A Machine Learning-Based Numerical Mapping Method for Marine Propellers
By constructing numerical maps of marine propellers through machine learning, the problem of high design time and cost in existing marine propeller technologies has been solved, enabling efficient and accurate propeller design and reducing development costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA SHIP SCIENTIFIC RESEARCH CENTER
- Filing Date
- 2022-11-29
- Publication Date
- 2026-05-26
AI Technical Summary
Existing marine propeller design methods rely on physical model testing, resulting in high time and cost, low efficiency, and hindering the widespread application of graphical design methods.
A numerical map of marine propellers is constructed using machine learning methods. By building a sample dataset, performing viscous flow numerical calculations, and training a performance surrogate prediction model using machine learning models, efficient prediction of propeller performance parameters is achieved.
It achieves high-precision, low-cost propeller design, improves design efficiency and accuracy, and reduces map development costs.
Smart Images
Figure CN115906644B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of marine technology, and in particular to a method for constructing numerical maps of marine propellers based on machine learning. Background Technology
[0002] Propellers are crucial components of ships, and their design significantly impacts a ship's speed, safety, vibration, and noise performance. Currently, the main design methods for marine propellers include graphical design and circulation theory design. Graphical design is not only simple to operate but also yields relatively accurate hydrodynamic results, allowing designers to quickly and effectively obtain mature solutions with relatively stable performance, which is beneficial for meeting the requirements of rapid and intelligent propeller design.
[0003] However, the development of marine propeller patterns currently relies mainly on physical model testing methods, which requires the processing of propeller physical models, water tank tests, and test data processing. A mature propeller pattern typically includes dozens or even hundreds of propeller schemes. Therefore, constructing marine propeller patterns often requires a lot of time and cost, is inefficient, and is not conducive to the promotion and application of pattern design methods. Summary of the Invention
[0004] To address the aforementioned problems and technical requirements, this applicant proposes a machine learning-based method for constructing numerical maps of marine propellers. The technical solution of this application is as follows:
[0005] A machine learning-based method for constructing numerical maps of marine propellers, the method comprising:
[0006] Construct a sample dataset, which includes several sets of sample design parameters for propellers;
[0007] For each set of sample design parameters in the sample dataset, the performance parameters of the propeller with the sample design parameters at each sample infeed speed are calculated using the viscous flow numerical calculation model.
[0008] Using the design parameters and sample feed rate of each group of samples as inputs and the corresponding performance parameters as outputs, a performance proxy prediction model is obtained by training a machine learning model using the sample dataset.
[0009] By using a performance surrogate prediction model, the performance parameters corresponding to different propeller schemes within the range of map parameters are obtained, and a numerical map of marine propellers is constructed.
[0010] A further technical solution involves using a viscous flow numerical calculation model to calculate the propeller's performance parameters, including open-water hydrodynamic performance parameters and cavitation performance parameters. The trained performance surrogate prediction model then includes:
[0011] Using the design parameters and sample inflow rate of each sample as inputs and the corresponding open water hydrodynamic performance parameters as outputs, a hydrodynamic proxy prediction model is obtained by training a machine learning model using the sample dataset.
[0012] Using the design parameters and sample infeed rate of each sample as input and the corresponding cavitation performance parameters as output, a cavitation surrogate prediction model is obtained by training a machine learning model using the sample dataset.
[0013] The constructed hydrodynamic proxy prediction model is used to obtain the open-water hydrodynamic performance parameters corresponding to different propeller schemes within the range of the map parameters. The constructed cavitation proxy prediction model is used to obtain the cavitation performance parameters corresponding to different propeller schemes within the range of the map parameters. Thus, the constructed numerical map of marine propellers reflects the open-water hydrodynamic performance parameters and cavitation performance parameters corresponding to different propeller schemes within the range of the map parameters.
[0014] A further technical solution involves using a performance surrogate prediction model to obtain the performance parameters corresponding to different propeller schemes within the range of spectrogram parameters, including:
[0015] Determine the disk surface ratio and skew parameters, and discretize the pitch ratio within the range of the pattern parameters. Based on the set disk surface ratio, skew parameters and the value of each pitch ratio, a set of candidate design parameters is constructed.
