Applied Thermal Engineering 80 (2015) 404e412

Contents lists available at ScienceDirect

Applied Thermal Engineering

journal homepage: www.elsevier.com/locate/apthermeng

Research paper

Condition-Based Maintenance for medium speed diesel engines

used in vessels in operation
Oihane C. Basurko a, *, Zigor Uriondo b
a
AZTI e Marine Research Division, Herrera Kaia, Portualdea z/g, 20110 Pasaia, Spain
b
Department of Thermal Engineering, University of the Basque Country UPV/EHU, Alameda Urquijo s/n, 48013 Bilbao, Spain

h i g h l i g h t s

Condition-Based Maintenance is defined for medium-speed diesel engines in operation.

ANNs have proved to be suitable to model engine performance of commercial vessels.

Fishing vessels with medium speed diesel engines are assessed as a case study.

A novel data monitoring and processing strategy is presented for fishing vessels.

An affordable onboard Condition-Based Monitoring approach for vessels is achieved.

a r t i c l e i n f o a b s t r a c t

Article history: Condition-Based Maintenance for diesel engines has contributed to reliability, energy-efficiency, and cost
Received 25 August 2014 reduction. Both, the modelling of engine performance and fault detection require large amounts of data;
Accepted 29 January 2015 usually, these are obtained on a test bench. In contrast, in operative engines, provoking faults onboard is
Available online 7 February 2015
not a viable proposition. Condition-Based Maintenance, fault detection and diagnosis need to be solved
on engines installed in commercial vessels: the present contribution answers this need. A medium-speed
Keywords:
diesel engine was monitored using thermocouples, pressure sensors, a propeller shaft torque meter and
Condition-based monitoring
fuel oil flow-meters, during more than 10,000 running hours. Monitored data were used to train a three-
Artificial neural network
Energy efficiency
layer feed-forward neural network, to generate the engine performance model; thus, determine the
Medium-speed diesel engines engine's fuel consumption and faulty conditions. The faulty conditions considered were: (1) a polluted
Fishing vessels turbine; (2) a dirty air filter/compressor; (3) a dirty air cooler; (4) and bad fuel injection, i.e. bad com-
bustion. The sensor's precision and the experience gained by monitoring the engine served as a baseline
to define the fault threshold values. The results proved the feasibility of installing a Condition-Based
Maintenance, for vessels in operation, by monitoring engine performance and analysing the data with
the aid of artificial neural networks.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction revise the way that fuel is consumed [2]. Research is growing
within the published literature regarding energy efficiency. For
International shipping emits some 2.7 % of the global green- example, in the fishing sector, research has focused upon: energy
house gas (GHG) emissions. Following the foreseen growth in audits; the monitoring of the engine and consumption parameters;
shipping, and in the absence of policies, it is expected that ship- the optimisation of fishing gears; and energy management and
borne GHG emissions will increase by a factor of 2e3 (compared saving solutions [3].
to emission in 2007) by 2050 [1]. As such, more effort is needed to Optimised maintenance of engines is another solution to
improve energy efficiency. Engine properties change in relation to
the years spent in operation, together with the reparations carried
out throughout those years. Hence an optimal maintenance helps
* Corresponding author. Tel.: þ34 667174394; fax: þ34 946572555. reducing unnecessary expenses and pollution. The only exception is
E-mail addresses: obasurko@azti.es (O.C. Basurko), zigor.uriondo@ehu.es
(Z. Uriondo).
NOx emission that it is reduced due to incorrect maintenance and

http://dx.doi.org/10.1016/j.applthermaleng.2015.01.075
1359-4311/© 2015 Elsevier Ltd. All rights reserved.
 O.C. Basurko, Z. Uriondo / Applied Thermal Engineering 80 (2015) 404e412 405

