KR102073810B1 - Method and system for predicting the failure of naval ship propulsion system using machine learning - Google Patents

Abstract

Provided are a method for predicting a failure of vessel propulsion equipment using machine learning and a system thereof. In order to use an RUL prediction model, characteristics of data are extracted and processed to generate a status indicator through the PCA. By dividing operation areas and standardizing result data for each operation area, the failure of equipment can be predicted based on a learning result input to machine learning.

Description

Method and system for predicting the failure of naval ship propulsion system using machine learning}

The present invention relates to a method and system for predicting a failure of a ship propulsion equipment using machine learning, and more particularly, to predict a failure point of a ship propulsion equipment in response to a real-time operating environment of the ship in the defense technology field. The present invention relates to a method and system for predicting failure of ship propulsion equipment to enable more accurate prediction by utilizing machine learning.

In the ship automation system, the Engineering Control System (ECS) is a system that allows the control from the ECS console by integrating the equipment that was separately controlled locally as the ship is modernized / automated. ECS integrates the propulsion, power, view and damage system of the ship based on network communication and maximizes the ship's operability, combat performance, automation and survivability through operation / control / monitoring.

The Integrated Condition Assessment System (ICAS) is one of the six technical elements of ECS.

ICAS is a technology to prevent critical failures of major equipment and to increase the reliability of equipment operation. Perform control and monitoring of the ship's propulsion system, power system, bogie system and damage system. Detailed functions include storage and display of performance data essential for equipment operation such as pressure, temperature and vibration, analysis of the organization's trend through performance evaluation, recommendations for maintenance procedures, generation and display of alarm conditions. do. These functions can be broadly categorized into diagnostics (alarms) that indicate a fault condition and predictive failures. The current ICAS level is a simple display level that provides sensor measurements attached to the equipment, and the equipment's maintenance forecasts and recommendations are not available. Complex interactions between components of the propulsion equipment are causing system degradation or failure. In order to ensure the reliability of the equipment, contribute to the improvement of vessel operation rate and manpower reduction, it is important to secure the equipment prediction function.

However, ICAS is being built by direct or foreign technology introduction, and advanced countries are avoiding technology transfer by classifying it as a core technology for enhancing ship readiness and availability. Since this technology is a core technology to be secured for the localization of the ship's integrated engineering control system (ECS) and the development of the ship's integrated ship computing environment (TSCE), domestic development is important.

In engineering systems, it is commonly used as the term Prognostics and Health Management (PHM), an extension of ICAS. By predicting the remaining life (RUL) of parts, PHM can be applied to areas where fatal damage is expected, areas where high cost is spent by preventive maintenance, and areas that are inaccessible for repair. have. PHM has emerged as the advances in machine learning that can process a lot of data have been actively studied in high technology fields such as automotive, aviation and railway.

At present, ECS's ICAS shows that ABB's predictive technology is only a trendline for individual sensor values, showing an average value for one sensor value, making it less reliable for fault determination and allowing the crew to directly combine sensor values to determine maintenance. As a result, there is a limit that the failure judgment is different depending on the ability of the crew.

Therefore, research on necessary technologies and procedures for securing ICAS predictability is necessary.

The technical problem to be solved by the present invention to solve the above problems is to utilize the machine learning, such as applying a residual life estimation algorithm to secure the core technology of ICAS, and to change in real time in predicting the failure of the propulsion equipment By increasing the accuracy of the status indicators by minimizing the influence of the operating environment, it is possible to prevent major catastrophic failures in advance and increase the availability of the equipments. To reduce system maintenance costs.

The problem of the present invention is not limited to those mentioned above, and other problems that are not mentioned will be clearly understood by those skilled in the art from the following description.

As a means for solving the above technical problem, according to an embodiment of the present invention, the method of predicting the failure of the ship propulsion equipment using the machine learning algorithm, in applying the integrated condition evaluation system (ICAS) to the ship propulsion system, Collecting raw data at a sensor of the system; Pre-processing the raw data; A feature engineering step of generating a status indicator by generating a feature from the preprocessed data; And inputting the feature into a machine learning algorithm to generate a residual life expectancy (RUL) prediction model.

The raw data includes temperature, pressure, output, torque, rotational speed, and frequency collected in real time from a sensor of the ship propulsion system, and the collecting step includes inputting an operation value and result data from the raw data. Separate by output value.

The operating environment data here includes outside temperature, outside pressure and propulsion lever, and the resulting data includes engine exhaust temperature, engine torque, engine vibration and propulsion shaft vibration.

The preprocessing step performs one or more of the following: integration, transformation, reduction, and cleaning of the collected data to counteract the impact on the operating environment.