[0016] The open-water hydrodynamic performance parameters of each set of candidate design parameters are predicted using a hydrodynamic proxy prediction model. Based on the open-water hydrodynamic performance parameters, the speed of the actual ship is predicted using the parameter matching method, and the speed performance and corresponding propeller diameter of each set of candidate design parameters are obtained.
[0017] Based on the rapid performance of each set of candidate design parameters and the corresponding propeller diameter, a set of candidate design parameters that meet the design requirements is selected as a propeller scheme, and the cavitation performance parameters of the propeller scheme are calculated using the cavitation proxy prediction model.
[0018] If the performance parameters of the propeller scheme meet the performance design requirements, output the propeller scheme; otherwise, adjust at least one of the disk ratio and skew parameters, and repeat the steps to determine the disk ratio and skew parameters.
[0019] A further technical solution involves selecting a set of candidate design parameters that meet the design requirements as a propeller design, including:
[0020] Alternatively, select a set of candidate design parameters that yields the best hydrodynamic efficiency in terms of high-speed performance as a propeller design, or select a set of candidate design parameters whose corresponding propeller diameter is the target diameter as a propeller design.
[0021] A further technical solution involves calculating the performance parameters of a propeller with the sample design parameters at different sample infeed speeds for each set of sample design parameters, including:
[0022] Within the infeed speed range corresponding to the sample design parameters, several fitting infeed speed points are selected according to the sampling interval, and the performance parameters of the propeller with sample design parameters at each fitting infeed speed point are calculated respectively.
[0023] Curve fitting is performed on the correspondence between several sets of fitting infeed points and performance parameters to obtain the performance parameter curves corresponding to the sample design parameters;
[0024] Within the advance speed range, several sample advance speeds are obtained by interpolation according to the working condition interval. The working condition interval is smaller than the sampling interval, and the number of sample advance speeds obtained is greater than the number of fitted advance speed points.
[0025] The performance parameters of the propeller with sample design parameters at each sample infeed speed are determined by the performance parameter curve.
[0026] A further technical solution involves using a sample dataset to train a machine learning model to obtain a performance proxy prediction model, including:
[0027] Multiple candidate models were trained using a sample dataset based on various machine learning models. The candidate model with the best forecast performance was selected as the performance proxy forecast model.
[0028] The further technical solution is that the machine learning model used is a non-linear machine learning regression model, and the machine learning model used includes random forest and feedforward neural network.
[0029] Its further technical solution includes methods for constructing sample datasets, including:
[0030] Based on the propeller design parameters of the parent propeller, at least one of the disk area ratio, pitch ratio, and skew parameters in the propeller design parameters is changed while keeping the other propeller design parameters unchanged, resulting in multiple sets of sample design parameters, and a sample dataset is constructed.
[0031] A further technical solution involves obtaining multiple sets of sample design parameters, including:
[0032] The parameters are expanded outward from 0.1Z as the center, and n1 uniformly distributed discrete points of disk surface ratio parameter are selected; n2 uniformly distributed discrete points of disk surface ratio parameter are selected in the range of 0.5 to 1.5; and n3 uniformly distributed discrete points of yaw parameter are selected in the range of 0 to 360 / Z degrees; where Z represents the number of propeller blades of the parent propeller, and n1, n2 and n3 all exceed the predetermined threshold.
[0033] The parameters of n1 disc ratio parameter discrete points, n2 disc ratio parameter discrete points and n3 skew parameter discrete points are combined, and combined with other propeller design parameters of the parent propeller, to obtain n1*n2*n3 sets of sample design parameters.
[0034] A further technical solution is that the method also includes:
[0035] By utilizing the propeller design parameters and performance parameters of the parent propeller, the model parameters of the viscous flow numerical calculation model that meet the calculation accuracy are obtained.
[0036] The beneficial technical effects of this application are:
[0037] This application discloses a method for constructing numerical maps of marine propellers based on machine learning. This method combines viscous flow numerical methods with machine learning models. CFD technology has high prediction efficiency and accuracy for performance parameters, while the machine learning model's superior nonlinear fitting capability far surpasses traditional regression methods. This allows the method to construct high-precision hydrodynamic and cavitation numerical maps of marine propellers without relying on a large number of physical model experiments. This enables rapid design and selection of marine propellers, with high numerical accuracy of propeller maps, reliable evaluation results, and significant engineering practical value. It helps to significantly reduce the development cost of propeller maps and improve the efficiency and accuracy of marine propeller map design. Attached Figure Description
[0038] Figure 1 This is a flowchart of a method for constructing a numerical map of a marine propeller according to one embodiment of this application.
