Scheduling Jobs Considering Physiological Factors

Transcription

1 Scheduling Jobs Considering Physiological Factors 1.1 Background In the classical scheduling problem, the optimal sequence o jobs on a single machine is equivalent to the optimal sequence o jobs done by a single worker. But this is not the case in the real situation because the algorithm ignores constraints related to a human s physiological actors and limitations. Workers get tired both physically and mentally while they are doing their job. So, this situation cause reduced perormance and productivity o the workers. Fatigue is one important reason or decreasing workers perormance. It can be caused by extended working hours, inadequate resting periods and unsuitable working conditions. According to Fatigue Management System Guidelines (FMSG), atigued workers ability to perorm their task, maybe lost or impaired. Workers who get atigued can have Reduced motivation Decreased speed o task Increase in memory errors Inability to concentrate Incorrect action Konz (1998) has declared that atigue increases exponentially with time. Thereore, it is important to get rest beore the atigue level becomes too high. To prevent atigue, workers need adequate resting periods during work periods. Breaks are designed to provide time or workers to overcome the atigue arising rom the work. Resting time can be classiied by Konz and Johnson (2004) as ormal breaks (lunch, coee), inormal break (interruptions, training) and micro breaks (short pauses o a minute). When a break is given, workers perormance is expected to increase because o recovery. The recovery value is dependent with how atigued the worker is when the rest begins and the length o the resting time. I the length o the break is small, the incremental amount o recovery is less than the incremental amount o time. For example, 4 breaks o 5 minutes are oten 1

2 more useul than 1 break o 20 minutes or two reasons; atigue will not be increased as much and the recovery will be better. In this study, we use rate- modiying activities (RMA) as a kind o resting activity or workers. RMA is irst introduced in literature by Lee and Leon (2001). They deined RMA as an activity which alters the production rate o machines. The RMA plays an important role in work-rest scheduling in the human actors literature. The main idea o work-rest scheduling is to obtain the number, place and duration o rest periods. While determining those decisions, productivity, saety and comort are considered. In literature, the objectives o the work-rest problem are to minimize atigue and to provide recovery o the worker while not reducing productivity. Boucsein and Thum (1997) ound that short breaks are more eective in promoting recovery rom both mental and emotional ways to workers. Table 1.1 shows that several researchers studied the work-rest problem or various physiological variables such as heart rate, blood pressure, oxygen uptake and electromyography (EMG) signals. They ound that resting time is important to recover rom atigue. Table 1.1 Work-Rest Problems Veltman and Gaillard (1993) Boucsein (1993) Imbeau et al. (1995) Wu and Wang (2002) Tiwari and Gite (2006) Hsie et al. (2009) Heart rate, blood pressure Heart rate variability (HRV), EMG signals, electro dermal activity Heart rate, Maximum aerobic capacity (%VO2 Max) %VO2 Max, heart rate Heart rate %VO2 Max Our model is extends the irst model which is in chapter one. The irst model is to determine the work sequence, the number o breaks and the place o each break without considering the physiological actors o the worker. In this study, we consider the human characteristics while determining the work sequence, the number o breaks and the optimal break schedule. 2

3 This study diers rom existing research in several ways. First, we deine rate- modiying activities as resting period o workers; whereas all other studies use rate-modiying activities as machine maintenance and repair time. Second, there is no research that considered the workers physiological actors when scheduling the jobs. Third, in our study, deteriorated jobs are caused by atigue o the worker, but in literature, jobs deteriorate while waiting to process. To the best o our knowledge, there is no research that proposed a task sequencing approach which combines with deteriorating jobs, RMA and considers the human characteristics. In the next section, we discussed the new model. In section 1.3, we present an extended mathematical model. Algorithms are in section 1.4. Special cases are presented in Section 1.5. We oer conclusions in section Discussion o New Model This chapter is motivated by the problem o manual order picking activities in warehousing systems. The problem is to determine the sequence in which the orders should be picked in minimum time and within acceptable levels o human physiological actors. In addition, breaks can be scheduled to allow workers to recover in the middle o the order picking sequence. We consider sequencing n orders (jobs) on a single worker (processor) with varying processing speed due to the eects o various ergonomic actors. We assume that a break or maintenance activity can be scheduled to improve the processor state any time ater the irst task is scheduled. In our irst mathematical model, we determined the optimal job sequence with the optimal number o breaks and the place o each break. But that model lacks considering human s physiological actors or the worker. Thereore, we added new constraints which relate to a human s physiological actors to calculate the optimal job schedule or a single worker. For this new model, we can include a physiological actor which aects the worker s perormance. Also, the usage o jobs and recovery rate aects the solution when compared to the result o the irst model. The new model gives more total completion 3