The preprocessing step may include applying a K-means algorithm to the extracted feature information to cluster data to classify an operation region; Calculating a center point of the data for each operation zone; Normalizing the result data corresponding to the operation area based on a data center point; And generating training data through the standardized result data.

The feature engineering may include: extracting a feature using the preprocessed data; Selecting (preprocessing data) a feature based on a criterion recognized by the developer (developer) from the extracted features; And generating (designing) a status indicator representing a health indicator of a device over time through a principal component analysis (PCA) that fuses the selected feature.

The generating of the model may include selecting an exponential function evaluation model based on a cumulative deterioration with time as the RUL estimation model.

The generating of the model includes predicting a failure point in real time according to a learning result by applying the health indicator to a RUL prediction algorithm using machine learning using the state of the equipment according to time derived through the state indicator as a health indicator. do.

On the other hand, according to another embodiment of the present invention, the trap propulsion equipment failure prediction system using the machine learning algorithm in collecting the raw data from the sensor of the ship propulsion system in applying the integrated condition evaluation system (ICAS) to the ship propulsion system part; A pre-processing unit for pre-processing the raw data; A feature engineering unit to generate a feature from the preprocessed data; And a modeling unit configured to input the feature into a machine learning algorithm to generate a residual life expectancy (RUL) prediction model.

The raw data includes temperature, pressure, output, torque, rotational speed, and frequency collected in real time from a sensor of the ship propulsion system, and the collection unit includes operating environment data among the raw data as input values and result data as output values. Separate.

The operating environment data here includes outside temperature, outside pressure and propulsion lever, and the resulting data includes engine exhaust temperature, engine torque, engine vibration and propulsion shaft vibration.

The preprocessing unit performs one or more of integration, transformation, reduction, and cleaning of the collected data to offset the impact on the operating environment.

Pre-processing unit, applying the K-means algorithm to the extracted feature information to cluster the data to distinguish the operating area; Calculating a center point of the data for each operation zone; Normalizing the result data corresponding to the operation area based on a data center point; And generating training data through the standardized result data.

The feature engineering unit may include: a feature extraction unit for extracting a feature using the preprocessed data; A feature selector configured to select a feature based on a criterion recognized by the developer from the extracted features (preprocessing data); And a status indicator generator for generating (designing) a status indicator representing a health indicator of a device over time through a principal component analysis (PCA) that fuses the selected feature.

The modeling unit selects an exponential function evaluation model based on deterioration accumulation over time as the RUL estimation model.

The modeling unit predicts a failure point in real time according to a learning result by applying the health indicator to the RUL prediction algorithm using machine learning using the state of the equipment according to the time derived through the state indicator as the health indicator.

According to the invention,

In order to secure the core technology of ICAS, it utilizes machine learning such as applying residual life estimation algorithm and increases the accuracy of the status indicator by minimizing the effect component of the operating environment that changes in real time in predicting the failure of propulsion equipment. Prevent equipment from catastrophic failures in advance to increase the availability of the equipment, and reduce system maintenance costs by ensuring reliability by quantitative numerical indicators rather than leaving failure predictions to the capacity of maintenance personnel.

The effects of the present invention are not limited to those mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a technical level analysis chart according to the Defense Technology Survey;
2 is a flowchart illustrating a method for predicting a failure of propulsion equipment of a ship using machine learning according to an embodiment of the present invention;
3 is a flowchart illustrating detailed steps of the preprocessing, feature engineering, and modeling steps of FIG. 2 in a failure prediction method according to an embodiment of the present invention;
4 is a block diagram showing a system for predicting the failure of the propulsion equipment of the ship using machine learning according to an embodiment of the present invention;
5 is an exemplary view illustrating an operation zone selection according to an embodiment of the present invention.

Objects, other objects, features and advantages of the present invention will be readily understood through the following preferred embodiments associated with the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided so that the disclosure may be made thorough and complete, and to fully convey the spirit of the present invention to those skilled in the art.

Where the terms first, second, etc. are used herein to describe the components, these components should not be limited by these terms. These terms are only used to distinguish one component from another. The embodiments described and illustrated herein also include complementary embodiments thereof.

In addition, when an element, component, apparatus, or system is referred to as including a component made up of a program or software, the element, component, apparatus, or system may be executed by the program or software even if there is no explicit mention. Or hardware (e.g., memory, CPU, etc.) or other programs or software (e.g., an operating system or drivers required to run the hardware, etc.) needed to operate.

