Modelling Worker Fatigue and Recovery in Dual-Resource Constrained Systems

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Computers & Industrial Engineering 59 (2010) 75–84

Contents lists available at ScienceDirect

Computers & Industrial Engineering


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

Modelling worker fatigue and recovery in dual-resource constrained systems


M.Y. Jaber *, W.P. Neumann
Department of Mechanical and Industrial Engineering, Ryerson University, Toronto, ON, Canada M5B 2K3

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

Article history: The operational benefits that dual-resource constrained (DRC) job shop systems bring have captured the
Received 14 June 2009 attention of researchers for some time. Although several studies that investigate DRCs are available in the
Received in revised form 30 January 2010 literature, none has investigated a DRC system for the effects of human fatigue and recovery, which poses
Accepted 10 March 2010
important parameters to avoiding overload and injury to employees. The purpose of this paper is to
Available online 12 March 2010
address this limitation by presenting a mixed-integer linear programming (MILP) model that describes
fatigue and recovery in a DRC system with one worker performing n tasks (flexibility level) within m
Keywords:
cycles. Later, the complexity of the MILP problem was reduced to four practical cases. These cases were
Modelling
Worker fatigue and recovery
investigated to evaluate several research questions. The results obtained from the MILP model and the
Dual-resource constrained systems four practical cases suggest that short rest breaks after each task, short cycle times and faster recovery
numerical analysis rates improve the system’s performance and that reduced force levels in the work tasks will reduce
Ergonomics recovery needs and further increase performance. Further research is still needed to identify or to develop
Human factors better models of physiological and mental fatigue that can be integrated to the modelling framework pre-
sented here.
Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction The first group of studies investigated the use of analytical mod-
els including mathematical, queuing and scheduling models in
Dual-resource constrained (DRC) systems are those where order to find optimal or near optimal solutions to questions regard-
capacity constraints on output come from both machines and hu- ing the degree of flexibility and the assignment (e.g., where and
mans (e.g., Treleven, 1989). Typically, the staffing level (ratio of when to transfer) of the workforce (e.g., Bobrowski & Park, 1993;
operators to machines) is less than 100%, where workers (opera- Fryer, 1976; Hottenstein & Bowman, 1998; Inman, Jordan, & Blu-
tors) are transferred from one task (machine) to another as needed; menfeld, 2004; Nelson, 1967; Treleven & Elvers, 1985). This forms
usually subject to certain limitations such as: batch size, worker’s the most investigated aspect of DRC systems. The second group of
idle time, delivery due dates, queue length, and upfront training studies investigated the effects of learning and forgetting on DRC
(e.g., Bokhorst, Slomp, & Gaalman, 2006; Kher, Malhotra, Phili- systems (e.g., Jaber, Kher, & Davis, 2003; Kher et al., 1999; McCre-
poom, & Fry, 1999; Treleven, 1989). Cross-training allows the ery & Krajewski, 1999). These studies indicated that when workers
workers to share workloads amongst themselves within any given are cross-trained, they incur productivity losses due to cycles of
team configuration (McCreery, Krajewski, Leong, & Ward, 2004). A learning and forgetting that can negatively affect system perfor-
cross-trained workforce is believed to offer several benefits to mance (e.g., Jaber et al., 2003; Kher et al., 1999; Wisner & Pearson,
firms. It can help the firm respond to unexpected and unbalanced 1993; Zamiska, Jaber, & Kher, 2007). These studies reported that
workloads by reducing manufacturing lead-times and inventories forgetting on the part of the system operators can have a negative
of work-in-process, accelerating the learning process of workers, impact on the system’s performance. It was generally found that as
improving quality and productivity and enhancing customer ser- the worker’s flexibility increases, forgetting increases since the
vice (e.g., Felan & Fry, 2001; Krajewski, King, Ritzman, & Wong, worker has to relearn each task he/she revisits. The third and last
1987; Treleven, 1989), as well as reducing fatigue, boredom or group of studies investigated the internal dynamics of DRC sys-
repetitive stress (e.g., Hopp, Tekin, & Van Oyen, 2004). There has tems. The components of a DRC system are machines, workers,
been considerable investigation of DRC systems and this literature materials and products; internal environmental dynamics here is
can be clustered into three groups, which are briefly surveyed specified as changes in any one of these components with respect
below. to their presence and behaviour such as the replacement of ma-
chines, worker attrition or absenteeism, introduction of new prod-
ucts or fluctuations in demand (Yue, 2005).
* Corresponding author. Tel.: +1 416 979 5000x7623; fax: +1 416 979 5265.
The authors’ survey of the literature did not identify any study
E-mail address: mjaber@ryerson.ca (M.Y. Jaber). that investigated the effects of operator fatigue and recovery when

0360-8352/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cie.2010.03.001
76 M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84

Nomenclature

Decision variables Lmax the maximum fatigue load a worker can accumulate in
tij time a worker spends on task i in cycle j, where i = 1, . . . , any task
n, and j = 1 , . . . , m LSij fatigue index at the start of task i in cycle j
bij break time following task i in cycle j Lij fatigue index generated while performing task i in cycle
Xij binary decision variable, 0 or 1, of having/not having a j, where Lij = Lmax  tij/METi
break following task i in cycle j LEij fatigue index at the end of task i in cycle j, where
LEij ¼ LSij þ Lij
Input parameters Rij fatigue index level at time t following tij
T length of the work-shift RAi recovery allowance required for task i as a fraction of
MLC maximum load capability METi
fi fraction of MLC applied when performing task i
METi mean endurance time for task i as a function of fi

