1、UST-IN-TI (JI) PRODOCY$OM PROCESS UNRELIAEILIY Menberu Lulu University of South Florida ABSTRACT. The effects of process unreliab,l,ty on the performance of a simula- tion mode of a Just-in-Time Production and Manufacturing System (J!TMPS) is in- vestigated. In a multi-component-fabrication and prod
2、uct-assembly environment, a relatively lowe ee of fabrication process unreliability results in a pro- nouced level of lowered sye utilization. Work-in-Process inventory-induced gains in po.dtion rates in he: JIPS are tas simi_a a maee- .is-ioed i rdit/uDal prodcti systems that esem-eraeder “as INTRO
3、DUCTION The oerational objective of the Just-in-Time Maci a Prlction System 3TP ! s to prdc the kd of units neeed, at the tie needed i.e. Hu:st-in-Time (JIT), and in the quantities needed. This production system is d%ep Kooted i Western concepts of quality improvement technique ran and Denning - USA
4、) and in systems-based (Group TechnoIogy) rato.alzation of batch prodction-cellular manufacturing (most notably: Kaling - Sweden, Mirofanov - USS and, Oitz - Germany). he ITMPS employs the “PuIl System“ of prodctiom contrI that as i- %egzated mporta functions of the production control system with th
5、at of the manufacturing system. Quality control, inventory control and short-%r produc- tion control have traditionally been staff functio,s. FK the JTMP$, these are new line functions. The pnedt workers ob is enlarged to include respon- sbil%ty for roduct quality assurance. The line foremans ob s e
6、mIared to include invetor cmntrol and short-term production plning a cerol. The JITMPS management approach and the cellular structure of ts manufacturing sys- tems haue generated new forms of worker organization, such as “Quality Circles“ that have led tD inmvations in product design, machine design
7、 e qsality im- provements 5. AnnualSimulation Symposinm/ 23T 238 LULU In the “Pull System“ of production control, the final assembly goes to the preceding process to obtain the necessary parts, at the necessary time, for a specific product assembly. This signals the preceding process to produce (i.e
8、., to replace) the parts withdrawn by the following process. For the production of these parts, the preceding process obtains the necessary parts from the process further preceding it. The Kanban system is a manual (nonautomated) implementa- tion of the “pul ! system“ of production control. A Kanban
9、 is a card usually put in a rectangular vinyl sack and attached to a part container (cart). The two primary kinds of Kanbans are the “Conveyance Kanban“ that is utilized by a succeeding process to withdraw parts from a preced- ing process, and the “Production Kanban“ that is utilized to order produc
10、tion of the portion withdrawn by the succeeding process. See references 3,5,7 for detailed explanation of the Kanban system. Using Kanbans between all the processes links the activities of the final process to the remaining preceding processes. This chain-linking is the mechanism for synchronizing J
11、IT production. Kimura and Terada 2 utilized simulation to show that demand fluctuation at the final production stage does not induce amplification in production and inven- tory levels in preceding production stages. The level of fluctuation in each of the preceding n-i production stages equals that
12、of the last stage. Kim i analyzed an operating policy for JIT production called Periodic Pull System (PPS). The essence of PPS is to review the status of material flow at all production stages at regular intervals. Consequently, only the exact amount of material that has been consumed at a succeedin
13、g stage since the last review time is allowed to be Withdrawn. PPS reduces production lead time by replacing manual processing of information (Kanbans) with computer information processing. RESEARCH OBJECTIVE AND METHODOLOGY PROBLEM DESCRIPTION In a Kanban-linked JITMPS, when a breakdown occurs at a
14、 preceding cell (process), the conveyance of work parts to the input stock point of a succeeding cell is delayed due to lack of work parts. As a consequence, the succeeding cell is starved and production halts. Furthermore, all the downstream processes eventually halt due to subsequent starving. Sim
15、ilarly, a breakdown at a succeed- ing stage delays movement of conveyance Kanbans to the output stock point of a preceding cell. This, in turn, delays production Kanban movement in the preced- ing cell. The net result is that the preceding cell is prevented from further Annual Simulation Symposium J
16、UST-IN-TIME PRODUCTION AND PROCESS UNRELIABILITY 239 production due to blocking by the succeeding cell. The eventual absence of conveyance Kanban movements between the upstream processes will induce a cor- responding absence of production Kanban movements. Hence, all the upstream processes will be p
