KR102621106B1 - Apparatus and method for fault diagnosis of ship engines - Google Patents

Abstract

An apparatus and method for diagnosing a ship engine failure are disclosed. The ship engine failure diagnosis method includes collecting vibration data for the ship engine and calculating power spectral density data or spectrogram data for the ship engine through Fourier transform of the collected vibration data. Step, generating learning data by reducing the dimension of power spectrum density data or spectrogram data, performing machine learning using a neural network on the generated learning data to generate a failure classification model, and generating the generated failure classification model. It includes the step of determining whether the ship engine is broken from power spectral density data or spectrogram data of the ship engine acquired in real time.

Description

Ship engine failure diagnosis apparatus and method {Apparatus and method for fault diagnosis of ship engines}

The present invention relates to a ship engine failure diagnosis device and method.

In order to operate unmanned or minimally manned ships such as autonomous ships, an effective status diagnosis method of the ship's engine system is required for safety. Vibration signals can be measured without affecting the condition of the engine and contain abundant information related to the health of the engine, so they are widely used in health diagnosis.

However, vibration data has the disadvantage of having a relatively high sample rate compared to other data. In other words, a high sample rate increases the number of dimensions of the data, and an increase in the number of data dimensions increases the size of the machine learning algorithm used for condition diagnosis, which can make training slow or make optimal search difficult.

Therefore, there is a need for a ship engine failure diagnosis technology that alleviates this problem.

Republic of Korea Patent Publication No. 10-1374840 (2014.03.10)

The present invention generates learning data by reducing the dimension of vibration data collected from ship engines, calculates a failure classification model through learning of the generated learning data, and uses the calculated failure classification model to diagnose ship engine failures. The purpose is to provide a ship engine failure diagnosis device and method.

According to one aspect of the present invention, a method for diagnosing a marine engine failure performed by a marine engine failure diagnosis apparatus is disclosed.

A marine engine failure diagnosis method according to an embodiment of the present invention includes collecting vibration data for a marine engine, power spectral density data for the marine engine through Fourier transform of the collected vibration data, or Calculating spectrogram data, generating learning data by reducing the dimension of the power spectral density data or the spectrogram data, and classifying faults by performing machine learning using a neural network on the generated learning data. It includes generating a model and using the generated failure classification model to determine whether the marine engine has failed from power spectral density data or spectrogram data of the marine engine obtained in real time.

The step of generating the learning data includes power spectral density data or spectrogram data using at least one of Principal Component Analysis (PCA), Nonnegative Matrix Factorization (NMF), and Autoencoder. Reduce the dimension of

Generating the training data includes generating first training data using the principal component analysis, generating second training data using the non-negative matrix decomposition, and third learning using the autoencoder. It includes generating data, and the step of generating the failure classification model includes performing machine learning on all of the generated first learning data, second learning data, and third learning data to generate an entire failure classification model. do.

The step of determining whether the marine engine has failed includes determining whether the marine engine has failed using all of the generated overall failure classification model, first failure classification model, second failure classification model, and third failure classification model, The determination result calculated through the overall failure classification model is considered the final determination result.

The method for diagnosing a ship engine failure further includes transmitting the determination result to a land control device, wherein the determination result includes information on whether the ship engine has failed and the learning data, and the shore control device determines the Upon receipt of the results, failure diagnosis is performed by measuring the distance between pre-stored normal data and the learning data to determine whether the ship engine is broken.

According to another aspect of the present invention, a marine engine failure diagnosis device is disclosed.

A marine engine failure diagnosis apparatus according to an embodiment of the present invention includes a memory for storing a command and a processor for executing the command, wherein the command includes: collecting vibration data for a marine engine, and the collected vibration data. Calculating power spectral density data or spectrogram data for the ship engine through Fourier transform, generating learning data by reducing the dimension of the power spectral density data or spectrogram data. A step of generating a failure classification model by performing machine learning using a neural network on the generated learning data, and using the generated failure classification model, power spectrum density data or spectrogram of the ship engine acquired in real time. A ship engine failure diagnosis method is performed including the step of determining whether the ship engine is broken from data.

The marine engine failure diagnosis apparatus and method according to an embodiment of the present invention generates learning data by reducing the dimension of vibration data collected from a marine engine, calculates a failure classification model through learning of the generated learning data, and calculates Failure diagnosis of ship engines can be performed using the fault classification model.

1 is a diagram schematically illustrating the configuration of a ship engine failure diagnosis system according to an embodiment of the present invention.
Figure 2 is a flow chart schematically illustrating a ship engine failure diagnosis method performed by a ship engine failure diagnosis device according to an embodiment of the present invention.
3 to 5 are diagrams for explaining a method for diagnosing a ship engine failure according to an embodiment of the present invention of FIG. 2.
Figure 6 is a diagram schematically illustrating the configuration of a ship engine failure diagnosis device according to an embodiment of the present invention.