4 time or makespan value than the irst model because o the changing recovery rate. In the new model, the worker doesn t recover completely. The job, which is already deteriorated, deteriorates again even i there is an RMA beore that job. Thus, this increases the realized processing times o jobs and then total completion time. Additionally, this also aects the number o RMA to be given by the new model. 1.3 Extended Model The new mathematical model decides the sequence in which jobs should be scheduled, how many breaks to use to avoid atigue, i any, and where to use them in the schedule. Those decisions are aected by human actors such as heart rate, oxygen consumption, and blood pressure. The main dierence o this model rom the irst one is that the worker doesn t recover completely, which means there is not 100% recovery o his/her lost energy and other actors, ater a break (RMA). Hence, in this model we assume that every job consumes worker s physiological capacity in some level. Hence, the recovery rate changes with the position o job when it is done. Also the types o jobs are aected the recovery rate o the worker. Heavy jobs need more eort than light jobs. Assumptions Breaks should be taken during work shits. The worker doesn t recover completely ater breaks. Consider at least one physiological actor. Worker can not process two or more jobs simultaneously. Parameters r Recovery rate o physiological actor, a j Usage rate o physiological actor by job j, s i Change in cumulative physiological actor caused by the i th RMA, R A lower (upper) limit on physiological actor R 4

5 Decision Variables R i Cumulative value o physiological actor at position i, Additional Constraints n r a j R 1 p j1x1 j0 (1.9) R i j1 R i, i1 p jk xi, j, ik s i yi i 2,..., n k1 j1 R R i R n r a j 1,..., F (1.10) i 1,..., n 1,..., F (1.11) (1.9) Constraints compute the cumulative level o physiological actor o the worker at the end o position 1. (1.10) Constraints compute the cumulative level o physiological actor o the worker at the end o position i. (1.11) The level o cumulative physiological actor is always within prescribed acceptable limits. Perormance o the Extended Model A computational experiment is conducted to test the perormance o the new model. To conduct our analyses on the proposed models, we identiy our experimental actors: deterioration rate ( ), RMA time (q), mean o processing time (M) and variance on mean o processing time (V). Also, we speciy three levels or each actor. So, this is called 3 4 -experimental design with six two-actor, our three- actor interactions and one our-actor interaction. The deined levels o those actors in the experiments are given below. 1) : 0.02, 0.04 and ) q : 5, 10 and 15 3) M : 20,40 and 80 4) V : 0.20, 0.40 and

6 Using the variance on mean o processing time, we produced an interval or mean o processing time o which it is assumed to be uniormly distributed. These are [18, 22], [10, 30], [1, 40], [36-44], [20-60], [1-80], [72-88], [40-120], [1-160]. As we mentioned beore, the processing time o a job in dierent positions is calculated i1 using p ji 1 p j. Besides, we test our models or 50 jobs and 810 instances are run or each experiment. In the extended model, we use three dierent physiological actors which are oxygen consumption ( VO 2 max ), heart rate ( HR max ) and blood pressure ( BP ). The new parameters are generated rom uniorm distribution. Table 1.2 Physiological actors Recovery rate ( r ) HR max r ~U[1-20 ] % VO 2 max r ~U[ 1-15] % BP r ~U[ 1-20] % Table 1.3 Physiological actors Usage rate ( a j ) HR max a j ~U[1-15 ] % VO 2 max a j ~U[ 1-15] % BP a j ~U[ 1-15] % Table 1.4 Physiological actors Changes based on actors ( s i ) HR max s i ~U[ 1-10] % VO 2 max s i ~U[ 1-10] % BP s i ~U[ 1-10] % 6