17、revented from further production. Western production managers have a policy of inserting buffer stocks between production stages to protect against production irregularity that is caused by process unreliability 6. While their approach guarantees the maintenance of higher production rates, it nevert
18、heless results in higher WIP inventory. Instead of adding buffer stocks at points of irregularity, JIT production managers deliberately expose the work force to the consequences of production irregularity. Each time workers correct the causes of a recent irregularity, the managers remove more invent
19、ory. The workers are never allowed to settle into a comfortable pattern of the status guo; or rather, the pattern becomes one of continually perfecting the production process. The thesis for such an approach is that WIP inventory hides problems which can cause production irregularity. This paper foc
20、uses on investigating the impact of process unreliability of a single process (production stage) on the JITMPS model performance. System utilization and average product cycle time are used as measures of system performance. SIMULATION MODEL DESCRIPTION Structure The structure of the hypothetical JIT
21、MPS model (see Figure i) constitutes a microcosm of a manufacturing system. It consists of two fabriction cells and an assembly cell. Within each fabrication cell there is one Input Stock Point (ISP) and one Output Stock Point (OSP). The ISP holds part inventories that are wait- ing to be processed
22、in the cell. The OSP holds processed part inventories that are waiting to be conveyed to the succeeding cell. Production Kanbans (PKs) control and facilitate the movement of part inventories from an ISP to its cor- responding process. Conveyance Kanbans (CKs) control and facilitate the movement of p
23、art inventories from the OSP of a preceding cell to the ISP of a succeeding cell. The CKs act as conveyors that link all the cells of the JITMPS. The model representation of the fabrication processes is homomorphic; i.e., each process may constitute more than one machine. On the other hand, the mode
24、ling framework for the assembly cell is isomorphic; i.e., the three-stage (Si, $2, $3) assembly cell has a one-to-one correspondence with the hypothetical JITMPS. Note that the assembly stages are directly linked. Annual Simulation Symposium 240 LULU RMSP ISPI OSPI fiP2 OSP2 FABRICATION CELL FABRICA
25、TION CELL . AA PRODUCT PRODUCTION, SCHEDULE : i , . -L- : I i ISP I .I t AS SEBLY CELL -I I sl, $2, $3 ASSEMBLY ST/LCES -4h. CONVEYANCE KANBANS : PRODUCTION KANBANS _ . WORKPIECE FLOW SUBASSEMBLY FLOW ItMSP RAW MATERIAL STOCK POINT ISP IIPUT ZTQCK POINT OS9 OUTP.U STOCK OINT Figure i. Schematic of t
26、he Simulation Model The system produces one product that is assembled from three different component types, identified as A, B and C, that are fabricated in Cell i and Cell 2. A product is assembled from one of Component Type A, two of Component Type :B, and four of Component Type C. A is added at S
27、I is added at $2, and C is added % at S3. Operation As is the case with “pull system“ production control, the only production facility that is scheduled is the last assembly state (S3). Once a production schedule is received at S3, the Kanhas system caxies out the production control unctiom for the
28、entire system. For an assembly operation to commence at any stage, the following two condi- tions must be met. I) A finished subassembly must be at a preceding assembly stage. 2) Parts must be available, in assembly requirement quantity ready to be added on to a subassembly from a precedi,g stage. W
29、hen these two conditions are met, assembly operation commences at a succeedLng stage. Since a product requires the three part types that are fabricated sequen- tially, aa assembly rate of one unit of product per unit time mnst correspond i.e., 1 AU E iA + 2B + CE .one unit of product. In the Kanban
30、system of production control, upstream inventory is pulled through the manufacturing processes by downstream inventory. To be operational, a Kanban-linked JITMPS requires a system-wide allocation of a minimum level WIP inventory. In a single product line, part inventories corresponding to a single u
31、nit of product constitute the minimum WIP inventory level. The Kanban system will not function without WIP inventory at each stock point. This minimum inven- tory level is an element of the Kanban-Liked UITMPS structure. Hence, with regard te the hypothetical model, the following statement is approp
32、riate. A Kanhan-linked JITMPS with component inventories corresponding to one (I) unit of product at each stock point is a zero-inventory Dr minimum inventory production system. In the hypothetical model, zero inventory corresponds to the allocation of three carts each for component Types A, B and C
