Ship performance coefficient estimation system and method using a convolutional neural network

Abstract

The present invention relates to an apparatus and a method for estimating various performance coefficients indicating the performance of a ship. According to the present invention, in the conventional method of predicting performance using the main specifications of a ship as input, in most cases, performance was predicted using only variables representing linearity, so there was a limitation in that it was difficult to reflect minute shape changes. To solve the problems of the prior art statistical-based ship performance estimation method, deep learning and convolutional neural network (CNN) technologies are utilized. Offset data representing the shape information (offset cross-section information) of the ship is converted into image data, and the main specification variables, speed variables, and valve shape information of the ship are matched to the image data, and learning is performed by applying 2D CNN or 3D CNN. In this process, dropout and regularization (L2 regularization) are added to prevent overfitting as necessary, and a convolutional neural network configured to output a desired performance variable with arbitrary linear offset data, main specification variables, bow valve variables, and speed variables as input is utilized to provide a ship performance coefficient estimation system and method.

Description

Technical Field

[0001] The present invention relates to an apparatus and a method for estimating various performance coefficients indicating the performance of a ship, such as Residual Resistance (Cr). More specifically, in conventional ship design, numerical analysis and model tests have to be repeatedly executed until the desired performance is obtained, which has the problem of increasing time and cost. To solve the problems of the conventional ship design method, a ship performance coefficient estimation system and method using a convolutional neural network are configured to be able to predict the performance of a ship by a statistical method using the specifications of the ship without performing numerical analysis or model tests at the initial design stage of the ship.

[0002] In addition, in order to perform more efficient ship design as described above, a method for predicting the performance of a ship by a statistical method using the specifications of the ship without performing numerical analysis or model tests is required. However, in the conventional method, when predicting the performance with the main specifications of the ship as input, in most cases, only variables representing linearity are used to predict the performance, so it is difficult to reflect minute shape changes. The present invention aims to solve the problems in the conventional statistical-based ship performance estimation method. Therefore, the present invention relates to a ship performance coefficient estimation system and method using a convolutional neural network, which is configured to derive an estimation formula for estimating the relationship between the linear cross-sectional shape and the performance coefficient by utilizing deep learning and convolutional neural network (CNN) technologies and considering minute linear changes.

[0003] Furthermore, the present invention utilizes deep learning and convolutional neural network (CNN) technology as described above to convert offset data representing ship shape information (offset cross-section information) into image data in order to estimate the relationship between the linear cross-sectional shape and the coefficient of performance. The main specification variables, speed variables, and valve shape information of the ship are matched to the image data, and learning is performed by applying 2D CNN or 3D CNN. At this time, if necessary, dropout and regularization (L2 regularization) are added to prevent overfitting. By constructing an artificial neural network model configured to output desired performance variables such as the residual resistance coefficient (Cr), which is a resistance performance index, the wake fraction (w), which is a propulsion-related index necessary for power estimation, and the thrust deduction coefficient (t), etc., by inputting arbitrary linear offset data, main specification variables, bow valve variables, and speed variables, not only can results considering fine linear changes be derived, but also the problem of overfitting, which is a major drawback of machine learning models, can be effectively avoided, and the position information regarding the cross-sectional shape of the ship can be prevented from being lost. The present invention relates to a ship performance coefficient estimation system and method using a convolutional neural network.

Background Art

[0004] Generally, during the design of a ship, operations such as estimating various performance coefficients related to the ship, for example, residual resistance (Cr), are required. To achieve this, conventionally, numerical analysis and model tests have been repeatedly performed until the desired performance is obtained, resulting in a problem that the time and cost required for estimating the performance coefficients significantly increase.

[0005] Consequently, recently, a ship performance estimation method has been used that is configured to predict the performance of a ship by using statistical methods without performing numerical analysis or model tests during the initial design of the ship.

[0006] Here, as described above, various performance coefficients indicating the performance of a ship are calculated. As an example of the prior art regarding an apparatus and method for evaluating the performance of a ship, first, for example, there is the "Method for Estimating Ship Resistance Performance Using Regression Analysis" described in Korean Registered Patent Publication No. 10-2158738.

[0007] More specifically, the above-mentioned Korean Registered Patent Publication No. 10-2158738 is a ship resistance performance estimation method using regression analysis, configured such that the process of estimating resistance performance using the main specifications of a ship is executed by a computer or dedicated hardware. The above process includes a database construction stage in which a process of collecting model test results related to the main specifications of a ship and constructing a database (DB) is executed; an interval classification stage in which a process of classifying intervals for regression analysis is executed based on the database constructed in the database construction stage; a variable selection stage in which a process of selecting candidates for the dependent variable and independent variables is executed for each interval classified in the interval classification stage; a regression equation definition stage in which a correlation analysis is performed for each interval for each selected candidate for the dependent variable and independent variable, and a process of defining the form of the regression equation is executed; a regression equation determination stage in which a non-linear regression analysis is performed using the form of the regression equation defined in the regression equation definition stage, and a process of completing the final regression equation by calculating the regression coefficients is executed; and a resistance performance estimation stage in which a process of calculating the residual resistance coefficient and estimating the resistance performance is executed using the final regression equation determined in the regression equation determination stage. The above interval classification stage is configured such that a process of classifying intervals for regression analysis is executed using histograms of the block coefficient (Cb) and the coefficient of fineness, based on the average values of the block coefficient and the coefficient of fineness. Specifically, intervals are classified into those where both the block coefficient and the coefficient of fineness are less than the average value, those where the block coefficient is less than the average value and the coefficient of fineness is greater than or equal to the average value, those where the block coefficient is greater than or equal to the average value and the coefficient of fineness is less than the average value, and those where both the block coefficient and the coefficient of fineness are greater than or equal to the average value. By deriving a regression analysis equation using a limited number of main ship specifications at the basic design stage of a ship, it becomes possible to calculate the residual resistance coefficient of a ship more quickly and easily than conventional methods and estimate the resistance performance.The present invention relates to a ship resistance performance estimation method using regression analysis, which is configured to solve the problems of a prior art Taylor chart-based ship residual resistance coefficient and resistance performance estimation method, where it was difficult to calculate accurate values for areas not shown on the chart.

