Professional Documents
Culture Documents
parts and key technical supports still rely on the OEM. maintenance, and 4) performance-based maintenance. The
Although MBC is a common practice in the power industry, evolution is driven by the continuous reduction of the cost of
the OEM has less motivation to improve the equipment ownership. The advent of PBC generates a strong momentum
reliability. This is because MRO generates a lucrative revenue pushing the transition towards the performance-based
stream owing to spare parts supply. Nevertheless, it creates maintenance paradigm.
financial pressures on the wind farmers who must maintain CM=Corrective Maintenance
high equipment availability to harness the intermittent wind
C. Determining Performance Metrics Gearboxes, generators, and transformers are typical repairable
As shown in Figure 2, a variety of metrics can be used to LRUs. In the following we estimate the reliability and
evaluate the system performance. This study recommends that downtime at the LRU level.
the wind power industry adopt operational availability (A), A. Modeling Subsystem Reliability
MTBF, and MDT (mean-down-time) as the key variables for
system performance assessment. These metrics can be easily The Weibull distribution is widely used to model the
monitored and estimated in practice, and they are also widely lifetime of repairable systems. The two-parameter Weibull
used in the equipment industry. Let A denote the WT reliability function is given as
operational availability, then
(
R (t ) = exp − (αt ) β , ) (2)
MTBF
A= . (1)
MTBF + MDT where α and β are scale and shape parameters, respectively. If
β=1, equation (2) simply becomes an exponential function.
MDT usually consists of two parts: MTTR (mean-time-to- Since a system failure corresponds to a subsystem failure,
repair) and MLDT (mean-logistics-delay-time). MTTR is the
equation (2) is indeed used to model the lifetime of a
hands-on time used to restore the system to the operational
particular subsystem. Let E[N] be the expected number of
state given the spare part is available. MLDT represents the
subsystem failures in [0, t0]. Then we have
waiting time for materials, labor, and repair tools.
t
D. Understanding Performance Drivers E[ N ] = 0 = λ t 0 , (3)
μ
When planning a PBC agreement, a fundamental question where
confronted by the OEM and the wind farmer is: What are the
1 1 1
key factors that drive the system performance? And how do μ= = Γ(1 + ) , (4)
they interact with each other and jointly control the λ α β
operational availability? Studies [12, 13] show that system
availability is dominated by five key performance drivers, that where, μ is the MTBF of the subsystem, and λ is the average
is, inherent reliability μ, spare parts level s, repair turn-around failure intensity rate in the steady-state condition.
time tr, the fleet size n, and the utilization rate θ (0≤θ≤1).
A. Estimating Mean-Down-Time
Inherent Fig. 4 depicts a typical spare parts supply chain system
MTBF
Reliability (μ) owned by the OEM to support the MRO service in different
OEM
regions. The OEM owns the repair center and the spare parts
Spare Parts Availability inventories located near the field systems. The repair center is
MTTR
Logistics (s, tr) (A) positioned in a central area with a relatively equal distance to
all WT fleets. A centralized repair center benefits the OEM
Wind Farm
System Fleet
MLDT
for achieving labor consolidation and high tool utilization. In
(n, θ) this study, the repair center is modeled as an M/G/∞ queuing
model, which is commonly used in operations management
Fig. 3. Performance Measures and Their Drivers
and service parts logistics literature [14].
Repair by
Fig. 3 depicts the relationship between three performance replacement
measures and their underlying drivers. For example, the Replenish Stockroom WT
MTBF for WT depends on the intrinsic reliability and the inventory with s1 units Fleet 1
usage rate. MTTR relies on the spares stocking level, repair Wind Farm 1
time, the fleet size, and usage. This indicates that both the OEM for Repair Center Stockroom WT
design and
OEM and the wind farm are responsible for the equipment manufacturing
M/G/∞ with s2 units Fleet 2
availability in the PBC regime. Wind Farm 2
Stockroom WT
IV. MODELING RELIABILITY AND DOWNTIME with s3 units Fleet 3
OEM Wind Farm 3
A wind turbine consists of multiple subsystems or items,
such as blades, the main bearing, the gearbox, the generator, Fig. 4. An Integrated Logistics Supply Chain for Wind Turbines
transformers, the main shaft, electrical controllers, and
mechanical brakes. The failure of one item often results in the Two independent scenarios will occur at the customer site
malfunction of the entire system. Upon failure, the defective upon failures. If a spare part is available in the local
item—i.e. the line replaceable unit (LRU) —is swapped out stockroom, the repair-by-replacement job can be performed
and replaced by a spare unit. If the item is repairable, it will be immediately. The average time for performing the
sent to the repair center for failure removal. Upon fix, the item replacement job is denoted as ts. On other hand, if an on-hand
is routed to a local stockroom for future replacement. spare unit is not available, the equipment downtime is
4
prolonged due to the extended waiting time for the spare unit. incremental manufacturing cost if the reliability is further
Let Td denote the MDT. Then we have improved. Assuming there are n subsystems operating in the
field, the total manufacturing cost for the entire subsystem
fleet is
Td = t s Pr{O ≤ s} + (t s + t r )(1 − Pr{O ≤ s}) , (5)
where M ( μ ) = nc ( μ ) (9)
s (nλt ) x e − nλt r
Pr{O ≤ s} = ∑ r
. (6) C. Logistics, Maintenance, and Repair Cost
x =0 x!
