Ship machinery vibration transmission path analysis method based on physical neural network

Abstract

The application relates to the technical field of mechanical vibration analysis, and discloses a ship mechanical vibration transmission path analysis method based on a physical neural network, which comprises the following steps: obtaining the difference between the displacement responses at the two ends of a vibration isolation device; obtaining the sound pressure response values in a radiation sound field under the working states of each device; obtaining the number of working conditions; establishing a transfer function, and equivalently taking the transfer function in a selected frequency band as a transfer function matrix with frequency points as the independent variables; constructing a full-connection complex deep neural network model fusing physical information, and outputting samples under different working condition combinations; verifying the samples, and evaluating the performance of the full-connection complex deep neural network model. The ship mechanical vibration transmission path analysis method based on the physical neural network realizes accurate analysis and contribution degree identification of the ship mechanical vibration transmission path by constructing a vibration source input, a sound pressure response equivalent transfer function and a complex deep neural network model fusing physical prior knowledge.

Description

Technical Field

[0001] This invention relates to the field of mechanical vibration analysis technology, specifically to a method for analyzing the transmission path of ship mechanical vibration based on physical neural networks. Background Technology

[0002] Vibrations generated during the operation of ship machinery are transmitted through the structure and excite hull vibrations, which in turn radiate underwater noise, making them a key factor affecting the overall performance of a ship. Accurately identifying the vibration-sound transmission path and its contribution is not only crucial for acoustic characteristic control and vibration reduction design, but also a core technological support for improving ship survivability and operational efficiency. However, the complex environment of ship cabins presents challenges in noise source identification, including dense multi-source excitation, highly coupled structural paths, nonlinear contact interfaces, and strong background noise interference.

[0003] Compared to traditional noise source identification methods, Transfer Path Analysis (TPA) methods can quantitatively analyze the complete energy transfer chain from the excitation source to the target acoustic response, revealing the causal mechanism between source, path, and response, and providing direct evidence for structural-level intervention. Operational Transfer Path Analysis (OTPA), compared to traditional TPA, can better simulate and analyze real-world working boundary conditions. It only requires measurement data from the system under operating conditions to assess the source contribution, eliminating the need for lengthy frequency response function (FRF) measurements. Furthermore, it can flexibly combine various sensor data (such as acceleration, sound pressure, and force), making it an effective method for rapidly assessing the proportion of structural and air conduction contributions.

[0004] However, existing technologies have the following problems in practical applications: 1) The traditional OTPA method treats each vibration source independently, without fully considering the cross-coupling problem between input vibration sources in dense multi-source excitation scenarios, which makes it difficult to separate the contributions between sources and to accurately identify the independent contribution of each vibration source to the target response; 2) The DOTPA (Decoupling Operational Transfer Path Analysis) method faces the ill-posed problem of matrix inversion in practical applications. When the number of vibration source path points is large, a large number of working cases are required. However, in actual engineering scenarios, due to the limitations of test conditions, the number of working cases is often insufficient. At the same time, segmented calculation leads to low computational efficiency. 3) Existing data-driven deep neural network models have poor physical interpretability, are difficult to obtain training samples, and are difficult to effectively handle the nonlinear and ill-posed problems of vibration and acoustic transmission in complex ship structures. 4) For the vibration cross-coupling problem caused by crosstalk between equipment in complex ship structures, existing methods are difficult to accurately identify the contribution value of each vibration source to the target point of the sound field in the complex domain, and cannot take into account both amplitude and phase information. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the above-mentioned technologies by providing a method for analyzing the vibration transmission path of ship machinery based on physical neural networks. By constructing an equivalent transfer function of vibration source input, sound pressure response, and a complex deep neural network model that integrates prior physical knowledge, the method achieves accurate analysis and contribution identification of the vibration transmission path of ship machinery.