[0008] Also, as another example of the prior art regarding an apparatus and method for calculating various performance coefficients indicating the performance of a ship and evaluating the performance of a ship, for example, there is "Ship Motion Performance Prediction System and Method, and a Computer-readable Recording Medium Recorded with a Computer Program for Executing the Method on a Computer" described in Korean Registered Patent Publication No. 10-2126838.

[0009] More specifically, the above-mentioned Korean Registered Patent Publication No. 10-2126838 relates to a system for predicting the motion performance of a ship using a ship motion performance prediction apparatus for predicting the motion performance of a ship and an apparatus database in which various data related to the ship motion performance prediction apparatus are stored. The apparatus database stores a two-dimensional motion response function obtained by measuring or analyzing the response of a ship according to frequency and direction, and two-dimensional environmental spectrum information representing the distribution of environmental factors considering the characteristics of the sea area based on frequency and direction at a specific position. To predict the motion performance of a ship, when the navigation route of the ship is determined and the position of the ship is specified on the navigation route, the ship motion performance prediction apparatus predicts the motion performance of the ship using the product of the two-dimensional environmental spectrum information representing the distribution of environmental factors at the specified position of the ship and the two-dimensional motion response function. The two-dimensional environmental spectrum information is composed of a two-dimensional spectrum considering the characteristics of the sea area, thereby improving the motion performance analysis method using a one-dimensional spectrum based on conventional frequency components. Thus, it relates to a ship motion performance prediction system and method configured to more accurately predict the motion performance of a ship by utilizing a two-dimensional spectrum that simultaneously considers the frequency and directivity that change according to the navigation route of the ship.

[0010] As described above, conventionally, various devices and methods for calculating various performance coefficients indicating the performance of a ship and evaluating the performance of a ship have been proposed. However, the content of these prior arts has the following limitations.

[0011] That is, generally, conventional ship performance coefficient estimation methods extract some linear variables indicating the characteristics of the ship shape, create an estimation formula through regression analysis, or use the Holtrop and Mennen method to predict the ship performance coefficient with the main specifications of the ship as input. However, since these methods perform predictions only based on variables representing linearity, there is a limitation in that it is difficult to reflect minute shape changes.

[0012] Furthermore, recently, a method of estimating the resistance coefficient and the self-propulsion coefficient using a simple neural network model, MLP (Multi-Layer Perceptron), has been used. However, this uses a plurality of linear variables (e.g., Length Between Perpendicular (LBP), Beam, Draft, Volume, Block Coefficient (Cb), Longitudinal Center of Buoyancy (LCB), etc.) as input values, arranges a plurality of hidden layers in the middle, and sets the regression target variables (e.g., residuary resistance coefficient, wake fraction, thrust deduction coefficient, etc.) in the output layer to construct a model through learning. However, in this method, since only some shape variables are used to reflect the overall linear shape, there is a limitation in that it is not possible to derive results considering minute linear changes.

[0013] Furthermore, when using a simple neural network model in the prior art as described above, there is also a problem that it is inevitable that the position information regarding the cross-sectional shape of the ship is lost.

[0014] Therefore, in order to solve the limitations of the above-described conventional statistics-based ship performance coefficient estimation method, for example, based on artificial neural network (ANN) technology using machine learning or deep learning, by utilizing the main specifications of the ship and the ship's shape information (offset cross-section information), and estimating performance coefficients such as residual resistance, it is desirable to propose a ship performance coefficient estimation system and method with a new configuration that can calculate performance coefficients reflecting fine linear changes. However, at present, no device or method that satisfies all of these requirements has been proposed yet.

Prior Art Documents

Patent Documents

[0015]

Patent Document 1

[0016]

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0017] The present invention aims to solve the problems of the prior art as described above. Therefore, the object of the present invention is to solve the problem of the prior art ship design method that in ship design, it is necessary to repeatedly perform numerical analysis and model tests to obtain desired performance, and accordingly, time and costs increase. For this reason, the present invention proposes a ship performance coefficient estimation system and method using a Convolutional Neural Network (CNN) that can predict the performance of a ship by a statistical method using the specifications of the ship without performing numerical analysis or model tests at the initial design stage of the ship.

[0018] Moreover, another object of the present invention is that in order to perform more efficient ship design as described above, a method for predicting the performance of a ship by a statistical method using the specifications of the ship without performing numerical analysis or model tests is required. However, in the conventional method, when predicting the performance with the main specifications of the ship as input, in most cases, the performance was predicted only with variables representing linearity, so it was difficult to reflect minute shape changes. The present invention proposes a ship performance coefficient estimation system and method using a convolutional neural network having a configuration capable of deriving an estimation formula for estimating the relationship between the linear cross-sectional shape and the performance coefficient in consideration of minute linear changes by utilizing deep learning and convolutional neural network (CNN) technologies.

