Real time fault detection and diagnosis system for onboard engine room

2.1 Characteristics of exhaust gas temperature of cylinder

The exhaust gas temperature of the main diesel engine and generator engine were obtained as time series data from the engine log book. Table 1 summarizes details of the ships used in this study. Relatively new and old ships and various types of engines were used to comprehensively understand the engine characteristics.

Table 1:

Details of ships used to gather engine data

Table 2 presents the statistical analysis of the M/E exhaust gas temperature of N_MS.

Table 2:

Descriptive statistics of M/E for N_MS

As shown in Table 2, the standard deviation of the No. 2 cylinder exhaust gas temperature was large in comparison with those of other cylinders. This implies that the temperature of this cylinder varies largely and the combustion is not even. There are several possible reasons that can explain the large variation in exhaust gas temperature; however, it is suspected that the spray of fuel or the exhaust gas valve and piston ring conditions are the most likely reasons. These types of value should be compared in the long term and it must be evaluated if the variation range is within a certain range. If the combustion statuses of other cylinders are similar, then the Pearson correlation coefficient (CC) of each cylinder should also be within a similar range. Table 3 shows the relation of Pearson correlation coefficient with other cylinders.

Table 3:

CC comparison of M/E exhaust gas temperature

The CC of the exhaust gas temperature for M/E cylinder No. 2 was much lower than the CCs of other cylinders. This indicates that the combustion status of the No. 2 cylinder may be unstable and the combustion characteristics of the cylinder may be different than those of other cylinders. It is necessary to carefully observe the change in cylinder No. 2.

Table 4 and Table 5 show the descriptive statistic No. 1 & 2 G/E of this ship.

Table 4:

Descriptive statistic of No.1 G/E for N_MS

Table 5:

CC comparison of exhaust gas temperature for No.1G/E

The standard deviation of the No. 5 cylinder was slightly higher than that of others; thus, its CC was compared with that of each cylinder. The CC comparison of No. 1 G/E is shown in Table 5. The lowest CC related with cylinder No. 5 was that related with cylinders No. 3 and No. 6. However, the CC was very high with respect to the significance level of 0.01; thus, the trends of CC change must be checked.

Table 6 shows statistical analysis for No.2 G/E. In comparison with that of No.1 G/E, SD of No.2 G/E is higher than that of No. 1 G/E.

Table 6:

Descriptive statistics of No. 2 G/E for N_MS

The average exhaust gas temperature and SD of the No. 4 cylinder was highest in comparison with that of others; the combustion of the No. 4 cylinder was suspicious. Investigation of CCs is shown in Table 7. As shown in Table 7, low CCs were related with cylinders No. 4 and No. 2, and with cylinders No. 1 and No. 2. Thus, it is important to observe the trends of cylinders No. 2 and No. 4.

Table 7:

CC comparison of exhaust gas temperature for No.2G/E

With the above investigation using statistical calculations of exhaust gas temperature, the combustion status inside each cylinder can be observed. In a diesel engine, abnormal combustion can occur due to various reasons such as lack of scavenging air and compression pressure, condition of piston ring, exhaust valve, fuel valve and fuel, lubrication of cylinder liner, and carbon deposit.

These reasons may be the cause and the results of reasons, because the reasons and results are to related each other. These reasons may be found using a fault diagnosis tree and decision making tree tracking phenomena. Therefore, it is very important that faults can be detected while engines are running.

If descriptive statistical analysis can be conducted using time series data from the alarm monitoring system (AMS), faults can be detected as shown in above investigation. Descriptive statistical analysis can be conducted using the following equations.

where,

where,

2.2 Structure of real time fault detection system

Engine room data can be categorized into the combustion system of the diesel engine such as exhaust gas temperature, T/C input and output gas temperature, and heat exchangers, including cooler and heaters, and pumps and motor to circulate cooling fluid. Data can be input in real time series from the network via AMS or data files from the database. These data can be input separately for each categorized fault detection system, as shown in Figure 1. The fault detection system detects fault from time series data using descriptive statistics. If the detected fault is confirmed with respect to relationships with other cylinders, then diagnosis is performed by the diagnosis and decision making tree. If the degree of degradation is obtained, fault prediction can be estimated.

Figure 1:

Structure of fault detection and diagnosis system

Even if the same type of machinery is installed and operated, the value of CC of each measurement may be different depending on the ship. Thus, it is necessary to decide the proper value of CC to be considered as a fault.

2.3 Fault detection algorithm

Time series data were input to the fault detection component shown in Figure 1 separately depending on their type. Each fault detection component calculates the moving average (MAVR) and moving SD (MVSD) for each type of data. When the MVSD is over the predetermined value (i.e., the fault detection SD (FDSD)), the CC between each measurement item is calculated. For example, for the main engine, the load or RPM is related to the output of each cylinder because the combustion of the cylinder is resulted in output of cylinder. To detect faults in the cylinder, the MAVR and MVSD of the load or RPM are continuously calculated. When the MVSD of load or RPM is over the FDSD, the CCs of exhaust gas temperature are calculated for all cylinders to identify the CC that is lower than the predetermined value (i.e., FDCC). In case of the generator engine, the load in terms of kilowatts can be used instead of that in terms of RPM, because the generator engine rotates with a constant RPM.