[0039] Figure 2 This is a flowchart of a method for training a performance proxy prediction model in one embodiment of this application.
[0040] Figure 3 This is a schematic diagram illustrating how, in an example, the thrust coefficient and torque coefficient curves are obtained by fitting the thrust coefficient and torque coefficient at the fitted advance point, respectively, and then interpolation is performed according to the operating condition intervals to obtain the thrust coefficient and torque coefficient corresponding to each sample advance speed.
[0041] Figure 4 This is a flowchart illustrating the method for obtaining a propeller design and its corresponding performance parameters within the range of spectrogram parameters. Detailed Implementation
[0042] The specific embodiments of this application will be further described below with reference to the accompanying drawings.
[0043] This application discloses a machine learning-based method for constructing numerical maps of marine propellers. Please refer to [link / reference]. Figure 1 The flowchart shown illustrates that the method includes the following steps:
[0044] Step 1: Construct a sample dataset, which includes several sets of sample design parameters for propellers.
[0045] In one embodiment, a sample dataset is constructed based on the propeller design parameters of a parent propeller. In this application, the parent propeller refers to a propeller with excellent overall performance, selected according to actual needs. Then, based on the propeller design parameters of the parent propeller, at least one of the disk area ratio, pitch ratio, and skew parameter is changed while keeping other propeller design parameters unchanged, resulting in multiple sets of sample design parameters, thus constructing the sample dataset.
[0046] Compared to traditional propeller numerical plots, this application adds consideration to the skew parameter. The disk ratio, pitch ratio, and skew parameter have a significant impact on propeller performance parameters, while other propeller design parameters have a smaller impact on performance parameters. Therefore, this application keeps other propeller design parameters unchanged in order to minimize the amount of computation without sacrificing accuracy.
[0047] Based on the propeller design parameters of the parent propeller, a series of multi-dimensional expansion designs are implemented for disk area ratio, pitch ratio, and skew. The expansion range of propeller parameters is designed to cover the common design requirements of marine propellers. In one embodiment, parameters are expanded outward from 0.1Z, selecting n1 uniformly distributed discrete points for disk area ratio parameters; n2 uniformly distributed discrete points for disk area ratio parameters are selected within the range of 0.5 to 1.5; and n3 uniformly distributed discrete points for skew parameters are selected within the range of 0 to 360 / Z degrees. Here, Z represents the number of propeller blades of the parent propeller. n1, n2, and n3 all exceed a predetermined threshold, for example, all exceeding 5. By combining the parameters of the n1, n2, and n3 discrete points for disk area ratio parameters with other propeller design parameters of the parent propeller, n1*n2*n3 sets of sample design parameters can be obtained.
[0048] Step 2: For each set of sample design parameters in the sample dataset, use the viscous flow numerical calculation model to calculate the performance parameters of the propeller with the sample design parameters at each sample infeed speed.
[0049] In this invention, the performance parameters of the propeller are calculated using the viscous flow numerical method. In one embodiment, in order to improve the accuracy of the calculation results, the model parameters of the viscous flow numerical calculation model that meet the calculation accuracy are adjusted using the propeller design parameters and performance parameters of the parent propeller.
[0050] Then, the propeller performance parameters for each sample design parameter at each sample infeed velocity are calculated using a viscous flow numerical calculation model. The performance parameters in this application include open-water hydrodynamic performance parameters and cavitation performance parameters. The open-water hydrodynamic performance parameters include the thrust coefficient K. T and torque coefficient K Q The cavitation performance parameter is the minimum pressure coefficient of the blade surface.
[0051] For each propeller with different sample design parameters, if too few sample infeed rates are selected, the sample size will be too small, affecting the accuracy of subsequent machine learning. However, if too many sample infeed rates are selected, the calculation time using the viscous flow numerical calculation model will be too long.