[0006] To achieve the above objectives, the present invention provides a method for analyzing the transmission path of ship mechanical vibrations based on physical neural networks, comprising the following steps: S1) Each device is connected to the base through a vibration isolation device. The difference in displacement response at both ends of the vibration isolation device is obtained to construct the vibration source input matrix. S2) Obtain the sound pressure response values ​​in the radiated sound field under the working conditions of each device, and construct the target response matrix; S3) Based on the activation status of each device, obtain the number of operating conditions; S4) Establish the transfer function between the difference in displacement response and the sound pressure response value, and convert the transfer function in the selected frequency band into a transfer function matrix with the frequency point as the independent variable; S5) Using the source input matrix, target response matrix and frequency information as input, construct a fully connected complex deep neural network model that integrates physical information, establish a nonlinear mapping relationship between the source input matrix and the target response matrix, and output samples under different working conditions. S6) Validate the samples, evaluate the performance of the fully connected complex deep neural network model, and calculate the contribution of the vibration source to the transmission path of the target response by combining the equipment operating conditions and the vibration source removal method.

[0007] Preferably, in step S1), the difference in displacement response is obtained by performing a second time-domain integral on the difference in vibration acceleration at both ends of the vibration isolation device. The difference in displacement response is regarded as an input vibration source, and the vibration source input matrix is ​​diagonally distributed to avoid cross-coupling between input vibration sources. By using the difference in displacement response at both ends of the vibration isolation device as an equivalent vibration source input, the cross-coupling problem between sources caused by the dense arrangement of multiple source excitations in the traditional OTPA method is eliminated from a physical perspective, so that the contribution of each vibration source can be calculated and separated independently.

[0008] Preferably, in step S2), the sound pressure response value is the superposition of the product of the corresponding FRF and the excitation force during operation.

[0009] Preferably, in step S3), the operating conditions include: individual equipment activation, partial equipment activation, and all equipment activation. By rationally planning the combination of operating conditions, the sample coverage is maximized under limited actual test conditions, ensuring that the transmission path analysis results can reflect the various states of the equipment in actual operation.

[0010] Preferably, in step S4), the transfer function within the selected frequency band is equivalent to a transfer function matrix with frequency points as independent variables. This includes dividing the frequency domain interval into multiple frequency bands, calculating the transfer function of the corresponding frequency band, and then connecting them to form the transfer function of the entire frequency band. The value of the transfer function at the intersection of adjacent frequency bands is taken as the arithmetic mean of the values ​​of the transfer function at the intersection of adjacent frequency bands. By segmenting the frequency domain, the transfer function matrix that originally required a large number of operating conditions to solve is transformed into an extended set of equations that can be solved by adjusting the number of frequency points under limited operating conditions. This effectively solves the problem of insufficient number of operating conditions and improves computational efficiency.

[0011] Preferably, in step S5), the fully connected complex deep neural network model that integrates physical information is based on the excitation-response physical information transmission path model and constructed based on the multilayer perceptron complex deep neural network model. It constructs a nonlinear mapping relationship between the source input matrix and the target response matrix at each frequency point. The frequency information, after normalization, is used as the network input encoding. The network structure includes an input layer, at least one hidden layer, and an output layer. The dimension of the input layer is determined by the number of sources and the number of frequency points within the frequency band, and the dimension of the output layer is determined by the number of target response measurement points and the number of frequency points within the frequency band. The activation function is the ReLU function. By embedding the physical model framework into the network structure constraints, the neural network training process is guided by physical laws, enhancing the physical interpretability of the model. Simultaneously, the frequency information, as the encoding input, enables the network to learn frequency-domain related nonlinear mapping features.

[0012] Preferably, during the training process of the fully connected complex deep neural network model, all input / output variables are normalized by Z-score, the optimizer is Adam, and an early stopping mechanism based on validation loss is introduced. That is, after each training round, the mean square error of the validation set is calculated, and training is terminated when the validation loss does not improve after a preset number of consecutive rounds, while maintaining the network weights with the best validation performance. The interference of different physical dimensions on network training is eliminated by Z-score normalization, and the early stopping mechanism effectively avoids the risk of overfitting under limited sample conditions, ensuring the generalization ability of the model.

