Anda di halaman 1dari 16

Annals of Operations Research 128, 251266, 2004 2004 Kluwer Academic Publishers. Manufactured in The Netherlands.

A Quantitative Method for Determining Proper Job Rotation Intervals


WIPAWEE THARMMAPHORNPHILAS

Department of Industrial Engineering, Chulalongkorn University, Pyathai Road, Bangkok 10330, Thailand banorman@engrng.pitt.edu Department of Industrial Engineering, University of Pittsburgh, 1048 Benedum Hall, Pittsburgh, PA 15261, USA
BRYAN A. NORMAN

Abstract. Job rotation is a useful tool that can be used in many work settings to produce task variety for workers and to reduce worker fatigue and injuries. Developing a job rotation plan requires determining the set of jobs to be included in the rotation, the rotation sequence, and the proper length of the rotation interval. Previous research has considered both the jobs to include in a rotation plan and how to sequence the rotation plan but no work has been conducted to quantitatively examine different rotation interval lengths. Using different rotation interval lengths may result in different stresses for the workers performing a particular set of tasks and therefore affect the quality of the job rotation implementation. This paper proposes a quantitative method based on mathematical programming to obtain a proper job rotation interval length. The merit of the method is determined by examining its effectiveness in reducing worker stress and the potential for injury as well as its applicability for use in real settings. Keywords: rotation interval, job rotation, integer programming

1.

Introduction

Job rotation is a method of rotating workers across various tasks that have different physical and mental demands over time. It is a promising method to manage worker fatigue. It is especially useful for dynamic tasks, which require variations in muscular load (Jonsson, 1988). Job rotation has been implemented in many settings including refuse collecting (Neter, Wasserman, and Kutner, 1990), cashiering (Hinnen et al., 1992), and poultry processing (Henderson, 1992). These studies claim that job rotation can help reduce the incidence of injuries due to performing repetitive tasks. It also helps reduce errors, and increase employee job satisfaction. Although job rotation is a promising means to manage daily exposure due to physical stress, a poorly designed job rotation plan can actually increase worker stress. To date, few researchers have explored the details of how to perform job rotation. Developing a job rotation plan requires determining the set of jobs to be included in the rotation, the rotation sequence, and the proper length of the rotation interval. In most
Corresponding author.

252

THARMMAPHORNPHILAS AND NORMAN

studies the authors assume that the set of jobs to be included in the rotation plan is already known. A few different methods have been applied to the rotation sequence problem. One simple way to sequence the jobs is to randomly rotate the workers through the jobs. A second simple method is to follow the task sequence by having workers move from the rst workstation to the second and so on until the worker at the last workstation moves to the rst one. Another possibility is to rotate workers based upon a specic policy such as choosing not to assign a worker to two stressful tasks in succession (Henderson, 1992). Carnahan, Redfern, and Norman (2000) proposed a method to reduce low back injury in a manual lifting environment by implementing a genetic algorithm to provide multiple good job rotation schedules and then used a clustering method to determine a general set of rules governing task exposure for each group of workers. Tharmmaphornphilas et al. (2003) used an integer programming method to develop a proper rotation plan to reduce noise exposure for workers working in a sawmill plant. Determining the correct rotation interval length is another important issue that should be considered when performing job rotation. The rotation interval length can be dened as the length of time that a worker performs a task before he/she is rotated to another task. To date, a constructive method has not been proposed in the literature to determine proper rotation frequency. Typically, simple ways to rotate workers are used such as rotating workers at random intervals or during break periods. However, neither of these has been demonstrated to be a proper time period to rotate workers. This paper proposes an integer programming based method to determine the proper length of rotation intervals, which can effectively reduce the potential for worker injuries. The method can be applied to various settings. In particular, this paper shows how to apply the integer programming method to two different settings. The rst concerns lifting tasks in a manual lifting environment and the second is lumber production in a sawmill plant. Background and detailed descriptions of these two settings are given in section 2. Section 3 describes the proposed methodology and section 4 provides a discussion and conclusions based on our ndings. 2. Problem settings

The proposed methodology was applied to two different types of problems. The rst is a setting where the tasks comprise manual lifting and the primary concern is low back injury. The second concerns minimizing workers noise exposure in a sawmill. Details of the manual lifting tasks and the methods that can be used to assess the potential for low back injuries are presented in section 2.1. Details of the sawmill operation and the method that can be used to assess the potential for hearing loss are included in section 2.2. 2.1. The manual lifting task setting A manual lifting environment is simulated in which task demands vary over time. The nature of the tasks and the task demands are similar to the ones used in Carnahan, Red-