engine malfunction, as found in a laboratory-scale experiment that an engine performance, the data need to be selected carefully and
simulated a fuel pump malfunction [4]; and in engines with faulty transformed.
injection pressures [5]. This study analyses how to implement a CBM for a medium-
Shipowners have adopted three maintenance strategies, to speed marine diesel engine, as used in operative fishing vessels.
ensure that engines function under their optimal conditions. The Since most of the models considered use ANN to develop suc-
traditional corrective maintenance, i.e. corrective repairs, takes cessfully prediction model of engine performance and the failure
place once a part is broken, or a machine has stopped working. detection, the same approach has been adopted in this contribu-
Secondly is the scheduled maintenance or the routing servicing, tion. Likewise, the strategies followed to establish a CBM onboard a
which replaces parts subjected to degradation or wear, after a commercial fishing vessel are explained. Also, the monitoring of the
certain period of operation. The last, the least applied, is Condition- engine parameters and the application of ANN method are intro-
Based Maintenance (CBM), which predicts when the failure will duced. The main difference in utilising published literature lays in
occur; this is based upon engine monitoring and performance in- the complexity of designing a CBM strategy, for an operative
dicators. Despite the fact that the scheduled maintenance and the engine.
CBM are complementary, the CBM is by far the most cost-effective
approach and the one which enhances the life expectancy of the
engine [6]. 2. Material and methods
Many authors have tried to monitor and model engine perfor-
mance, to predict parameters such as fuel consumption. Examples 2.1. The engine and its use
are the performance maps developed for an eight cylinder four
stroke diesel engines proposed by Çelik [7], and the prediction The engine monitored here is a typical medium-speed diesel
model for the brake specific fuel consumption, effective power, engine, which is used commonly in fishing vessels and tugboats, as
average effective pressure and exhaust gas temperature of an eight the main engine; likewise, in shipping, as an auxiliary engine
cylinder four stroke methanol diesel engine proposed by Çay [8], burning heavy fuel oil. The engine lies within the low power range
both using intelligent algorithms. However, only a few have linked of medium-speed engines; as such, the methodology applied in this
both, i.e. the modelling of the engine performance and the CBM, as study can be applicable to larger engines, which are being used in
the way to reduce costs and optimise the energy consumption operative vessels. The basic technical details of the engine are listed
onboard. in Table 1.
The input and output variables, faults, and modelling methods The engine monitored is used in a commercial bottom-otter
used in such models are varied. Regarding the variables the engine trawler. More information regarding the vessel and its working
speed, engine power, cooling water temperature and the combus- pattern are available elsewhere [3]. Summertime is used to perform
tion pressure of the engine cylinders are the most common; all but any required reparations and maintenance.
the cooling water temperature that is used as an input variable, The vessel is equipped with a shaft generator (or alternator),
appear interchangeably as input and outputs in the consulted which is connected to the main engine. The power demand of the
literature [7e9]. However, there are other variables, such as engine engine originates from the propeller and the alternator: the first is
vibration and exhaust temperature [10]. A similar diversity can be used to move the vessel; the second to run the onboard electric-
found in the selection of faults, which include leak failures [9] as consuming equipment, including the fishing gears.
well as shaft imbalance, cylinder misfire, and clogged intake [10]. The electric load is somewhat constant whilst the vessel is
However other researchers have adopted a more comprehensive cruising (from the port to the fishing ground and vice-versa, and
approach including the previously mentioned faults and those between two fishing grounds); it suffers small changes, due to the
related to exhaust gas, turbine and compressor performance [11]. larger electric-consuming equipment used only during fishing ac-
Likewise, the techniques used to design the engine performance tivities. Furthermore, during cruising, the alternator is coupled
predictive model and the maintenance model go from the usage of most of the time. When uncoupled, the power demand comes from
intelligent algorithms, such as: Artificial Neural Networks (ANN) the propeller to move the vessel. For certain cruising speed, such
[10,12], genetic algorithms [13], and fuzzy logic [14]; to less so-
phisticated approaches, e.g. non-linear multivariate statistics [15];
and experimental formulae based upon thermodynamics [16]. One
Table 1
of the most innovative studies is probably that presented by Mes- Technical details of the engine studied.
bahi [12]. Here the author proposes not only a model that predicts
Parameter Value
the fuel consumption, based upon several inputs, but a failure
detection system is also presented; this is based upon pattern Number of cylinders 8
Cycle Medium-speed 4 stroke
recognition, which optimises the input data using reverse
Induction system Turbocharged
modelling. Compression ratio 15.5:1
Most of the contributions on the CBM of marine diesel engines Stroke  bore 210 mm  290 mm
are based upon data collected from tests produced under controlled Displacement 80,36 L
laboratory conditions. In such cases, data acquisition is not sub- Mean Effective Pressure 1923 MPa
Max. Combustion Pressure 18.14 MPa
jected usually to field operation problems; thus, quality data are Nominal Speed 800 rpm
easier to obtain. More significantly, the faults are always induced by Net power 1030 kW
the authors; this facilitates the definition of ‘failure threshold Cooling system Water cooled (fresh water)
values’. In contrast, onboard commercial ships data collection is not Fuel system Mechanical
Type Injector/1 pump per cyl.
as straightforward. Sensors and measuring equipment are sub-
Injection pump Bosch Type
jected to harsh environments, where inducing faults onboard is Injector 8 orifices
unviable. Likewise, the activity pattern of fishing vessels is more Nozzle opening pressure 35 MPa
complex than that of commercial ships. Furthermore, data acqui- Governor type Hydraulic
sition in adverse conditions hinders data quality; as such, data Fuel type Marine Diesel Oil
Activity Used within Fishing Vessel (stern trawler)
anomalies are common. Since data quality is essential for modelling
406 O.C. Basurko, Z. Uriondo / Applied Thermal Engineering 80 (2015) 404e412

