Spray pump composite fault diagnosis method based on multi-source signal migration

Abstract

The invention belongs to the technical field of spray pump mechanical composite fault diagnosis, and particularly discloses a spray pump composite fault diagnosis method based on multi-source signal migration, which comprises the following steps: respectively carrying out FFT (Fast Fourier Transform) on original vibration and displacement signals of a source domain and a target domain to obtain amplitude spectrums and normalize the amplitude spectrums; respectively inputting the normalized amplitude spectrums of the vibration and displacement signals into a deep transfer learning model for diagnosis so as to respectively obtain posterior probability values corresponding to the deep transfer learning model; and fusing the posterior probability values by using a weighted voting method so as to obtain a final diagnosis result and realize diagnosis analysis of the compound fault of the spray pump. According to the method, a multi-source information fusion migration diagnosis method is utilized, the problem of composite fault diagnosis of the spray pump under variable working conditions is solved, and diagnosis analysis of composite faults of the spray pump under the variable working conditions, weak fault information and strong interference conditions is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of composite fault diagnosis of a jet pump mechanism, and in particular to a composite fault diagnosis method for a jet pump based on multi-source signal migration. Background Art

[0002] As a new type of propulsion method, the jet pump (water jet propulsion device) has been widely used in a variety of ship propulsion equipment. A typical jet pump is mainly composed of three parts: mechanical system, control system, and hydraulic system. During the long-term operation of the jet pump, it is affected by the leakage of seawater, sewage, and other complex and changeable working conditions. The complexity of the jet pump structure makes it have a wide variety of faults, often showing a phenomenon of multiple fault superposition. Composite faults such as bearing failure, impeller imbalance, and shaft movement often occur, which seriously affects the operating efficiency and safety and stability of the jet pump. Therefore, the diagnosis and research of composite faults of the jet pump is of great significance.

[0003] Since the bearing is placed under the guide vane of the jet pump, the vibration sensor cannot be installed close to the inner ring of the guide vane of the fixed bearing outer ring, but can only be installed on the outer casing. However, due to the long vibration transmission path, the picked-up vibration signal is weak. At the same time, affected by other vibration sources such as wave impact, pressure pulsation, engine vibration and other factors, the picked-up vibration signal contains a strong coupling interference signal and is also affected by noise, which makes it difficult to extract the bearing fault information, and also makes it difficult to extract the fault information of impeller imbalance and axial movement. The displacement sensor installed at the impeller shaft end can easily monitor the displacement signal of axial movement, but due to the interference signal, noise and long transmission path, it is difficult to extract the displacement signal for impeller imbalance and guide vane bearing fault information ( Figure 1 The measuring point location information is given). At the same time, affected by the variable working conditions, the data sample distribution under the variable working conditions is inconsistent. Therefore, it is extremely difficult to diagnose and analyze the compound fault of the jet pump under the conditions of variable working conditions, weak fault information and strong interference.

[0004] Data-driven fault diagnosis methods for jet pumps include: 1) When using a single sensor signal in combination with a deep learning model to diagnose and analyze complex faults such as jet pump bearing faults, impeller imbalance, and shaft movement, although the advantage of deep learning in automatically extracting feature parameters can be used for diagnosis, the acquired fault information is weak, which can easily lead to low diagnostic accuracy for complex faults; 2) Although multi-source information fusion technology can effectively make up for the shortcomings of the diagnosis method based on a single sensor signal, it is still difficult to effectively solve the fault diagnosis problem under variable operating conditions; 3) Although transfer learning can effectively solve the fault diagnosis problem under variable operating conditions, it is affected by the coupling of excitations such as fluid pressure pulsation, main engine vibration, and wave fluctuations, resulting in the feature parameters extracted by deep learning affecting the diagnostic accuracy, which in turn leads to the diagnostic accuracy of the deep learning diagnosis method based on transfer still has room for improvement. Knowledge-based methods for fault diagnosis of jet pumps include: 1) Traditional diagnostic knowledge based on signal analysis, such as the skewness and kurtosis of time domain signals, the center frequency and average frequency of frequency domain signals, can well describe fault information such as bearing and impeller imbalance. However, when using only these diagnostic knowledge for diagnostic analysis, it is affected by interference information and weak fault information, and the diagnostic accuracy is not high.