As used herein, singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as “consists of” or “comprises” should not be construed as necessarily including all of the various components or steps described in the specification, and some of the components or steps may be included in the specification. It may not be included, or it should be interpreted as including additional components or steps. In addition, terms such as "... unit" and "module" used in the specification refer to a unit that processes at least one function or operation, which may be implemented as hardware or software, or as a combination of hardware and software. .

Hereinafter, various embodiments of the present invention will be described in detail with reference to the attached drawings.

1 is a diagram schematically illustrating the configuration of a ship engine failure diagnosis system according to an embodiment of the present invention.

Referring to FIG. 1, the marine engine failure diagnosis system according to an embodiment of the present invention includes a marine engine failure diagnosis device 100 installed in an autonomous ship 10 and a land control device installed in the land control center 20. It may be configured to include (200).

For example, communication between the autonomous ship 10 and the land control center 20 can be accomplished through a maritime communication network such as LTE-Maritime.

Here, the ship engine failure diagnosis device 100 and the land control device 200 may be implemented through a network-connectable server. For example, the server may perform functions such as control, management, and fault diagnosis of the autonomous vessel 100, or may perform functions to support the operation of the autonomous vessel 100, and may be used in embodiments of the present invention. It may be configured to include a ship engine failure diagnosis device 100 or a land control device 200.

In this specification, a server is a computing device that performs a ship engine failure diagnosis method according to an embodiment of the present invention, and may be one or two or more physical entities. When a server is divided into multiple physical entities and implemented, the management entity of each physical entity may be different. The server may include a DB, which refers to a functional and structural combination of software and hardware that stores information corresponding to each database. The DB may be implemented as at least one table, and the information stored in the database can be searched, stored, and It may also include a separate DBMS (Database Management System) for management. In addition, it can be implemented in various ways, such as in the form of a linked-list, tree, or relational database, and includes all data storage media and data structures that can store information corresponding to the database.

This marine engine failure diagnosis device 100 collects vibration data of the marine engine of the autonomous ship 10, performs failure diagnosis, and transmits the failure diagnosis result to the shore control device 200.

The operation of the marine engine failure diagnosis device 100 according to an embodiment of the present invention will be described in detail later with reference to FIGS. 2 to 5.

Figure 2 is a flowchart schematically illustrating a method of diagnosing a marine engine failure performed by a marine engine failure diagnosis apparatus according to an embodiment of the present invention, and Figures 3 to 5 are a flow chart showing a marine engine failure diagnosis method according to an embodiment of the present invention in Figure 2. This diagram is intended to explain the fault diagnosis method. Hereinafter, a method for diagnosing a ship engine failure according to an embodiment of the present invention will be described, focusing on FIG. 2, with reference to FIGS. 3 to 5.

In step S210, the marine engine failure diagnosis device 100 collects vibration data about the marine engine.

That is, the marine engine failure diagnosis apparatus 100 may collect time-series vibration data of the marine engine by acquiring a vibration signal of the marine engine for a preset time at a preset sample rate using an accelerometer. At this time, the vibration data may be collected separately into vibration data in a normal state and vibration data in a fault state.

Such vibration data can be measured non-destructively, contains a lot of information related to machine health, can be used as an indicator of early abnormalities, and has a high sample rate.

In step S220, the marine engine failure diagnosis apparatus 100 calculates power spectral density data or spectrogram data for the marine engine through Fourier transform of the collected vibration data.

For example, referring to FIG. 3, a total of 10 minutes of vibration data can be divided into 0.16 second units and converted into a power spectral density through Fourier transform of the collected vibration data. This power spectral density shows different shapes in normal and fault states in a specific frequency range.

Figure 4 shows an example of a power spectrum density graph for 5 second long sample data for each failure type.

This power spectral density can be applied to data that is not synchronized to the engine cycle, has limitations in detecting temporal transitions, and is characterized by a large information dimension. Since this increases the learning time and amount of calculation, it is necessary to summarize or reduce the dimension of information while minimizing information loss.

In step S230, the marine engine failure diagnosis apparatus 100 generates learning data by reducing the dimension of power spectrum density data or spectrogram data for the marine engine.

That is, as shown in FIG. 5, the marine engine failure diagnosis device 100 uses at least one of Principal Component Analysis (PCA), Nonnegative Matrix Factorization (NMF), and Autoencoder. You can use this to reduce the dimensionality of power spectral density data or spectrogram data.