power demand is dependent strongly upon the sea and vessel was developed with LabVIEW. Based upon monitored data, the fuel
conditions. oil consumption and specific fuel oil consumptions (SFOC) were
Whilst the vessel is fishing, the speed is reduced and the pro- calculated and, subsequently, incorporated into the LabVIEW soft-
pulsion load is used to move the trawling gears. Such power de- ware and shown in real-time. As such, data collection incorporated
mand is usually variable and depends upon several variables, such both the monitored and derived data.
as fishing condition, depth, species, otter doors [17], and cable The data acquisition system was synchronised with the main
length [18]. engine; once the engine was started, the system was initiated: the
quantity of data registered was extensive. The engine was moni-
2.2. Condition-Based Maintenance for vessels in operation tored during two complete fishing seasons, i.e. 2010e2011 and
2011e2012, registering at a frequency of 0.1 Hz. In total, more than
The procedure followed to establish the CBM for the engine of a 10,500 h were recorded. Data were abstracted from the vessel every
commercial vessel consisted of four steps (Fig. 1): (1) data acqui- two weeks. At the same time, the efficient functioning of the data
sition; (2) data selection, purging and conditioning; (3) develop- acquisition system was checked. When needed, such system and/or
ment of engine's performance model; and (4) fault diagnosis. malfunctioning sensors were replaced, to minimise data errors and
anomalies.
2.2.1. Step 1 e data acquisition
The engine monitoring consisted in collecting data related to: 2.2.2. Step 2 e data selection, purging and conditioning
the combustion of the engine; the turbocharger; the fuel condition; Only the fishing trips which registered correctly were analysed.
and the electric energy consumption of the alternator. Since the Data analysis consisted of determining the duration and energy
engine studied is installed onboard an operative vessel, the vari- consumption of each trip and each of the fishing activities, i.e.
ables monitored had to guarantee the collection of sufficient data; cruising to the fishing ground, fishing. The fuel consumption (L/h),
this was in an event of failure onboard, to provide enough infor- the shaft power (with the alternator coupled and uncoupled), the
mation to calculate fuel consumption, the simulation and faults power of the alternator (kW), the SFOC (g/kWh) (with the alter-
diagnosis. In total, 45 engine-related variables were monitored, nator coupled and uncoupled), and the total power demand (incl.
using intrusive and non-intrusive sensors and meters. The type of the propulsion power and the alternator) were calculated from the
sensors and meters used, together with their accuracy, are listed in data registered; this was once data were collected from the vessels
Table A.1 of the Appendix. and analysed. The monitored and derived variables are listed in
All input signals were recorded using National Instruments’ Table 2. Likewise, from all of the data, only those referred to the
CompacDAQ equipment and collected in a rugged industrial per- engine in cruising mode, i.e. when the ship is sailing from port to
sonal computer, located in the engine room. The acquisition system the fishing ground, was considered to develop the models. This

Fig. 1. Steps (4) followed to develop the Condition-Based Maintenance (CMB) of vessels in operation.
 O.C. Basurko, Z. Uriondo / Applied Thermal Engineering 80 (2015) 404e412 407

Table 2 first months after the break, were used to develop the ANN model,
Monitored and derived variables. i.e. more than 500 engine working hours.
Variable code Unit Description Whilst sailing, the engine required 60e75 % of its total load. The
Monitored variables
ship had a fixed-pitch propeller, which did not require full engine
Day e Present date power at nominal engine speed. This meant that the engine speed,
Hour hh:mm:ss Present time power and boost pressure values would not deviate too far from the
Torque kNm Propeller shaft torque usual values. If such values were considerably far from the usual
Shaft power kW Propeller shaft power
spread, they were considered as outliers. Data purging consisted in
RPM_shaft rpm Shaft's speed
BoostP Bar Boost pressure deleting rows that contained one or more of the criteria listed in
DP Air filter T/C mmH2O Pressure loss in the air filter Table 3.
DP Air cooler mbar Pressure loss in the air cooler Purged data were conditioned by deriving running averages
Amb. Press mbar Ambient pressure, measured
every 10 measurements, in order to dampen the variations and
in engine room
RH % Relative humidity, measured
instantaneous peak values. Likewise, data were randomised, in
in engine room order to avoid local minima and to maximise the learning process.