[0019] Furthermore, another object of the present invention is to propose a system and method configured as follows to utilize deep learning and convolutional neural network (CNN) technologies to estimate the relationship between the linear cross-sectional shape and the coefficient of performance. First, offset data indicating the shape information (offset cross-sectional information) of a ship is converted into image data, and the main specification variables, speed variables, and valve shape information of the ship are matched to this image data. Then, learning is performed by applying a two-dimensional CNN or a three-dimensional CNN. At this time, dropout and regularization (L2 regularization) for preventing overfitting are added as necessary. As a result, an artificial neural network model is constructed that outputs desired performance variables such as the residual resistance coefficient (Cr), which is an index of resistance performance, the wake fraction (w), which is a propulsion-related index necessary for power estimation, and the thrust deduction coefficient (t), etc., with arbitrary linear offset data, main specification variables, bow valve variables, and speed variables as inputs. This is configured not only to derive results considering fine linear changes, but also to effectively avoid the problem of overfitting, which is a major drawback of machine learning models, and to prevent the loss of position information regarding the cross-sectional shape of the ship. It is to propose a ship performance coefficient estimation system and method using the convolutional neural network constructed as described above.

Means for Solving the Problems

[0020] In order to achieve the above object, according to the present invention, in a ship performance coefficient estimation system, there is provided a data collection unit configured to execute a process of inputting various information including the specifications and shape information of a ship and constructing a database; a data processing unit configured to execute a process of imaging the shape information of the ship collected through the data collection unit; a performance coefficient estimation unit configured to match the offset data imaged through the data processing unit with the information collected through the data collection unit, execute learning using an artificial neural network, and execute a process of estimating the performance coefficient of the ship based on the learning result; and an output unit configured to execute a process of outputting the overall processing process and processing results of the estimation system including the data collection unit, the data processing unit, and the performance coefficient estimation unit. A ship performance coefficient estimation system is provided, characterized in that it includes the above components.

[0021] Here, the data collection unit is configured to collect various information including main specification information such as the length (Length; L), breadth (Breadth; B), draft (Draft; T), L / B, B / T, L / T of the ship, speed information including the Froude number (Fn), bulb shape information including the bulb length (Bulb Length; BulbL) and bulb height (Bulb Height; BulbH), ship shape information including offset cross-section information, and various performance variables including the residual resistance coefficient (Residual Resistance Coefficient; Cr), wake coefficient (Wake Coefficient; w), and thrust deduction coefficient (Thrust Deduction Coefficient; t) of the corresponding ship, and save it in the form of a database, thereby executing a process of constructing a database related to various ships.

[0022] In addition, the data processing unit executes an image conversion stage in which two-dimensional data (y, z) regarding each station included in the offset cross-section information collected through the data collection unit is converted into image data based on a predetermined setting; and a ship shape image data generation stage in which the ship shape information data converted into an image through the image conversion stage is added with the main specification information (L / B, B / T, L / T), the speed information (Fn), and the valve shape information (valve length, valve height) to generate ship shape image data in which the main information regarding the corresponding ship is reflected. The data processing unit is configured to execute a process including these stages.

[0023] In addition, the image conversion stage is configured to execute a process of converting the offset cross-section information into an image by using a binary method in which a portion corresponding to the hull boundary is set to 1 and the rest is set to 0, or an SDF (Signed Distance Function) method in which the hull boundary is set to 0 and the inside of the hull is represented by the shortest distance from a boundary having a negative sign and the outside of the hull is represented by a positive sign.

[0024] Furthermore, the performance coefficient estimation unit is configured to execute learning on the ship shape image data generated through the data processing unit by using a pre-constructed Convolutional Neural Network (CNN) model, and execute a process of estimating a predetermined performance variable from an input value based on the learning result.

[0025] Here, the above CNN model takes as input the offset (cross-sectional image) data of the ship to be estimated, main specification information (L / B, B / T, L / T), speed information (Fn), and bulb shape information (bulb length (BulbL), bulb height (BulbH)), treats the entire hull as an n×n image with m channels, and is configured to execute a process that outputs preset performance variables. It is composed of a two-dimensional CNN model, and in order to prevent overfitting, it is further configured to include a preset dropout and regularization (L2 regularization) process.

[0026] Alternatively, the above CNN model takes as input the offset (cross-sectional image) data of the ship to be estimated, main specification information (L / B, B / T, L / T), speed information (Fn), and bulb shape information (bulb length (BulbL), bulb height (BulbH)), treats the entire hull as an n×n×m image with one channel, and is configured to execute a process that outputs preset performance variables. It is composed of a three-dimensional CNN model, and in order to prevent overfitting, it is further configured to include a preset dropout and regularization (L2 regularization) process.

[0027] Also, the above output unit is configured to include display means including a monitor or a display for visually displaying various information including the overall operation, processing process, and processing results of the above estimation system.

[0028] Furthermore, the above estimation system is configured to include a communication unit that communicates with external devices including a server and a user terminal in at least one of wireless or wired communication manners and executes a process of transmitting and receiving various data; and a control unit that is configured to execute a process of controlling the overall operation of the above estimation system.

[0029] Furthermore, according to the present invention, in a ship performance coefficient estimation method configured using the ship performance coefficient estimation system described above, a data collection stage in which various information regarding a ship is collected to construct a database is executed through the data collection unit of the ship performance coefficient estimation system; a data processing stage in which the offset data of the ship collected in the data collection stage is imaged is executed through the data processing unit of the ship performance coefficient estimation system; a performance coefficient estimation stage in which the offset data imaged through the data processing stage is matched with the information collected through the data collection stage, learning using an artificial neural network is performed, and the performance coefficient of the ship is estimated based on the learning result is executed through the performance coefficient estimation unit of the ship performance coefficient estimation system; and an output stage in which the overall processing process and processing results of the data collection stage, the data processing stage, and the performance coefficient estimation stage are output is executed through the output unit of the ship performance coefficient estimation system. A ship performance coefficient estimation method is provided, which is characterized by being configured to include these stages.