33、, with a capacity of part inven- tories of one unit of A, two units of B and four units of C, respectively, at each inventory stock point. “Zero inventory“ is synonymous with “lotless production“ 5. Specification of Model arametes Failure probability parameters for Process 1 and assembly tages Si, S
34、2 and 3 (Figure i) are set to zero. For Process 2, failure probabilities (p) of .i, 0.2 and 0.3 are consiered he c:ell downtime (Td) associated with each failure is 36 minutes (one product cycle time). It is assumed dbat roces fa:lure occurs only once .duling the fabrication of a component. Specific
35、ally, Process :cycle time = 36 min. per assembly unit T d = 36 min. per failure, p = (0.1, 0.2, 0.3) WIP inventory := (0, 5, 10, 15, 10, 15, 30) AU Random failures of Process 2 render the JITMPS model tochskc. A ttal of 27 model configurations representing specific combinations of levels of process
36、reliability and WIP inventory were used in the simulation experiment. Each model conEigurtio was run 500Z simulated, minutes n replicated time adh replication yielded an average of 72 Observations. Thus the experimental design constitutes a two-factor factorial design. The WIP inventory factor is at
37、 seven Annual Simulation Sympos/mm 242 LULU levels (0, 5, 10, 15, 20, 25, 30 AOs) and the reliability factor is at three levels (0.9, 0.8, 0.7) per 36-minute cycle. Performance measures used are: a) Cell 1 utilization (pl), b) Cell 3 utilization (p3), c) Product cycle time (Tc). The statistical anal
38、yses are based on the analysis of variance (fixed ef- fects model - two-way classification). Model adequacy checks (normal probability plots and homoscedasticity tests) showed that the residuals are normally dis- tributed with mean zero and common variance. EXPERIMENTAL RESULTS AND ANALYSES EFFECT O
39、F WIP INVENTORY LEVEL ON AVERAGE PRODUCT CYCLE TIME Table 1 reveals that, for a given level of reliability, T c is a nonincreas- ing function of WIP inventory. Consider the curve for R = 0.8 (see Figure 2). When the WIP inventory level is lowered to 15 AU, T c jumps from 72.65 to 78.95 minutes. For
40、WIP inventory levels beyond 15 AU (20, 25, 30), the value of T c remains unchanged at 72.65 minutes. The jumps in T c that are associated with the lowering of WIP inventory can be equated to the exposing of process reliability problems that are effectively hidden by WIP inventory. Table 1 Product Cy
41、cle Time in Minutes Reliability 0 R = 1.0 36 R = 0.9 60.94 R = 0.8 84.70 Work-In-Process Inventory 5 36 57.14 78.95 i0 36 57.14 78.95 15 36 57.14 72.65 20 25 30 36 36 36 57.14 57.14 57.14 72.65 72.65 72.65 87.48 87.48 87.48 R = 0.7 I05.59 I00.i 93.46 93.46 The large increase in T c from 36 minutes t
42、o 84.70 minutes as R is lowered from 1.0 to 0.8 may surprise the reader. However, as discussed below, an ex- amination of the component requirement for assembly in conjunction with the stochastic processes that result from failure probability of 0.2 for Process 2, justified the corresponding large i
43、ncrease in product cycle time, T c. Annual Simulation Symposium JUST-IN-TIME PRODUCTION AND PROCESS UNRELIABILITY 243 % E o Figure 105.59 I00.01 93.46 87.48 85.56 84.70 78.95 72.65 60.94 -. 57.14 36.0 O. 0- 0 2 . I R = 0.7 Rffi0.8 L R = 0.9 Ideal T c R= 1.0 WIP Inventory in Assembly Units Effect of
44、Process Reliability and WIP Inventory on Product Cycle Time Annual Simulation Symposium ,/ 244 LULU To assemble one-.unit of “product eveny, 366 mutes, the preced-ing cells must fabricate seven. (7,) eompDnents: - 1 un.t off Chmporent A., 2 units of Cbmpenent B and 4 units of Component C, in:.exacly
45、 36 minutes .If Rroces 2_ 9aiil, s once during component fabricain;, the acua fliation cyclie time is lenthe%ne ,13,y. 36 m/ilnus (Td)., Two, tree, etc., process f_ai,luuduring .ro-d,uc.crin, o-these seven components would .result in a coTre.sponding lerrghening of actual -fabrica- tion cycle tmes b
46、y 76, I12, e tc%. minutTes., Basedi on the assumption of one process failu.redhring the fabrication off a component, the :pr.obab:li1y of .eaactly x process failures and (7-x) process nenfailures during :fabribsti)n of the- sev.en components can be caculaed, (m.dl:e), b Z rising .binomial distribut.
47、ibn, i.e., P ,(off e.xaCtly x machine failues) Table 2_ give, s ne ,theoz.,et-kca.,l: isn$u.cion a:s:s_ocia.td: .wit the, occurrences of pzocess failures and corresponding acta-l“ fbr!ietion cycle times in Cell 2. Table Eail.uKe Distribution and Corresponding X P(N = x) T F 0 I 2 3 -4 !5 -6 _ I 36 1
48、 76 II 148,! 18 220 256 I, 292 The resulting avemage fabriica%iowcye%/e-.time, l.s.:. ET F = 88.84. Note the closeness: of the 88.84-minu,d theoretical fabr:iation cycle time and the 84!.30minute .empirilcal produc._ cycle ,t;/Tme (:see Table _Z, R = 0.8, WIP = 0)., Besies_ veriiyi! he seemingl, , nIdinace i.n,crease, thi-s er:.ati.on .(of average fahr;ication ,cycle t:ime .fox Pzseess ) i:likst.raes a, ey important Kanban system function. Cell 1 (fabrication cell) and Cell 3 (assembly cell), b