QUANTITATIVE METHOD

253

Table 1 Task demand description for task A. Task descriptions Lifting weight (kg) Lifting rate (lifts/minute) Time 8:0012:00 1:005:00 8:0011:00 11:0012:00 1:002:00 2:005:00 8:0012:00 1:005:00 Subtasks (lifting ranges) FK SR KS 3537 2530 0.11.0 1.03.0 1.03.0 0.11.0 3040 1820 2327 2731 7.011.0 3.06.0 3.06.0 4.07.0 1520 1722 1822 1316 2.06.0 5.08.0 5.08.0 2.04.0 2027 1020

Lifting distance (cm)

fern, and Norman (2000) and include stochastic lifting weight, lifting frequency, lifting height, and the horizontal lifting distance. Each lifting task is comprised of smaller subtasks that are determined by the lifting height. Six different types of subtasks are considered: subtask FK is the task of lifting from oor level to knuckle height, subtask FS is the task of lifting from oor level to shoulder height, subtask FR is the task of lifting from oor level to reach height, subtask KS is the task of lifting from knuckle height to shoulder height, subtask KR is the task of lifting from knuckle height to reach height, and subtask SR is the task of lifting from shoulder height to reach height. The number and types of subtasks within each task can be different. We assume that the lifting weight, the lifting frequency and the horizontal lifting distance are uniformly distributed with different ranges during each time period. An example description for task A is shown in table 1 to demonstrate the variation in task demand. Task A is composed of 3 subtasks, which are subtasks FK, SR and KS. The lifting weight, the lifting rate, and the horizontal lifting distance are uniformly distributed with different ranges during different time periods. For example, from 8 a.m. to 12 p.m. the lifting weight of subtask FK is uniformly distributed between 35 and 37 kg while from 1 p.m. to 5 p.m. it is uniformly distributed between 25 and 30 kg. The ranges for the lifting rates and the horizontal lifting distances are shown in table 1. The method for generating the lifting weights, the lifting rates, and the lifting distances is similar for every subtask. Non-identical worker lifting proles are considered in order to simulate a real setting. Workers may vary in gender, anthropometry, age, strength, etc. Workers are grouped into percentiles based upon their gender and lifting capacity. Lifting capacity is determined by the maximum acceptable weight that workers can lift repeatedly for long periods without undue stress or over-tiring (Ayoub et al., 1978, 1980). Workers are assumed to perform their tasks by lifting with two hands in the sagittal plane. Lifting with only one hand or lifting in other planes accrues different amounts of stress in other parts of the body but is not considered in this research. We consider balanced rotation, which assumes that the number of workers and the number of tasks are equivalent. At any given time a worker performs only one task and a

254

THARMMAPHORNPHILAS AND NORMAN

task is done by only one worker. Workers can be rotated to perform different tasks at the end of each rotation interval and a working period is eight hours per day. The proposed methods can be extended to consider the case where multiple workers perform the same task or where workers perform multiple tasks during a single work interval. The Job Severity Index (JSI) is utilized as a measure to represent worker stress due to lifting (Liles and Deivanayagam, 1984). The JSI is a unitless measure that relates the lifting required for a task to a workers lifting capacity. The potential for low back injury is higher among workers who have higher JSI values than among workers who have lower JSI values. Statistical analysis shows that the potential for low back injury is dramatically increased when JSI values are greater than 1.5. Details of how to calculate JSI values can be obtained from Liles and Deivanayagam (1984). 2.2. The sawmill setting The second application in this study is based on applying job rotation to minimize noise exposure of workers performing sawmill operations at a cabinet manufacturer located in the southeastern United States. Initially, a noise evaluation consisting of area monitoring of eleven different workstations within the sawmill was conducted. The sawmill itself is only a part of the total cabinet-making operation at this plant. Raw timber from the yard outside the sawmill is debarked and sent into the mill. The sawmill operation involves the production of lumber from these trees, and the nished boards are sent to a storage building and are eventually used as raw material for the cabinet-production portion of the facility. Specically, the sawmill itself is responsible for the sawing, edging, trimming and grading of these boards. There are nine different operations that comprise the sawmill and these are presented in gure 1. Note that there are eleven workstations that required noise monitoring because two of the operations require two workers. The rst task, the head saw, involves cutting the debarked trees into small boards and large, square canes. The head saw tailor then sorts these boards and canes (and any scrap pieces of wood), sending the boards and scrap to the edger and chipper, respectively, while sending the canes to the gang saw. The gang saw cuts the canes into smaller boards and sends them to the descrambler and sorter while the edger trims the edges of its boards, also sending them to the descrambler and sorter. The descrambler and sorter arranges the boards onto a conveyor that takes them to the trim saw. The trim saw trims the ends of the boards and releases them to the green chain. There, the boards are graded and marked by the board marker, and the board stackers at the end of the green chain load the boards onto carts to be taken out of the sawmill. Operators generally work four 10-hour days (MondayThursday), and occasionally an eight-hour day on Friday. Currently, there is a job rotation plan within the sawmill based upon three jobs where workers can be rotated every 2.5-hour period. These jobs are the descrambler and sorter, the trim saw, and the 2-saw edger and chipper. The Occupational Safety and Health Administration of the US Department of Labor (OSHA) (Montgomery, 1997) and the National Institute for Occupational Safety and Health (NIOSH) (NIOSH, 1998) have determined threshold limit values for noise