Amb. Temp. C Ambient temperature
RPM rpm Main engine speed
Texh.Cyl_No_1 
C Exhaust temperature of Cylinder n 1 2.2.3. Step 3 e development of engine's performance model

Texh.Cyl_No_2 C Exhaust temperature of Cylinder n 2

Texh.Cyl_No_3 C Exhaust temperature of Cylinder n 3
Input and output dataset

Texh.Cyl_No_4 C Exhaust temperature of Cylinder n 4

Texh.Cyl_No_5 C Exhaust temperature of Cylinder n 5
Texh.Cyl_No_6 
C Exhaust temperature of Cylinder n 6 The selection of the variables to be used is a normal procedure in
Texh.Cyl_No_7 
C Exhaust temperature of Cylinder n 7 the development of any mathematical model. The non-linearity,

Texh.Cyl_No_8 C Exhaust temperature of Cylinder n 8 the inherent complexity and the characteristic non-

Texh.T/C 1278 C Exhaust gas temperatures in the manifolds parameterisation of the ANN-regression makes difficult the use of
of the cylinders 1/2/7/8
Texh.T/C 3456 
C Exhaust gas temperatures in the manifolds
traditional analytical methods, to select variables [19]. Despite the
of the cylinders 3/4/5/6 ANN having been considered often as a ‘black-box’, the under-
 standing of the variables influence on the model has been discussed
Texh.T/C C Turbo-Charger exhaust gas temperature

Toutlet_air T/C C Air temperature at the outlet of the for some time, such as in biomedical applications [19], and in
turbo-compressor, inlet to the air cooler

modelling ecological phenomena [20]. A sensitivity analysis, as
Tair_inlet_Cyl C Cylinders inlet air temperature
Fuel.Temp 
C Fuel temperature proposed in [11], was used to determine the weighting of each
T fuel in 
C Engine inlet fuel temperature variable, in relation to the remainder, to select the most sensitive

T fuel out C Engine outlet fuel temperature for use in the ANN model.
L in L/h Fuel volumetric flow, engine inlet
L out L/h Fuel volumetric flow, engine outlet
M fuel in kg/h Fuel mass flow, engine inlet

ANN network
M fuel out kg/h Fuel mass flow, engine outlet
Speed kn Vessel speed The feed-forward network containing three layers of neurons

Latitude Geo-positioning can approximate any function to an arbitrary accuracy, given a

Longitude Geo-positioning
sufficient number of neurons in each layer [21]. Hence, a three-
kWh kWh Active electric energy consumption
of the alternator layer feed-forward network, with a sigmoid function as activation
Calculated variables function and back-propagation method for network training was

Aver_Texha C Exhaust temperature of all the cylinders used to develop the model. The weights were updated in an on-line
FuelConsa L/h Fuel consumption manner, i.e. after each data run, instead of at the end of the training
Acc.FuelConsa L Accumulated consumption
set. The amount of data used in developing ANN models consisted
Palternatorb kW Alternator power
Pshaft (uncoupled)b kW Shaft power without the alternator of 31,324 patterns; from these, 70 % is allocated for training, 25 % for
Pshaft (coupled)b kW Shaft power coupled with the alternator testing, and the remaining 15 % for cross-validation. Neither the
(Propeller shaft power þ alternator power) momentum nor the step size was modified; both were maintained
g/kWh (uncoupled)b g/kWh Engine's SFOC with alternator uncoupled
with values of 0.7 and 0.1 respectively. All of the networks were
g/kWh (coupled)b g/kWh Engine's SFOC with alternator coupled
Total.power kW Total engine power, including the power
trained with an epoch of 1000. The number of neurons was the only
demand of the alternator and the one parameter changed each time, varying from 20 to 50 hidden
needed for the propulsion.

Key.
a
Variable calculated in real time. It is incorporated in the software.
b
Variable calculated during the data analysis at the office. Table 3
Criteria followed for data purging.

Criteria Reason
approach was adopted to guarantee the repeatability of the engine
Any variable Empty cell Error in the acquisition
conditions, throughout all the selected trips. The fishing and system
manoeuvring phases were very irregular; hence, data referred to Alternator Palternator >100 kW Outlier (larger than 95
these modes were excluded. power percentile of data)
Fuel density was used to calculate consumption. In order to Palternator < 0 kW Error in the acquisition
system
maintain the results as accurately as possible, a sample was taken
Shaft power Pshaft (uncoupled) > 1500 kW Outlier (larger than 95
from each bunkered fuel. If the density differed from the original, (uncoupled) percentile of data)
the new value was updated into the acquisition system software. Engine speed RPM < 400 For excluding data related to
It was considered that the engine during its first month of fishing or manoeuvring.
operation, following the summer break, was in its best operational Boost pressure BoostP >1.5 bar Outlier (larger than 95
percentile of data)
conditions. Hence, the four fishing trips, taking place during the
408 O.C. Basurko, Z. Uriondo / Applied Thermal Engineering 80 (2015) 404e412