Also, the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. In this specification, the singular also includes the plural unless specifically stated otherwise in the phrase. As used herein, the words 'comprises' and / or 'comprising' do not exclude the presence or addition of one or more other components.

In describing the specific embodiments below, various specific details are written to more specifically explain and help understand the invention. However, a person of ordinary skill in the art can understand that the present invention can be used without these various specific contents.

In some cases, it is mentioned in advance that parts of the invention that are commonly known in the description of the invention and which are not highly related to the invention are not described in order to prevent confusion in explaining the invention without cause.

Hereinafter, specific technical contents to be implemented in the present invention will be described in detail with reference to the accompanying drawings.

1 is a technical level analysis diagram according to the Defense Technology Survey, and the technical level of the survivability technology (BDCS) and the propulsion technology (IPMS, OBTS) among the ECS (Engineering Control System) technology as of 2013 is the combat system technology. It can be seen that it is significantly lower than the level. Among these, the invention is an integrated condition assessment system (ICAS) that belongs to the propulsion technology. As one of the predictive maintenance methods of the engineering system, we will use the Remaining Useful Life (RUL) algorithm.

Various industries, such as medical, automotive, aerospace, and manufacturing, are introducing predictive maintenance, including residual life estimation technology. During predictive maintenance, RUL technology makes quantitative forecasts for the future. RUL technology can be divided into technology to make status indicator through data analysis and technology to make RUL evaluation model suitable for system. To predict ship propulsion equipment failures, a health indicator must be derived mathematically. Status indicators are calculated by PCA (dimension reduction) of the sensor data (temperature, pressure, output, torque, rotational speed, vibration, etc.) of the ship propulsion equipment that have a major impact on a particular failure. However, due to the operational characteristics of the ship, the sensor data of the propulsion equipment changes according to the operating environment (user input lever position, outside temperature, outside pressure, etc.). Therefore, in order to minimize the impact components of the operating environment, it is necessary to classify the operating zones by using the operating environment impact removal algorithm for predicting the failure of the ship propulsion equipment and reorganize the center data accordingly. Specifically, it is as follows.

2 is a flowchart illustrating a method of predicting a failure of a propulsion equipment of a ship using machine learning according to an embodiment of the present invention, and FIG. 4 is a flowchart of a propulsion equipment of a ship using machine learning according to an embodiment of the present invention. It is a block diagram which shows the system which predicts a failure.

The method of predicting the failure of the ship propulsion equipment using machine learning starts at step S110 of collecting raw data from the sensors of the propulsion system of the shrine in the collecting unit 210 as shown. It collects a plurality of data such as outside temperature, outside pressure, propulsion lever, engine exhaust temperature, engine torque, engine vibration and propulsion shaft vibration from sensors of the propulsion system for a specific failure. Examples of data collected from propulsion system sensors are shown in Table 1.

Time (s) Outside temperature Outside pressure Propulsion lever Engine exhaust temperature Engine torque Engine vibration Propulsion Shaft Vibration One 30 1.08 3 400 50 10 11 2 30 1.09 3.1 410 52 10 11 3 31 1.09 3.3 410 58 11 12 4 31 1.09 3.8 430 62 12 13 5 30 1.08 3.8 440 65 15 15 6 30 1.08 4 450 66 16 17

Collect data in the form of databases, spreadsheets, etc., and determine if the data is suitable for analysis. The data configuration to determine whether it is appropriate may be the time stamp, size, number, etc. of the data.

In the collecting step S110, the operation environment data among the collected raw data is divided into an input value and a result data. An example of such data classification is the input of the outside temperature, outside pressure and propulsion lever value (blue) as shown in [Table 2], and the engine exhaust temperature, engine torque, engine vibration, and propulsion shaft vibration values are output (red) values. Can be distinguished.

Time (s) Outside temperature Outside pressure Propulsion lever Engine exhaust temperature Engine torque Engine vibration Propulsion Shaft Vibration One 30 1.08 3 400 50 10 11 2 30 1.09 3.1 410 52 10 11 3 31 1.09 3.3 410 58 11 12 4 31 1.09 3.8 430 62 12 13 5 30 1.08 3.8 440 65 15 15 6 30 1.08 4 450 66 16 17

Next, the raw data collected in this way is pre-processed in the preprocessor 220 (S120). Preprocessing is the transformation of collected data into a format suitable for analysis and processing, which requires data preprocessing to attenuate the impact of the various operating environments of the ship. Standardization and normalization of clustered data populations can offset the impact of the operational environment.

FIG. 3 is a detailed flowchart of the preprocessing, feature engineering, and modeling steps of FIG. 2 in a failure prediction method according to an exemplary embodiment of the present invention. Data preprocessing may include data integration, transformation, reduction, and cleaning.

