] The concatenated features are input into the classification layer of the fault diagnosis model to obtain the fault diagnosis results output by the classification layer.

[0075] Specifically, the forward propagation layer can be understood as the forward WN-LSTM layer. After the vibration feature is input into the forward propagation layer, the first hidden state feature is obtained. as follows:

[0076]

[0077] In addition, the back propagation layer can be understood as the reverse WN-LSTM layer. After the vibration feature is input into the back propagation layer, the second hidden state feature is obtained. as follows:

[0078]

[0079] The hidden layer of the fault diagnosis model is the first hidden state feature and the second hidden state feature Perform matrix splicing and output the splicing matrix as follows:

[0080]

[0081] The above concatenated features are input into the softmax classification layer, and the softmax function can be used to calculate the probability values ​​of different vibration features belonging to various fault category labels (i.e., fault diagnosis results), as shown in the following formula:

[0082] FD j =softmax(wx+b) j

[0083] Among them: FD j It represents the predicted probability distribution of different fault categories, w is the weight, b is the bias, and x is the probability value that different vibration features output by the fault diagnosis model belong to various fault category labels.

[0084] Based on any of the above embodiments, considering the problem of uneven distribution of sample data corresponding to different types of faults (for example, there are more samples corresponding to some types of faults, and this category is easy to classify and identify; there are fewer sample data corresponding to some types of faults, and this category is difficult to classify and identify), in order to further improve the fault identification ability of the fault diagnosis model, the loss value of the fault diagnosis model designed in the embodiment of the present invention is:

[0085] FL=-α t (1-p t ) γ log(p t );

[0086]

[0087] Among them, FL represents the loss value of the fault diagnosis model, α trepresents the shared weight used to control the loss value of the positive sample vibration feature and the negative sample vibration feature, γ represents the modulation coefficient, α t and γ can reduce the weight of easy-to-classify samples, so that the model training can focus more on difficult-to-classify samples. y can be understood as the ground truth label, which indicates the accuracy of the fault prediction results based on the sample vibration characteristics. For example, y = 1 indicates correct classification, and p indicates the probability corresponding to the fault prediction result.

[0088] It can be seen that the embodiment of the present invention takes into account the problem of uneven sample distribution and introduces an improved loss value to measure the contribution of balanced difficult and easy classification samples and positive and negative classification samples to the total loss, which effectively improves the universality and accuracy of the model.

[0089] Based on any of the above embodiments, in order to achieve a more efficient network structure of the optimization model, an Adam optimizer with a momentum term and weight decay is adopted, and the learning rate lr gradually decreases with the iteration steps, that is, the fault diagnosis model solves the minimum loss value based on the first-order exponential smoothing value of the historical gradient, the first-order exponential smoothing value of the square of the historical gradient, and the updated value of the variable.

[0090] Among them, the first-order exponential smoothing value of the historical gradient is used to obtain the gradient value with momentum, which is determined based on the following formula:

[0091] m t =beta 1 ×m t-1 +(1-beta 1 )×g;

[0092] The first-order exponentially smoothed value of the square of the historical gradient is used to obtain the learning rate weight parameter of each weight parameter, which is determined based on the following formula:

[0093] v t =beta 2 ×v t-1 +(1-beta 2 )×g×g;

[0094] The updated value of the variable is proportional to the first-order exponential smoothing value of the historical gradient and inversely proportional to the first-order exponential smoothing value of the square of the historical gradient, and is determined based on the following formula:

[0095]

[0096] Among them, m t represents the first-order exponential smoothing value of the historical gradient (i.e., the first-order moment of the gradient g, i.e., the expectation of the gradient g), beta 1 represents the first-order moment attenuation coefficient, g represents the gradient obtained by taking the derivative of the optimization objective function with respect to X, and v trepresents the first-order exponential smoothing value of the square of the historical gradient (i.e., the second-order moment of the gradient g, i.e., the expectation of g×g), beta 2 represents the second-order moment attenuation coefficient, X represents the updated value of the variable, and ε is a constant, which is used to prevent the denominator from being 0 during the calculation process.