neurons, in order to find the optimum network to develop the threshold of error. The validation of the model was undertaken by
model. In addition to the correlation coefficient and the mean comparing data simulated by the model, with the data obtained
squared error (MSE) used usually in ANN models, the error of the from fishing trips, from the season 2010e2011.
model was calculated in each trial; that with lowest error was Engine performance maps were developed also for: any input
considered as the optimal. An error of 5 % was considered accept- data that fell within the observed sensor readings range; and any
able, following the published literature on ANN models [7]. combinations of output variables. A sample map is presented as
The engine performance model was built using the normalised Fig. 5.
input and output, together with vectors of weights of the best ANN
network structure, obtained by the software Neurosolutions. Lab- 3.3. The toolkit
VIEW programming was used to develop the performance model
and the user interface of the CBM. The interface includes: the en- The ANN model was used to design a computing tool, in Lab-
gine performance model; the fault diagnosis tool; and the data VIEW. The tool is expected to be installed in the engine room
acquisition system. alongside the data acquisition system used by the chief engineer
and the shipowner to control the fuel oil consumption of the vessel.
2.2.4. Step 4 e fault diagnosis Fig. 6 shows the screenshot of such a tool. The tool calculates the
The output of the performance model was used to determine percentage difference between the simulated value and the sensor
the failure condition. The fault diagnosis was based upon the main reading, for the same input values.
four areas of the engine that could generate an increase of the
instantaneous fuel consumption: (1) dirty turbine; (2) dirty air 3.4. Economic balance
filter/compressor; (3) dirty air cooler; and (4) bad fuel injection.
Table 4 lists the faults, variables and threshold values considered to The monitoring of the engine shown in this contribution,
define faulty conditions. together with the data acquisition system, has cost approx. 25,000
Faulty conditions were considered to occur when faults took V. Some of the equipment installed, such as the flow and torque
place together with a 3 % threshold increase of fuel consumption meters, can be used also for different purposes; these include
and the situation remained steady over time (discussed further in navigation optimisation and the reduction in fuel consumption.
Section 4). The faults were defined based upon the precision of the Likewise, monitoring different signals permits the establishment of
sensors and the experience gained by monitoring the engine. Fig. 2 an alarm system, which can warn the user of engine failure, before
represent a flow chart that shows the integration of the variables, any serious breakdowns or accidents occur.
ANN model and the fault conditions of the CBM developed for the The main engine's maintenance cost, in such a vessel type, de-
vessel studied. pends strongly upon the time passed since the last major overhaul.
The monitored engine has an annual fuel consumption of approx.
3. Results 840,000 L [3], i.e. 560,000 V at a current fuel price of 0.7 V/L. The
common major overhaul interval for the shipowner is every two
3.1. Selected variables years (10,000 running hours). The main engine's maintenance cost
is approx. 20,000 to 30,000 V [22], man work cost for major
Out of the 45 variables monitored, 6 variables were selected as overhaul is 15,000 V, and the cost of fuel injection pumps equip-
input and 15 as output. Fig. 3 shows such variables. ment 8000 V (Pers. Comm., shipowner and engine supplier).
Nevertheless, the monitoring of engine condition let the shipowner
3.2. Engine performance model extend the time elapsed between overhauls by at least an addi-
tional one year due to performance parameters were within
The optimal ANN architecture identified contains 6 input neu- acceptable values (below 3 % threshold). This increase of time
rons, 15 output neurons, and 49 neurons in the hidden layer (Fig. 3). elapsed between overhauls let the shipowner save a considerable
A sample of the prediction undertaken by the model, together with amount of money, reducing the payback time of the monitoring
sensor readings for the same input data are shown in Fig. 4. The equipment investment.
optimal network has demonstrated ‘good learning’, on the basis Considering a hypothetical 2 % of annual fuel saving, in response
that 97.1 % of the data used for testing was lower than the 3 % to the CBM [1], i.e. 11,200 V, the monitoring system could be repaid,

Table 4
Definition of fault conditions.