[0005] Although the above-mentioned data-based and knowledge-based diagnostic methods have their own advantages and disadvantages when performing diagnostic analysis, these fault diagnosis methods cannot effectively solve the diagnostic analysis of composite faults such as jet pump bearing faults, impeller imbalance and axial movement when used alone. Therefore, a fault method of multi-source signal migration driven by knowledge and data fusion is proposed to achieve the diagnostic analysis of composite faults such as jet pump bearing faults, impeller imbalance and axial movement.

[0006] For typical faults such as jet pump bearing failure, impeller imbalance and shaft movement, the sensor installation diagram is as follows: Figure 1 As shown, 1 is a guide vane, 2 is an impeller, 3 is a bearing, 4 is a guide vane, 5 is a unidirectional vibration sensor, 6 is an impeller, 7 is a displacement sensor, and 8 is a shaft. Summary of the invention

[0007] The purpose of the present invention is to solve the problem of composite faults such as jet pump bearing failure, impeller imbalance and shaft movement. Due to the long vibration transmission path and multiple vibration sources, the vibration signals and displacement signals contain weak fault information, many coupling components and strong noise, which makes the diagnosis of composite faults difficult. To this end, a composite fault diagnosis method for jet pumps based on multi-source signal migration is provided.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] A composite fault diagnosis method for a jet pump based on multi-source signal migration, comprising:

[0010] Perform FFT transformation on the original vibration and displacement signals in the source domain and target domain respectively to obtain the amplitude spectrum and then normalize it;

[0011] The normalized amplitude spectra of the vibration and displacement signals are respectively input into the deep transfer learning model for diagnosis, thereby obtaining the posterior probability values ​​of the corresponding deep transfer learning model respectively;

[0012] The weighted voting method is used to fuse the posterior probability values ​​to obtain the final diagnosis result and realize the diagnosis and analysis of the compound fault of the injection pump.

[0013] The following is a technical solution further defined by the present invention, which performs FFT transformation on the original time domain signals of the source domain and a small amount of labeled target domain to obtain amplitude spectra and then normalize them, and input them into the deep metric learning model for automatic extraction of feature parameters.

[0014] The following is a technical solution further defined by the present invention, which extracts skewness, kurtosis, average frequency and center frequency from the original time domain signals of the source domain and the target domain with a small amount of labels, respectively.

[0015] The following is a technical solution further defined by the present invention, in which the extracted skewness, kurtosis, mean frequency and center frequency are fused in parallel with the feature parameters automatically extracted by the deep metric learning model from the perspectives of the source domain and the target domain respectively.

[0016] The following is a technical solution further defined by the present invention, which utilizes the hidden layer sharing of the deep metric learning model and the migration mechanism based on minimizing the maximum mean difference, and combines the classifier of the deep metric learning model to construct a deep transfer learning model.

[0017] The following is a technical solution further limited by the present invention, which performs FFT transformation on the original time domain signal of the unlabeled target domain to obtain the amplitude spectrum and then normalizes it, and inputs it into the deep metric learning model for automatic extraction of feature parameters, extracts skewness, kurtosis, average frequency and center frequency from the original time domain signal of the unlabeled target domain, and fused the extracted skewness, kurtosis, average frequency and center frequency in parallel with the feature parameters automatically extracted by the deep metric learning model from the perspective of the target domain.

[0018] The following is a technical solution further defined by the present invention, in which the result of parallel fusion of the original time domain signal of the unlabeled target domain is input into the constructed deep transfer learning model to obtain the posterior probability value of the corresponding deep transfer learning model.

[0019] The following is a technical solution further defined by the present invention, in which a vibration sensor is used to collect source domain vibration signals and target domain vibration signals.

[0020] The following is a technical solution further defined by the present invention, in which a displacement sensor is used to collect a source domain displacement signal and a target domain displacement signal.

[0021] The following is a technical solution further defined by the present invention. The composite faults of the jet pump include jet pump bearing damage, impeller imbalance and axial movement.