[0097] It can be seen that the embodiment of the present invention adopts the Adam optimizer with momentum term and weight decay. At the same time, the learning rate lr gradually decreases with the iteration steps, which can not only sparse the gradient, but also alleviate the problem of gradient oscillation, enhance the convergence stability of the model training, and further improve the accuracy of the model.

[0098] Based on any of the above embodiments, the parameter vector is updated based on the following formula:

[0099]

[0100] The parameter scalar is updated based on the following formula:

[0101]

[0102] Among them, v represents the parameter vector, g represents the parameter scalar, and w represents the weight vector.

[0103] Based on any of the above embodiments, the forward propagation layer and the backward propagation layer are each provided with an input gate, a first update gate, a forget gate, a second update gate and an output gate in sequence.

[0104] Specifically, the forward propagation layer and the backward propagation layer are sequentially provided with an input gate, a first update gate, a forget gate, a second update gate, and an output gate, wherein the output value of the update gate (the first update gate or the second update gate) at time t is for:

[0105]

[0106] Among them, W u Indicates the current input value X <t>< / t> With the historical data stream h at the previous moment <t-1>< / t-1> The weight vector input to the update gate, b u Indicates the current input value X <t>< / t> With the historical data stream h at the previous moment <t-1>< / t-1> The bias value input to the update gate, σ(·) represents the sigmoid activation function, WN(W u ) represents the weight vector W u Decoupled into parameter vector and parameter scalar.

[0107] The output value of the forget gate at time t for:

[0108]

[0109] Among them, W f Indicates the current input value X <t>< / t> With the historical data stream h at the previous moment <t-1>< / t-1> The weight vector during the input to the forget gate, b f Indicates the current input value X <t>< / t> With the historical data stream h at the previous moment <t-1>< / t-1> The bias value in the process of inputting to the forget gate, WN(W f ) represents the weight vector W f Decoupled into parameter vector and parameter scalar.

[0110] The state value of the memory unit at time t is c <t>< / t> for:

[0111]

[0112] in, represents the input data at time t after hyperbolic tangent activation, W c represents the weight vector in the hyperbolic tangent layer, b c represents the bias value in the hyperbolic tangent layer, tanh(·) represents the hyperbolic tangent activation function, WN(W c ) represents the weight vector W c Decoupled into parameter vector and parameter scalar.

[0113] Memory unit output value h <t>< / t> for:

[0114] Output value of the output gate for:

[0115] Among them, W o Indicates the current input value X <t>< / t> With the historical data stream h at the previous moment <t-1>< / t-1> The weight vector during the input to the output gate, b o Indicates the current input value X <t>< / t> With the historical data stream h at the previous moment <t-1>< / t-1> The bias value during the input to the output gate, WN(W o ) represents the weight vector W o Decoupled into parameter vector and parameter scalar.

[0116] Based on any of the above embodiments, the vibration characteristics are input into the fault diagnosis model to obtain the fault diagnosis result output by the fault diagnosis model, which also includes: normalizing the vibration characteristics, such as using the min-max normalization method for normalization.

[0117] Based on any of the above embodiments, the present invention also provides a fault diagnosis model training method, such as Figure 2 As shown, the method includes:

[0118] First, a vertical vibration sensor is used to collect centrifugal pump fault vibration data samples from the pump end of the centrifugal pump fault experiment. The collected samples include normal samples (fault-free samples), bearing faults (inner ring, outer ring, rolling element, lubrication failure), rotor faults (imbalance, misalignment, friction), pump rotor faults, mechanical looseness faults, etc. The sampling frequency of the vibration sensor can be 51.2kHz, and the rotation frequency of the centrifugal pump used in the test can be 25Hz.