QUANTITATIVE METHOD

255

Figure 1. Sawmill ow chart.

pressure levels that represent conditions which workers may be exposed to repeatedly without undue risk of noise induced hearing loss. OSHA has set a permissible exposure limit, using an 8-hour criterion, at the level of a time-weighted average (TWA) of 90 dBA while NIOSH uses an 8-hour criteria level of a TWA of 85 dBA. The time-weighted average represents the average noise pressure level that a worker is exposed to where the duration of the exposure is considered in forming the average. It is widely accepted that a worker can tolerate noise levels higher than the threshold levels for shorter periods of time. In this paper, we consider only the NIOSH criterion. According to NIOSH recommendations, the maximum exposure duration (Tj ) for task j can be calculated as Tj = 8 2(Lj 85)/3

where Lj represents the noise level (in dBA) of task j . Noise level and exposure duration are two critical factors considered by safety professionals. However, the noise level may vary during the work shift since workers

256

THARMMAPHORNPHILAS AND NORMAN

move about the workplace. Hence, to assess the noise exposure, the worker exposure duration (Cj ) to each noise level (Lj ) must be measured, and the exposure dosage (Dj ) for each noise level is calculated using the following equation: Dj = Cj . Tj

The total exposure dose (D) for the entire workday is calculated by summing all of the partial doses based on the tasks a worker performs during the day,
n

D = 100
j =1

Cj Tj

where n is the number of tasks in the work environment. Once the total exposure dose, D, has been found then the TWA can be calculated as shown below, D + 85 100 and the TWA is compared with the 85 dBA NIOSH guideline. Since the NIOSH recommendation was developed based upon an 8-hour criterion and the sawmill operates 10 hours a day, modications of the maximum exposure duration (Tj ) and TWA formulations are required. By substituting Tj with 10 hours and calculating Lj , the new exposure limit equals 84 dB. The maximum exposure duration (Tj ) for task j is modied to be TWA = 10 log Tj = 10 2(Lj 84)/3

where Lj represents the noise level (in dBA) of task j . The TWA is modied as shown below: D + 84. TWA = 10 log 100 Because the sawmill operates using 10-hour shifts, we use these modied formulations to calculate the noise exposure in the sawmill plant. 3. Methodology

Selecting a proper rotation interval length is an important consideration in implementing a job rotation plan. The proposed methodology investigates worker schedules based upon different rotation interval lengths to determine the length of time that is proper for rotating workers. A desirable rotation interval is one that minimizes the potential for worker injury and can be implemented realistically (e.g., not rotating workers every ve minutes). Before describing the integer programming model, the criteria used to determine the effectiveness of different rotation intervals is discussed. In the current research the

