KR Webzine Vol.166

1. Background

Big data can predict the future and provide multi-faceted insights for decision-making. It allows flexible responses to changes in circumstances by examining various possibilities through data analysis. Efforts to create new value using data have been the driving force behind a change in the technological paradigm of the shipbuilding and offshore industry. 

 

All-round technology development, such as smart ships and smart ports, is actively progressing. Digital twin technology is an intensive result of various 4th industrial revolution technologies. It is the core of future technology, encompassing core technology development to enable ship level integration, operational, management and control technology. 

 

In order to provide “remote inspection services using digital twin core technology and regularization”, KR will develop core element technology to create the digital twin of ships for condition monitoring and maintenance technology, remote inspection technology, engineering information visualization technology, and digital model and simulation. 

 

We have been developing core element technologies for many years, and will be establishing an independent technology base, preparing classification inspection rules and certification standards to provide more advanced digital services for unique inspection tasks. 

 

As mentioned above, Condition Based Maintenance (CBM) is a technology that can be converted using the existing maintenance system of a ship to become the CBM-based maintenance system of future autonomous ships, as shown in Figure 1. The need for CBM is emerging in terms of guaranteeing the stable operational performance of ships, prevention of engine failure, and efficiency of operational management.

 

The application of the technology improves the efficiency of ship management and at the same time enables soundness evaluation and accident prevention through real-time information analysis, so it can be expected to reduce the incidence of ship accidents due to engine failure. Through this technology, ship managers will receive operational and decision-making support that can realize operational efficiency and operating expense (OPEX) reduction.

 

Figure 1. The conversion of ship maintenance technology system

 

 

2. Technical Definitions and Procedures

CBM technology is defined as a technology that proactively optimizes maintenance measures, unlike the existing calendar-based or running time-based methods.  It does this by monitoring the state of machinery and diagnosing and predicting failures. KR has established its own development framework for CBM technology as shown in Figure 2, and is developing an engineering-based failure analysis report, data production/acquisition, data pre-processing and feature analysis, and an AI-based failure diagnosis and prediction algorithm are in development. 

 

In this way, data frames are derived through data collection/transmission/storage and pre-processing analysis, and a failure diagnosis and prediction algorithm is developed using the data frames. The diagnosis or prediction result derived from the algorithm is transmitted to the crew, who are the user, and based on the result, the decision-making method to establish the maintenance method and detailed plan can be supported.

 

Figure 2. CBM technology definition and development process

 

 

3. Selection of core machinery systems

The ship's engine system is intensively systemized with the core equipment required for the propulsion power and power supply. In order to confirm the reliability and operational safety of the engine operations, a proactive failure diagnosis and prediction function based on real-time condition monitoring data is required. 

 

The engine system is composed of the main propulsion engine system and the generator engine system that directly affects the operation safety and performance of the ship. When an accident occurs due to the failure of the system, it threatens the safety of the ship and can lead to serious marine accidents at the same time. As a result, it is necessary to detect the signs of failure in advance using the CBM system and to establish a maintenance plan to ensure the safety of vessel operations and to ensure excellent quality control. 

 

Considering the importance of the equipment constituting the engine system of the ship, the main propulsion engine and generator engine of the ship, the pump that supplies fuel oil, lubricating oil, and various coolants, the purifier, and the pipe that is the system supply line were selected as the target equipment for CBM technology. 

 

In addition, the piping system was considered for the seawater piping and steam piping, which carry a high possibility of damage and parts of the piping are very important and directly related to the core engine system, except for fresh water, bilge (drain), and fire extinguishing piping. In the case of seawater piping, since it is an open circuit system that is discharged overboard after passing through a heat exchanger, it is impossible to apply corrosion protection through water treatments such as boiler water and fresh water cooling water, so corrosion by chlorine ions and erosion inside the pipe according to the flow rate is accelerated. This means there is a high possibility of pipe damage. 

 

In addition, the steam pipe is commonly damaged in the form of physical erosion due to water hammering, which leads to a very high potential risk of damage to adjacent equipment or personal injury when high-temperature and high-pressure steam is leaked or released due to rupture. 

 

Since the inspection of the pipe is limited to valve inspection and replacement and the replacement of the gasket at the pipe connection part during regular docking, it is necessary to introduce a CBM system to diagnose damage based on real-time condition monitoring. In this study, as shown in Figure 3, a total of 5 types (main propulsion engine, generator engine, pump, purifier, and piping) were selected including the ship's core engine system, auxiliary equipment, and piping.