Fault Reason Variable involved Thresholda

1 Dirty turbine Texh T/C1278 or Texh T/C3456 >8 %

TexhT/C e [(Texh T/C1278 þ TexhT/C 3456)/2] >8 %
2 Dirty air filter/compressor DP Air filter/TC >NA
Toutlet_air T/C <5 %
BoostP <5 %
Tair_inlet_Cyl <8 %
3 Dirty air cooler DP Air cooler >NA
Tair_inlet_Cyl >8 %
BoostP <5 %
4 Bad fuel injection Texh.Cyl_No_1, Texh.Cyl_No_2, Texh.Cyl_No_3, Texh.Cyl_No_4, >8 %
Texh.Cyl_No_5; Texh.Cyl_No_6, Texh.Cyl_No_7, Texh.Cyl_No_8
TexhT/C1278, TexhT/C3456 >8 %
Toutlet_airT/C >5 %

Key.
a
Difference between simulated values by the ANN model and sensor reading for the same input data.
 O.C. Basurko, Z. Uriondo / Applied Thermal Engineering 80 (2015) 404e412 409

Fig. 2. Flow chart of the Condition-Based Maintenance for onboard marine engines installed in vessels in operation.

to the shipowner, within three years. On the basis that the invest-
ment is repaid in such a short time, together with the other added
benefits, in terms of safety onboard and breakdown prevention, the
idea of introducing a CBM in similar fishing vessels can lead to
important benefits.

4. Discussion

Medium-speed four stroke engines are installed in some 35 % of

the global fleet [23]. The fuel consumption of this type of engines
accounts for approx. 10e20 % of total ship fuel consumption [1].
Thus, their contribution to the global GHG emission of shipping
could be estimated to account for approximately 0.3e0.6 % of the
consumption. Based upon published literature, an efficient main-
tenance of engines, focused upon energy efficiency, could provide a
saving in annual fuel consumption of 1e2 % [1]. According to sales
records, some 6,500 units of this particular engine model have been
sold since the 1990's (Pers. Comm., Engine Manufacturer). Thus, the
CBM could present a market for medium-speed diesel engines;
similarly, contribute substantially to the reduction in air pollution
of marine applications, e.g. oceangoing shipping, short-sea ship-
ping, fishing, and tugboats.
The data acquisition onboard an operative vessel represents a
complex task. The fuel consumption monitoring onboard is difficult
to ascertain, due to the configuration of a fuel injection system and
the pressure pulses in a fuel system. There is also limitation caused
by the harsh electronic environment in which the acquisition sys-
tem is to be installed; this produced interferences, whilst electronic
noise is very common. Thus, performance data collected within
commercial vessels require an appropriate data selection, purging,
conditioning and pre-treatment, before they are ready to be used to
Fig. 3. Best ANN architecture obtained to model the performance of the engine (using
feed intelligent algorithms. Furthermore, the engine performance
6 input neurons, 15 output neurons and 49 neurons in the hidden layer), including the and the design of the faults need to be envisaged, considering the
statistical results of the network. realities occurring onboard the vessel.
410 O.C. Basurko, Z. Uriondo / Applied Thermal Engineering 80 (2015) 404e412

Fig. 4. Matching of real sensor reading values with ANN model predicted values for the same input values, and all output variables.

In contrast to commercial shipping, the engines of fishing

vessels are submitted to different engine loads and activities
during a trip, in relation to fishing activity. The vessel studied
here left port on Mondays, in very similar navigation conditions;
it was observed that the changes in power demand were insig-
nificant during the different trips examined, as can be appreciated
by the standard deviation of the shaft power, published by the
authors elsewhere [3]. In order to compare different trips and
instigate repeatability in the engine conditions, it is advisable to
select data from the beginning of the trip, i.e. whilst the ship is
cruising to the fishing ground, to develop the CBM model for
fishing vessels in operation. Likewise, to ensure that faults are
detected correctly.
The engine power demand changes with the propeller power
demand. In cruising mode, the propeller follows the propeller
‘cubic law’. Moreover, in rough weather conditions, under waves,
the load is changing constantly and the fuel consumption readings
change in line with the power demand of the propeller; if the po-
wer changes, so do the efficiency and the SFOC. The change is
usually very rapid, as the speed governor introduces immediately
more or less fuel, depending upon the power demand to maintain
Fig. 5. Engine's performance map for constant values: room temperature 25.5  C; the engine speed constant. In contrast, the change in the remainder
room pressure at 1017 bar; relative humidity 32 %; and with alternator uncoupled. of the variables is less noticeable. Therefore, small deviations could
 O.C. Basurko, Z. Uriondo / Applied Thermal Engineering 80 (2015) 404e412 411

Fig. 6. Screenshot of the engine's Condition-Based Maintenance computing tool, as installed in the engine room for the use of chief engineer/shipowner.