QUANTITATIVE METHOD

257

focus is on criteria related to worker health and well-being. The criteria that have been used to evaluate worker schedules are the Job Severity Index (JSI) values for lifting tasks and the Time-Weighted Average (TWA) for noise exposure. Using these criteria, generally, a shorter rotation interval will provide a better worker schedule. (This is true only when the length of time increases proportionally. For example worker schedules based upon 2-hour rotations are always better than ones based upon 4 hours. In contrast, schedules based upon 3 hours may be worse than ones based upon 4 hours.) However, in some settings, such as the ones that require a signicant amount of set up time, learning, or transportation between workstations, worker schedules with very short rotation interval lengths such as those rotating workers every 5 or 10 minutes may not be realistically applicable. Therefore, the trade-offs between JSI and the TWA values with rotation interval lengths must be considered. A proper rotation interval should not only be able to provide small JSI or TWA values but should also be practical. After the decision criteria were determined, the relevant data was collected. The data collection methods used in this study are described in section 3.1. Lifting task information was simulated based upon realistic settings while data for the hearing exposure was collected from an operating sawmill. Then an integer programming model was developed. The model can be applied to various settings by adjusting some of the input data and parameters, which will be explained in more detail later. Different rotation interval lengths were evaluated using this model. Each rotation length provides different potential for low back injury and different noise exposure. In the nal step a statistical analysis was performed to evaluate the effectiveness of the different rotation interval lengths. The details of the integer programming model are presented in section 3.2 and the statistical analysis is presented in section 3.3. 3.1. Data collection Task demands for the lifting tasks can be obtained from a simulation program. The simulation program is used to generate task demands multiple times and to calculate the corresponding JSI values that result from each possible worker, task, and hour combination. Then, based upon the workers schedules, the entire days JSI values can be computed by simply adding up the JSI values of each period. This research simulates 1000 replications of each worker, task and hour combination in order to provide good estimates of the corresponding JSI values. Forty test problems were taken from (Tharmmaphornphilas, 2001) including twenty four-worker and four-task problems and twenty eight-worker and eight-task problems. The rst twenty problems in table 3 are the four-worker and four-task problems and the last twenty are the eight-worker and eight-task problems. The task demand characteristics and worker proles in each problem are different (see (Tharmmaphornphilas, 2001) for details). The test problems all consider balanced rotation where the number of workers and the number of tasks in the settings are equivalent and workers can perform any task. The rst step in the occupational noise evaluation at the sawmill was to perform area noise surveys to measure noise levels in terms of intensity (sound pressure level

258

THARMMAPHORNPHILAS AND NORMAN

Table 2 Mean sound pressure levels recorded across sawmill operations. Job 8:0011:00 Head Saw Booth (Job 1) Head Saw Tailor (Job 2) Gang Saw Booth (Job 3) Two Saw Edger & Chipper (Job 4) Descrambler & Sorter (Job 5) Trim Saw (Job 6) Board Marker (Job 7) Green Chain Middle (Jobs 8 & 9) Green Chain End (Jobs 10 & 11) 77.9 98.2 81.6 100.5 96.0 94.2 91.8 89.9 87.6 Time period 11:303:30 77.0 99.5 77.9 100.2 95.9 93.0 92.5 88.3 86.3 8:003:30 77.5 98.9 80.1 100.4 96.0 93.6 92.2 89.2 87.0

SPL). To ensure the greatest data accuracy, the sound level meter was calibrated at the start of the sampling days and rechecked again at the end of each day. Noise samples were taken in the morning and afternoon at each job position, with roughly 30 minutes of data collection per job location. Data collection at each job position appeared to represent a typical workload situation. The sound pressure level readings per job position are listed in table 2. The values are in units of decibels and are morning and afternoon averages of two 2-minute SPL values. The third column shows the log average SPL over the course of the day, which represents the sound exposure of each operation. The highest A-weighted sound level recorded was for the two-saw edger and chipper (for both the morning and afternoon samples). All of the jobs except the head saw operation and gang saw operation have average SPL values greater than the 80 dBA recommended by NIOSH as the maximum noise level without hearing protection. The gang saw and head saw operators are both located in a booth, and it should be noted that the morning SPL for the gang saw operator is greater than the 80 decibel recommended limit. Each operation requires one worker except the green chain middle and the green chain end which each require two workers. To simplify the problem, we split the green chain middle (end) into two sub jobs, green chain middle (end) I and II, each of which can be performed by one worker. Therefore, a total of 11 workstations and 11 workers are included in the problem. 3.2. Integer programming models An integer programming method was used to analyze the effect of different rotation interval lengths on the workers. One integer programming model was developed to assess the potential of back injury due to lifting. The system performance measure of this model is to minimize the maximum JSI value. The other model is used to assess the hearing exposure in a sawmill plant by adjusting the system performance measure to minimizing the maximum TWA. The minmax criterion is chosen because this objective function focuses on reducing the likelihood and severity of a worker injury. Moreover,