[0013] Preferably, in step S5), the number of frequency bands and the number of frequency points within them are adjusted to expand the number of samples.

[0014] Preferably, when evaluating the performance of a fully connected complex deep neural network model, the coefficient of determination is defined as an evaluation index. It is calculated uniformly after concatenating all frequency points and all output dimensions into a one-dimensional vector to reflect the overall fitting effect.

[0015] Preferably, the vibration source removal method includes: setting the input data corresponding to a certain vibration source device to zero, calculating the change in the target response before and after the vibration source removal, so as to evaluate the contribution spectrum and contribution ranking of the vibration source to the target point. Through the virtual vibration source removal strategy, the independent contribution of each vibration source can be quantitatively evaluated without interrupting the operation of the actual equipment, providing a direct basis for vibration source priority ranking and vibration reduction and noise reduction design.

[0016] Compared with the prior art, the present invention has the following advantages: 1. By treating the difference in displacement response at both ends of the vibration isolator as the input vibration source, the vibration source input matrix is ​​diagonally distributed, which avoids the cross-coupling between input vibration sources in a physical sense and solves the problem of cross-coupling between sources faced by the traditional OTPA method. 2. The transfer function is equivalent to a frequency domain polynomial expression. By adjusting the number of frequency bands and the number of frequency points within them, the effective sample set is expanded under limited operating conditions, which solves the problem of insufficient number of operating conditions in the application of the DOTPA method, and at the same time significantly improves the computational efficiency. 3. It integrates physical models and prior knowledge, and is based on the decoupled vibration source matrix and target response matrix data under varying working conditions. It considers frequency information as encoding, which effectively solves the analysis errors caused by the poor physical interpretability of the model, the difficulty in obtaining training samples, nonlinearity and ill-posedness. 4. By combining the optimal equipment operating condition combination with the vibration source removal method, the network model's ability to handle vibration cross-coupling problems caused by crosstalk between equipment is enhanced, and it can accurately identify the contribution value of the vibration source to the sound field target point in the complex domain. 5. By integrating the physical interpretability of frequency domain modeling with the nonlinear fitting capability of data-driven models, it not only achieves high-precision vibration response prediction, but also provides a traceable end-to-end mapping basis for subsequent transmission path contribution analysis, laying the theoretical and algorithmic support for engineering applications such as intelligent fault vibration, virtual sensing and active vibration isolation control. Attached Figure Description

[0017] Figure 1 The sensor locations are shown in this embodiment of the invention (black dots are vibration acceleration measurement points on the equipment, and blue dots are acceleration measurement points on the base, all arranged above and below the vibration isolation device). Figure 2 The spectrum of the coherence coefficient of each device is selected from 1 vibration source measurement point (m=3) and 2 vibration source measurement points (m=6); Figure 3To calculate the linear fitting model between the vibration source input and the target point response output, R 2 ; Figure 4 The actual and predicted sound pressure amplitude and phase curves of the response when device No. 2 is turned on; Figure 5 Contribute a cloud map to the spectrum when device number 2 is turned on; Figure 6 Contribute a bar chart when device number 2 is turned on; Figure 7 The actual and predicted sound pressure amplitude and phase curves of the three devices when they are turned on; Figure 8 Calculate the magnitude error curves between predicted and actual values ​​for different network models; Figure 9 Spectrum contribution cloud map when three devices are turned on simultaneously; Figure 10 A bar chart showing the contribution of three devices when they are turned on simultaneously. Detailed Implementation