addressing issues associated with flexibility and assignment of the eratures. For the purposes of modeling work, we draw especially on
workforce in DRC systems. Large scale surveys have identified the ergonomics literature (e.g. Åhsberg, 1998, El Ahrache, Imbeau, &
trends of increasing intensification of work (Docherty, Forslin, & Farbos, 2006) and in particular focus on the injury related concept
Rami Shani, 2002) and the total costs of work-related ill health of muscle fatigue (Bigland-Ritchie, Rice, Garland, & Walsh, 1995;
are similar to those of all cancer combined (Leigh, Markowitz, Fahs, Edwards, 1981; Enoka & Duchateau, 2008). Muscle fatigue develops
Shin, & Landrigan, 1997). These problems are often integral to sys- when the rate and level of muscular demands on the individual ex-
tem design choices made at each stage of the development process ceed their muscle’s physiological capacity to recover from those de-
(Neumann, Kihlberg, Medbo, Mathiassen, & Winkel, 2002; Neu- mands. The work physiology research has largely focussed on
mann, Winkel, Medbo, Mathiassen, & Magneberg, 2006). Musculo- predicting individuals Maximum Endurance Time (MET) – the point
skeletal disorders form the largest part of work-related ill health at which individuals can no longer exert the required force level
and are known to be caused by fatigue inducing work conditions (Niebel & Freivalds, 2003; Rohmert, 1973). This end point is a func-
including high loads and frequent repeated and prolonged loading, tion of the force level – usually studied as a fraction of the individ-
especially if combined with awkward working postures (Hagberg ual’s maximum capability (f) under continuous static loading
et al., 1995). Beyond the health and safety imperative, attending conditions (Van Dieen & Oude Vrielink, 1994). For an overview of
to worker fatigue and injury can greatly impact system perfor- available models relating MET to f, we recommend the review of
mance in terms of both quality (Eklund, 1997) and productivity El Ahrache et al. (2006). Most of the models in this review describe
(Oxenburgh, Marlow, & Oxenburgh, 2004). The total cost of em- the relationships between f and MET as having either a linear or
ployee injury to the firm includes aspects such as reduced quality power form:
and quantity of work performance before the period of absentee-
MET ¼ A  f a ð1Þ
ism as well as after the absenteeism as injured workers ‘ramp
up’ to full capacity. These costs along with administrative and
other overheads are typically many times greater than just the di- MET ¼ B  ebf ð2Þ
rect costs of treating the injured worker (Oxenburgh et al., 2004). where A, a, B and b are model-specific parameters.
Retrofitting systems to correct either human or machine design er- In order to monitor fatigue levels in the planning of DRC sys-
rors can be very expensive and time consuming. It is better to con- tems we establish a ‘fatigue index’ based on the MET and the load
sider human factors and operator risks during early DRC system impulse (force  time), with MLC being the ‘maximum load capa-
design stages where the costs for improvement are lowest and po- bility’. The maximum fatigue index for task i is defined as:
tential benefits greatest (Alexander, 1998; Miles & Swift, 1998).
Since fatigue is a pre-cursor to a number of problems, it would Limax ¼ MLC  fi  MET i ð3Þ
be useful if operator fatigue levels could be modelled and thus
managed during the design stage of DRC system development. The total limit for maximum fatigue index is presented as
The purpose of this paper is to investigate the effects of human
Lmax ¼ minfLimax ji ¼ 1; 2; :::; ng ð4Þ
fatigue and recovery on the performance of dual-resource con-
strained (DRC) systems. Section 2 provides readers with concise The shape of the fatigue function presents a modelling chal-
overview of fatigue and recovery models. In Section 3, a mixed- lenge. The available literature has focused mainly on developing
integer linear programming (MILP) problem is developed that de- models to estimate MET (or maximum fatigue) as a function of f
scribes DRC systems with the inclusion of fatigue and recovery. (e.g., El Ahrache et al., 2006). So, these models estimate the end
Section 4 discusses four practical cases that are derivatives, and point of a function at which maximum fatigue is reached, but never
simplified versions, of the MILP model described in Section 3. Sec- reveal the form of fatigue accumulation function (exponential, lin-
tion 5 provides several numerical examples and addresses some ear, etc.). While some researchers assert that fatigue increases
research questions. Section 6 provides a general discussion of the exponentially over time (e.g., Konz, 1998), others assumed it in-
models presented, and Section 7 concludes this modelling effort. creases linearly (e.g., Kaneko & Sakamoto, 2001; Soo, Nishino, Sugi,
et al., 2009). The evidence presented by Bigland-Ritchie, Cafarelli,
and Vøllestad (1986a), Bigland-Ritchie, Furbush, and Woods
2. Fatigue and recovery models (1986b), for example, provides some indication of a linear reduc-
tion in force muscle capability as fatigue develops – but there is
2.1. Workload and fatigue no consensus available and this problem may also depend on
how ‘fatigue’ is operationalised and which level, from whole body
The concept of ‘fatigue’ has been widely studied and discussed in to biochemical reactions, of the fatigue phenomenon is being
the physiological, psychological, sport-science, and ergonomics lit- considered. For the initial modelling efforts presented here, we
M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84 77

assumed that fatigue increases linearly over time. This aspect of load as an indicator, then the term ðLSij þ LEij Þ  t ij should be summed
the model can be readily modified as new data on the shape of up over n tasks and m cycles, then divided by n  m  T to determine
the fatigue accumulation function become available. the weighted average physical work load for any task i in any cycle
j, and by Lmax to normalize this measure over a 0–1 scale, with
2.2. Rest and recovery Z2 = 0 and Z2 = 1 representing the best and worst ergonomic perfor-
mance respectively. The objective function is then written from (5)
Rest breaks allow for the physiological recovery of a worker by to (6) as:
overcoming fatigue. Rest breaks are either ‘‘off-the job” (i.e., eve-
nings and weekends) or ‘‘on-the-job”, which is the concern of this Z ¼ w1 Z 1 þ w2 Z 2
!
paper. On-the-job breaks can be classified as either ‘‘scheduled m  
1X n X m
w2 Xn X

breaks” or ‘‘random breaks”. The ‘‘at-work” recovery can further ¼ w1 1 tij þ LS þ LEij  t ij ð7Þ
T i¼1 j¼1 2nmLmax T i¼1 j¼1 ij
be divided into ‘‘formal breaks”, which are scheduled, and ‘‘infor-
mal breaks”, which are unscheduled interruptions, and ‘‘working P2
where 0 < wk < 1 is the weight of objective k and k¼1 wk ¼ 1;
rest” in which rotation to tasks with different demands provides
emphasising the importance of Z1 and Z2 to the decision maker.
recovery time to the newly unloaded musculature (Konz & John-
For example, a decision maker may view that increasing productiv-
son, 2004). The muscular recovery time required has been mod-
ity (Z1) as equally important to reducing a worker’s physical loading
elled in terms of rest allowance (RA) models and has recently
(Z2), i.e., w1 = w2 = 0.5, or that productivity is twice as important as
been reviewed by El Ahrache and Imbeau (2008). These models
loading; w1 = 0.75 w2 = 0.25. A decision maker assigns preferences
use Maximum Endurance Time (MET), which is a function of load
(i.e., weights) to objectives (Z1 and Z2) based on his/her aspiration
level, to calculate rest allowance or recovery time following a static
level, instructions of management, or in conformance with the labor
muscular work.
contract between management and the workforce. The objective
function can take some other forms (e.g., maximization of produc-
3. DRC system model of productivity and fatigue tive time) depending on the emphasis and policies of the model’s
users.
This section presents an analytical model to address the work-
force flexibility issues when fatigue and recovery are taken into
consideration in DRC system design. A mixed-integer linear pro- 3.2. Constraints
gramming (MILP) model is developed to formulate and solve the
above described problem of allocating one worker to n tasks (flex- 3.2.1. Time
ibility level) in m cycles within a constrained time period taking Time-related constraints include the sum of processing times
into account fatigue and recovery. and the break times must be equal to length of the work-shift;
The model presented herein is deterministic. That is, a worker no task can last beyond the MET value as this is a physiological lim-
spends exactly tij units of time on task i in cycle j, which is the time it to performance; and, in order to prevent impractical high-fre-
required to perform all the required activities in that task. In prac- quency rotation solutions, we add the constraint that no worker
tice, a worker may either leave the task before or after tij, however, is to be transferred to the next task or to take a break before the
this assumption is beyond the scope of this paper. This remains a worker spending a minimum time tmin on the job. These time con-
limitation that will be addressed in a future work. straint equations are written as:

3.1. The objective function X


n X
m n X
X m
tij þ X ij bij ¼ T; 8i and j P 1 ð8Þ
i¼1 j¼1 i¼1 j¼1
In any organization, usually, there are conflicting objectives.
Management wants to increase the productivity of its workforce
tmin;ij 6 t ij 6 MET i ; 8i and j P 1 ð9Þ
by reducing idle (unproductive) time. This productivity sub-objec-
tive is translated as follows:
3.2.2. Fatigue and recovery
1X n X m
Z1 ¼ 1  t ij ð5Þ The fatigue index must always be less than or equal to the
T i¼1 j¼1
maximum fatigue that a worker can attain in any cycle at any
P P time. The minimum fatigue index attained at the start of a task
where the term ni¼1 m j¼1 t ij in (5) represents the productive time in
P P is larger than or equal to zero. The fatigue level at the start of a
T. Although impossible to attain, when ni¼1 m j¼1 t ij = T, Z1 = 0 corre-
task plus fatigue generated during the execution of the task is
sponding to 100% efficiency. Increasing the percentage of the pro-
equal to fatigue level at the end of the execution of that task. Fa-
ductive time increases the physical load on the workforce and
tigue load at the end of task execution minus recovery in a given
may not be desirable due to reasons related to the fatigue level of
break/rest period is equal to the fatigue load at the start of exe-
the worker, performance quality targets, or occupational health
cution of the next task. Since there is no point in resting longer
and safety issues. When the productive time increases, the average
than required for complete recovery, the recovery time is to be
workload on the worker in task i, increases, increasing the weighted
less than or equal to the duration recommended by the RA mod-
average fatigue index, which is represented by the following sub-
els. The fatigue load at the start and the end of a task must satisfy
objective:
the following:
Xn X m  
1
Z2 ¼ LSij þ LEij  tij ð6Þ LSij P 0; LSiþ1;j 6 LEij and LEij 6 Lmax ; 8i and j P 1 ð10Þ
2nmTLmax i¼1 j¼1

where ðLSij þ LEij Þ=2 is the average loading on job (task) i in cycle j, where LSij þ Lij ¼ LEij , Lij ¼ Lmax  t ij =MET i , and LSiþ1;j ¼ LEij  Rij P 0. Fati-
sustained for a period of tij, where LSij 6 LEij : gue is recovered over a period of bij P 0, where Rij 6 LEij reduces to a
So, the term ðLSij þ LEij Þ  tij =2 represents the total physical work lower value over bij, i.e. LSiþ1;j ¼ LEij  Rij . Therefore, LSiþ1;j ¼ LEij  Rij ¼
load for job i in task j. Using the weighted average physical work LEij  ai bij ; where ai ¼ ðLmax  t ij =MET i ÞX ij =ðRAi  tij Þ, Xij = 1, and
78 M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84

bij 6 RAi  MET i ð11Þ stant unmoving force, which results in relatively short METs and
are not realistic for extended, dynamic production tasks. In order
when LEij 6 Lmax .
to position our model in a realistic operational range we calibrate
the exponential form of the MET equation with the assumption
3.2.3. Binary variables
that operators will be fully fatigued performing a task of 30% of
We define a recovery period to be present when its correspond-
their maximum capability (0.3 of MLC) after 75 units of time and
ing decision variable gets the value of 1 and on the contrary there
that the RA for such a task is 50%. If the worker is assumed to work
shall be no break when its related binary variable is assigned the
at this load level for a maximum time (e.g., maximum endurance
value of 0.
time or MET = 75), this value can be derived from an exponential
X ij ¼ 0; 1; 8i and j P 1 ð12Þ MET Eq. (2) when the parameters in equation are set as
MET1 = MET2 = 80 exp ( 0.21513  0.3) = 75. The corresponding
Therefore, the mixed-integer linear programming (MILP) prob-
Lmax is calculated from (4) as Lmax = 1  0.3  75 = 22.5. Further, if
lem could be written from (7)–(12) as:
! it is assumed that the recovery allowance (RA) is 50%, this value
1X n X m
w2 Xn can be derived from a power recovery allowance function (5) when
Minimize Z ¼ w1 1 t ij þ the parameters in equation are set as RAi = 3  MET0.41479 = 0.5 or
T i¼1 j¼1 2nmLmax T i¼1
50%. These assumptions allow the exploration of model behaviour
m 
X  within an operationally realistic range.
 LSij þ LEij  t ij ð13Þ
In this section, we examine four cases that may occur in prac-
j¼1
tice, which are: (1) A break is enforced after each task, (2) a break
n X
X m n X
X m is enforced only after the second task, (3) a break to full recovery is
Subject to t ij þ X ij bij ¼ T required after each task, and (4) a break until full recovery is re-
i¼1 j¼1 i¼1 j¼1 quired after the second task. We use these four hypothetical cases
tmin;ij 6 tij 6 MET i to examine the following specific questions on the behaviour of the
developed general DRC fatigue-recovery model’s behaviour:
LEij 6 Lmax ; LSiþ1;j 6 LEij and LSij P 0
bij 6 RAi  MET i (1) What effect does changing the preferences of the sub-objec-
bij P 0; t ij P 0; and X ij ¼ 0; 1 8i and j P 1 tive have on the system’s performance?
(2) Does having a minimum residence time on the job affect sys-
Although solving the above MILP problem provides the optimal tem performance?
solution, its implementation in industry may be impractical and (3) What effect does the speed of recovery from fatigue have on
therefore limited. In the next section we investigate four practical the system’s performance?
cases that are simplifications of the problem presented in (13). (4) How does the load level (fraction of MLC) affect the system’s
performance?
4. Practical cases (5) What effect does assuming equal task times have on system
performance?
Consider a DRC system with a flexibility of two, where the (6) How suboptimal are the simplified solutions from the above
worker alternates between the two tasks for m cycles over T = cases?
480 min. The worker has to spend a minimum time, (e.g., tmin = (7) How does forcing lunch breaks into the work schedule affect
30 min), working on each task before being transferred to the the results?
other task. For a given task, the operator will be required to work (8) How sensitive is the model to the form of the MET function?
at a fraction of his/her maximum load capability (MLC), say
f1 = 0.3 and f2 = 0.3. In this model, the fraction of MLC is assumed
to be applied continuously by the worker when performing a task 4.1. Case I: force a break after each task
for a period equivalent to the task’s duration.
While we use the form of MET equations that are based on In this case, it is assumed that the length of the break
empirical data, these models are based on static tasks, with con- between tasks is not long enough to result in full recovery