 

4. The need for high-quality data

The introduction of CBM technology has various advantages in terms of ship operation and management, but it is necessary to recognize and overcome the limitations in the current technology level. First, the configuration and installation of a monitoring system for data measurement may incur significant costs. In addition, since the CBM system is also a system that needs to be maintained, the workload of crew members and ship managers is inevitably increased. 

 

In the onshore or aviation industry, the operational data acquisition infrastructure and communication environment conditions are favorable, so high-quality data is abundant, and it has the technical and empirical know-how and research results accumulated over the past decades, so the CBM technology field is quite mature. On the other hand, the shipbuilding and marine industry (especially in the case of ships) have limitations in terms of external environmental factors such as data acquisition, management, transmission and analysis for various reasons such as the continuation of the existing maintenance system, low infrastructure level, and communication restrictions. 

 

Nevertheless, since CBM technology can be applied to existing operating ships and is the core technology of future autonomous ships, it is required to develop inspection standards and rules in connection with the ship life cycle and failure diagnosis and prediction technology (H/W & S/W) with the Classification inspection system. 

 

The accuracy and reliability of the algorithm must be guaranteed first, in order to derive the result that supports the final decision through the algorithm analyzing the data acquired from the hardware of the state monitoring system. In other words, algorithms trained on inaccurate data with no guaranteed quality do not produce reliable results. 

 

It is easy to develop an algorithm with excellent performance when high-quality data, that is, normal and abnormal, are clearly classified and, in particular, when the abnormalities are clearly defined as to what kind of failure. As a result, the need for high-quality data in the development of CBM technology is vital. 

 

However, high-quality data for the ship's engine system is insufficient. The reasons are, first of all, for several years existing ships have been performing systematic and planned maintenance on board ships. The problem is that, rather than accurately recognizing and analyzing the root cause in the event of a failure, immediate repair measures are taken to secure ship operation safety and minimize losses. Data that can diagnose abnormal conditions, that is, data that can reflect the failure phenomenon very sensitively, is practically insufficient. 

 

On the other hand, if we look at the case of directly producing data onshore, manufacturers generally have onshore test beds for marine engine systems, but mainly use them for product performance tests. Paying for and acquiring data will inevitably have a negative impact on corporate productivity. As a result, KR intends to continuously develop CBM system hardware and software, starting with a failure mode simulation experiment to secure high-quality normal and abnormal data for five types of ship machinery systems (engine, generator engine, pump, purifier and piping).

 

5. Failure mode data production 

The failure mode data production research for five core ship engine systems consists of failure analysis and simulation scenario development, normal and abnormal state data production experiments, and preprocessing and feature analysis for data reliability evaluation. In this article, only failure mode data production research for the main propulsion engine and generator engine among the five core engine systems will be dealt with. 

 

For data production experiments, it is necessary to acquire data by starting the actual marine engines and generator engines owned by KR, so it is essential to conduct a detailed failure analysis and to establish an experimental plan in advance. First of all, the main occurrence of certain failures and risks in the event of failure were reviewed for ship 2stroke low-speed engines and 4stroke medium-speed generator engines. A Fault Tree Analysis (hereinafter referred to as FTA) was completed to specify the failure items, causes of failures, and the phenomena that can identify failures by organizing each failure item and analyzing its causes and phenomena. The FTA largely classified the internal combustion engine systems into machine driving systems, combustion and gas emission systems, consumption and gas supply systems, and analyzed the correlation by specifying failure items, causes, and phenomena for each sub-component, and identified additional measurement items to monitor failure phenomena. 

 

Based on the final derived FTA results, and through consultation with external experts with many years of engineering experience, 10 main propulsion engines and 5 generator engines were selected as the failure modes to be simulated, as shown in Table 1. In addition to the existing monitoring system for the main propulsion engine and generator engine, additional monitoring systems such as cylinder combustion state measurement sensor, vibration feature sensor, and high-speed camera were established to actually implement each failure mode, in order to secure high-quality, large-capacity data.

Table 1. Main propulsion engine and generator engine system failure mode.

 

2stroke main engine system (10 cases)
Failure mode	Failure mode item	Major failures phenomenon
FM1	Gas leakage at exhaust valve	exhaust gas leak
FM2	Non-return valve leakage of FIV & suction valve	Decreased fuel oil injection quantity, leakage of suction valve
FM3	Increased back pressure	Increased exhaust gas flow resistance (back pressure)
FM4	Blocked orifice of exhaust valve	Residual fuel oil bubbles, knocking
FM5	Broken piston rings	Blow by
FM6	Blocked grating in the exh. gas receiver	Poor T/C exhaust gas flow, T/C performance degradation
FM7	Low scav. air pressure	Scavenging pressure drop
FM8	Chain slack for CAM shaft	Poor fuel injection and exhaust valve timing
FM9	Nozzle leakage	Fuel oil leakage, incomplete combustion
FM10	Low opening Pressure at FIV	Fuel oil leakage, incomplete combustion