[0052] Therefore, in order to balance computational load and sample size, in one embodiment, the method for calculating the performance parameters of a propeller with the sample design parameters at different sample infeed velocities for each set of sample design parameters includes:
[0053] (1) Select several fitting infeed points within the infeed range corresponding to the sample design parameters according to the sampling interval, and calculate the performance parameters of the propeller with sample design parameters at each fitting infeed point.
[0054] Different sample design parameters may correspond to different advance speed ranges. The standard is that the thrust coefficient should not be less than 0. That is, the thrust coefficient of a propeller with the sample design parameters is always greater than or equal to 0 in the corresponding advance speed range. The sampling interval is set according to the actual situation.
[0055] (2) Perform curve fitting on the correspondence between several sets of fitted infeed points and performance parameters to obtain the performance parameter curve corresponding to the sample design parameters. The performance parameter curve corresponding to the sample design parameters is a continuous curve of the performance parameters of the propeller with the sample design parameters at different infeed speeds. The curve fitting in this step can be performed using polynomial fitting.
[0056] As described above, in this application, the performance parameters include open-water hydrodynamic performance parameters and cavitation performance parameters. Therefore, the performance parameter curves obtained in this step include hydrodynamic curves and cavitation curves. The hydrodynamic curve is a continuous curve of the open-water hydrodynamic performance parameters of the propeller with the sample design parameters at different advance speeds, and the cavitation curve is a continuous curve of the cavitation performance parameters of the propeller with the sample design parameters at different advance speeds.
[0057] (3) Several sample velocities are obtained by interpolation within the velocity range according to the working condition interval. The working condition interval is smaller than the sampling interval. Therefore, the number of sample velocities obtained is greater than the number of fitted velocities.
[0058] (4) The performance parameters of the propeller with sample design parameters at each sample infeed speed are directly determined by the performance parameter curve. This eliminates the need to calculate the performance parameters for each sample infeed speed, but achieves the effect of expanding the sample size, which is beneficial for the training of subsequent machine learning models.
[0059] For example, please refer to Figure 3 In the example shown, 12 fitting infeed points were first selected according to a sampling interval ΔJ = 0.1, and the thrust coefficient K at each fitting infeed point was calculated. T and torque coefficient K Q Then, using the advance velocity J as a variable, the thrust coefficient K is obtained by fitting. T The thrust coefficient curve as the advance speed J changes, and the torque coefficient K obtained by fitting. Q The torque coefficient curve varies with the advance speed J. Then, interpolation is performed at operating intervals ΔJ = 0.025 to obtain 46 sample advance speeds. The thrust coefficient curve is used to obtain the thrust coefficient K corresponding to these 46 sample advance speeds. T The torque coefficient K corresponding to the feed speed of 46 samples was obtained using the torque coefficient curve. Q .
[0060] Step 3: Using the design parameters and sample feed rate of each sample as input and the corresponding performance parameters as output, a performance proxy prediction model is trained based on a machine learning model using the sample dataset.
[0061] This includes: using the design parameters and sample inrush velocity of each sample group as input, and the corresponding open-water hydrodynamic performance parameters as output, a hydrodynamic surrogate prediction model is trained using a machine learning model based on the sample dataset. Also, using the design parameters and sample inrush velocity of each sample group as input, and the corresponding cavitation performance parameters as output, a cavitation surrogate prediction model is trained using a machine learning model based on the sample dataset.
[0062] Regardless of whether a hydrodynamic surrogate prediction model or a cavitation surrogate prediction model is trained, multiple candidate models are first trained using sample datasets based on various different machine learning models. For example... Figure 2 Taking the training of P different machine learning models as an example. In one embodiment, the machine learning model used is a nonlinear machine learning regression model, and common machine learning models include random forests and feedforward neural networks. Then, the candidate model with the best prediction performance is selected as the performance surrogate prediction model. The prediction performance of the candidate model can be measured by the mean squared error (MSE) of the prediction, that is, the candidate model with the smallest MSE is selected as the performance surrogate prediction model.
[0063] Step 4: Using the performance surrogate prediction model, obtain the performance parameters corresponding to different propeller schemes within the range of the graph parameters, thus constructing the numerical graph of marine propellers. Each propeller scheme represents a set of propeller design parameters.