Figure 2 shows the fault detection algorithm. When a fault is detected and remains over a predetermined time, the fault diagnosis system attempts to identify its cause using expert knowledge and a diagnosis tree. The diagnosis tree consists of if-then components and expert judgement and knowledge.

Figure 2:

Fault detection algorithm

2.4 Fault diagnosis tree with expert judgement and knowledge

Reasons of faults are composed with various reason and combination of those reasons. To identify out the exact reason, information regarding the fault as well as expert judgement and knowledge are necessary. Figure 3 shows the fault diagnosis tree when the fault detection system detects a specific cylinder fault. The scavenging air pressure is dependent on the RPM or load of the engine; in addition, the T/C RPM is dependent on the temperature of the gas input to the T/C. When the system detects a cylinder fault, values of scavenging air pressure are compared and used to diagnose the fault of the scavenging or turbocharger system.

Figure 3:

Fault diagnosis tree

When the engine rotates at a certain RPM, exhaust gas flows into the T/C. The T/C drives the blower connected to it. The blower creates air pressure to scavenge the cylinder and provide air super charged in the cylinder with fresh air. A scavenging air pressure is properly established by the T/C rotation in a normal case. The engine RPM, T/C RPM, and scavenging air pressure are correlated by the cause of the fault and the result chain. Combustion in the cylinder is related to the quantity of air, the fuel and air mixture in the cylinder, the spray status of the fuel valve, the injection time of the fuel, and the compression pressure, which is related to the status of the piston ring and the lubrication of the cylinder wall. These reasons may be results of combined reasons.

When a cylinder fault is detected, the diagnosis tree confirms whether the scavenging air pressure and T/C rotation are appropriately established. The most significant reasons for cylinder faults are T/C and piston problems. Another reason for unstable combustion could be exhaust and fuel valve leakage, combustion and injection time faults. Because the injection time is principally the most significant reason and can affect all cylinders, cylinder faults can occur frequently due to fuel valve leakage and spray conditions. These types of faults change frequently depending on the conditions. Expert engineers can check their status and plan maintenance measures. Figure 3 shows the series of steps in identifying the reasons behind the faults. If the diagnosis tree finds significant faults such as T/C faults and Pcomp (compressed pressure by piston movement) faults, which involve lack of compression of air in the compression stroke of the piston, then an expert engineer needs to make a decision by comparing and considering more information.

3.1 Implementation of real-time fault detection and diagnosis system

An automatic real-time fault detection and diagnosis system (RTFDDS) is developed using C++ based on the description in Section 2. The RTFDDS reads data from spreadsheet files, which are acquired from a real ship’s engine logbook during a time period of 6 months, and calculate the MAVR and MVSD. According to the fault detection algorithm described in Figure 2, if the MVSD of the M/E load is over the predetermined value, then the CCs of all cylinders are calculated. If the CC is lower than the predetermined FDCC, then the faulty cylinder can be identified. The system is implemented only on the M/E for the time being and it will be developed in the near future for the G/E, head exchanger, and the system of pumps and motors.

The predetermined value of the MVSD and FDCC are quite different depending on the ship even if the same machine is used. Therefore, the proper degree of value should be investigated and tests should be carried out after onboard installation.

Figure 4 shows how the RTFDDS reads data from the spreadsheet and calculates and displays online raw data (red), MAVR (yellow) values, and MVSD (green) values for a selected item. The upper graph shows the M/E load or RPM and the lower graph shows the selected item, which, in this case, is the exhaust temperature of the No. 1 cylinder.

Figure 4:

Real-time monitoring of RTFDDS

The graph on the left of Figure 5 shows the CC and MVSDs of selected items 1 and 2; however, in reality, the MVSD of the load or RPM for the M/E is calculated continuously in the background and when the MVSD is over the FDSD, the system calculates the CCs of all cylinders.

Figure 5:

CC calculated between exhaust temperatures of No. 1 & 2 cylinder

When the CCs of the cylinders are lower than the FDCC, then the LED lamps shown in the bottom part of Figure 5 turn red. The FDCC can be adjusted by changing the input value as shown in the middle box of Figure 5.

Figure 6 shows an enlarged version of the graph in Figure 5. The graph in Figure 5 only shows the change in the CC for the selected item; however, all the CCs are calculated when the MVSD of the load or RPM of the M/E is over the FDSD in the background of the system.

Figure 6:

Enlarged graph of Figure 5, red: item 1 (No. 1 cylinder), yellow: item 2 (No. 2 cylinder), green: CC between Nos. 1 & 2

Author Contributions

Conceptualization, Y. YU; Methodology, Y. YU; Software, Y. YU and S. R. Baloch; Validation, S. R. Baloch and Y. YU; Formal Analysis, S. R. Baloch; Investigation, Y. YU; Resources, Y. YU and S. R. Baloch; Data Curation, Y. YU and S. R. Baloch; Writing—Original Draft Preparation, S. R. Baloch; Writing—Review & Editing, Y. YU; Visualization, Y. YU and S. R. Baloch; Supervision, Y. YU; Project Administration, Y. YU; Funding Acquisition, Y. YU;