[0030] Also, according to the present invention, in a ship performance coefficient estimation service providing system, a user terminal for each user to request and receive a ship performance coefficient estimation service; a server configured to execute a process of providing a ship performance coefficient estimation service based on data received through each user terminal is included. The server is configured to include the ship performance coefficient estimation system described above, so that each user can perform a remote performance coefficient estimation operation, and an online-based service can be realized in which a large number of users can perform a performance coefficient estimation operation simultaneously. A ship performance coefficient estimation service providing system is provided, which is characterized by being configured in this way.

[0031] Here, the user terminal is configured by installing a dedicated program that interacts with the ship performance coefficient estimation system on an information processing device including a PC or a notebook, or by installing a dedicated application program that interacts with the ship performance coefficient estimation system on a personal mobile terminal device including a smartphone or a tablet PC.

Effect of the Invention

[0032] As described above, according to the present invention, offset data representing ship shape information (offset cross-section information) is converted into image data, and major specification variables, speed variables, and valve shape information of the ship are matched to the image data, and learning is performed by applying a two-dimensional CNN or a three-dimensional CNN. Then, by adding dropout and L2 regularization to prevent overfitting as needed, with arbitrary linear offset data, major specification variables, bow valve variables, and speed variables as inputs, for example, the residual resistance coefficient (Cr), which is a resistance performance index, and the wake coefficient (w), which is a propulsion-related index necessary for power estimation, a ship performance coefficient estimation system and method using a convolutional neural network configured to output desired performance variables such as the thrust reduction coefficient (t) are provided. As a result, it becomes possible to derive a result considering fine linear changes, and at the same time, effectively avoid the problem of overfitting, which is a major drawback of the machine learning model, and prevent the loss of position information regarding the cross-sectional shape of the ship.

[0033] In addition, according to the present invention, there is provided a ship performance coefficient estimation system and method using a convolutional neural network, which is configured to derive an estimation formula for estimating the relationship between a linear cross-sectional shape and a performance coefficient by utilizing deep learning and convolutional neural network (CNN) technologies as described above, taking into account fine linear changes. Thereby, in the conventional method, when predicting performance by inputting the main specifications of a ship, since it was based only on variables representing linearity, it was difficult to reflect fine shape changes, and the problem of the conventional statistical-based ship performance estimation method can be solved.

[0034] In addition, according to the present invention, there is provided a ship performance coefficient estimation system and method using a convolutional neural network, which is configured to be able to predict the performance of a ship using the specifications of the ship without performing numerical analysis or model tests at the initial design stage of the ship as described above. Thereby, since there has been no proposed method for predicting ship performance by a statistical method using the specifications of a ship without performing numerical analysis or model tests, there has been a problem that numerical analysis or model tests have to be repeatedly performed until the desired performance is obtained, and the time and cost involved increase, and the problem of the conventional ship design method can be solved.

Brief Description of the Drawings

[0035]

Figure 1

[0036]

Figure 2

[0037]

Figure 3

[0038]

Figure 4

[0039]

Figure 5

[0040]

Figure 6

[0041]

Figure 7

[0042]

Figure 8

[0043]

Figure 9

[0044]

Figure 10

MODE FOR CARRYING OUT THE INVENTION

[0045] Hereinafter, with reference to the accompanying drawings, a specific embodiment of a ship performance coefficient estimation system and method using a convolutional neural network according to the present invention will be described.

[0046] It should be noted here that the content described below is only one embodiment for implementing the present invention, and the present invention is not limited only to the content of the embodiment described below.

[0047] Also, in the following description of the embodiments of the present invention, for parts that are the same as or similar to the content of the prior art, and parts that are judged to be easily understood and implemented at the level of those skilled in the art, detailed descriptions are omitted in order to simplify the description. It should also be noted.

[0048] That is, as will be described later, the present invention solves the problem of the prior art ship design method in which numerical analysis and model tests have to be repeated in order to obtain desired performance in ship design, and the time and cost increase accordingly. In order to achieve this, a ship performance coefficient estimation system and method using a convolutional neural network that can predict the performance of a ship by statistical methods using the ship specifications without performing numerical analysis or model tests during the initial design of the ship.

[0049] Furthermore, as described below, in order to perform more efficient ship design, there is a need for a method to predict ship performance using statistical methods based on ship specifications without performing numerical analysis or model tests. However, in the conventional method, when predicting performance with the main specifications of the ship as input, the performance was predicted only based on variables representing linearity, so there was a limitation in that it was difficult to reflect minute shape changes. In order to solve the problems of such a conventional statistical-based ship performance estimation method, by utilizing deep learning and convolutional neural network (CNN) technology, it is possible to derive an estimation formula for estimating the relationship between the linear cross-sectional shape and the performance coefficient while considering minute linear changes. The present invention relates to a ship performance coefficient estimation system and method using a convolutional neural network having a configuration capable of doing so.