[0064] As described above, the trained performance surrogate prediction model includes a hydrodynamic surrogate prediction model and a cavitation surrogate prediction model. The hydrodynamic surrogate prediction model is used to obtain the open-water hydrodynamic performance parameters corresponding to different propeller schemes within the spectrum parameter range, and the cavitation surrogate prediction model is used to obtain the cavitation performance parameters corresponding to different propeller schemes within the spectrum parameter range. Therefore, the constructed marine propeller numerical spectrum reflects the open-water hydrodynamic performance parameters and cavitation performance parameters corresponding to different propeller schemes within the spectrum parameter range. Considering that the open-water hydrodynamic performance parameters of the propeller scheme are relatively more important performance parameters in the propeller design process, and the issue of operating point matching also needs to be considered, in one embodiment, the method for obtaining the performance parameters corresponding to different propeller schemes within the spectrum parameter range using the performance surrogate prediction model includes:
[0065] First, determine the disk surface ratio and skew parameters, and input their values as specified values. Then, discretize the pitch ratio within the range of the pattern parameters. The discretization density of this step reaches the density threshold, that is, the pitch ratio is densely discretized.
[0066] Based on the set disk ratio, the set skew parameter, and the discrete values of each pitch ratio, a set of candidate design parameters is constructed. The same disk ratio and skew parameter can be combined with different pitch ratios to obtain multiple sets of different candidate design parameters.
[0067] The open-water hydrodynamic performance parameters for each set of candidate design parameters are predicted using a hydrodynamic proxy prediction model, and then K is used based on these parameters. T / J 2 The parameter matching method is used to predict the speed of the actual ship, and the speed performance and corresponding propeller diameter of each set of candidate design parameters are obtained. The speed performance includes hydrodynamic efficiency, speed, speed and operating point.
[0068] Based on the rapid performance of each set of candidate design parameters and the corresponding propeller diameter, a set of candidate design parameters that meets the design requirements is selected as a propeller scheme. In one embodiment, the set of candidate design parameters with the best hydrodynamic efficiency in rapid performance is selected as a propeller scheme. Alternatively, in another embodiment, the required target diameter of the propeller is also input as a specified value, and a set of candidate design parameters with the corresponding propeller diameter as the target diameter is selected as a propeller scheme. After selecting a pitch ratio to obtain a propeller scheme, the cavitation performance parameters of the propeller scheme can be calculated using a cavitation surrogate prediction model.
[0069] If the performance parameters of the obtained propeller design meet the performance design requirements, the propeller design is output, along with its open-water hydrodynamic performance parameters, cavitation performance parameters, rapid propulsion performance, and corresponding propeller diameter. If the performance parameters of the obtained propeller design do not meet the performance design requirements, at least one of the disk ratio and skew parameters is adjusted, and the steps for determining the disk ratio and skew parameters are repeated until the propeller design is output. The performance design requirements indicate the requirements that the open-water hydrodynamic performance parameters and cavitation performance parameters must meet.
[0070] In practical applications, the methods described in this application can be implemented using programming languages such as C / Python.
[0071] The above descriptions are merely preferred embodiments of this application, and this application is not limited to the above embodiments. It is understood that other improvements and variations that can be directly derived or conceived by those skilled in the art without departing from the spirit and concept of this application should be considered to be included within the protection scope of this application.