In general data preprocessing, data processing is performed prior to data analysis such as outlier removal, missing data replacement, and noise removal using a filter. Since the data collected from the ship was collected under various conditions, it is necessary to offset the impact on the operating environment in order to analyze the effects on the failure.

5 is an exemplary view illustrating an operation zone selection according to an embodiment of the present invention. As shown in Fig. 5 (a), the characteristics including the operation environment data are classified into the operation zones by using the K-means algorithm for the data collected first (S121). In order to eliminate the operating condition which is unnecessary data, k-means is applied to cluster the data. Subsequently, as shown in FIG. 5- (b), the center point of the data for each operation area is calculated (S122), and standardized result data corresponding to the operation area based on the center point for each operation area (S123). Normalization of each clustered data group (operational zone) is used to calculate the abnormal state rate for the normal for each group. Each abnormality rate is normalized to eliminate the effects of operating conditions and to uniformly compare sensor data from different criteria.

 The formula used for data normalization is shown in [Table 3].

(

: 요소값,

: 전체 데이터 평균,

: 전체 데이터 표준편차)Equation :

(

: Element value,

: Overall data average,

: Total data standard deviation)

In this way, standardization is performed on each data of each operation area to remove the scale difference of each area.

 The training data is generated through the standardized result data (S124), and the training data is then input to machine learning to utilize the RUL algorithm for predicting the failure of the trap propulsion equipment according to the learning result (S142). The results of standardizing the output data based on the examples of [Table 1] and [Table 2] are as shown in [Table 4].

Time (s) Engine exhaust temperature Engine torque Engine vibration Propulsion Shaft Vibration One -1.29987 -1.44301 -0.98995 -0.98837 2 -0.74278 -1.11629 -0.98995 -0.98837 3 -0.74278 -0.13613 -0.56569 -0.5322 4 0.371391 0.517306 -0.14142 -0.07603 5 0.928477 1.007386 1.131371 0.836315 6 1.485563 1.170745 1.555635 1.748658

The feature engineering unit 230 undergoes feature engineering (S130) to generate a feature from the preprocessed data. Feature engineering is a process of processing and generating features from initial data, and is a step of generating a health indicator essential for estimating remaining life.

The impact on residual life expectancy models is significant and requires a lot of expertise and time. This step results in a health indicator, which becomes the input function of the remaining life prediction model. The first step in feature engineering is feature extraction (S131), which adds information using existing data, and converts raw data into data suitable for machine learning algorithms. For example, it includes features such as the mean, kurtosis, and skewness of the data. The second step of feature engineering is a step of selecting feature (Preprocess Data) (S132) that the developer selects through ranking among the extracted features. The feature selection is performed because the dimension is too large to use all the extracted features as the status indicator. Set the criteria to be used by the developer and use the corresponding features in the analysis. At this stage, the sensor itself can be selected as a feature for critical sensors that have a significant impact on failure.

The third step of feature engineering is the status indicator generation step (S133). In this case, Principal Component Analysis (PCA), which fuses the features selected in the previous step, is a component of a status indicator that indicates the state of the equipment, and the entire data set is provided to provide an intuitive graph to the user. In operation S133, a plurality of features are fused to express the status indicator. If you select multiple features in the previous step, if you represent all of them as a graph over time, a multidimensional graph is drawn (for example, if you extract 15 features and select 5 of them from the feature extraction step), The graph needs a five-dimensional graph) It is difficult for the user to understand the current state of the main equipment. Therefore, it is possible to express a plurality of selected features, and to create a status indicator to fuse the features in one dimension. This technique is called principal component analysis. Use the function of monotonicity to identify the status indicator value to predict the RUL using the status indicator value over time.

 The modeling unit 240 undergoes modeling (S140) of generating a residual valid life (RUL) prediction model by inputting a feature selected in the previous step and fused to the machine learning algorithm. RUL analysis establishes a residual life estimation model by analyzing a system or data, and develops a model that can perform estimation based on time variation or statistical characteristics of status indicator values to estimate the remaining useful life of a system. In general, in engineering systems, the residual life expectancy evaluation model is selected as an exponential function because deterioration over time accumulates system errors. In an embodiment of the present invention, an exponential function evaluation model may be selected (S141) due to deterioration over time. Such an exponential function evaluation model updates the data to enable real-time RUL prediction (S142). In the modeling step, a failure point is predicted in real time by applying the indicator of the equipment displayed in the step S133 generation step S133 to a RUL prediction algorithm using machine learning. Since the ship operates as an independent system, it should be possible to predict the remaining useful life based on real-time data. The exponential function evaluation model provided by Matlab enables real-time updating of data, so that the remaining useful life can also be estimated in real time.