4 stroke generator engine system (5 cases)
Failure mode	Failure mode item	Major failures phenomenon
FM1	Turbocharger nozzle ring damage	T/C speed & scavenging pressure decrease, incomplete combustion and deterioration of T/C performance
FM2	Tappet clearance difference	Poor combustion condition, increased fuel consumption rate, lowered Pcomp, Pmax
FM3	Exhaust valve spindle damaged & leakage	Heat load increase, exhaust gas temperature rise, Pcomp & Pmax decrease
FM4	Fuel nozzle leakage & hole blockage	Lower engine performance, lower fuel injection characteristics
FM5	Deteriorated cooling performance for air cooler	Increase in scavenging air temperature, decrease in scavenging air supply, incomplete combustion

6. Failure diagnosis algorithm development

The normal and abnormal state data obtained through failure mode simulation was analyzed and preprocessed using various exploratory analysis techniques such as principal component analysis and correlation coefficient analysis, and the core feature data was extracted through multivariate data statistical analysis. 

 

Using engineering knowledge and mathematical evidence, the multivariate feature factors of cylinder pressure data and vibration data were extracted, and operation data were merged based on measurement time to generate the data frames necessary for algorithm development. 

 

KR applied various machine learning and deep learning techniques such as AutoML, RP+CNN, and LSTM using the data, and derived algorithms that could classify the failure mode of the target equipment with an accuracy of more than 95%. In the simulation experiment, the start-up of the main propulsion engine and generator engine system was implemented similarly to the RPM and load conditions operated by the actual ship and ran continuously for more than five hours. 

 

Therefore, the learning and evaluation of the algorithm model should be made in consideration of the interaction and time series characteristics between multiple data variables. In addition, in the event of an engine failure, not only one data changes, but also multiple data changes simultaneously, so multivariate analysis is required, and it is essential to structurally understand the pattern changes of multivariate data for each failure mode. Therefore, an algorithm that can understand the dynamic and organic interactions of hundreds of time series data produced through simulation tests of failure modes of main propulsion engines and generator engines was applied. 

 

Fig. 3 illustrates the algorithm performance evaluation for binary classification that distinguishes normal and abnormal by using simulation data of the failure mode of the main propulsion engine and generator engine. This is a preliminary evaluation of data reliability verification and application of abnormal diagnosis algorithms. An excellent diagnostic performance was ensured with F1 score of 98% or more among each performance indicator, and as a result, the reliability of the main propulsion engine and generator engine failure mode data was verified. 

 

Fig. 4 uses a data frame configured by integrating cylinder internal pressure, vibration characteristics, and operation information as the result of diagnostic algorithm performance evaluation for each failure mode of the main propulsion engine and generator engine. Excellent diagnostic performance was derived for each of 10 and 5 failure modes, and excellent performance was secured over 96% for the main propulsion engine and over 99% for the generator engine on the F1 score performance index. Based on the analysis results, the possibility of qualitative evaluation to distinguish pattern changes between time series data variables was confirmed, and the algorithm to be developed in the future will be applied to actual ships to enhance reliability and evaluate applicability of the algorithm itself.

 

Figure 3. Performance evaluation of normal and abnormal binary classification
for main propulsion engines and generator engines.

Figure 4. Performance evaluation of diagnostic algorithm
by failure mode of main propulsion engine and generator engine.

 

7. Consideration and future plans

The use of CBM technology for ship machinery systems requires a step-by-step transition from the existing maintenance system. In addition, there is a need for a platform that can provide various levels of functions so that users, that is, sailors and ship managers, can selectively apply and utilize the applicability of the technology comprehensively.

 

Accordingly, KR will link the CBM system with the shipping company's general matters and requirements and the classification inspection rules, matching the failure items of the shipping company with the classification rules for a gradual transition from the existing maintenance system to the CBM-based maintenance system. Through this, we intend to establish a list of failure items and related data to be monitored. 

 

In the future, the CBM system should be applicable in consideration of the level of data acquisition infrastructure for the finals, and should be able to provide functions such as detection, failure diagnosis, failure prediction, maintenance plan, and cost analysis depending on the level of CBM system. As a result, shipping companies can selectively apply the CBM system without improving the infrastructure construction of the ship data acquisition system. Regarding ship inspection services in the future, KR intends to establish CBM technical advancement and classification inspection rules to enable a phased transition from the existing decades-old ship management system to the CBM-based maintenance system.

 

Figure 5. CBM Technical Level and Functional Requirements