Claims
1. A method for constructing a numerical map of a marine propeller based on machine learning, characterized by, The method includes: Construct a sample dataset, which includes several sets of sample design parameters for propellers; For each set of sample design parameters in the sample dataset, the performance parameters of the propeller with the sample design parameters at each sample inlet velocity are calculated using a viscous flow numerical calculation model. The performance parameters include open water hydrodynamic performance parameters and cavitation performance parameters. Using the design parameters and sample advance velocity of each sample as input and the corresponding performance parameters as output, a performance proxy prediction model is trained using the sample dataset based on a machine learning model. This includes training a hydrodynamic proxy prediction model using the design parameters and sample advance velocity of each sample as input and the corresponding open-water hydrodynamic performance parameters as output, and training a cavitation proxy prediction model using the sample dataset based on a machine learning model. The constructed hydrodynamic proxy prediction model is used to obtain the open-water hydrodynamic performance parameters corresponding to different propeller schemes within the range of map parameters, and the constructed cavitation proxy prediction model is used to obtain the cavitation performance parameters corresponding to different propeller schemes within the range of map parameters. Thus, the constructed marine propeller numerical map reflects the open-water hydrodynamic performance parameters and cavitation performance parameters corresponding to different propeller schemes within the range of map parameters. The disk ratio and skew parameters are determined, and the pitch ratio is discretized within the range of the spectrum parameters. A set of candidate design parameters is constructed based on the set disk ratio, skew parameters, and the value of each pitch ratio. The open-water hydrodynamic performance parameters of each set of candidate design parameters are predicted using the hydrodynamic proxy prediction model. Based on the open-water hydrodynamic performance parameters, the speed prediction of the actual ship is performed using the parameter matching method to obtain the speed performance and corresponding propeller diameter of each set of candidate design parameters. According to the speed performance and corresponding propeller diameter of each set of candidate design parameters, a set of candidate design parameters that meets the design requirements is selected as a propeller scheme, and the cavitation performance parameters of the propeller scheme are calculated using the cavitation proxy prediction model. When the performance parameters of the propeller scheme meet the performance design requirements, the propeller scheme is output; otherwise, at least one of the disk ratio and skew parameters is adjusted, and the steps of determining the disk ratio and skew parameters are repeated. A numerical spectrum of marine propellers is constructed.
2. The method of claim 1, wherein, Methods for selecting a set of candidate design parameters that meet the design requirements as a propeller scheme include: Alternatively, select a set of candidate design parameters that yields the best hydrodynamic efficiency in terms of high-speed performance as a propeller design, or select a set of candidate design parameters whose corresponding propeller diameter is the target diameter as a propeller design.
3. The method according to claim 1, characterized in that, For each set of sample design parameters, the method for calculating the performance parameters of the propeller with the sample design parameters at different sample infeed velocities includes: Within the infeed speed range corresponding to the sample design parameters, several fitting infeed speed points are selected according to the sampling interval, and the performance parameters of the propeller with the sample design parameters at each fitting infeed speed point are calculated respectively. Curve fitting is performed on the correspondence between several sets of fitting infeed points and performance parameters to obtain the performance parameter curves corresponding to the sample design parameters; Within the specified speed range, several sample speeds are obtained by interpolation at working condition intervals, where the working condition interval is smaller than the sampling interval, and the number of sample speeds obtained is greater than the number of fitted speed points. The performance parameters of the propeller with the sample design parameters at each sample advance speed are determined by the performance parameter curve.
4. The method according to claim 1, characterized in that, The methods for obtaining a performance agent prediction model by training a machine learning model using the aforementioned sample dataset include: Multiple candidate models are trained using the sample dataset based on various machine learning models. The candidate model with the best prediction performance is selected as the performance proxy prediction model.
5. The method according to claim 4, characterized in that, The machine learning model used is a non-linear machine learning regression model, which includes random forest and feedforward neural network.
6. The method according to claim 1, characterized in that, Methods for constructing sample datasets include: Based on the propeller design parameters of the parent propeller, at least one of the disk area ratio, pitch ratio, and skew parameter in the propeller design parameters is changed while keeping the other propeller design parameters unchanged, to obtain multiple sets of sample design parameters, and the sample dataset is constructed.
7. The method according to claim 6, characterized in that, Methods for obtaining design parameters for multiple sets of samples include: The parameters are expanded outward from 0.1Z as the center, and n1 uniformly distributed discrete points of disk surface ratio parameter are selected; n2 uniformly distributed discrete points of disk surface ratio parameter are selected in the range of 0.5~1.5; and n3 uniformly distributed discrete points of yaw parameter are selected in the range of 0~360 / Z degrees; where Z represents the number of propeller blades of the parent propeller, and n1, n2 and n3 all exceed a predetermined threshold. By combining the discrete points of the n1 disk aspect ratio parameter, the n2 disk aspect ratio parameter, and the n3 skew parameter, and in conjunction with other propeller design parameters of the parent propeller, n1 is obtained. n2 n3 group sample design parameters.
8. The method according to claim 6, characterized in that, The method further includes: By using the propeller design parameters and performance parameters of the parent propeller, the model parameters of the viscous flow numerical calculation model that meet the calculation accuracy are adjusted.