The OEM is responsible for sustaining the operation of a
In equation (5), tr is the repair-turn-around time between the WT fleet at a pre-determined availability level written in the
stockroom and the repair center. O is a random variable PBC contract. The sustainment costs comprise spare parts
representing the steady-state inventory on order. Failures of capitals, inventory holding cost, and all the repair expenses
field working items are approximated as a Poisson process incurred in [0, t0]. We assume a spare part corresponds to one
with a rate of λ which is given in equation (4). For detailed subsystem unit. Let L(μ, s) be the total sustainment costs, then
derivations of Td , readers are referred to [12].
L( μ , s ) = sc( μ ) + sht 0 + nE[ N ](r1 p1 + r2 p 2 ) (10)
V. WIND TURBINE LIFECYCLE COST ASSESSMENT
The first term in equation (10) is the spare parts inventory
The lifecycle costs of a wind turbine include all the cost. The second term is the spare parts holding cost, and h is
expenses associated with design, manufacturing, operation the holding cost per unit time. The last term is the cost for
and maintenance. In the following, we present several parts replacement and repair. E[N] is the expected failures in
analytical models to describe subsystem level costs. The [0, t0] as given equation (3). Notice that r1 the repair cost
system cost can be easily estimated by aggregating all when an on-hand spare part is available, and r2 is repair cost
subsystem costs. in case that a spare part is backordered. Similarly, p1 and p2
A. Design Cost vs. Reliability are the probabilities to estimate whether an on-hand spare unit
is available or not. From equation (6), we can immediately
In general, it has been agreed that the design cost increases obtain
with the reliability, yet there is no consensus on how to
describe such a relationship. Some researchers [13, 15] argue
p1= Pr{O ≤ s} , and p 2 = 1 − p1 (11)
that the design cost will grow exponentially with the
reliability. Without loss of generality, we use the exponential D. Fleet Lifecycle Cost
model to characterize the design cost versus the reliability. Let
By aggregating all the cost items in equations (7), (9) and
D(μ) be the subsystem design cost, then (10), the lifecycle cost (LCC) for the subsystem fleet can be
expressed as
⎡ ⎛ μ − μ min ⎞⎤
D( μ ) = B1 ⎢exp⎜⎜ k ⎟⎟⎥ , for μmin≤μ≤μmax, (7)
⎢⎣ ⎝ μ ∞ − μ ⎠⎥⎦ C(μ, s) = D(μ ) + M (μ ) + L(μ, s) (12)
where, μmax and μmin are the best achievable and the least A WT system consists of multiple subsystems, say m
acceptable MTBF, respectively. μ∞ is considered an MTBF repairable subsystems. Each subsystem has different reliability
level which in theory leads to infinitely design cost, and and spare parts stocking levels. Then the LCC for the WT
fleet can be estimated by
μ∞>μmax. Notice that B1 is the baseline design cost for μ=μmin,
and k describes how difficult to further increase the reliability m m m
subject to design, material, and resource constraints. C(μ, s) = ∑ Di (μi ) + ∑ M i (μi ) + ∑ Li (μi , si ) . (13)
i =1 i =1 i =1
B. Manufacturing Cost vs. Reliability
To build the reliability into the system, OEM must adopt Where μ=[μ1, μ2, …, μm] and s=[s1, s2, …, sm]. The goal of
new materials, novel technologies, or advanced manufacturing the OEM is to minimize the fleet LCC while meeting the
processes in the production phase. These activities imply performance criteria (e.g. reliability and availability).
additional investment in the production phase. We adopt the Similarly, the goal of the wind farmer is to monitor the system
manufacturing cost model in [13] to estimate the unit reliability and devise payment schemes to incentivize OEM in
production cost, achieving the performance goal.
c( μ ) = B2 + B3 ( μ v − μ min
v
) , for μmin≤μ≤μmax. (8) VI. PERFORMANCE COMPENSATION
In equation (8), B2, B3 and ν are model parameters. In A. Cost Plus Payment Method
particular, B2 is the baseline manufacturing cost for producing There are three types of cost-plus oriented payment
one piece of an item for μ=μmin, and B3 captures the schemes: cost plus fixed fee (CPFF), cost plus incentive fee
(CPIF), and cost plus award fee (CPAF).