Fatigue

T
LT

cycle m

cycle 1
L1

t1 b1 t2 b2 Time

Fig. 1. Schematic illustrating work-rest policy over time for Case I with a recovery break after each task.
M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84 79

according to the RA. So, for this case and as described in Fig. 1, Subject to mðt1 þ t2 þ b1 þ b2 Þ  b2 ¼ T
fatigue builds over the duration of the shift. Fig. 1, as well as mf1 t 1  ma1 b1 þ mf2 t 2  ðm  1Þa2 b2 6 Lmax
Figs. 2–4 that describe the remaining cases are illustrative sche-
t min;i 6 t i 6 MET i ; i ¼ 1; 2
matics of how fatigue may accumulate under each policy. These
figures were constructed for the case where the fatigue accumu-
RAi t i  bi P 0
lation rate in task 1 is faster than that in task 2 (task 1 has a bi P 0
higher f value than 2; i.e., f1 > f2). However, the same figures m P 1; m is an integer
could as well be constructed for the case where the fatigue accu-
mulation rate is the same for both tasks (i.e., f1 = f2) and for the where i = 1, 2. Z2 could be determined by calculating the area
case where the fatigue accumulation rate in task 1 is slower beneath the fatigue line in Fig. 1 and then dividing it by T and Lmax;
than that of task 2 (i.e., f1 < f2). So, these figures must not be which is approximated as LT/2Lmax.
confused with results shown in the numerical results section.
In Fig. 1, LT is the fatigue index at the end of T, where LT 6 Lmax
P
and T ¼ m j¼1 ðt 1j þ b1j þ t 2j þ b2j Þ  b2m . For practicality and sim-
4.2. Case II: a break follows the second task
plicity, let t1j = t1, b1j = b1, t2j = t2, and b2j = b2 for j = 1, . . . , m. Then,
T = m(t1 + b1 + t2 + b2)  b2. Note that the fatigue index reached by In this case, it is assumed that a break is introduced following
the end of task 1 in cycle 1 (m = 1) is L1 = f1t1, which is followed task 2. In Fig. 2, the length of this break is not significant enough
P
by a break of length b1 over which some of the accumulated fatigue to result in full recovery; i.e., LT 6 Lmax. T ¼ m j¼1 ðt 1j þ t 2j Þ  b2m .
is recovered. The fatigue index then reduces from L1 to L1  a1b1, In this equation the final break is not included in the shift as the
where a1 = Lmax/(RA1  MET1)= a1 = f1/RA1. Following this break, recovery can occur when the operator is at home between shifts.
the worker performs task 2 in cycle 1 for t2 during which the fati- Since t1j = t1, t2j = t2, and b2j = b2 for j = 1, . . . , m, then,
gue accumulates to the level f1t1  a1b1 + f2t2. Task 2 is followed by T = m(t1 + t2 + b2)  b2. Fatigue builds over T and it is computed as
a break of length b2, which reduces the fatigue index to f1t1  LT = m(f1t1 + f2t2)  (m  1)a2b2 6 Lmax.
a1b1 + f2t2  a2b2. This pattern is repeated for m cycle with the In a similar approach to that adopted in modelling CASE I, prob-
exception in the last cycle that the last b2 falls beyond T and so it lem (13) can be written for this case as
is ignored. Then, LT is computed as
Minimize Z ¼ w1 Z 1 þ w2 Z 2
LT ¼ f1 t1  a1 b1 þ f2 t 2  a2 b2 þ     
mðt1 þ t 2 Þ
¼ mf1 t 1  ma1 b1 þ mf2 t 2  ðm  1Þa2 b2 ; ¼ w1 1 
T
where LT 6 Lmax.  
mðf1 t 1 þ f2 t2 Þ  ðm  1Þa2 b
The average fatigue over T can be determined in a similar manner þ w2 ð15Þ
2Lmax
to determining average stock level in inventory management litera-
ture (see for instance Khouja, 2005). By doing so, the size of the prob-
lem presented in (13) is reduced significantly as the constraint Subject to mðt1 þ t2 þ b2 Þ  b2 ¼ T
equations in (10) are no longer needed. Besides, assuming t1j = t1,
mðf1 t1 þ f2 t2 Þ  ðm  1Þa2 b2 6 Lmax
b1j = b1, t2j = t2, and b2j = b2 reduces the number of decision variables
to 5 from 6  2  m as the Xij binary decision variables are no longer
t i P t min;i ; i ¼ 1; 2
needed. Therefore, problem (13) can be re-written for this case as: t 1 þ t 2 6 minðMET 1 ; MET 2 Þ
  RAðt 1 þ t 2 Þ  b2 P 0
mðt1 þ t2 Þ
Minimize Z ¼ w1 Z 1 þ w2 Z 2 ¼ w1 1  b2 P 0; m P 1; m is an integer
T
 
mf1 t1  ma1 b1 þ mf2 t2  ðm  1Þa2 b2
þ w2 ð14Þ
2Lmax where RA being a weighted value of RA1 and RA2.

Fatigue

T
LT

cycle m
cycle 1

t1 t2 b2

Fig. 2. Schematic illustrating work-rest policy for Case II with a recovery break only after task 2.
80 M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84

Fatigue
T

cycle 1 cycle m

L1≤ Lmax
L2≤ Lmax

t1 b1,m t2 b2,m
Time
Fig. 3. Schematic illustrating work-rest policy for Case III with full recovery after each task.

4.3. Case III: force full recovery (break) after each task 4.4. Case IV: force full recovery (break) after the second task

This case assumes that a break follows each task. The length of This case assumes that a break is forced after the second task.
this break is significant enough to result in full recovery. The work- The length of this break is set to result in full recovery. The worker
er may or may not have reached their maximum fatigue index level may or may not have reached maximum fatigue index in each task;
P
in each task; i.e., L1 6 Lmax and L2 6 Lmax. For this case, i.e., L 6 Lmax. For this case, T ¼ m j¼1 ðt 1j þ t 2j þ b2;max Þ  b2;max , by
P
T¼ m j¼1 ðt 1j þ b1;max þ t 2j þ b2;max Þ  b2;max , by letting t1j = t1 and letting t1j = t1 and t2j = t2, then, T = m(t1 + t2 + b2,max)  b2,max.
t2j = t2, then, T = m(t1 + b1,max + t2 + b2,max)  b2,max. The average fa- The average fatigue over T is determined by calculating the
tigue over T is determined by calculating the areas of the triangles areas of the triangles and the trapezoids in Fig. 4 and then dividing
in Fig. 3 and then dividing the sum of these areas by T. Here, b1,max the sum of these areas by T. Here, b2,max is not a decision variable as
and b2,max are not decision variables as b1,max = a1t1 and b2,max = b2,max = a2t2. The maximum fatigue level attained in T is
a2t2. The maximum of maximum fatigue levels in T are attained L ¼ f1 t1 þ f2 t2 6 Lmax :
when maxðL1 ; L2 Þ 6 Lmax and minðL1 ; L2 Þ P 0 respectively. In a similar manner to modelling the previous cases, problem
In a similar manner to modelling the previous cases, problem (13) can be written for this case as
(13) can be written for this case as  
mðt 1 þ t 2 Þ
  Minimize Z ¼ w1 Z 1 þ w2 Z 2 ¼ w1 1 
mðt1 þ t 2 Þ T
Minimize Z ¼ w1 Z 1 þ w2 Z 2 ¼ w1 1 
T 
 w2 f1 t 1 f2 t 2
f1 t 1 þ m t1 þ mðf1 t 1 þ Þt2
þ w2 m ðt1 þ b1;max Þ Lmax T 2 2
2TLmax
 
f2 t 2 f2 ðf1 t 1 þ f2 t 2 Þ
þ ðm  1Þ ðt2 þ b2;max Þ þ t2 ð16Þ þ ðm  1Þ b2;m ð17Þ
2TLmax 2TLmax 2