QUANTITATIVE METHOD

259

the injury cost due to one major injury is generally high compared to two or three minor injuries. In both settings we have not sought to develop multiple day or cyclic scheduling models. In the lifting environment we have assumed that the 16 hours that workers have off between shifts represents enough time for them to recover fully. Additionally, the JSI values from different days cannot be combined in a meaningful way because the JSI measure is designed to measure the effects of tasks occurring within the same shift. In the sawmill setting we were trying to minimize the maximum daily noise exposure in order to meet the NIOSH recommendations, which are based on noise exposure during a shift. Thus, it does not make sense to try and minimize the TWA over multiple days. For the lifting tasks, different rotation interval lengths were investigated including allowing workers to be rotated at any instant in time (LP), rotating workers hourly (IP), rotating workers every 2 hours (2P), rotating workers every 4 hours (4P), and no rotation (8P). Note that the LP relaxation can be used to determine the optimal amount of time that each worker should perform each task assuming that the workers could be rotated at any point in time. However, the rotation intervals implied by the LP solution may not be realistically implementable since workers may be shifting jobs every few minutes. For the sawmill setting, different lengths of rotation intervals were applied including allowing workers to be rotated at any fraction of time (LP), rotating workers every 2.5 hours (2.5P), rotating workers every 5 hours (5P), and no rotation (10P). We now dene the variables and parameters used for the lifting task integer program and then present the model. Variable and parameter denitions I : set of workers i = 1, 2, . . . , n, J : set of jobs j = 1, 2, . . . , n, T : set of working periods t = 1, 2, . . . , m, xij t = 1 0 if worker i does job j at time t, otherwise,

cij t = the JSI value accrued from assigning worker i to task j at time t, Zmax = the maximum JSI value accrued among all workers. IP(1) formulation min Zmax subject to
n m

(1)

Zmax
j =1 t =1 n

cij t xij t

i, j, t,

(2)

xij t = 1
i=1

(3)

260
n

THARMMAPHORNPHILAS AND NORMAN

xij t = 1 i, t, xij t (0, 1)


j =1

(4) (5)

i, j, t.

The model aims to minimize the maximum JSI value. The simulation program generates the cij t parameters, which are the JSI values accrued from assigning worker i to task j at time t. The number of workers and number of tasks (n) is 4 for the rst 20 problems and 8 for the last 20 problems. The number of rotating periods (m) depends on the rotation interval. It equals 8 for rotating workers every hour, 4 for rotating workers every 2 hours, 2 for rotating workers every 4 hours and 1 for no rotation. Two constraint sets are included in the model. The rst set, constraint (2), guarantees that Zmax is the maximum JSI value among all of the workers. The second set consists of the assignment constraints (3) and (4). Constraint (3) ensures that only one worker is assigned to task j at time t. Constraint (4) ensures that only one task is performed by worker i at time t. The mathematical model was solved using CPLEX and the experimental results are shown in table 3. A similar integer programming model was developed and applied to minimize the maximum daily noise exposure encountered among the sawmill workers. The average noise exposure level for each job was based upon the noise sampling data collected within the sawmill. NIOSH noise regulations concerning safe duration of noise exposure were used to nd the solution. The integer programming model and variable and parameter denitions are shown below. Since this model is similar to the lifting model, many of the same variables can be used. Here, we present only the additional variables that are used only in this model. Variable and parameter denitions Dmax = the maximum daily noise dose among all workers, Tj = the maximum allowable exposure duration for job j given its average SPL. IP(2) formulation min Dmax subject to
n m

(6)

Dmax
j =1 t =1 n

2.5 xij t Tj

i, j, t, i, t, i, j, t.

(7) (8) (9) (10)

xij t = 1
i=1 n

xij t = 1 xij t (0, 1)