[0018] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] A method for analyzing the vibration transmission path of ship machinery based on physical neural networks includes the following steps: S1) Each piece of equipment is connected to the base via a vibration isolation device. The difference in displacement response between the two ends of the vibration isolation device is obtained to construct the vibration source input matrix. Specifically, according to the complex stiffness method, the active side (equipment) and the passive side (base) are connected by an elastic element (vibration isolation device). Displacement occurs at both ends of the connection between the vibration isolation device and the active and passive sides. In the low-frequency band below 1kHz, the contact unit between the equipment and the base, as well as the entire vibration reduction device, is approximated as point contact. The difference in displacement response is obtained by performing a second time-domain integral on the difference in vibration acceleration between the two ends of the vibration isolation device. : , In the formula, and These are the acceleration responses on the equipment side and the base side, respectively. S2) Obtain the sound pressure response value in the radiated sound field under the working conditions of each device. Construct the target response matrix, where the sound pressure response value is the superposition of the product of the corresponding FRF and the excitation force during operation: , In the formula, Let be the excitation force vector under working conditions. According to the complex stiffness method, the active side (equipment) and the passive side (base) are connected by an elastic element (vibration isolation device). Displacement occurs at both ends of the vibration isolation device and the active and passive sides. The equipment acts on the base through the vibration isolation device, and its working force can be obtained according to the dynamic stiffness method. , In the formula, It is a diagonal negative stiffness matrix. Let be the frequency response function of the sound pressure response from each noise source to the radiated sound field; S3) Based on the on / off status of each device, obtain the number of operating conditions. r Operating conditions include: individual equipment operation, partial equipment operation, and full equipment operation. p When a device is turned on individually, the input vibration source is determined. , m The number of differences in displacement response under each working condition, i.e., the number of vibration sources; S4) Establish the transfer function between the difference in displacement response and the sound pressure response value, and convert the transfer function within the selected frequency band into an equivalent transfer function matrix with frequency points as independent variables. Specifically, this is combined with the measured vibration response. , can be obtained In the formula, It is a diagonal matrix. Furthermore, we can obtain: , In the formula, To measure the fluctuation values ​​caused by nonlinearity in the transmission of noise and vibration, For the transfer function matrix, for r Different working conditions are satisfied: , Right now In the formula, n The number of sound pressure responses under each operating condition r ≥ m , For a transfer function, a large number of operating conditions are required. r At that time, the transfer function within a certain frequency band can be expressed in the frequency domain with the independent variable as the frequency domain. The polynomial representation of is , In the formula, i The index represents the difference in displacement response. j This is the sequence number of the sound pressure response. h For the initial frequency band, s The number of frequency points after the initial frequency band.l For frequency points, For coefficients; set up , The formula can be Represented as: ,set up ,get: ,satisfy The unknown parameters within this frequency range can be obtained. The value is calculated. Transfer function within the interval This transforms the transfer function within a selected frequency band into an equivalent transfer function matrix with frequency points as independent variables. This involves dividing the frequency domain into multiple frequency bands, calculating the transfer function for each corresponding band, and then connecting these bands to form the transfer function for the entire frequency band. The value of the transfer function at the intersection of adjacent frequency bands is taken as the arithmetic mean of the values ​​at that intersection. Specifically, the entire frequency domain is segmented as follows: The transfer function for each frequency band is calculated by applying the formula to each segment, and then these segments are concatenated to form the transfer function for the entire frequency band, thus obtaining the transfer function matrix. m , k Given the quantity under operating conditions r When the frequency is low, adjust the number of frequency points within the frequency band. s To solve the equation; S5) Using the source input matrix, target response matrix, and frequency information as inputs, a fully connected complex deep neural network model integrating physical information is constructed. A nonlinear mapping relationship between the source input matrix and the target response matrix is ​​established. The frequency information is normalized and used as the network input encoding. The network structure includes an input layer, at least one hidden layer, and an output layer. The dimension of the input layer is determined by the number of sources and the number of frequency points within the frequency band, and the dimension of the output layer is determined by the number of target response measurement points and the number of frequency points within the frequency band. The ReLU function is used as the activation function, and finally, samples under different operating condition combinations are output. Specifically, each frequency point is constructed... l Place, n Response matrix of external targets and m Each vibration source input matrix Nonlinear mapping relationship between them: Constructing different working condition combinations using the OTPA complex matrix input / output framework. Z With a limited number of samples, the constructed nonlinear mapping requires only a small number of equipment operating conditions to generate a large number of samples, thus reducing the number of operating conditions. r When the sample data is insufficient, a linear mapping can be constructed by adjusting the number of frequency points within the frequency band. ; S6) Validate the samples, evaluate the performance of the fully connected complex deep neural network model, and define the coefficient of determination. R 2 As an evaluation indicator: , In the formula, z The sample number. It is the first z Each sample at frequency point l The actual output vector, It is the first z Each sample at frequency point l The predicted output vector, It is the mean vector of all sample output vectors, considering the global mean. R 2 This involves concatenating all frequency points and all output dimensions into a one-dimensional vector and then performing a unified calculation to reflect the overall fitting effect. Additionally, it deliberately combines equipment operating conditions with a vibration source removal method to calculate the contribution of the vibration source to the target response's transmission path. The vibration source removal method includes setting the input data corresponding to a specific vibration source device to zero and calculating the change in the target response before and after the removal of the vibration source, in order to evaluate the contribution spectrum and contribution ranking of the vibration source to the target point. Taking three pieces of equipment on a ship as an example, namely equipment No. 1, equipment No. 2, and equipment No. 3, all of which are mounted on bases using vibration isolation devices. The experiment employed 50 vibration sensors and 2 hydrophones. One vibration acceleration sensor was placed on the base and on each machine foot of each piece of equipment, while the remaining vibration acceleration sensors were arranged on the hull to assist in selecting vibration source measurement points and analyzing vibration transmission characteristics. The sensor locations are distributed as follows: Figure 1 As shown.