7 Table 1.5 Physiological actors Acceptable limits ( R R ) HR 160 bpm- 220 bpm max VO 15 % - 33% 2 max BP 75 mmhg -115 mmhg The proposed model is also coded using AMPL and solved by CPLEX 9.1 on a computer with Pentium IV 2.8 GHz processor and 1GB o RAM. Again 81 replications were run or total completion time objective unction. The average computation times are given in Table 1.6 and Table 1.7. Table 1.6 Average Run Time or Total Completion Time (sec.) Det. Rate RMA time Run Time Table 1.7 Average Run Time or Makespan (sec.) Det. Rate RMA time Run Time The eicacy o the model is based on the average run time in seconds. The average time or total completion and the makespan problem is seconds and 41.7 seconds. The run time or the problem o minimizing total completion time decreases, when deterioration rate increases. Also, it increases when RMA time goes up. These results are the same as the result or the irst model. Hence, we say the new model is consistent. When we compare the run time, the extended model is aster than previous model. Because now we have more constraint and our easible region gets smaller. Ater 90 replications, the main dierence between the irst model and the extended new model is the place o the breaks. This is what we expected because in the extended model we 7

8 consider the workers physiological actors. As a result, the extended new model relects a real industry situation. 1.4 The Heuristic Algorithm In this section, we now give a polynomial time algorithm or the problem 1 p (1 ) p, rm, C with considers to physiological actors o workers. ij j j max The general problem is known as NP hard. In order to solve it we have developed a heuristic algorithm which is given in the below. Santos and Resnick (2005) developed an equation which calculates worker s predicted atigue rate. We use atigue as the single physiological actor, other actors could be used as well. The parameters and the equation are given below; m The mass o the handled load in kg t The time into shit in minutes p The production rate in units per minute Fatigue( F) t 0.53m 0. 08p In the proposed heuristic model, we irst provide the worker s basic inormation such as gender, age, height and weight. Then, sort the jobs according to SPT rule. The job which has the largest processing time is assigned to position 1. Ater assigning the job, then calculate atigue rate o the worker which is equal to worker s physiological capacity. Fatigue can be used as an upper limit or energy expenditure during the work. According to Borg scale, determine upper limit. I this amount is bigger than acceptable limits, then do an RMA and assign the next largest job to the current position immediately ater the RMA. Otherwise assign the smallest job to the current position without doing an RMA. Stop the heuristic when all jobs are assigned all positions. The procedure is described as ollows: Step 1 Determine the basic inormation that will be used in other step. (e.g. mass o the load, production rate and time) 8

9 Step 2 Number the jobs in SPT order. Step 3 Schedule the job with largest processing time to position one and no RMA at the beginning. Step 4 For R 0, J n [ 1] k 2... n [ 1] Set C[ 1] Pn [1] b n 1 and s 1 Calculate the Fatigue Fatigue( F) t 0.53m 0. 08p Step 5 Determine the acceptable limit o worker s atigue. Step 6 i atigue is within the acceptable limits or not. I F acceptable limit ( R ) Then; R 0, s [ k ] J [ k ] Step 7 R 1 (give RMA), b [ k ] Set C[ k] C[ k 1] Pb [ m], s s 1 and m m 1 Otherwise go to step 7. J[ k ] Set C[ k ] C[ k 1] Pb [1] + q, b b 1 and m 1 I k n Go to Step 4. Otherwise go to Step 8. Step 8 I k n, Stop the algorithm, Makespan C [n ]. 9