j =1

QUANTITATIVE METHOD

261

Table 3 Comparison of the maximum JSI values for different rotation interval lengths. Problems Problem 1 Problem 2 Problem 3 Problem 4 Problem 5 Problem 6 Problem 7 Problem 8 Problem 9 Problem 10 Problem 11 Problem 12 Problem 13 Problem 14 Problem 15 Problem 16 Problem 17 Problem 18 Problem 19 Problem 20 Problem 21 Problem 22 Problem 23 Problem 24 Problem 25 Problem 26 Problem 27 Problem 28 Problem 29 Problem 30 Problem 31 Problem 32 Problem 33 Problem 34 Problem 35 Problem 36 Problem 37 Problem 38 Problem 39 Problem 40 Average Std. Dev. Min Max LP 1.61 1.79 1.95 2.42 1.69 1.87 2.03 2.50 1.62 1.80 1.96 2.43 1.38 1.55 1.69 2.05 1.56 1.72 1.87 2.28 0.90 1.00 1.09 1.34 1.34 1.48 1.60 1.92 1.36 1.48 1.60 1.91 1.51 1.67 1.81 2.17 1.53 1.67 1.80 2.15 1.73 0.36 0.90 2.50 1P 1.63 1.82 2.00 2.47 1.75 1.89 2.04 2.52 1.63 1.82 1.99 2.46 1.42 1.64 1.75 2.11 1.59 1.73 1.88 2.30 0.91 1.01 1.10 1.35 1.35 1.49 1.62 1.93 1.38 1.50 1.61 1.93 1.53 1.68 1.82 2.19 1.56 1.70 1.82 2.17 1.75 0.37 0.91 2.52 2P 1.67 1.88 2.12 2.55 1.79 1.98 2.11 2.60 1.71 1.90 2.05 2.54 1.42 1.65 1.83 2.13 1.60 1.77 1.93 2.32 0.92 1.04 1.13 1.38 1.39 1.53 1.65 1.97 1.39 1.52 1.63 1.95 1.55 1.71 1.85 2.21 1.58 1.71 1.84 2.21 1.79 0.38 0.92 2.60 4P 1.78 1.97 2.14 2.73 2.03 2.19 2.27 2.67 1.71 2.00 2.23 2.71 1.71 1.89 1.99 2.42 1.61 1.79 1.95 2.46 1.06 1.21 1.32 1.62 1.45 1.66 1.83 2.10 1.50 1.62 1.73 2.04 1.65 1.85 1.99 2.36 1.69 1.84 1.92 2.25 1.92 0.38 1.06 2.73 8P 1.94 2.13 2.42 3.24 2.15 2.35 2.65 3.48 1.88 2.13 2.42 3.07 1.71 1.94 1.99 2.55 2.03 2.19 2.27 2.77 1.42 1.68 1.91 2.31 1.72 1.94 1.99 2.31 1.83 1.89 1.95 2.34 1.91 2.02 2.17 2.56 1.94 2.00 2.10 2.76 2.20 0.43 1.42 3.48

262

THARMMAPHORNPHILAS AND NORMAN

Table 4 Comparison of the maximum daily noise dose for different rotation interval lengths. Measure Dose dBA LP 10.7 94.3 2.5 h 11.6 94.6 5h 22.0 97.4 No rotation 43.9 100.4

The objective function (6) minimizes the maximum daily noise dose among all workers. Two constraint sets are included in this formulation. The rst set (constraint (7)) guarantees that Dmax is the maximum daily noise dose among all workers. The coefcient in this constraint depends on the length of the rotation interval. For example, if the rotation interval is 2.5 hours the coefcient will be 2.5 (as shown in (7)), it is 5.0 when workers are rotated every 5 hours and 10.0 when there is no rotation. For the case where workers can be rotated at any point in time, the LP relaxation is used as in the case of the lifting tasks. The number of rotation periods, m, varies based upon the rotation frequency. For example, if workers are rotated at the end of each 2.5-hour time period, the number of rotated periods (m) will be 4. However, if workers are rotated at the end of each 5-hour time period, m will be 2. For the LP relaxation only one period of length 10 hours is considered and then the xij values represent the percentage of the shift that worker i does task j . The number of workers included in the model depends on the number of jobs that the workers can rotate through. In the experiments presented here, 11-job rotation was performed and thus, n equals 11. The maximum allowable exposure duration for task j, (Tj ) is calculated using the NIOSH criterion. The second set of constraints ((8) and (9)) includes assignment constraints. Constraint (8) ensures that each task is done by only one worker during each time period, while constraint (9) ensures that each worker only performs one task during each time period. After the mathematical model was formulated, CPLEX was used to solve the problem. The experimental results are shown in table 4. 3.3. Statistical analysis to nd a proper rotation frequency Data analysis was done to evaluate the results of using different rotation interval lengths. Since multiple levels of rotation interval lengths are considered, a randomized complete block design (Montgomery, 1997) was used to determine the difference among the mean JSI values and then nd a proper rotation interval. Here, we use a randomized complete block design to remove the variation between test problems and to determine whether or not different rotation interval lengths result in different maximum JSI values. The hypothesis test used in this analysis is H0 : Mean JSI values resulting from different rotation interval lengths are the same. H1 : At least one mean JSI value is different from the others.

QUANTITATIVE METHOD

263

Table 5 An analysis of variance for the rotation interval effect. Source Rotation interval Block Error Total SS 6.086 27.974 1.079 35.138 Df 4 39 156 199 MS 1.521 0.717 0.006 F 220 Sig. 0.000

Table 6 A pairwise comparison for the treatment means. Method Tukey Pairs LP1P LP2P 1P2P 1P4P 2P4P 4P8P Results of comparison No difference (0.679) Difference (0.005) No difference (0.200) Difference (0.000) Difference (0.000) Difference (0.000)