Here, principal component analysis calculates the principal component corresponding to the direction vector in which the data has a large variance and projects the data onto the direction vector. It is a method of reducing and summarizing the dimension of the data with highly correlated variables.

For example, in the case of principal component analysis, a basis matrix U is calculated from the collected vibration data matrix X through singular value decomposition of the collected vibration data using the equation below.

here, are orthogonal matrices, is a diagonal matrix containing the nonzero eigenvalues of X ^T

Next, using U calculated from the vibration data matrix, a training data matrix Q _PCA with reduced dimension can be calculated as shown in the following equation.

here, is a projection matrix selected from the first k columns of U. The columns of Q _PCA have k dimensions, and the columns of the original X _test have n dimensions.

And, non-negative matrix decomposition is an algorithm that decomposes a matrix that does not contain negative numbers into the product of two matrices that do not contain negative numbers.

For example, in the case of non-negative matrix decomposition, the vibration data matrix X is approximated as the product of two non-negative matrices W and H, as shown in the equation below.

here, is the basis matrix, is the coefficient matrix.

The learning data matrix Q _PCA with reduced dimensionality is obtained through factorization of X _test using the learned basis matrix W, as shown in the equation below.

Like principal component analysis, the columns of Q _NMF have k dimensions, and the columns of the original X _test have n dimensions.

And, an autoencoder is a neural network that aims to extract a high-dimensional input vector into low-dimensional encoding with valid data components and then reconstruct an output vector from the encoding.

For example, as shown in FIG. 5, an autoencoder consisting of three layers can be used, and a learning data matrix with reduced dimension can be calculated through 1-dimensional encoding to 30-dimensional encoding.

In step S240, the marine engine failure diagnosis apparatus 100 generates a failure classification model by performing machine learning using a neural network on the generated learning data.

For example, the ship engine failure diagnosis apparatus 100 generates first learning data through dimensionality reduction of power spectral density data or spectrogram data using principal component analysis, and generates first learning data as power spectral density data or spectrogram data using non-negative matrix decomposition. Second learning data can be generated through dimensionality reduction of gram data, and third learning data can be generated through dimensionality reduction of power spectral density data or spectrogram data using an autoencoder. In addition, the marine engine failure diagnosis apparatus 100 may perform machine learning on all of the generated first learning data, second learning data, and third learning data to generate an overall failure classification model.

At this time, the marine engine failure diagnosis device 100 may separately generate a first failure classification model, a second failure classification model, and a third failure classification model for each of the first learning data, second learning data, and third learning data. there is.

In step S250, the marine engine failure diagnosis apparatus 100 uses the generated failure classification model to determine whether the marine engine has failed from power spectrum density data or spectrogram data of the marine engine obtained in real time.

For example, the marine engine failure diagnosis device 100 uses all of the generated overall failure classification model, the first failure classification model, the second failure classification model, and the third failure classification model to determine whether the marine engine is broken, The determination result calculated through the overall failure classification model can be considered the final determination result. Thereafter, the marine engine failure diagnosis device 100 determines the failure classification model with the highest frequency of producing the same determination result as the final determination result among the first failure classification model, the second failure classification model, and the third failure classification model, Learning data can be generated using only the dimensionality reduction method used to generate the determined failure classification model. Through this, the amount of calculation and computational complexity can be reduced.

In step S260, the ship engine failure diagnosis device 100 transmits the determination result of whether the ship engine is broken to the land control device 200.

For example, the ship engine failure diagnosis device 100 may transmit the determination result to the land control device 200 in one of a normal state, an unusual state, and an abnormal state. At this time, in the case of principal component analysis, if in a normal state, only the coefficients of principal component analysis 1 and principal component analysis 2 are included in the discrimination result; if in a singular state, only the coefficients of principal component analysis 1 to principal component analysis 10 are included; and if in an abnormal state, only coefficients of principal component analysis 1 to principal component analysis 10 are included. Only coefficients from analysis 50 can be included.

In other words, the ship engine failure diagnosis device 100 transmits a determination result including information on whether the ship engine has failed and learning data with reduced dimensions to the land control device 200, so that the land control device 200 determines whether the ship engine has failed. The amount of information transmitted can be varied.

In addition, upon receipt of the determination result, the land control device 200 performs precise failure diagnosis by measuring the distance between the pre-stored normal matrix and the received learning data matrix with reduced dimensionality, thereby determining whether the ship engine is broken. You can.

Figure 6 is a diagram schematically illustrating the configuration of a ship engine failure diagnosis device according to an embodiment of the present invention.

Referring to FIG. 6, the marine engine failure diagnosis apparatus 100 according to an embodiment of the present invention includes a processor 110, a memory 120, a communication unit 130, and an interface unit 140.