occur during operation, which could be due to both variable 5. Conclusions

operation conditions and sensor accuracy limits. Against the
abovementioned situations and to avoid incurring errors, during The use of intelligent algorithms has proved to be useful for
the definition of fault conditions, it is recommended that three key developing a CBM of vessels in operation. ANN has been used
aspects are considered when defining the faulty conditions that: (1) previously for this purpose: nevertheless, all previous models were
the fault threshold value lies outside the measuring uncertainty; (2) based upon laboratory tests, under controlled conditions. In
the same situation is retained steady over time; and (3) fuel con- contrast to the laboratory-based tests engine models in which
sumption increase is >3 % steady over time. faults have been induced in commercial vessels have to be detected
The selection of the 3% of threshold for fuel consumption was from registered data. The fault diagnosis has been established on
based upon different criteria. In order to define such a threshold, the four main areas of the engine, which can generate an increase in
the difference between simulated and sensor reading registered the instantaneous fuel consumption: (1) clogged turbine; (2)
values for variable “Fuel Consumption” was calculated for all trips clogged air filter/compressor; (3) clogged air cooler; and (4) bad
registered correctly. Average and standard deviations values below fuel injection.
3% were obtained in the fishing trips succeeding the summer break. A good approximation of the model suggests that the diagram
After that period, values ranged from 4 to 5 % sporadically. shown in Fig. 1 is appropriate. Thus, such a diagram can be used for
Furthermore, the difference between simulated and sensor establishing a CBM for installation in the engine room. Based upon
reading values were plotted as a histogram, to identify the fre- the diagram, a computing tool has been developed in LabVIEW. The
quency of occurrence of a particular difference range. The histo- tool will permit shipowners and chief engineers to control the
gram showed that most frequent range was 0e5 %, which occurred maintenance of the engine; likewise, optimise the engine, based
in 96 % of the data assessed. Moreover, 82.7 % of the cases presented upon energy efficiency. The failure modes should be refined for the
a difference below 3 %. Such data referred to discrete measure- future CBM of such engines.
ments and provided no indication of duration of a particular situ-
ation over time. Hence, it may be expected that if the command of
having the situation steady with time was introduced, the fre-
quency of occurrence for the 0e5 % range would reduce signifi- Acknowledgements
cantly. Therefore, adopting the threshold of 3 % steady was
considered the most realistic threshold, to define a failure event in The work presented in this contribution has been supported by
the engine and avoid misleading false alarms. the European Fisheries Fund (ref. 351BI20090040). We would like
Spectrometric oil analysis has been used successfully in the past to express our sincere gratitude to: the ship owner, skipper and
to establish the CBM of an engine [24]. Occasional oil analysis may crew of the vessel for their helpful support in this project; Gorka
play a key role in determining when the engine runs in normal and Merino and Manuel Gonzalez (AZTI) for improving the quality of
unusual conditions. Such analyses may assist in defining the faulty the figures; and Prof Michael Collins (Bizkaia Xede Fellow (PIE,
conditions of the engine and the faulty threshold values of the UPV/EHU)), for his constructive comments. This is contribution n
selected variables. 699 of AZTI (Marine Research Division).
412 O.C. Basurko, Z. Uriondo / Applied Thermal Engineering 80 (2015) 404e412