Using the statistical software SPSS, it was found that the mean JSI values from each rotation interval length were statistically different with a p-value of 0.000. Table 5 provides the results of the completed analysis. The analysis of variance shows that the mean JSI values are not the same. However, it does not specify which treatment levels bring about the difference. Therefore, a comparison of each pair of treatment means was also performed. The Tukey method (see (Montgomery, 1997) and (Neter, Wasserman, and Kutner, 1990)) is chosen as a multiple comparison technique. Using SPSS, the results of the multiple mean comparisons can be summarized as in table 6. The test shows no difference in the number of lost days resulting from rotating workers hourly and allowing workers to be rotated at any fraction of time. Therefore, there is no benet to rotating workers more frequently than every 1 hour. The test also shows no statistical difference between rotating workers every 1 hour versus every 2 hours. Thus, it can be claimed that rotation every 2 hours is an acceptable rotation interval and can be applied if an hourly rotation cannot be used practically. The Tukey tests indicate that there is a statistically signicant difference between rotating workers every 4 or 8 hours versus the shorter rotation intervals. To help clarify the signicance of reducing the maximum JSI value for the set of workers, note that one way to quantify the effects of low back injuries is to examine the number of lost worker days resulting from lifting exertion. Therefore, a function relating the number of lost working days to workers JSI values was developed (see (Tharmmaphornphilas, 2001) and (Tharmmaphornphilas and Norman, 2002)). Recall that table 3 shows that on average the JSI value reduces from 2.20 (no rotation) to 1.92 (allowing rotation every 4 hours) to 1.79 (allowing rotation every 2 hours) to 1.75 (allowing hourly rotation) to 1.73 (allowing rotation continuously). This is equivalent to

264

THARMMAPHORNPHILAS AND NORMAN

reducing the number of worker lost days per hundred full-time equivalent worker years from 62.97 (no rotation) to 57.51 (allowing rotation every 4 hours) to 54.97 (allowing rotation every 2 hours) to 54.19 (allowing hourly rotation) to 53.80 (allowing rotation continuously). From an occupational safety perspective, the reductions from no rotation to rotation every 4 hours and then down to 2 hours are signicant. Thus, based on the statistical analysis one could make a clear recommendation that a maximum rotation interval of 2 hours should be used. However, a one-hour interval may be slightly preferable if it does not represent operational difculties. For the sawmill setting there is only one problem instance. The noise dose rises signicantly if workers are rotated only once per day (every 5 hours) or are not rotated at all. Similarly, the dBA values rise signicantly for rotation intervals of 5 hours or longer. Note that while the dBA reductions may seem small, they are actually quite signicant because depending on the criterion being used reductions of 35 dBA permit a worker to double the time that they can spend in the sawmill. Thus, the 2.8 dBA difference between the 2.5 and 5-hour rotation plans is practically quite important. These results show the clear benets of using job rotation to reduce potential hearing loss in this setting. The results in table 4 indicate that at most a rotation interval of 2.5 hours should be used. The LP dosage result is 7.8% less than the 2.5-hour rotation interval result but it would create signicant operational difculties to have the workers rotating at intervals as small as only a few minutes. The test results also indicate that the variation in noise exposure among the workers is reduced if more frequent rotation is used. For example, there is no difference in the worker exposure if the workers are rotated continuously (LP). Every worker has an exposure dosage of 10.7, which is equivalent to 94.3 dBA. If the workers are rotated every 2.5 hours, the maximum exposure dosage is 11.6, which is equivalent to 94.6 dBA, while the minimum exposure dosage is 7.9, which is equivalent to 93.0 dBA. In this case, the sound pressure level difference is 1.6 dBA. For rotating workers every 5 hours, the maximum exposure dosage is 22.0, which is equivalent to 97.4 dBA, while the minimum exposure dosage is 0.4, which is equivalent to 80.0 dBA. The sound pressure level difference is 17.4 dBA. For no rotation, the maximum exposure dosage is 43.9, which is equivalent to 100.4 dBA, while the minimum exposure dosage is 0.2, which is equivalent to 77.5 dBA. The sound pressure level difference is 22.9 dBA. Thus, in addition to reducing the maximum exposure, more frequent rotation also reduces the magnitude of the exposure differences among the workers. 4. Discussion and conclusion

This paper presents an integer programming approach for developing job rotation schedules and a method for analyzing different job rotation interval lengths in order to determine what the proper job interval is for different workplace settings in order to reduce the potential for worker injuries. A decision maker could utilize our injury risk information related to different rotation intervals in their staff scheduling decision process. The method has been applied to two specic work settings but could be readily adapted to