[0020] Simultaneously, underwater noise testing was conducted using a single microphone in the far field to determine the far-field noise spectrum characteristics of the ship in a stationary state, providing a basis for determining the main frequency range for sound source identification. Target measurement points were placed at positions 1m, 10m, and 60m from the ship's engine room to test the vibration transmission patterns at the interfaces between the hull and internal equipment, analyzing the excitation vibration and noise response of the hull structure. Hydrophones were deployed in a highly directional direction outside the hull to collect sound field data, determining the noise spectrum characteristics of the ship in a stationary state, providing a basis for determining the main frequency range for sound source identification. The hydrophone deployment depth was maintained at approximately the same horizontal plane as the equipment location.

[0021] In this embodiment, the radiated noise outside the ship and the structural vibration data of the ship are collected simultaneously. The test conditions mainly include tests in which the equipment in the cabin is turned on individually and all of them when the ship is at sea. The test measures the dynamic data of each main excitation source of the equipment in different states being turned on individually and all of them.

[0022] Based on the coherence coefficient between any two measuring points, one of the measuring points is selected as the vibration source. This process is repeated to determine the candidate vibration source measuring points. Specifically, based on the candidate vibration source measuring points for each device, the recoherence coefficient from the candidate vibration source measuring points to the target point is calculated using a partial coherence recoherence analysis method. If the recoherence coefficient is >0.8, then all vibration source measuring points can be considered to reflect the response of the target point. Figure 2 Select one vibration source measurement point for each device (vibration source) m =3) and 2 vibration source measurement points ( m The spectrum of the re-coherence coefficient (=6).

[0023] When the vibration source m When =3, the re-coherence coefficient values ​​below 0.1 and in the 0.5~0.7 frequency band are... This indicates that an important transmission path has been missed. The recoherence coefficient is also low at other frequencies. In this case, the missed vibration source should be located and preferably selected. m =6 vibration sources (equipment 1 uses measuring points 1 and 2, 3 and 4; equipment 2 uses measuring points 3 and 4, 11 and 12; equipment 3 uses measuring points 3 and 4, 7 and 8), and then use the coherence coefficient again for testing until... It can be assumed that no important vibration sources have been overlooked.

[0024] In this embodiment, 15 operating conditions are preferred, which can form a sample size. .like Figure 3 The calculation yields a linear fitting model between the vibration source input and the target point response output. R 2 It can be seen that at more frequency points R 2 <0.9.

[0025] Using the method proposed in this invention, the number of vibration sources is selected. m =6, Number of working conditions r =6, number of sound pressure response n =3, select the low frequency band below 1kHz for analysis (frequency normalization processing).