Subject to mðt 1 þ t2 þ b1;m þ b2;m Þ  b2;m ¼ T Subject to mðt1 þ t2 þ b2;m Þ  b2;m ¼ T


t min;i 6 t i 6 MET i ; i ¼ 1; 2 t 1 þ t 2 6 minðMET 1 ; MET 2 Þ
maxðf1 t 1 ; f2 t2 Þ 6 Lmax t i P t min;i ; i ¼ 1; 2
minðf1 t 1 ; f2 t 2 Þ P 0 m P 1; m is an integer f1 t 1 þ f2 t2 6 Lmax ; m 6 1; m is an integer

Fig. 4. Schematic illustrating work-rest policy for Case IV with full recovery after task 2.
M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84 81

5. Numerical examples and results that allowing for a minimum residence time, tmin, improves the
overall system performance for all cases (Low Z values) since it al-
The four cases described in Section 4 were solved with EXCEL lows shorter cycles and more frequent transfers (large m values).
SOLVER enhanced with VBA codes using the above input data with This does not allow fatigue to accumulate, thus reducing break
results summarized in Table 1. The results show that Case I per- times to low values and improving productivity. Further, Case I
formed the best when the Z values are compared. Case III performed out-performed the other cases when the average Z values in Table 3
better than cases II and IV but slightly lower than Case I. Case II and are compared.
III performed better on the productivity scale (lowest Z1 value),
while Cases I and III performed better on the fatigue scale. 3. What effect does the speed of recovery from fatigue have on the
system’s performance?
1. What effect does changing the preferences of the sub-objective
have on the system’s performance? The numerical examples in Table 1 were solved for varying RAi,
i = 1, 2, with the results summarized in Table 4. The results show
The above numerical examples were solved for Z1 (productivity) that the longer it takes the worker to recover following a break
being preferred over Z2 (fatigue) (w1 = 0.75 > w2 = 0.25). The results (Large RA value) the worse would be the system performance (high
are summarized in Table 2 and show that Case I performed the best Z values). Higher RA values reduced the systems’ productivity (Z1
when the Z values are compared. Case IV performed the best on increases), but has little effect on the fatigue index (Z2 is insensi-
productivity (lowest Z1 value), while Case III was the best on fati- tive). The results show that Case I performed better then the other
gue. The change in sub-objective weights has no affect on the gen- cases when the Z values for each case in Table 4 were averaged
eral results. (ZI = 0.346, ZII = 0.448, ZIII = 0.353, ZIV = 0.669). The numerical
examples in Table 4 were solved once more for different weight
2. Does having a minimum residence time on the job affect system values with preference given to productivity; w1 = 0.75 and
performance? w2 = 0.25. Again, the results show that CASE I out-performed the
other cases (ZI = 0.418, ZII = 0.450, ZIII = 0.422, ZIV = 0.565).
The numerical examples in Table 1 were solved for tmin = 10, 20,
and 30 with the results summarized in Table 3. The results show 4. How does the load level (fraction of MLC) affect the system’s
performance?
Table 1
Optimal work/recovery policies for the four practical cases.
It was shown earlier that having a minimum residence time on
Cases m Task durations Break durations Optimisation terms the job before transferring the worker to the next job improved the
t1 t2 b1 b2 Z1 Z2 Z system’s performance for all four cases. The numerical examples in
Table 1 were solved for varying the value f1(0.05–0.5), while hold-
I 4 52.50 30.00 26.25 15.00 0.156 0.100 0.256
II 5 34.29 34.29 0.00 34.29 0.143 0.229 0.372 ing that of f2 (i.e., f2 = 0.3), as well as the other input parameters
III 5 31.38 34.95 15.69 17.48 0.155 0.111 0.266 used in the examples of Table 1 unchanged, with the results shown
IV 5 30.00 38.57 0.00 34.29 0.143 0.444 0.587 in Fig. 5. It can be shown from Fig. 5 that the system’s performance
Note: w1 = w2 = 0.5. improves as f1 – f2, with the best performance attained when
f1 < f2. This recommends that a worker be assigned to tasks with
different fraction MLC in order to attain better system perfor-
Table 2 mance. These results also indicate that in a DRC system with flex-
Optimal work/recovery policies for the four practical cases. ibility of 2 it is better to start with a task that has a lower fraction
Cases Cycles Task durations Break durations Optimisation terms MLC than the other task. The results also show that Case III is the
least sensitive to changes in fraction MLC. This suggests that, for
m t1 t2 b1 b2 Z1 Z2 Z
production situations where tasks have similar fraction MLC, Case
I 4 52.50 30.00 26.25 15.00 0.234 0.050 0.284
III may be the recommended one. Furthermore, the results indicate
II 5 34.29 34.29 0.00 34.29 0.215 0.114 0.329
III 5 30.00 36.43 15.00 18.21 0.232 0.057 0.289 that productivity will tend to increase as workload decreases.
IV 5 30.00 38.57 0.00 34.29 0.214 0.223 0.437 Similarly, the same numerical examples were investigated for
f1 = 0.3, while f2 (0.05–0.3) was varied. Similar results to those
Note: w1 = 0.75, w2 = 0.25.
shown in Fig. 5 were found.

Table 3
Optimal work/recovery policies for the four practical cases for a varying task residence time.

Cases Minimum residence time Cycles Task durations Break durations Optimisation terms
tmin m t1 t2 b1 b2 Z1 Z2 Z
I 10 16 10.21 10.00 5.10 5.00 0.163 0.034 0.197
20 8 20.83 20.00 10.42 10.00 0.160 0.066 0.226
30 4 52.50 30.00 26.25 15.00 0.156 0.100 0.256
II 10 16 10.21 10.21 0.00 10.21 0.160 0.068 0.228
20 8 20.87 20.87 0.00 20.87 0.152 0.139 0.291
30 5 34.29 34.29 0.00 34.29 0.143 0.229 0.372
III 10 16 10.00 10.21 5.00 5.11 0.163 0.034 0.197
20 8 20.00 20.87 10.00 10.43 0.159 0.069 0.228
30 5 31.38 34.95 15.69 17.48 0.155 0.111 0.266
IV 10 5 10.00 58.57 0.00 34.29 0.143 0.293 0.436
20 5 20.00 48.57 0.00 34.29 0.143 0.353 0.496
30 5 30.00 38.57 0.00 34.29 0.143 0.444 0.587

Note: w1 = w2 = 0.5.
82 M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84

Table 4
Optimal work/recovery policies for the four practical cases for varying recovery allowances.