QUANTITATIVE METHOD

265

other work settings with different evaluation criteria. For the two work settings tested, the results show that rotation intervals of approximately 2 hours accrue nearly all of the possible benets that can be obtained from job rotation. Two hours also represents a job rotation interval that is very practical from the standpoint of being able to implement it in real settings. This approach can be used to determine the proper rotation interval length in other settings once data related to the job characteristics, such as lifting intensity or noise level in the cases examined, has been gathered. The authors have also administered a survey to gather information concerning job rotation and its use in the workplace to reduce occupational injuries. One question in the survey was developed with the purpose to gather information on rotation interval lengths that have been seen implemented in real settings. The survey responses indicate that the job rotation interval lengths that have been implemented vary based upon the types of jobs involved. The survey results indicate that rotating workers every two hours is used often in real settings which agrees with the ndings for the two settings evaluated in this research. However, more settings would have to be evaluated before one could make the conclusion that in general a two hour rotation interval is a good choice. There are several possible extensions to this research. The rst is to explicitly model productivity in the math programming model. This has the benet of creating a more general model but is likely to create a multi-criteria problem where ergonomic and productivity factors must be balanced in any nal solution. Second, we will consider applying our methodology to settings that have an unbalanced workload (some or all workers do more than one task). Third, to consider the problem of scheduling workers over multiple days and creating cyclic schedules. This would be important for ergonomic considerations related to cumulative trauma disorders and for productivity measures related to cumulative production over multiple shifts. Fourth, we will conduct more case study analysis by working with companies to examine proper rotation intervals for specic job rotation implementations.

References
Ayoub, M.M., N.J. Bethea, S. Deivanayagam, S.S. Asfour, G.M. Bakken, D.H. Liles, A. Mital and M. Sherif. (1978). Determination and Modeling of Lifting Capacity. National Institute for Occupational Safety and Health, Grant No. 5RO1OH00945-02, 1978. Ayoub, M.M., A. Mital, G.M. Bakken, S.S. Asfour and N.J. Bethea. (1980). Development of Strength and Capacity Norms for Manual Materials Handling: The State of the Art. Human Factors 22, 271283. Carnahan, B.J., M.S. Redfern and B.A. Norman. (2000). Designing Safe Job Rotation Schedules Using Optimization and Heuristic Search. Ergonomics 43(4), 543560. Henderson, C.J. (1992). Ergonomic Job Rotation in Poultry Processing. Advances in Industrial Ergonomics and Safety 4, 443450. Hinnen, U., T. Laubli, U. Guggenbuhl and H. Krueger. (1992). Design of Check-Out Systems Including Laser Scanners for Sitting Work Posture. Scandinavian Journal of Work, Environment and Health 18, 186194. Jonsson, B. (1988). Electromyographic Studies of Job Rotation. Scandinavian Journal of Work, Environment and Health 14, Suppl. 1, 108109.

266

THARMMAPHORNPHILAS AND NORMAN

Liles, D. and S. Deivanayagam. (1984). A Job Severity Index for the Evaluation an Control of Lifting Injury. Human Factors 26(6), 683693. Montgomery, D.C. (1997). Design and Analysis od Experiments, 4th ed., New York: Wiley. Neter, J., W. Wasserman, and M.H. Kutner. (1990). Applied Linear Statistical Models, 3rd ed., Irwin, IL. NIOSH. (1998). Criteria for a Recommended Standard: Occupational Noise Exposure. DHHS (NIOSH) Publication No. 98-126, Department of Health and Human Services, National Institute of Occupational Safety and Health, Washington, DC. OSHA. (1983). CPL 2-2.35A CFR 1910.95(b)(l) Guidelines for Noise Enforcement: Appendix A. OSHA Directive No. CPL 2-2.35A (December 19, 1983), U.S. Department of Labor, Occupational Safety and Health Administration, Washington, DC. Paul, P., F.M. Kuijer, B. Visser and H.C.G. Kemper. (1999). Job Rotation as a Factor in Reducing Physical Workload at a Refuse Collecting Department. Ergonomics 42(9), 11671178. Tharmmaphornphilas, W. (2001). A Robust Job Rotation Schedule to Minimize Worker Injuries. Ph.D. Dissertation, University of Pittsburgh. Tharmmaphornphilas, W. and B.A. Norman. (2002). Robust Job Rotation Methodologies to Reduce Worker Injuries. Technical Report 02-03, Department of Industrial Engineering, University of Pittsburgh. Tharmmaphornphilas, W., B. Green, B.J. Carnahan and B.A. Norman. (2003). Applying Mathematical Modeling to Create Job Rotation Schedules for Minimizing Occupational Noise Exposure. AIHA Journal 64, 401405.

Anda mungkin juga menyukai