[0050] Furthermore, as described below, the present invention utilizes deep learning and convolutional neural network (CNN) technology and is configured as follows to estimate the relationship between the linear cross-sectional shape and the coefficient of performance. Offset data representing the shape information (offset cross-sectional information) of a ship is converted into image data, and the main specification variables, speed variables, and valve shape information of the ship are matched to the image data, and learning is performed by applying a two-dimensional CNN or a three-dimensional CNN. At this time, dropout and regularization (L2 regularization) for preventing overfitting are added as necessary. As a result, an artificial neural network model is constructed that is configured to output desired performance variables such as the residual resistance coefficient (Cr), which is a resistance performance index, the wake fraction (w), which is a propulsion-related index required for power estimation, and the thrust deduction coefficient (t), etc., with arbitrary linear offset data, main specification variables, bow valve variables, and speed variables as inputs. With this configuration, it is possible to derive results considering fine linear changes, effectively avoid the problem of overfitting, which is a major drawback of machine learning models, and prevent the loss of position information regarding the cross-sectional shape of the ship. The present invention relates to a ship performance coefficient estimation system and method using a convolutional neural network configured in this way.

[0051] Subsequently, with reference to the drawings, the specific content of a ship performance coefficient estimation system and method using a convolutional neural network (CNN) according to the present invention will be described.

[0052] More specifically, first, refer to FIG. 1. FIG. 1 is a block diagram schematically showing the overall configuration of a ship performance coefficient estimation system (10) using a convolutional neural network according to an embodiment of the present invention.

[0053] As shown in FIG. 1, a ship performance coefficient estimation system (10) using a convolutional neural network according to an embodiment of the present invention mainly includes a data collection unit (11) configured to execute a process of constructing a database by inputting various information including ship specifications and shape information, a data processing unit (12) configured to execute a process of imaging the shape information of the ship collected through the data collection unit (11), a performance coefficient estimation unit (13) configured to match the offset data imaged through the data processing unit (12) with the information collected through the data collection unit (11), execute learning through an artificial neural network, and execute a process of estimating the performance coefficient of the ship based on the learning result, and an output unit (14) configured to execute a process of outputting the overall processing process and processing results of the system (10) including the data collection unit (11), the data processing unit (12), and the performance coefficient estimation unit (13), a communication unit (15) configured to communicate with external devices such as servers and user terminals and execute a process of transmitting and receiving various data, and a control unit (16) configured to execute a process of controlling the overall operations of the above-mentioned units and the system (10).

[0054] More specifically, the above-mentioned data collection unit (11) collects various information including the main specification information of the ship (length (Length; L), breadth (Breadth; B), draft (Draft; T), L / B, B / T, L / T, speed information (Froude number; Fn)), bow bulb shape information (bulb length (Bulb Length; BulbL), bulb height (Bulb Height; BulbH)), ship shape information (offset cross-section information), and various performance variables of the corresponding ship (residual resistance coefficient (Residual Resistance Coefficient; Cr), wake coefficient (Wake Coefficient; w), thrust deduction coefficient (Thrust Deduction Coefficient; t), etc.), and constructs a database for various ships by saving it in a database format.

[0055] Here, L (length), B (width), T (draft), and Fn (Froude number) are data provided as the main information of the ship, and the offset cross-section information is usually defined as the (y, z) coordinates at each station. Also, the target variable such as the residual resistance coefficient (Cr) is data obtained as the result of a model test.

[0056] Also, the above-described data processing unit (12) performs a process of converting the offset cross-section information collected through the data collection unit (11) into an image. The method of imaging this offset data is as follows.

[0057] Generally, the offset information of a ship is provided in the form of (y, z) coordinates for the cross-sectional shape at each station in the longitudinal direction. In the present invention, the (y, z) coordinates of each cross-section are made dimensionless by the width (B) and draft (T) of the ship, and are converted into (n×n) pixel image data for use.

[0058] Furthermore, as a method of converting the offset information as described above into image data, a binary method in which a portion corresponding to the hull boundary is set to 1 and the rest is set to 0, or an SDF (Signed Distance Function) method in which the hull boundary is set to 0 and the inside of the hull is represented by the shortest distance from a boundary having a negative sign and the outside of the hull has a positive sign can be used for configuration.

[0059] That is, referring to FIGS. 2 and 3, FIGS. 2 and 3 are drawings showing examples of the results of converting offset information into an image using the binary method and the SDF (Signed Distance Function) method, respectively.

[0060] As shown in FIGS. 2 and 3, the binary method is a method of simply setting a line to 1 and the rest to 0 when imaging the offset, which is simple but has low performance. On the other hand, it can be confirmed that the SDF shows better performance.

[0061] Therefore, as described above, it is possible to convert the offset of the hull into n×n image data at m stations. Further, in consideration of the shape of the bow valve, variables L and T obtained by dimensionlessizing the length and height of the valve are additionally applied to the converted image data.

[0062] Here, regarding more specific contents of the above-mentioned binary method and SDF method, reference is made to the prior art image conversion method, which is obvious to those skilled in the art. Therefore, it should be noted that in the present invention, the detailed description of the contents that can be easily understood and implemented by those skilled in the art through the prior art documents as described above is omitted.

[0063] As described above, in the present invention, among the overall shape of the ship, the part from the after perpendicular (AP) to the forward perpendicular (FP) is divided into m cross-sections (stations), and the two-dimensional data (y, z) of each cross-section is converted into image data using the binary method or the SDF method. At this time, since the converted cross-sectional image information is dimensionless data, linear variables (L / B, B / T, L / T), speed information (Fn), and valve shape information (valve length, valve height) are added.

[0064] In addition, after converting the cross-sectional shape into image data through the data processing unit (12) as described above, a process of applying CNN (Convolutional Neural Network) technology and estimating performance variables through the pattern of the cross-section can be configured to be executed through the performance coefficient estimation unit (13).

[0065] That is, the present invention is not a general method of simply inputting linear shape information in the form of (x, y, z) and learning through artificial intelligence. Instead, as described above, it is characterized by a method of converting offset data into an image, matching the converted image with performance variables, and performing learning using a CNN model. And based on the learning results, a series of processes for estimating desired performance variables from arbitrary input values are configured to be executed.