Appendix [7] V. Çelik, E. Arcakliog lu, Performance maps of a diesel engine, Appl. Energy 81
(2005) 247e259.
[8] Y. Çay, A. Çiçek, F. Kara, S. Sag irog lu, Prediction of engine performance for an
The type of sensors and meters used, together with their accu- alternative fuel using artificial neural network, Appl. Therm. Eng. 37 (2012)
racy, are listed in Table A.1. 217e225.
[9] A.J.C. Sharkey, G.O. Chandroth, N.E. Sharkey, A multi-net system for the fault
diagnosis of a diesel engine, Neural Comput. Appl. 9 (2000) 152e160.
Table A.1 ~ o, J.L. Míguez, Diesel engine condition monitoring
[10] J. Porteiro, J. Collazo, D. Patin
Type of sensors and meters used for engine monitoring using a multi-net neural network system with nonintrusive sensors, Appl.
Therm. Eng. 31 (2011) 4097e4105.
Instrument type Accuracy [11] D.T. Hountalas, Prediction of marine diesel engine performance under fault
Ambient relative humidity Vaisala HMW70Y ±1% conditions, Appl. Therm. Eng. 20 (2000) 1753e1783.
Ambient temperature Vaisala HMW70Y ±0.1  C [12] E. Mesbahi, An intelligent sensor validation and fault diagnostic technique for
diesel engines, J. Dyn. Syst. Meas. Control 123 (2001) 141e144.
Ambient pressure Vaisala PTB110 ±0.15 mbar
[13] N. Togun, S. Baysec, Genetic programming approach to predict torque and
Propeller shaft power Binsfeld Torque Trak Revolution ±1 %a
brake specific fuel consumption of a gasoline engine, Appl. Energy 87 (2010)
Engine RPM Magnetic pick-up ±1% (FS)
3401e3408.
Shaft generator CIRCUTOR EDMK ±1% [14] J.A. Twiddle, N.B. Jones, A high-level technique for diesel engine combustion
(Alternator) power system condition monitoring and fault diagnosis, Proc. Institution Mech. Eng.
Boost Air Pressure DANFOSS MBS 1050 ±0.3% (FS) Part I: J. Syst. Control Eng. 216 (2002) 125e134.
Fuel Oil Consumption KRAL OMG20 ±0.2% [15] D. Antory, U. Kruger, G. Irwin, G. McCullough, Fault diagnosis in internal
Operation temperatures K-type thermocouple (class 1) ±1.5  C combustion engines using non-linear multivariate statistics, Proc. Institution
Fuel Oil Density METTLER TOLEDO DENSIO 30PX ±0.001 g/cm3 Mech. Eng. Part I: J. Syst. Control Eng. 219 (2005) 243e258.
Location and speed GPSb NA [16] V.T. Lamaris, D.T. Hountalas, A general purpose diagnostic technique for
of the vessel marine diesel engines e application on the main propulsion and auxiliary
diesel units of a marine vessel, Energy Convers. Manag. 51 (2010)
NA: Not Available. 740e753.
Key. [17] R. Khaled, D. Priour, J.-Y. Billard, Numerical optimization of trawl energy ef-
a
Estimated value, strong dependence upon installation. ficiency taking into account fish distribution, Ocean Eng. 54 (2012) 34e45.
b
Value obtained from the GPS of the vessel. [18] R. Khaled, D. Priour, J.-Y. Billard, Cable length optimization for trawl fuel
consumption reduction, Ocean Eng. 58 (2013) 167e179.
[19] R. May, G. Dandy, H. Maier, Review of input variable selection methods
for artificial neural networks, in: K. Suzuki (Ed.), Artificial Neural Net-
References works e Methodological Advances and Biomedical Applications, InTech,
2011, p. 362.
[1] Ø. Buhaug, J.J. Corbett, Ø. Endresen, V. Eyring, J. Faber, S. Hanayama, D.S. Lee, [20] J.D. Olden, D.A. Jackson, Illuminating the “black box”: a randomization
D. Lee, H. Lindstad, A.Z. Markowska, A. Mjelde, D. Nelissen, J. Nilsen, C. Pålsson, approach for understanding variable contributions in artificial neural net-
J.J. Winebrake, W. Wu, K. Yoshida, Second IMO GHG Study 2009, International works, Ecol. Model. 154 (2002) 135e150.
Maritime Organization (IMO), London, United Kingdom, 2009, p. 240. [21] E. Mesbahi, Application of Artificial Neural Networks in Modelling and Control
[2] Lloyd's Register, Ship Energy Services, 2012. www.lr.org/sectors/marine/ of Diesel Engines, PhD thesis, Newcastle University, Newcastle upon Tyne, UK,
Services/Consultancy/ES/ (accessed: 05/11/2012). 2000, p. 196.
[3] O.C. Basurko, G. Gabin~ a, Z. Uriondo, Energy performance of fishing vessels and [22] O. Cabezas, J.M. Ferarios, J. Franco, G. Gabin ~ a, Z. Uriondo, Development and
potential savings, J. Clean. Prod. 54 (2013) 30e40. Implementation of Energy Audits and Technical Actions to Improve Energy
[4] J. Kowalski, An experimental study of emission and combustion characteristics Efficiency in Fishing Vessels, AZTI-Tecnalia, Sukarrieta, Spain, 2012, p. 425 (in
of marine diesel engine with fuel pump malfunctions, Appl. Therm. Eng. 65 Spanish).
(2014) 469e476. [23] IHS, IHS Fairplay World Register of Ships, 2014. www.ihs.com/products/
[5] V. Duran, Z. Uriondo, J. Moreno-Gutie rrez, The impact of marine engine maritime-information/ships/world-register.aspx (accessed: 08/04/2014).
operation and maintenance on emissions, Transp. Res. Part D: Transp. Environ. [24] H.J. Wei, Research of marine diesel engine condition detecting base on BP
17 (2012) 54e60. neural network and spectrometric analysis, in: International Conference on
[6] IAEA, Implementation Strategies and Tools for Condition Based Maintenance Multimedia Technology (ICMT), Ningbo, China, 2010, pp. 1e4.
at Nuclear Power Plants, International Atomic Energy Agency, Vienna, Austria,
2007, p. 187.