5
CPFF is used when the cost and pricing risk is very high, We solved the gearbox asset management problem using
especially in the development of new systems. The idea is to the decision model in Problem P1. Fig. 5 shows the
reimburse the contractor or OEM for the level of effort or annualized cost for a 10-year service contract computed under
work accomplished, plus certain amounts of reasonable profit. various MTBF and spare part levels. It is found that adding an
CPIF is used in circumstances when metric baseline extra spare part in the stockroom is able to improve 1% of
identification is immature. CPAF is used when subjective availability, regardless of the gearbox MTBF. The chart also
assessments of service performance are desired (e.g. customer shows how to attain the availability goal by trading off the
satisfaction surveys). In many cases, CPAF is used in reliability and spare parts stocking. For example, if Amin=0.98
conjunction with CPIF to achieve the incentive goal as well as is required by the wind farm, the OEM can choose {s=0,
accommodate the subjective judgment on the OEM μ=87,000}, {s=1, μ=64,500}, or {s=2, μ=50,000}. Hence the
performance. Readers are referred to [16] for more discussion OEM is empowered with the flexibility to achieve the
on cost-plus payment schemes. availability criterion under the performance-based service
This study derived the PBC cost model from the OEM framework.
perspective. We assume the technology is relatively mature Table I
and a fixed cost payment plan is adopted by the WT owner. Reliability and Cost Parameters of Gearbox (n/a=not applicable)
Parameter Value Unit Parameter Value Unit
This type of payment plan actually puts the OEM in the
μ∞ 100,000 hours B3 2,000 $
highest risk while lowering the risk of the equipment user. μmin 50,000 hours h 0.30 $/hour
k 0.2 n/a r1 5,000 $/repair
B. Fixed Price Payment Method
ν 0.2 n/a r2 10,000 $/repair
Under a fixed price contract, the OEM prefers to minimize t0 10 or 20 years n 50 units
the lifecycle cost of the committed service. This can be B1 250,000 $/design tr 90 days
translated into the following optimization model: B2 50,000 $/unit ts 216 hours
Problem P1:
Min C(μ , s) = D(μ ) + M (μ ) + L(μ , s) (14) Gearbox MTBF vs. Operational Availability
Subject to (10-Year Contract)
1.00
μmin≤μ≤μmax, (15)
0.99
A( μ , s) ≥ Amin . (16)
Availability
0.98
0.97
This formulation minimizes the LCC for a subsystem fleet. s=2
0.96 s=1
Constraint (15) ensures that the subsystem reliability falls into 0.95 s=0
the desired range. Constraint (16) states the minimum
0.94
expected operational availability by the equipment user. A(μ, 40,000 50,000 60,000 70,000 80,000 90,000 100,000
s) can be easily estimated by substituting equations (4) and MTBF (hours)
(5) into (1).
If we consider the cost minimization at the WT system Fig. 5. Gearbox Availability under the 10-Year Service Contract
level, Problem P1 can be expanded to accommodate such a
requirement. The following formulation minimizes the LCC of
the WT fleet. Gearbox MTBF vs. Annualized Cost
(10-Year Contract)
Problem P2:
380,000
m m m
s=2
Min C(μ, s) = ∑ Di (μi ) + ∑ M i (μi ) + ∑ Li (μi , si ) (17) 370,000 s=1
i =1 i =1 i =1
s=0
Cost ($)
360,000
Subject to
μmin, i≤μi≤μmax,i, for i=1,2, …., m (18) 350,000
m 340,000
∏ Ai ( μi , si ) ≥ Amin (19)
330,000
i =1
40,000 50,000 60,000 70,000 80,000 90,000 100,000
MTBF (hours)
Equation (19) is the system availability obtained through
the multiplication of individual subsystem availabilities.
Fig. 6. Annualized Cost under the 10-Year Service Contract
VII. CASE STUDIES
Fig. 6 plots the annualized cost for 50 gearboxes under the
In this section, the proposed PBC scheme is used to manage
10-year service contract. The lowest service cost is realized if
the reliability and maintenance of gearbox in a WT fleet
the gearbox MTBF is designed between 70,000 and 80,000
consisting of 50 turbines. Costs are calculated and compared
hours. This cost is relatively independent of the spare parts
between a 10-year service contract and a 20-year contract.
quantity. If the MTBF is above 80,000 hours, both the design
Reliability and cost related to gearbox design, manufacturing,
and manufacturing costs dramatically increase. On the other
spares inventory and repair are listed in Tables 1. These
hand, when the MTBF drops below 70,000 hours, the logistics
values are estimated based on the studies in [1, 5].
and repair expenses quickly rise.
6