[0066] To achieve this, the above-mentioned CNN model has a structure that includes multiple CNN layers and dense layers. Also, in order to prevent overfitting, dropout technology, L2 regularization technology, and the Keras functional API for applying multi-input are used, and it can be configured as a 2D CNN or a 3D CNN.

[0067] More specifically, refer to FIGS. 4 and 5. FIG. 4 is a drawing schematically showing the overall configuration of a 2D CNN model, and FIG. 5 is a drawing schematically showing the overall configuration of a 3D CNN model.

[0068] As shown in FIGS. 4 and 5, an artificial neural network is composed of multiple layers connecting an input layer and an output layer. Inside these layers, through linear combination and non-linear transformation, it is a kind of black box model with coefficients that can effectively represent the non-linear relationship between the input and the output. This is not an equation inferred from a specific physical relationship, but a linear combination model that minimizes the error of the data itself.

[0069] Here, the artificial neural network structure applied to the embodiments of the present invention can apply both 2D CNN and 3D CNN. First, 2D CNN is a method generally used when learning a 2D image with three RGB channels. Taking an n×n×m tensor as the input, it goes through a plurality of 2D Convolution layers and 3D maxpooling layers, and then through a Flatten process that converts this into 1D data.

[0070] Furthermore, 3D CNN is a more suitable method for shapes with volume. The hull data is defined as a 3D tensor n×n×m×1 with one channel, and it has a structure that goes through multiple 3D Convolution layers, 3D maxpooling layers, and a Flatten process. In this way, a model can be constructed in the same way as a 2D CNN.

[0071] Here, the number of layers and the number of filters to be used should be determined during the learning process according to the characteristics of the problem, and there is no specific law.

[0072] Furthermore, in order to prevent overfitting, it is necessary to add dropout or regularization (L2 regularization) in the middle of the model. The values of these hyperparameters are determined through tuning work to construct the final model.

[0073] Here, regarding the more specific content of the above-mentioned 2D CNN and 3D CNN, refer to the content related to the prior art artificial neural network, which is obvious to those skilled in the art. Therefore, in the present invention, for the content that can be easily understood and implemented by those skilled in the art through the prior art documents as described above, it should be noted that the detailed description thereof is omitted.

[0074] Therefore, the CNN model completed through learning as described above will have the function of receiving any linear offset data, main specification variables, bow bulb variables, and speed variables as inputs and outputting pre-set performance variables.

[0075] That is, referring to Figure 6, Figure 6 is a drawing schematically showing the overall configuration of the CNN model applied to the ship performance coefficient estimation system (10) using the convolutional neural network according to an embodiment of the present invention.

[0076] Here, in the embodiment shown in FIG. 6, a configuration example of a residual resistance coefficient (Cr) prediction model using a 2D CNN is shown.

[0077] As shown in FIG. 6, in the present invention, in order to apply a two-dimensional CNN to hull data, the entire hull is treated as an n×n image with m channels, and a model is constructed.

[0078] Therefore, taking a tensor of (n×n×m) as input, it passes through a plurality of two-dimensional Convolution layers and three-dimensional maxpooling layers, and then undergoes a Flatten process of converting the data into one dimension. By combining the main specification variables, speed variables, and bow bulb variables with this compressed hull shape information and applying a plurality of MLPs, a model corresponding to the hull performance variable (e.g., Cr) which is the final output layer can be constructed.

[0079] Here, depending on the types of variables set in the output layer of the artificial neural network, the coefficients that can be estimated based on the ship shape data are determined. That is, since the performance coefficients of a ship are closely related to the shape of the ship, by setting various types of performance variables in the output layer, it is possible to construct an estimation model corresponding thereto.

[0080] As a typical example, not only the residual resistance coefficient (Cr) which is an index of resistance performance as described above is estimated, but also the wake fraction coefficient (w), thrust deduction coefficient (t), etc. which are propulsion-related indices necessary for power estimation can be mentioned.

[0081] Also, as in the case of the above-described embodiment, instead of a single performance variable, for example, by arranging field information such as the wake distribution of the stern propeller plane in the output layer and proceeding with learning, it is also possible to construct a model for estimating these wake distributions.

[0082] Referring to FIGS. 7 to 9, FIGS. 7 to 9 are drawings showing the results of constructing a CNN for estimating ship performance coefficients as described above and testing the actual performance based on the embodiments of the present invention. First, FIG. 7 is a drawing showing the relationship between predicted values and true values (experimental values), FIG. 8 is a drawing graphically showing the error distribution for verification data, and FIG. 9 is a drawing showing the comparison results of predicted values and true values for three sample lines, respectively.

[0083] Here, in the results shown in FIGS. 7 to 9, the artificial neural network model is composed of a 2D CNN using four convolutional layers and three Dense layers, and 2337 speed-Cr data related to 271 ships are used as total data. 80% of the total data is used as training data, and 20% of the total data is used as verification data. The inputs are offset (cross-sectional image), L / B, B / T, L / T, Fn, BulbL, and BulbH, and the output is Cr (residual resistance coefficient).

[0084] Also, in FIG. 7, the horizontal axis represents the predicted value, and the vertical axis represents the true value (experimental value).

[0085] As shown in FIGS. 7 to 9, it is shown that for the learning results, the average error for the training data is 0.038 and the coefficient of determination (R2) is 0.997, and the average error for the verification data is 0.084 and the coefficient of determination (R2) is 0.854, confirming that relatively good results are shown.