The processor 110 may be a CPU or a semiconductor device that executes processing instructions stored in the memory 120.

Memory 120 may include various types of volatile or non-volatile storage media. For example, memory 120 may include ROM, RAM, etc.

For example, the memory 120 may store instructions for performing a ship engine failure diagnosis method according to an embodiment of the present invention.

The communication unit 130 is a means for transmitting and receiving data with other devices through a communication network.

The interface unit 140 may include a network interface and a user interface for accessing a network.

Meanwhile, the components of the above-described embodiment can be easily understood from a process perspective. In other words, each component can be understood as a separate process. Additionally, the processes of the above-described embodiments can be easily understood from the perspective of the components of the device.

Additionally, the technical contents described above may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the embodiments or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. A hardware device may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

The above-described embodiments of the present invention have been disclosed for illustrative purposes, and those skilled in the art will be able to make various modifications, changes, and additions within the spirit and scope of the present invention, and such modifications, changes, and additions will be possible. should be regarded as falling within the scope of the patent claims below.

10: Autonomous ship
20: Land control center
100: Ship engine failure diagnosis device
110: processor
120: memory
130: Department of Communications
140: Interface unit
200: Land control device

Claims (6)

In the ship engine failure diagnosis method performed by the ship engine failure diagnosis device,
collecting vibration data for a marine engine;
Calculating power spectral density data or spectrogram data for the ship engine through Fourier transform of the collected vibration data;
generating learning data by reducing the dimension of the power spectral density data or the spectrogram data;
Generating a failure classification model by performing machine learning using a neural network on the generated learning data;
Using the generated failure classification model, determining whether the ship engine is broken from power spectral density data or spectrogram data of the ship engine acquired in real time; and
Including transmitting the determination result to a land control device,
The step of generating the learning data is,
Generating first learning data through dimensionality reduction of power spectral density data or spectrogram data using principal component analysis (PCA);
Generating second learning data through dimensionality reduction of power spectral density data or spectrogram data using nonnegative matrix factorization (NMF); and
Generating third learning data through dimensionality reduction of power spectral density data or spectrogram data using an autoencoder,
The step of generating the failure classification model is,
Machine learning is performed on all and each of the generated first learning data, second learning data, and third learning data to create an overall failure classification model, first failure classification model, second failure classification model, and third failure classification model. create,
The step of determining whether the ship engine is broken is,
The generated overall failure classification model, the first failure classification model, the second failure classification model, and the third failure classification model are all used to determine whether the ship engine has failed, and the determination result calculated through the overall failure classification model is used. Considered as the final determination result,
The determination result transmitted to the land control device includes information on whether the ship engine is broken and the learning data,
The land control device performs a fault diagnosis by measuring the distance between pre-stored normal data and the learning data upon receipt of the determination result to determine whether the ship engine is broken. A method for diagnosing a ship engine failure.
delete delete delete delete In a ship engine failure diagnosis device,
Memory for storing instructions; and
Including a processor that executes the instructions,
The command is:
collecting vibration data for a marine engine;
Calculating power spectral density data or spectrogram data for the ship engine through Fourier transform of the collected vibration data;
generating learning data by reducing the dimension of the power spectral density data or the spectrogram data;
Generating a failure classification model by performing machine learning using a neural network on the generated learning data;
Using the generated failure classification model, determining whether the ship engine is broken from power spectral density data or spectrogram data of the ship engine acquired in real time; and
Performing a ship engine failure diagnosis method including transmitting the determination result to an onshore control device,
The step of generating the learning data is,
Generating first learning data through dimensionality reduction of power spectral density data or spectrogram data using principal component analysis (PCA);
Generating second learning data through dimensionality reduction of power spectral density data or spectrogram data using nonnegative matrix factorization (NMF); and
Generating third learning data through dimensionality reduction of power spectral density data or spectrogram data using an autoencoder,
The step of generating the failure classification model is,
Machine learning is performed on all and each of the generated first learning data, second learning data, and third learning data to create an overall failure classification model, first failure classification model, second failure classification model, and third failure classification model. create,
The step of determining whether the ship engine is broken is,
The generated overall failure classification model, the first failure classification model, the second failure classification model, and the third failure classification model are all used to determine whether the ship engine is broken, and the determination result calculated through the overall failure classification model is used. Considered as the final determination result,
The determination result includes information on whether the ship engine is broken and the learning data,
The land control device is a ship engine failure diagnosis device, characterized in that upon receipt of the determination result, the ship engine failure diagnosis is performed by measuring the distance between pre-stored normal data and the learning data to determine whether the ship engine is broken.

Images