[0026] Define a complex matrix neural network structure: 72-dimensional input, 48-dimensional output, one hidden layer containing 96 neurons. Introduce the ReLU activation function to consider moderate nonlinear expressive power and avoid the risk of overfitting caused by deep structures with limited samples.

[0027] During training, all input / output variables were Z-score normalized. The optimizer used was Adam, the initial learning rate was set to 5e-4, and the maximum number of training epochs was 200. An early stopping mechanism based on validation loss was introduced, whereby the mean squared error of the validation set was calculated after each epoch, and training was terminated if no improvement was observed for a preset number of consecutive epochs, while maintaining the network weights with the best validation performance.

[0028] Three prediction network models were established based on the method of this invention: based on single frequency (Δ) f =1Hz), based on frequency band △ f =5Hz, based on frequency Δ f =10Hz. The performance of the three prediction network models was compared using validation samples, including the amplitude determination coefficient calculated across the entire frequency band. R 2 The analysis included complex relative RMSE, deviation between predicted and actual average response values, and training time. As shown in Table 1, it was found that the accuracy was higher when Δf=1Hz, and subsequent analyses will primarily focus on the results calculated using this network model.

[0029] Table 1. Performance comparison of three prediction network models on the validation set. Select the operating condition where only device #2 is turned on, such as Figure 4 As shown, this method calculates the actual and predicted sound pressure amplitude and phase of the target point response across the entire frequency band. The trends in amplitude and phase variation between the predicted and actual values ​​are then observed to ensure consistency. Figure 5 , 6 The contribution spectrum cloud diagram and the contribution diagram of the maximum response frequency point are given. It can be seen that the method accurately identifies the contribution of the two vibration sources of only equipment #2 and successfully realizes the separation of cross-coupled mechanical vibration sources in the cabin.

[0030] Select the operating condition where three devices are turned on simultaneously, and the number of vibration sources. m =6. This method calculates the actual and predicted sound pressure amplitude and phase of the target point response across the entire frequency band. The trends in amplitude and phase changes between the predicted and actual values ​​are then observed to ensure consistency. Figure 7 The curves showing the actual and predicted sound pressure amplitude and phase changes of the target point response across the entire frequency band demonstrate excellent prediction accuracy. Figure 8 The amplitude error curves between the predicted and actual values ​​calculated by different network models at various major peak frequencies under this operating condition are given. It can be seen that Δ f =1Hz and Δ f At 5Hz, the prediction error is generally within 2dB, which meets the requirements for practical engineering applications.

[0031] In the constructed network prediction model, the contribution spectrum cloud map of the six vibration sources to the target point under this working condition is evaluated by combining the vibration source removal method, such as... Figure 9 As can be seen, each path has a significant contribution in the low-frequency band. This figure can show the contribution of any vibration source to the target response at any frequency. Figure 10 The contribution of a single vibration source device to the target point (located at the peak frequency) is calculated. When all three devices are turned on simultaneously, the contribution at the highest peak frequency is ranked as follows: 5#>6#>2#>1#>3#>4#.

[0032] As can be seen, the present invention can accurately identify the contribution of the vibration source of the cross-coupled mechanical equipment in the ship's cabin to the target point, and thus locate the key vibration source and its frequency range that affect the response of the target measurement point in the water, which can provide strong support for the precise control of the vibration source and its path in the ship's cabin.