[0086] Also, the above-described output unit (14) can be configured to include display means such as a monitor or a display for visually displaying various information including the overall operation, processing process, and results of the ship performance coefficient estimation system (10) using the convolutional neural network based on the embodiments of the present invention executed as described above.

[0087] Furthermore, the above-described communication unit (15) communicates with external devices including servers, user terminals, etc. in at least one of the wireless or wired communication methods. For example, it can be configured to perform transmission and reception processing of various data, such as receiving ship specification information as input and transmitting the performance coefficient estimated through an artificial neural network. Thereby, an online-based service can be realized in which each user can remotely perform the estimation work of the performance coefficient, or multiple users can simultaneously perform the estimation work of the performance coefficient.

[0088] That is, referring to FIG. 10, FIG. 10 is a drawing schematically showing the overall configuration of a ship performance coefficient estimation service providing system (20) using a convolutional neural network-based ship performance coefficient estimation system (10) according to an embodiment of the present invention.

[0089] As shown in FIG. 10, the ship performance coefficient estimation service providing system (20) according to an embodiment of the present invention can be configured to mainly include a user terminal (21) for each user to request and receive a ship performance coefficient estimation service, and a server (22) configured to execute a process of providing a ship performance coefficient estimation service based on the data received through each user terminal (21).

[0090] Here, the above-described server (22) can be configured to include the ship performance coefficient estimation system (10) using a convolutional neural network based on an embodiment of the present invention configured as described above.

[0091] In addition, the above-described user terminal (21) can be configured by installing a dedicated program that interacts with the above-described ship performance coefficient estimation system (10) in an information processing device such as a PC or a notebook, or by installing a dedicated application program that interacts with the above-described ship performance coefficient estimation system (10) in a personal portable information processing terminal device having a communication function such as a smartphone or a tablet PC.

[0092] Therefore, with the configuration as described above, each user can easily obtain a desired ship performance coefficient regardless of time and location by connecting to the server (22) through their own user terminal (21).

[0093] Here, in the above-described embodiment of the present invention, as an example, the present invention has been described in the case where the offset data of the ship is converted into image data, the main specification variables and speed variables of the ship are matched with the valve shape information, and the residual resistance coefficient (Cr) of the ship is estimated based on the result of learning using CNN. However, the present invention is not necessarily limited to the configuration shown in the above-described embodiment.

[0094] That is, in addition to the above-described residual resistance coefficient (Cr), the present invention can also be configured to estimate the wake fraction coefficient (w) and the thrust deduction coefficient (t), which are propulsion-related indicators necessary for power estimation. Further, instead of a single performance variable, for example, field information such as the wake distribution on the propeller plane at the stern can be placed in the output layer and learning can be advanced to estimate the wake distribution. It should be noted that the present invention can be variously modified and changed as necessary by those skilled in the art without departing from the spirit and essence of the present invention.

[0095] Therefore, as described above, it is possible to implement a ship performance coefficient estimation system and method using a convolutional neural network according to an embodiment of the present invention. According to the present invention, offset data representing ship shape information (offset cross-section information) is converted into image data, and main specification variables, speed variables, and valve shape information of the ship are matched to the image data, and learning is performed by applying a two-dimensional CNN or a three-dimensional CNN. At this time, dropout and L2 regularization for preventing overfitting are added as necessary, so that, with arbitrary linear offset data, main specification variables, bow valve variables, and speed variables as inputs, for example, residual resistance coefficient (Cr), which is a resistance performance index, wake fraction (w), which is a propulsion-related index required for power estimation, thrust deduction coefficient (t), and other desired performance variables can be output. Thus, it is possible to derive a result considering fine linear changes, effectively avoid the problem of overfitting, which is a significant drawback of machine learning models, and prevent the loss of position information regarding the cross-sectional shape of the ship.

[0096] Further, according to the present invention, by providing a ship performance coefficient estimation system and method using a convolutional neural network, which is configured to derive an estimation formula for estimating the relationship between a linear cross-sectional shape and performance coefficients considering fine linear changes by utilizing deep learning and convolutional neural network (CNN) technologies as described above, the problem of the prior art statistical-based ship performance estimation method, in which the conventional method of predicting performance with the main specifications of a ship as inputs can hardly reflect fine shape changes because it mostly predicts performance based only on variables representing linearity, can be solved.

[0097] Moreover, according to the present invention, as described above, there is provided a ship performance coefficient estimation system and method using a convolutional neural network configured to be able to predict the performance of a ship using the specifications of the ship without performing numerical analysis or model tests during the initial design of the ship. Since there has been no method presented for predicting the performance of a ship by a statistical method using the specifications of the ship without performing numerical analysis or model tests, numerical analysis or model tests have to be repeatedly performed until the desired performance is obtained. As a result, there has been a problem of increased time and cost, and the problems of the conventional ship design method can be solved.

[0098] As described above, through the embodiments of the present invention, the detailed contents of the ship performance coefficient estimation system and method using a convolutional neural network according to the present invention have been described. However, the present invention is not limited only to the contents described in the above-described embodiments. Therefore, it is natural that various modifications, changes, combinations, and substitutions are possible for the present invention by those having ordinary knowledge in the technical field to which the present invention belongs according to design requirements and other various factors.