Cases Recovery allowance Cycles Task durations Break durations Optimisation terms
RA1 RA2 m t1 t2 b1 b2 Z1 Z2 Z
I 2.00 2.00 3 30.00 30.00 60.00 60.00 0.313 0.100 0.413
1.50 1.50 3 40.00 30.00 60.00 45.00 0.281 0.100 0.381
1.00 1.00 3 55.00 30.00 55.00 30.00 0.234 0.100 0.334
0.50 0.50 4 52.50 30.00 26.25 15.00 0.156 0.100 0.256
II – 2.00 3 30.00 38.57 0.00 137.14 0.286 0.229 0.515
– 1.50 4 30.00 30.00 0.00 90.00 0.262 0.200 0.462
– 1.00 4 34.29 34.29 0.00 68.57 0.214 0.229 0.443
– 0.50 5 34.29 34.29 0.00 34.29 0.143 0.229 0.372
III 2.00 2.00 3 30.00 30.00 60.00 60.00 0.313 0.100 0.413
1.50 1.50 3 32.22 39.72 48.33 59.58 0.275 0.120 0.395
1.00 1.00 4 30.00 34.29 30.00 34.29 0.232 0.107 0.339
0.50 0.50 5 31.38 34.95 15.69 17.48 0.155 0.111 0.266
IV – 2.00 3 30.00 38.57 0.00 137.14 0.286 0.444 0.730
– 1.50 3 34.40 40.60 0.00 112.50 0.266 0.433 0.699
– 1.00 4 38.57 30.00 0.00 68.57 0.214 0.444 0.658
– 0.50 5 30.00 38.57 0.00 34.29 0.143 0.444 0.587

Note: w1 = w2 = 0.5.

than 2% for Cases II and III. This suggests that for some cases, sim-
0.700
CASE I plifying the problem further can make the system easier to manage
CASE II without compromising system performance.
0.600 CASE III
System's Performance (Z)

CASE IV
6. How suboptimal are the simplified solutions from the above
cases?
0.500

The MILP problem given by (13) was solved for the same input
0.400 parameters used to solve those examples in Table 1. The optimal
solution obtained using LINDO is summarized in Table 6. The re-
sults show that a break is forced after each task allowing for full
0.300
recovery. There is no break following the second task in the last cy-
cle, which is consistent with the assumption adopted in all four
0.200 cases. The optimal policy in Table 6 is similar to that of Case I,
which has been shown above to be better than the other cases
(Cases II, III, and IV), where it suggests forcing a break after each
0.100
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 task allowing partial recovery from fatigue.
Note that Z2 in (15) is an approximation of the one used in (13),
Fraction MLC (f1)
where it was computed in a similar manner to determining aver-
Fig. 5. The behaviour of Z for different loading levels on task 1. age stock level in inventory management (LT  T/2). Besides, the
duration of the second task in the last cycle is larger than those
of task 2 in the previous cycles, whereas Case I assumes that cycle
5. What effect does assuming equal task times have on system time is constant. These are the reasons why the value of Z in Table 6
performance? is lower than that of Z for Case I in Table 1. Similar numerical
examples to the one in Table 6 were solved and the results were
Here we investigate the numerical examples in Table 1 for the in favour of Case I as it produced closest Z values to the optimal
assumption that t1 = t2, a situation that would be easier for produc- policy determined from (13).
tion supervisors to manage on the shop floor. The results in Table 5
suggest that this assumption provides a good approximation to the 7. How does forcing lunch breaks into the work schedule affect the
four cases when comparing the values of Z1, Z2, and Z in Table 5 to results?
those in Table 1. The numerical examples in Fig. 5 were also solved
for this assumption and the results showed that the values of Z Here, we investigate Case I as it appears to be the most promis-
were on the average higher by about 7% for Cases I and IV, and less ing of the cases discussed. Assume a lunch break of B = 60 is im-
posed in a work-shift of T = 480. This requires altering Z2 in the
Table 5 objective function and the first constraint to:
Optimal work/recovery policies for the four practical cases when t1 = t2.  
m1 a2 b2 a2 B
Cases Cycle Task durations Break durations Optimisation terms Z2 ¼ ðf1 t 1 þ f2 t 2  a1 b1  a2 b2 Þ þ 2 
2TLmax 2TLmax 2TLmax
m t1 t2 b1 b2 Z1 Z2 Z
I 5 33.10 33.10 16.55 16.55 0.155 0.111 0.266 mðt 1 þ t 2 þ b1 þ b2 Þ þ B  2b2 ¼ T or mðt 1 þ t 2 þ b1 þ b2 Þ  2b2
II 5 34.29 34.29 0.00 34.29 0.143 0.229 0.372
III 5 33.10 33.10 16.55 16.55 0.155 0.111 0.266 ¼T B
IV 5 34.29 34.29 0.00 34.29 0.143 0.500 0.643
where m = m1 + m2 with m1 and m2 being the number of cycles be-
Note: w1 = w2 = 0.5. fore and after the break respectively. Solving for the modified Case I
M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84 83

Table 6 6. General discussion


Optimal work/recovery policy obtained from solving the MILP.