[0033] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0034] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0035] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0036] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0037] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0038] If the aforementioned functions are implemented as 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 this invention, essentially, or the part that contributes to the prior art, or a portion of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0039] Furthermore, it should be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0040] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing device. The computing device can be a known general-purpose device. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0041] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for analyzing the transmission path of ship mechanical vibrations based on physical neural networks, characterized in that: Includes the following steps: S1) Each device is connected to the base through a vibration isolation device. The difference in displacement response at both ends of the vibration isolation device is obtained to construct the vibration source input matrix. S2) Obtain the sound pressure response values ​​in the radiated sound field under the working conditions of each device, and construct the target response matrix; S3) Based on the activation status of each device, obtain the number of operating conditions; S4) Establish the transfer function between the difference in displacement response and the sound pressure response value, and convert the transfer function in the selected frequency band into a transfer function matrix with the frequency point as the independent variable; S5) Using the source input matrix, target response matrix and frequency information as input, construct a fully connected complex deep neural network model that integrates physical information, establish a nonlinear mapping relationship between the source input matrix and the target response matrix, and output samples under different working conditions. S6) Validate the samples, evaluate the performance of the fully connected complex deep neural network model, and calculate the contribution of the vibration source to the transmission path of the target response by combining the equipment operating conditions and the vibration source removal method.

2. The method for analyzing the transmission path of ship mechanical vibration based on physical neural networks as described in claim 1, characterized in that: In step S1), the difference in displacement response is obtained by performing a second time-domain integral on the difference in vibration acceleration at both ends of the vibration isolation device.

3. The method for analyzing the transmission path of ship mechanical vibration based on physical neural networks as described in claim 1, characterized in that: In step S2), the sound pressure response value is the superposition of the product of the corresponding FRF and the excitation force during operation.

4. The method for analyzing the transmission path of ship mechanical vibration based on physical neural networks as described in claim 1, characterized in that: In step S3), the operating conditions include: individual equipment startup, partial equipment startup, and all equipment startup.

5. The method for analyzing the transmission path of ship mechanical vibration based on a physical neural network as described in claim 1, characterized in that: In step S4), the transfer function within the selected frequency band is equivalent to a transfer function matrix with frequency points as independent variables. This includes dividing the frequency domain interval into multiple frequency bands, calculating the transfer function of the corresponding frequency band, and then connecting them to form the transfer function of the entire frequency band. The value of the transfer function at the junction of adjacent frequency bands is taken as the arithmetic mean of the values ​​of the transfer function at the junction of adjacent frequency bands.

6. The method for analyzing the transmission path of ship mechanical vibration based on physical neural networks as described in claim 1, characterized in that: In step S5), a fully connected complex deep neural network model that integrates physical information is constructed based on the excitation-response physical information transmission path model and a multilayer perceptron complex deep neural network model. At each frequency point, a nonlinear mapping relationship is constructed between the source input matrix and the target response matrix. The frequency information is normalized and used as the network input encoding. The network structure includes an input layer, at least one hidden layer, and an output layer. The dimension of the input layer is determined by the number of sources and the number of frequency points within the frequency band. The dimension of the output layer is determined by the number of target response measurement points and the number of frequency points within the frequency band. The activation function is the ReLU function.

7. The method for analyzing the transmission path of ship mechanical vibration based on physical neural networks as described in claim 6, characterized in that: During the training of the fully connected complex deep neural network model, all input / output variables are normalized by Z-score. The optimizer is Adam, and an early stopping mechanism based on validation loss is introduced. That is, after each training round, the mean square error of the validation set is calculated, and training is terminated when the validation loss does not improve after a preset number of rounds, while maintaining the network weights with the best validation performance.

8. The method for analyzing the transmission path of ship mechanical vibration based on a physical neural network as described in claim 1, characterized in that: In step S5), the number of frequency bands and the number of frequency points within them are adjusted to expand the number of samples.

9. The method for analyzing the transmission path of ship mechanical vibration based on physical neural networks as described in claim 1, characterized in that: When evaluating the performance of a fully connected complex deep neural network model, the coefficient of determination is defined as an evaluation metric. It is calculated by concatenating all frequency points and all output dimensions into a one-dimensional vector to reflect the overall fitting effect.

10. The method for analyzing the transmission path of ship mechanical vibration based on a physical neural network as described in claim 1, characterized in that: The vibration source removal method includes: setting the input data corresponding to a certain vibration source device to zero, calculating the change in the target response before and after the vibration source is removed, so as to evaluate the contribution spectrum and contribution ranking of the vibration source to the target point.

Images