Explanation of Reference Numerals

[0099] 10 Ship performance coefficient estimation system using a convolutional neural network 11 Data collection unit 12 Data processing unit 13 Performance coefficient estimation unit 14 Output unit 15 Communication unit 16 Control unit 17 Ship performance coefficient estimation service providing system 18 User terminal 19 Server

Claims

1. In a ship performance coefficient estimation system, a data collection unit configured to execute a process of constructing a database by inputting various information including the specifications and shape information of a ship; a data processing unit configured to execute a process of imaging the shape information of the ship collected through the data collection unit; a performance coefficient estimation unit configured to match the offset data imaged through the data processing unit with the information collected through the data collection unit, perform learning through an artificial neural network, and estimate the performance coefficient of the ship based on the learning result; and an output unit configured to execute a process of outputting the overall processing process and processing results of the estimation system including the data collection unit, the data processing unit, and the performance coefficient estimation unit. A ship performance coefficient estimation system, characterized in that it includes the above.

2. The data collection unit collects various information including main specification information such as the length (Length; L), width (Breadth; B), draft (Draft; T), L / B, B / T, L / T of the ship, speed information including the Froude number (Fn), bulb shape information including the bulb length (Bulb Length; BulbL) and bulb height (Bulb Height; BulbH), ship shape information including offset cross-section information, and various performance variables including the residual resistance coefficient (Residual Resistance Coefficient; Cr), wake coefficient (Wake Coefficient; w), and thrust deduction coefficient (Thrust Deduction Coefficient; t) of the corresponding ship, and stores it in a database format, thereby constructing a database for various ships. The ship performance coefficient estimation system according to Claim 1.

3. The data processing unit an image conversion stage in which two-dimensional data (y, z) regarding each station included in the offset cross-section information collected through the data collection unit is converted into image data according to preset conditions; and Through the above image conversion stage, the main specification information (L / B, B / T, L / T), the speed information (Fn), and the valve shape information (valve length, valve height) are added to the ship shape information data converted into an image, and a process of generating ship shape image data reflecting the main information of the corresponding ship is executed, and a process including a ship shape image data generation stage is configured to be executed. The ship performance coefficient estimation system according to claim 2.

4. The above image conversion stage is configured to execute a process of converting the offset cross-sectional information into image data by using a binary method of setting a portion corresponding to the hull boundary to 1 and the rest to 0, or an SDF (Signed Distance Function) method of setting the hull boundary to 0, setting the inside of the hull to a negative sign, setting the outside of the hull to a positive sign, and expressing it by the shortest distance from the boundary. The ship performance coefficient estimation system according to claim 3.

5. The above performance coefficient estimation unit is configured to use a pre-constructed Convolutional Neural Network (CNN) model to perform learning on the ship shape image data generated through the above data processing unit, and based on the learning result, estimate a pre-set performance variable from the input value. The ship performance coefficient estimation system according to claim 3.

6. The above CNN model is a 2D CNN model configured to take as input the offset (cross-sectional image) data of the ship to be estimated, the main specification information (L / B, B / T, L / T), the speed information (Fn), and the valve shape information (valve length (BulbL), valve height (BulbH)), treat the entire hull as an n×n image with m channels, and execute a process of outputting a pre-set performance variable. further including a pre-set dropout and regularization (L2 regularization) process to prevent overfitting. The ship performance coefficient estimation system according to claim 5.

7. The above CNN model is Using the offset (cross-sectional image) data of the ship to be estimated, main specification information (L / B, B / T, L / T), speed information (Fn), and valve shape information (valve length (BulbL), valve height (BulbH)) as inputs, the entire hull is treated as an n×n×m image with one channel, and a 3D CNN model configured to execute a process of outputting pre-set performance variables. Characterized in that, in order to prevent overfitting, it is further configured to include a pre-set dropout and L2 regularization process. The ship performance coefficient estimation system according to claim 5.

8. The above output unit Characterized in that it includes display means including a monitor or a display for visually displaying various information including the overall operation, processing process, and processing results of the above estimation system. The ship performance coefficient estimation system according to claim 1.

9. The above estimation system A communication unit configured to communicate with external devices including a server and a user terminal by at least one of wireless or wired communication methods and execute a process of transmitting and receiving various data, and Characterized in that it further includes a control unit configured to execute a process of controlling the overall operation of the above estimation system. The ship performance coefficient estimation system according to claim 1.

10. In a ship performance coefficient estimation method configured using the ship performance coefficient estimation system according to any one of claims 1 to 9, A data collection stage in which a process of collecting various information about the ship and constructing a database is executed through the data collection unit of the above ship performance coefficient estimation system. A data processing stage in which a process of imaging the offset data of the ship collected in the above data collection stage is executed through the data processing unit of the above ship performance coefficient estimation system. A performance coefficient estimation stage in which the offset data imaged through the above data processing stage is matched with the information collected in the above data collection stage, learning is performed through an artificial neural network, and the performance coefficient of the ship is estimated based on the learning result, and An output stage is included, in which a process of outputting the overall processing procedures and processing results of the data collection stage, the data processing stage, and the performance coefficient estimation stage is executed through an output unit of the ship performance coefficient estimation system. A method for estimating a ship performance coefficient.

11. In a ship performance coefficient estimation service providing system, A user terminal for each user to request and receive a ship performance coefficient estimation service. It includes a server configured to execute a process of providing a ship performance coefficient estimation service based on data received through each user terminal. The server is configured to include the ship performance coefficient estimation system according to any one of Claims 1 to 9, characterized in that an online-based service is realized such that each user can perform a performance coefficient estimation operation remotely or multiple users can perform a performance coefficient estimation operation simultaneously. A ship performance coefficient estimation service providing system.

12. The user terminal is configured by installing a dedicated program that interacts with the ship performance coefficient estimation system in an information processing device including a PC, a notebook, etc., or configured by installing a dedicated application program that interacts with the ship performance coefficient estimation system in a personal portable terminal such as a smartphone or a tablet PC. The ship performance coefficient estimation service providing system according to Claim 11.

Images