[0022] Compared with the prior art, the present invention has the following technical effects:

[0023] The present invention utilizes a migration diagnosis method of multi-source information fusion to solve the difficult problem of composite fault diagnosis of jet pumps under variable operating conditions, and realizes the diagnostic analysis of composite faults of jet pumps under variable operating conditions, weak fault information and strong interference conditions; establishes a fault diagnosis method that integrates diagnostic knowledge such as skewness and kurtosis, characteristic parameters such as average frequency and center frequency, and data diagnostic models, realizes the complementarity of diagnostic knowledge and data-driven diagnostic models, and solves the problem of insufficient accuracy of single diagnostic methods such as knowledge diagnostic methods or data-driven diagnostic methods under weak fault information and strong interference.

[0024] The present invention is further described below in conjunction with the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only 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.

[0026] Figure 1 It is a schematic diagram of the sensor installation location;

[0027] Figure 2 It is a schematic diagram of the overall process of the method of the present invention;

[0028] Figure 3 It is a schematic flow chart of a method for diagnosing composite faults of a jet pump using vibration signals according to the present invention;

[0029] Figure 4 It is a schematic diagram of the diagnosis result of the present invention. DETAILED DESCRIPTION

[0030] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0031] This embodiment provides a method for composite fault diagnosis of a jet pump based on knowledge and data-driven multi-source signal migration. Figure 2 As shown, the details are as follows:

[0032] 1) Perform FFT transformation on the original vibration and displacement signals in the source domain and target domain respectively to obtain the amplitude spectrum and then normalize it;

[0033] 2) The normalized amplitude spectra of the vibration and displacement signals are respectively input into the knowledge and data fusion driven deep transfer learning model (Euc_DTML) for diagnosis, so as to obtain the preliminary diagnosis results of each Euc_DTML model, i.e., the posterior probability value;

[0034] 3) Then, the weighted voting method is used to fuse these posterior probability values ​​to obtain the final diagnosis result and realize the diagnosis and analysis of the composite fault of the injection pump.

[0035] Among them, the construction process of the deep transfer learning model (Euc_DTML) driven by knowledge and data fusion specifically includes:

[0036] 1) If Figure 3 As shown in the figure, taking the vibration signal as an example, the original time-domain vibration signals of the source domain and the target domain with a small amount of labels are respectively transformed by FFT to obtain the amplitude spectrum and then normalized, and then input into the deep metric learning model (Euc_DML) based on Euclidean distance for automatic extraction of feature parameters; at the same time, four feature parameters such as skewness and kurtosis, average frequency and center frequency are extracted from the original time-domain vibration signals of the source domain and the target domain with a small amount of labels, and they are fused in parallel with the feature parameters automatically extracted by the previous Euc_DML model from the perspective of the source domain and the target domain respectively;

[0037] 2) Using the hidden layer sharing of the Euc_DML model and the migration mechanism based on the minimization of the maximum mean difference (MMD), combined with the classifier of the Euc_DML model, a knowledge- and data-driven deep transfer learning model (Euc_DTML) is constructed.

[0038] While constructing a knowledge and data fusion driven deep transfer learning model (Euc_DTML), the original time domain signal of the unlabeled target domain is transformed by FFT to obtain the amplitude spectrum and then normalized, and then input into the deep metric learning model for automatic extraction of feature parameters. The skewness, kurtosis, average frequency and center frequency are extracted from the original time domain signal of the unlabeled target domain, and the extracted skewness, kurtosis, average frequency and center frequency are fused in parallel with the feature parameters automatically extracted by the deep metric learning model from the perspective of the target domain. The result of the parallel fusion of the original time domain signal of the unlabeled target domain is input into the constructed deep transfer learning model to obtain the posterior probability value of the corresponding deep transfer learning model.

[0039] Vibration sensors are used to collect source domain vibration signals and target domain vibration signals. Displacement sensors are used to collect source domain displacement signals and target domain displacement signals.

[0040] The fault diagnosis method of the present invention is used to diagnose and analyze the composite faults of bearing damage, impeller imbalance and axial movement of the jet pump under variable working conditions, wherein the fault types are normal, bearing damage, impeller imbalance and axial movement composite faults.