Z1 = 0.1406 w1 = 0.5 This paper investigated the effects of human fatigue and recov-
Z2 = 0.0101 w2 = 0.5 ery on the system’s performance of a dual-resource constrained
Z = 0.1507 (DRC) system. A mixed-integer linear programming (MILP) prob-
j t1j t2j X1j X2j X1jb1j X2jb2j lem was developed that is capable of accommodating n tasks and
m cycles. Increasing n and m increases the computational complex-
1 38.57 38.57 1 1 19.29 19.29
2 38.57 38.57 1 1 19.29 19.29
ity of the problem. However, the literature on DRC systems shows
3 38.57 38.56 1 1 19.29 19.29 that a workforce flexibility of 2 or 3 is recommended (e.g., Kher
4 38.56 75.00 1 0 19.29 0.00 et al., 1999), although fatigue aspects have not been considered
Total 154.27 190.70 77.16 57.87 in these works. In this paper, a flexibility of 2 was assumed and
four practical cases of the MILP were considered. Such a model
Task 1 Task 2 has the potential to help secure productive job rotation schedules
j LS1j L1j LE1j LS2j L2j LE2j that minimize risks of overload and associated health disorders in
employees. More detailed analyses of model behaviour under dif-
1 0.00 11.57 11.57 0.00 11.57 11.57
2 0.00 11.57 11.57 0.00 11.57 11.57 ferent circumstances, including different combinations of task load
3 0.00 11.57 11.57 0.00 11.57 11.57 amplitudes, and different policies is required. In particular, the
4 0.00 11.57 11.57 0.00 22.50 22.50 choice of weight factors between fatigue and productivity (w1,
T = 154.27 + 190.70 + 77.16 + 57.87 = 480 min. w2) is a non-trivial matter. While this also requires examination
in terms of model behaviour, there is a separate issue of manage-
rial attitudes and beliefs about what kind of working conditions
should be offered to employees. Both qualitative and quantitative
for the same input parameters as those for Table 1, the results show investigation techniques are indicated for such a topic.
the optimal policy is m = 5 (m1 = 4, m2 = 1), t1 = 36.00, t2 = 30.00, While the model presented here provides a general frame for
b1 = 18.00, b2 = 0, Z1 = 0.156, Z2 = 0.100, Z = 0.256. When solving this assessment, there remain weaknesses with the available fatigue
example, m2 was set to 1, 2, etc., while m1 (decision variable) was and recovery models which currently predict MET for static (iso-
optimised. Comparing these values with those of Case I in Table 1, metric) loading only. The model could be refined with better
permitting a lunch break does not impede the system performance knowledge on the shape of the fatigue accumulation and recovery
as Z remained unchanged. curves which were assumed to be linear in this preliminary model.
Further research is also needed to understand the ‘level’ of fatigue
8. How sensitive is the model to the form of the MET function? to be most usefully modelled, where options include psychological
fatigue, physical fatigue at the level of the whole body, a specific
The human factors literature suggests that the MET function for body joint (e.g. elbow), a muscle group or individual muscle, a mo-
muscular fatigue may be of either power or exponential forms as tor unit within a muscle, motor-neuron activity, or even the bio-
described in (1) and (2) (El Ahrache et al., 2006). Although this pa- chemistry of the contractile units. The relationship between load
per adopted an exponential form similar to that of Rose, Ericsson, and fatigue may be a complex mix of effects from these various
Glimskar, Nordgren, and Ortengren (1992), here, we will investi- system levels. While these aspects do not invalidate the modelling
gate a few of the above examples for the alternative assumption procedures here, which are based on the form of the MET functions,
that MET is of a power form instead. For the comparison to be there remains a need for more usable predictive fatigue-recovery
meaningful, the parameters used with the exponential form (1) models that can accommodate a range of activities and dynamic
(MET = 75 when fMLC = f = 0.30) are adapted to the power form (2) muscular demands for each real-world task under consideration.
which was found to be MET = 83.032  f0.0699, R2 = 0.87. The Other extensions to this work could include the learning and for-
numerical examples in Table 1 were solved for this MET function getting effects that occur as operators move between different
with results summarized in Table 7. tasks in the system and the inclusion of variable error making rates
Similar to what has been observed in Table 1, the results show that can occur under conditions of high fatigue and poor learning.
that Case I performed the best when the Z values are compared. While it might seem obvious that more errors will be made as fa-
Case III performed better than cases II and IV but slightly lower tigue increases the form of these relationships requires investiga-
than Case I. Case II and III performed better on productivity (lowest tion. The interaction effects of these factors may greatly influence
Z1 value), while Cases I and III performed better on fatigue. Similar overall system performance. The real-world validity of these mod-
results to those in Tables 2–4 and Fig. 5 were obtained concluding elling approaches needs to be confirmed through comparison to
that assuming a MET of either a power or exponential form does real running systems. Finally, studies are needed of the utility of
not greatly affect the general behaviour of the DRC system de- these models in supporting design teams to create systems with
scribed and analyzed in this paper. sustainable performance that is not compromised by inattention
to human factors and that does not threaten the health and safety
of employees.

Table 7
Optimal work/recovery policies for the four practical cases when MET is of a power
form.
7. Conclusions

Cases Cycle Task durations Break durations Optimisation terms It is possible to include fatigue and recovery aspects into models
m t1 t2 b1 b2 Z1 Z2 Z of DRC systems to support the identification of task schedules that
I 5 36.00 30.00 18.00 15.00 0.156 0.084 0.240 optimise productivity without overloading system operators.
II 5 34.29 34.29 0.00 34.29 0.143 0.190 0.333 Adapting this model as an inventory problem reduced mathemat-
III 5 31.03 35.32 15.52 17.66 0.154 0.093 0.247 ical complexity without biasing results. Of the four practical cases
IV 6 30.00 30.00 0 30.00 0.125 0.531 0.656
investigated Case I, in which partial recovery is permitted after
Note: w1 = w2 = 0.5. each task, performed best. The optimal durations of the tasks were
84 M.Y. Jaber, W.P. Neumann / Computers & Industrial Engineering 59 (2010) 75–84