[0119] Next, the sample is sliced ​​using a sliding window to obtain the sliding window overlapping slice data. The time-frequency domain feature extraction is performed on the sliding window overlapping slice data to obtain the sample vibration feature, and the fault label corresponding to the sample vibration feature is one-hot encoded. The sliding window size is set to 4096 and the overlap is 60%.

[0120] The sample vibration characteristics and the encoded fault labels are used as data samples, and the training set and test set are divided into 70% and 30% respectively. At the same time, the initial fault diagnosis model is constructed based on the Bi-WNLSTM network model, and the initial fault diagnosis model parameters are initialized. Among them, the network structure of the Bi-WNLSTM network model is constructed based on BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer are decoupled into parameter vectors and parameter scalars.

[0121] After initializing the initial fault diagnosis model parameters, model training is performed on the initial fault diagnosis model parameters based on the training set. Table 1 is a comparison table of model training parameters.

[0122] Table 1

[0123]

[0124]

[0125] After the training is completed, a 10-fold cross validation is performed based on the test set to perform statistics on the fault diagnosis accuracy. Table 2 is a statistical table of fault diagnosis accuracy. As shown in Table 2, the fault diagnosis model provided by the embodiment of the present invention has high accuracy and strong algorithm model stability.

[0126] Table 2

[0127]

[0128] After verification, the optimal model is obtained, and the verification set (which can be understood as the vibration characteristics corresponding to the vibration signal of the centrifugal pump to be diagnosed) is input into the optimal model to accurately obtain the model diagnosis result.

[0129] The centrifugal pump fault diagnosis device provided by the present invention is described below. The centrifugal pump fault diagnosis device described below and the centrifugal pump fault diagnosis method described above can be referenced to each other.

[0130] Based on any of the above embodiments, the present invention also provides a centrifugal pump fault diagnosis device, such as Figure 3 As shown, the device comprises:

[0131] An extraction unit 310 is used to extract features from the vibration signal of the centrifugal pump to be diagnosed, and obtain vibration features; the vibration features include time domain features, frequency domain features, and time-frequency domain features;

[0132] A diagnosis unit 320, configured to input the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model;

[0133] Among them, the fault diagnosis model is trained based on sample vibration characteristics and fault labels corresponding to the sample vibration characteristics, the fault diagnosis model is constructed based on a bidirectional long short-term memory network BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

[0134] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 4 As shown, the electronic device may include: a processor 410, a memory 420, a communication interface 430 and a communication bus 440, wherein the processor 410, the memory 420 and the communication interface 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 420 to execute the centrifugal pump fault diagnosis method, which includes: extracting features from the vibration signal of the centrifugal pump to be diagnosed to obtain vibration features; the vibration features include time domain features, frequency domain features and time-frequency domain features; inputting the vibration features into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model; wherein the fault diagnosis model is obtained by training based on sample vibration features and fault labels corresponding to the sample vibration features, the fault diagnosis model is constructed based on a bidirectional long short-term memory network BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

[0135] In addition, the logic instructions in the above-mentioned memory 420 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0136] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the centrifugal pump fault diagnosis method provided by the above methods, and the method includes: extracting features from the vibration signal of the centrifugal pump to be diagnosed to obtain vibration features; the vibration features include time domain features, frequency domain features, and time-frequency domain features; inputting the vibration features into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model; wherein the fault diagnosis model is trained based on sample vibration features and fault labels corresponding to the sample vibration features, and the fault diagnosis model is constructed based on a bidirectional long short-term memory network BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

[0137] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the above-mentioned centrifugal pump fault diagnosis methods, the method comprising: extracting features from the vibration signal of the centrifugal pump to be diagnosed to obtain vibration features; the vibration features include time domain features, frequency domain features and time-frequency domain features; inputting the vibration features into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model; wherein the fault diagnosis model is trained based on sample vibration features and fault labels corresponding to the sample vibration features, the fault diagnosis model is constructed based on a bidirectional long short-term memory network BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