10 1.5 Numerical Example o Heuristic In this section, we use several numerical examples to illustrate the behavior o the model. In all the examples we assume the ollowing parameters values which is seen in Table 1.8 and Table 1.9. Table 1.8 Basic inormation Mass o the hand load Time into the shits Production Rate U~[3-20] kg U~[1-10]minutes U~[1-10]unit /minute Table 1.9 Experimental actors Factors Levels Number o jobs 10, 25, 50 Processing time U~[10-100] Deterioration rate 0.04, 0.06, 0.08 RMA time 5, 10, 15 For larger problems, we compare heuristic solution with a random solution which is a solution taken randomly rom the solution sets. We use ollowing equation; e F h F F r r *100% In this equation, F is the makespan value o the heuristic model, F is the makespan h r value o the random solution which is selected randomly rom the set o heuristic solutions and e is the average perormance o the heuristic. The heuristic is replicated 10 times or each problem cases and it developed by JAVA with Pentium IV, 2.8 GHz processor and 1GB o RAM. When we test our models or 10 replications or each speciic set o conditions, we obtain the average percentage error o the heuristic model which is given in the Table

11 Table 1.10 Comparison o Heuristic and Mathematical Model or Total completion or one actor Number o jobs α RMA time Ave. Error o Heuristic (%)

12 Table1.11 Comparison o Heuristic and Mathematical Model or Makespan or one actor Number o jobs α RMA time Ave. Error o Heuristic (%) Ave:

13 3.6 Summary In this study, we studied a scheduling problem with deterioration jobs and RMA simultaneously with consider to physiological actors. Firstly, we present a mathematical model with the objective o minimizing the makespan and total completion time. Our model can decide the sequence o jobs in which jobs should be scheduled, how many RMAs to use, i any, and where to use them in the schedule. Those decisions are aected by human actors such as heart rate, oxygen consumption, and blood pressure. Secondly, we propose a heuristic model or makespan. Then, we present some computational experiments. According to the experiment results, the perormance o the proposed model and the heuristic are quite satisactory. We show that, when the number o jobs increases the computational time to solve the problem increases dramatically or the mathematical model. Besides, the heuristic model gives a reasonable error or 50 jobs when compared to the mathematical model. 13

14 Reerences Boucsein, W., (1993) Psychophysiology in the computer work-place goals and methods, Elsevier North Holland, Amsterdam, Boucsein, W, Thum, M. (1997) Design o work/rest schedule or computer work based on psycho-physiological recovery measures, International Journal o Industrial Ergonomics, 20, Christensen H., Sogaard, K., Pilegaard M. and Olsen, H. B.(2000) The importance o the work/rest pattern as a risk actor in repetitive monotonous work International Journal o Industrial Ergonomics, 25, H.C. Wu, M.J.J. Wang, (2001) Determining the maximum acceptable work duration or high-intensity work, European Journal o Applied Physiology, 85 (3 4), H.C. Wu, M.J.J. Wang, (2002) Relationship between maximum acceptable work time and physical workload, Ergonomics, 45, Hsie, M., Hsiao, Cheng, T. and Chen, H. A model used in creating a work-rest schedule or laborers, Automation in Construction, in press. Imbeau, D., Desjardins, L., Dessureault, P.C., Riel, P. and Fraser, R. (1995) Oxygen consumption during scaold assembling and disassembling work: Comparison between ield measurements and estimation rom heart rate International Journal o Industrial Ergonomics, 15, Karvonen, M.J., Kentala, E. and Mustala, O. (1957) The eects o training on heart rate: a longitudinal study, AnnMed Exper Fenn, 35(3), Lee, C.Y. and Leon, V.J. (2001) Machine scheduling with a rate-modiying activity, European Journal o Operational Research, 128, Tiwari, P.S. and Gite, L.P. (2006) Evaluation o work-rest schedules during operation o a rotary power tiller, International Journal o Industrial Ergonomics, 36,

15 S. Konz, (1998) Work/rest: Part II - The scientiic basis (knowledge base) or the guide, International Journal o Industrial Ergonomics, 22, Veltman, J.A. and Gaillard, A.W.K., (1993) Indices o mental workload in a complex task environment. Neuropsychobiology, 28, Veltman, J.A. and Gaillard, A.W.K.(1996) Physiological indices o workload in a simulated ligt task, Biological Psychology, 42,