such that maximum fatigue was never reached. The analysis of Hagberg, M., Silverstein, B., Wells, R., Smith, M. J., Hendrick, H. W., Carayon, P., &
Pérusse, M. (1995). Work related musculoskeletal disorders (WMSDs): A reference
these cases suggests our findings are not affected by changing
book for prevention. London: Taylor & Francis.
the preference of the fatigue and productivity sub-objective func- Hopp, W. J., Tekin, E., & Van Oyen, M. P. (2004). Benefits of skill chaining in serial
tions. A shorter task residence time, with more frequent alterna- production lines with cross-trained workers. Management Science, 50(1), 83–98.
tion between tasks, increased system performance for all cases Hottenstein, M. P., & Bowman, S. A. (1998). Cross training and worker flexibility: A
review of DRC system research. The Journal of High Technology Management
due to less fatigue accumulation. Faster recovery rates are associ- Research, 9(2), 157–174.
ated with improved system performance, mostly due to productiv- Inman, R. R., Jordan, W. C., & Blumenfeld, D. E. (2004). Chained cross-training of
ity improvements. System performance was generally worst when assembly line workers. International Journal of Production Research, 42(10),
1899–1910.
loading was equal for both tasks and was best when loading for the Jaber, M. Y., Kher, H. V., & Davis, D. J. (2003). Countering forgetting through training
first task was lower than that of the second task. Simplifying the and deployment. International Journal of Production Economics, 85(1), 33–46.
allocation schedule with equal task times had little difference on Kaneko, K., & Sakamoto, K. (2001). Spontaneous blinks as a criterion of visual fatigue
during prolonged work on visual display terminals. Perceptual and Motor Skills,
the system’s behaviour. The schedule generated from the more 92(1), 234–250.
complex optimal solution was in conformance with Case 1. Adding Kher, H. V., Malhotra, M. K., Philipoom, P. R., & Fry, T. D. (1999). Modeling
a lunch break in the work-shift (with the same active work time) simultaneous worker learning and forgetting in dual resource constrained
systems. European Journal of Operational Research, 115(1), 158–172.
did not reduce system performance. The form of the fatigue func- Khouja, M. (2005). The use of minor setups within production cycles to improve
tion, be it of exponential or power form, did not affect the system product quality and yield. International Transactions in Operational Research,
behaviour. 12(4), 403–416.
Konz, S. (1998). Part II the scientific basis (knowledge base) for the guide.
This paper is believed to be the first in presenting an approach
International Journal of Industrial Ergonomics, 22(1), 67–99.
to accounting for operator fatigue and recovery in the design of Konz, S., & Johnson, S. (2004). Work design-occupational ergonomics (6th ed.).
DRC systems, considerable research and development is needed Scottsdale, AZ: HolcombHathaway.
Krajewski, L. J., King, B. E., Ritzman, L. P., & Wong, D. S. (1987). Kanban, MRP, and
to make it more accessible to system designers and managers.
shaping the manufacturing environment. Management Science, 33(1), 39–57.
Leigh, J. P., Markowitz, S. B., Fahs, M., Shin, C., & Landrigan, P. J. (1997). Occupational
injury and illness in the United States. Estimates of costs, morbidity, and
Acknowledgements mortality. Archives of Internal Medicine, 157(14), 1557–1568.
McCreery, J. K., & Krajewski, L. J. (1999). Improving performance using workforce
flexibility in an assembly environment with learning and forgetting effects.
The authors thank the Social Sciences and Humanities Research International Journal of Production Research, 37(9), 2031–2058.
Council of Canada (SSHRC)-Standard Grant-for supporting their re- McCreery, J. K., Krajewski, L. J., Leong, G. K., & Ward, P. T. (2004). Performance
search. They also thank Mr. Khashayar Hojjati-Emami for his re- implications of assembly work teams. Journal of Operations Management, 22(4),
387–412.
search assistance work. Miles, B. L., & Swift, K. (1998). Design for manufacture and assembly. Manufacturing
Engineer, 77(5), 221–224.
Nelson, R. T. (1967). Labor and machine limited production systems. Management
References Science, 13(9), 648–671.
Neumann, W. P., Kihlberg, S., Medbo, P., Mathiassen, S. E., & Winkel, J. (2002). A case
study evaluating the ergonomic and productivity impacts of partial automation
Åhsberg, E. (1998). Perceived Fatigue related to Work. Ph.D. Thesis. Sweden,
strategies in the electronics industry. International Journal of Production
University of Stockholm & National Institute for Working Life.
Research, 40(16), 4059–4075.
Alexander, D. C. (1998). Strategies for cost justifying ergonomic improvements. IIE
Neumann, W. P., Winkel, J., Medbo, L., Mathiassen, S. E., & Magneberg, R. (2006).
Solutions, 30(3), 30–35.
Production system design elements influencing productivity and ergonomics –
Bigland-Ritchie, B., Cafarelli, E., & Vøllestad, N. K. (1986a). Fatigue of submaximal
A case study of parallel and serial flow strategies. International Journal of
static contractions. Acta physiologica Scandinavica, 128(suppl. 556), 137–148.
Operations and Production Management, 26(8), 904–923.
Bigland-Ritchie, B., Furbush, F., & Woods, J. J. (1986b). Fatigue of intermittent
Niebel, B. W., & Freivalds, A. (2003). Methods standards and work design (11th ed.).
submaximal voluntary contractions: Central and peripheral factors. Journal of
Boston, MA: McGraw Hill.
Applied Physiology, 61(2), 421–429.
Oxenburgh, M., Marlow, P., & Oxenburgh, A. (2004). Increasing productivity and profit
Bigland-Ritchie, B., Rice, C. L., Garland, S. J., & Walsh, M. L. (1995). Task-dependent
through health & safety: The financial returns from a safe working environment.
factors in fatigue of human voluntary contractions. Advances in Experimental
Boca Raton, FL: CRC Press.
Medicine and Biology, 384, 361–380.
Rohmert, W. (1973). Problems in determining rest allowances, Part 2: Determining
Bobrowski, P. M., & Park, P. S. (1993). An evaluation of labor assignment rules when
rest allowances in different tasks. Applied Ergonomics, 4(3), 158–162.
workers are not perfectly interchangeable. Journal of Operations Management,
Rose, L., Ericsson, M., Glimskar, B., Nordgren, B., & Ortengren, R. (1992). Ergo-index.
11(3), 257–268.
Development of a model to determine pause needs after fatigue and pain
Bokhorst, J. A. C., Slomp, J., & Gaalman, G. J. C. (2006). Cooperation in a distributed
reactions during work. In M. Mattila & w. Karwowski (Eds.), Computer
dual resource-constrained manufacturing system. International Journal
applications in ergonomics. Occupational safety and health (pp. 461–468).
Computer Applications in Technology, 26(1–2), 83–90.
North-Holland: Elsevier Science Publishers B.V.
Docherty, P., Forslin, J., & Rami Shani, A. B. (2002). Creating sustainable work systems
Soo, Y., Nishino, M., Sugi, M., et al. (2009). Evaluation of frequency band technique
(2nd ed.). London: Routledge.
in estimating muscle fatigue during dynamic contraction task. In IEEE
Edwards, R. H. T. (1981). Human muscle function and fatigue. Ciba Foundation
international conference on robotics and automation, May 12–17, Kobe
Symposium, 82, 1–18.
International Conference Center Kobe, Japan.
Eklund, J. (1997). Ergonomics, quality and continuous improvement – Conceptual
Treleven, M. D. (1989). A review of the dual resource constrained system research.
and empirical relationships in an industrial context. Ergonomics, 40(10),
IIE Transactions, 21(3), 279–287.
982–1001.
Treleven, M. D., & Elvers, D. A. (1985). An investigation of labor assignment rules in
El Ahrache, K., & Imbeau, D. (2008). Comparison of rest allowance models for static
a dual constrained job shop. Journal of Operations Management, 16(1), 51–68.
muscular work. International Journal of Industrial Ergonomics, 39(1), 1–8.
Van Dieen, J. H., & Oude Vrielink, H. H. E. (1994). The use of relation between
El Ahrache, K., Imbeau, D., & Farbos, B. (2006). Percentile values for determining
relative force and endurance time. Ergonomics, 37(2), 231–243.
maximum endurance times for static muscular work. International Journal of
Wisner, J. D., & Pearson, J. N. (1993). An exploratory study of the effects of operator
Industrial Ergonomics, 36(2), 99–108.
relearning in a dual resource constrained job shop. Production and Operations
Enoka, R. M., & Duchateau, J. (2008). Muscle fatigue: What, why and how it
Management Journal, 2(1), 55–68.
influences muscle function. The Journal of Physiology, 586(1), 11–23.
Yue, H. (2005). Worker flexibility in dual resource constrained (DRC) shops. PhD
Felan, J. T., & Fry, T. D. (2001). Multi-level heterogeneous worker flexibility in a dual
Thesis, Netherlands, University of Groningen.
resource constrained (DRC) job shop. International Journal of Production
Zamiska, J. R., Jaber, M. Y., & Kher, H. V. (2007). Worker deployment in dual resource
Research, 39(14), 3041–5059.
constrained systems with a task-type factor. European Journal of Operational
Fryer, J. S. (1976). Organizational segmentation and labor transfer policies in labor
Research, 177(3), 1507–1519.
and machine limited production systems. Decision Sciences, 7(4), 725–738.

You might also like