[0138] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0139] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A centrifugal pump fault diagnosis method, characterized in that: include: Extracting features of the vibration signal of the centrifugal pump to be diagnosed to obtain vibration features; the vibration features include time domain features, frequency domain features and time-frequency domain features; Inputting the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model; Among them, the fault diagnosis model is trained based on sample vibration characteristics and fault labels corresponding to the sample vibration characteristics, the fault diagnosis model is constructed based on a bidirectional long short-term memory network BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

2. The centrifugal pump fault diagnosis method according to claim 1, characterized in that: The step of inputting the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model includes: Inputting the vibration feature into the forward propagation layer to obtain a first hidden state feature output by the forward propagation layer; Inputting the vibration feature into the back propagation layer to obtain a second hidden state feature output by the back propagation layer; The first hidden state feature and the second hidden state feature are used to form a hidden layer of the fault diagnosis model, thereby obtaining a concatenated feature output by the hidden layer; The splicing features are input into the classification layer of the fault diagnosis model to obtain the fault diagnosis result output by the classification layer.

3. The centrifugal pump fault diagnosis method according to claim 1 or 2, characterized in that: The loss value of the fault diagnosis model is: FL=-a t (1-p t ) γ log(p t ); Among them, FL represents the loss value of the fault diagnosis model, α t represents the shared weight used to control the loss value of the positive sample vibration feature and the negative sample vibration feature, γ represents the modulation coefficient, y represents the accuracy of the fault prediction result obtained based on the sample vibration feature, and p represents the probability corresponding to the fault prediction result.

4. The centrifugal pump fault diagnosis method according to claim 3, characterized in that: The fault diagnosis model solves the minimum value of the loss value based on the first-order exponential smoothing value of the historical gradient, the first-order exponential smoothing value of the square of the historical gradient and the updated value of the variable; The first-order exponential smoothing value of the historical gradient is determined based on the following formula: m t =beta1×m t-1 +(1-beta1)×g; The first-order exponential smoothing value of the historical gradient square is determined based on the following formula: v t =beta2×v t-1 +(1-beta2)×g×g; The updated value of the variable is determined based on the following formula: Among them, m t represents the first-order exponential smoothing value of the historical gradient, beta1 represents the first-order moment attenuation coefficient, g represents the gradient, v t represents the first-order exponential smoothing value of the historical gradient square, beta2 represents the second-order moment attenuation coefficient, X represents the updated value of the variable, and ε is a constant.

5. The centrifugal pump fault diagnosis method according to claim 3, characterized in that: The parameter vector is updated based on the following formula: The parameter scalar is updated based on the following formula: Among them, v represents the parameter vector, g represents the parameter scalar, and w represents the weight vector.

6. The centrifugal pump fault diagnosis method according to claim 1 or 2, characterized in that: The forward propagation layer and the backward propagation layer are each provided with an input gate, a first update gate, a forget gate, a second update gate and an output gate in sequence.

7. The centrifugal pump fault diagnosis method according to claim 1 or 2, characterized in that: The step of inputting the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model also includes: normalizing the vibration characteristics.

8. A centrifugal pump fault diagnosis device, characterized in that: include: An extraction unit is used to extract features from the vibration signal of the centrifugal pump to be diagnosed, and obtain vibration features; the vibration features include time domain features, frequency domain features, and time-frequency domain features; A diagnosis unit, used for inputting the vibration characteristics into a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model; Among them, the fault diagnosis model is trained based on sample vibration characteristics and fault labels corresponding to the sample vibration characteristics, the fault diagnosis model is constructed based on a bidirectional long short-term memory network BiLSTM, and the weight vectors in the forward propagation layer and the reverse propagation layer of the fault diagnosis model are decoupled into parameter vectors and parameter scalars.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the centrifugal pump fault diagnosis method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the centrifugal pump fault diagnosis method according to any one of claims 1 to 7 is implemented.

Images