[0041] The A data set of working condition 1 is the source domain, and its speed is 100r / min. The B data set of working condition 2 is the target domain, and its speed is 200r / min. The number of labeled training set samples in the source domain is 450, the number of labeled target domain training set samples is 90, and the number of unlabeled target domain test set samples is 480 (see Table 1):

[0042] Table 1: Description of the bearing fault, impeller imbalance and axial play datasets

[0043]

[0044] The method of the present invention is trained using labeled training samples from the source domain and a small number of target domains, and diagnostic analysis is performed on the test samples from the target domain. The results are shown in Figure 4 As shown in the figure, the diagnostic accuracy reaches 100%, which verifies the effectiveness of the method of the present invention. Its accuracy is significantly better than the Euc_DML diagnostic model based on a single sensor signal, the Euc_DML migration diagnostic model based on a single sensor signal, and the Euc_DTML model of multi-source signal fusion (no knowledge fusion), as shown in Table 2:

[0045] Table 2: Diagnostic accuracy of each diagnostic model

[0046]

[0047] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Any technician familiar with the art can make many possible changes and modifications to the technical solution of the present invention by using the above disclosed methods and technical contents without departing from the scope of the technical solution of the present invention, or modify it into an equivalent embodiment of equivalent changes. Therefore, all equivalent changes made according to the shape, structure and principle of the present invention without departing from the content of the technical solution of the present invention should be included in the protection scope of the present invention.

Claims

1. A composite fault diagnosis method for a jet pump based on multi-source signal migration, characterized in that: include: Perform FFT transformation on the original vibration and displacement signals in the source domain and target domain respectively to obtain the amplitude spectrum and then normalize it; The normalized amplitude spectra of the vibration and displacement signals are respectively input into the deep transfer learning model for diagnosis, thereby obtaining the posterior probability values ​​of the corresponding deep transfer learning model respectively; The weighted voting method is used to fuse the posterior probability values ​​to obtain the final diagnosis result and realize the diagnosis and analysis of the compound fault of the injection pump.

2. A method for composite fault diagnosis of a jet pump based on multi-source signal migration as claimed in claim 1, characterized in that: The original time domain signals in the source domain and a small amount of labeled target domain are transformed by FFT to obtain the amplitude spectrum, which is then normalized and input into the deep metric learning model for automatic extraction of feature parameters.

3. A method for composite fault diagnosis of a jet pump based on multi-source signal migration as claimed in claim 2, characterized in that: The skewness, kurtosis, mean frequency and center frequency are extracted from the original time domain signals in the source domain and a small amount of labeled target domain, respectively.

4. A method for composite fault diagnosis of a jet pump based on multi-source signal migration as claimed in claim 3, characterized in that: The extracted skewness, kurtosis, mean frequency and center frequency are fused in parallel with the feature parameters automatically extracted by the deep metric learning model from the perspective of the source domain and the target domain respectively.

5. A method for composite fault diagnosis of a jet pump based on multi-source signal migration as claimed in claim 4, characterized in that: A deep transfer learning model is constructed by utilizing the hidden layer sharing of the deep metric learning model and the migration mechanism based on minimizing the maximum mean difference, combined with the classifier of the deep metric learning model.

6. A method for composite fault diagnosis of a jet pump based on multi-source signal migration as claimed in claim 5, characterized in that: The original time domain signal of the unlabeled target domain is transformed by FFT to obtain the amplitude spectrum and then normalized, and then input into the deep metric learning model for automatic extraction of feature parameters. The skewness, kurtosis, mean frequency and center frequency are extracted from the original time domain signal of the unlabeled target domain, and the extracted skewness, kurtosis, mean frequency and center frequency are fused in parallel with the feature parameters automatically extracted by the deep metric learning model from the perspective of the target domain.

7. A method for composite fault diagnosis of a jet pump based on multi-source signal migration as claimed in claim 6, characterized in that: The result of parallel fusion of the original time domain signals of the unlabeled target domain is input into the constructed deep transfer learning model to obtain the posterior probability value of the corresponding deep transfer learning model.

8. A method for composite fault diagnosis of a jet pump based on multi-source signal migration as claimed in claim 1, characterized in that: Vibration sensors are used to collect source domain vibration signals and target domain vibration signals.

9. A method for composite fault diagnosis of a jet pump based on multi-source signal migration as claimed in claim 1, characterized in that: Displacement sensors are used to collect source domain displacement signals and target domain displacement signals.

10. The method for composite fault diagnosis of a jet pump based on multi-source signal migration according to claim 1, characterized in that: Compound failures of jet pumps include jet pump bearing damage, impeller imbalance and axial movement.

Images