Remaining Useful Life (RUL) of parts can be predicted and applied in Prognostics and Health Management (PHM), an extension of ICAS. Because PHM technology has to process a lot of data, research is being actively conducted as machine learning technology is advanced.

Therefore, in the embodiment of the present invention, the clustering (data clustering), in particular, the K-Means clustering algorithm was used in order to find a suitable method for applying the trap to the traps. In the life prediction, the machine learning technology is used to enable efficient and reliable modeling.

100: How to predict the failure of ship propulsion equipment using machine learning
200: System predicting failure of ship propulsion equipment using machine learning
210: data collector
220: data preprocessing unit
230: Features Engineering
240: modeling unit

Claims (16)

In the failure prediction method of applying the Integrated Condition Assessment System (ICAS) to the ship propulsion system,
Collecting raw data including temperature, pressure, output, torque, rotational speed, and frequency in real time from a sensor of the ship propulsion system, and dividing operating environment data among input data and input data into output values;
The K-means algorithm is applied to the raw data to cluster the data to divide the operating area,
Calculate the center point of the data for each operation area,
Standardize the result data corresponding to the operation area based on the data center point,
Pre-processing the data, wherein the training data is generated through the standardized result data;
A feature engineering step of generating a status indicator by generating a feature from the preprocessed data;
Inputting the feature to a machine learning algorithm to generate an RUL prediction model by selecting an exponential function evaluation model as a cumulative deterioration with time as a RUL estimation model;
Trap propulsion equipment failure prediction method comprising a. delete The method of claim 1,
The operating environment data includes outside temperature, outside pressure, propulsion lever,
The result data includes engine exhaust temperature, engine torque, engine vibration, propulsion shaft vibration. The method of claim 1,
The pretreatment step,
A method for predicting a failure of a ship propulsion equipment comprising performing one or more of the following data: integration, transformation, reduction, and cleaning in order to offset the impact on the operating environment. delete The method of claim 1,
The feature engineering step,
Feature Extraction using the preprocessed data;
Selecting (preprocessing data) a feature based on a criterion recognized by the developer (developer) from the extracted features;
Generating (designing) a status indicator representing a health indicator of a device over time through a principal component analysis (PCA) that fuses the selected feature;
Ship propulsion equipment failure prediction method comprising a. delete The method of claim 1,
Generating the model,
Using the state of the equipment according to the time derived through the status indicator as a health indicator, the health indicator is applied to the RUL prediction algorithm using machine learning to predict the failure point in real time according to the learning result. How to predict equipment failure. In the failure prediction system applying the integrated condition evaluation system (ICAS) to the ship propulsion system,
A collecting unit which collects raw data including temperature, pressure, output, torque, rotational speed, and frequency from a sensor of the ship propulsion system in real time, and divides operating environment data among input data and output data into output values;
The K-means algorithm is applied to the raw data to cluster the data to divide the operating area,
Calculate the center point of the data for each operation area,
Standardize the result data corresponding to the operation area based on the data center point,
A pre-processing unit generating pre-processing data by generating learning data through the standardized result data;
A feature engineering unit to generate a feature from the preprocessed data;
A modeling unit generating the RUL prediction model by inputting the feature into a machine learning algorithm and selecting an exponential function evaluation model as a cumulative deterioration with time as a RUL estimation model;
Ship propulsion equipment failure prediction system comprising a. delete The method of claim 9,
The operating environment data includes outside temperature, outside pressure, propulsion lever,
The result data includes engine exhaust temperature, engine torque, engine vibration, propulsion shaft vibration. The method of claim 9,
The preprocessing unit,
A ship propulsion equipment failure prediction system comprising performing one or more of the following data: integration, transformation, reduction, and cleaning, in order to offset the impact on the operating environment. delete The method of claim 9,
The feature engineering unit,
A feature extractor for extracting a feature using the preprocessed data;
A feature selector configured to select a feature based on a criterion recognized by the developer from the extracted features (preprocessing data);
A status indicator generator for generating (designing) a status indicator representing a health indicator of a device over time through a principal component analysis (PCA) that fuses the selected feature;
Ship propulsion equipment failure prediction system comprising a. delete The method of claim 9,
The modeling unit,
Using the state of the equipment according to the time derived through the status indicator as a health indicator, the health indicator is applied to the RUL prediction algorithm using machine learning to predict the failure point in real time according to the learning result. Equipment failure prediction system.

Images