COLLABORATIVE MANUFACTURING:CASE EXAMPLES

CASE EXAMPLES

Global markets are increasingly demanding that organizations collaborate and coordinate efforts for coping with distributed customers, operations, and suppliers. An important aspect of the collaboration

process of distributed, often remote organizations is the coordination cost. The coordination equip- ment and operating costs limit the benefit attainable from collaboration. In certain cases, this cost can render the interaction among distributed organizations nonprofitable. Previous research investi- gated a distributed manufacturing case, operating under a job-shop model with two distributed collaborating centers, one for sales and one for production. A new model incorporating the communication cost of coordination has been developed (Ceroni et al. 1999) yields the net reward of the total system, determining the profitability of the coordination. Two alternative coordination modes are examined: (1) distributed coordination by the two centers and (2) centralized coordination by a third party. The results indicate that distributed and centralized coordination modes are com- parable up to a certain limit; over this limit, distributed coordination is always preferred.

Coordination Cost in Collaboration

In a modern CIM environment, collaboration among distributed organizations has gained importance as companies try to cope with distributed customers, operations, and suppliers (Papastavrou and Nof 1992; Wei and Zhongjun 1992). The distributed environment constrains companies from attaining operational efficiency (Nof 1994). Furthermore, coordination becomes critical as operations face real- time requirements (Kelling et al. 1995). The role of coordination is demonstrated by analyzing the coordination problem of sales and production centers under a job-shop operation (Matsui 1982, 1988). Optimality of the centers’ cooperative operation and suboptimality of their noncooperative operation have been demonstrated for known demand, neglecting the coordination cost (Matsui et al. 1996). We introduce the coordination cost when the demand rate is unknown and present one model of the coordination with communication cost. The communication cost is modeled by a message-passing protocol with fixed data exchange, with cost depending on the number of negotiation iterations for reaching the system’s optimal operating conditions. The model developed here is based on the re- search in Matsui et al. (1996) and the research developed on the integration of parallel distributed production systems by the Production Robotics and Integration Software for Manufacturing Group (PRISM) at Purdue University (Ceroni 1996; Ceroni and Nof 1999).

Job-Shop Model

The job-shop model consists of two distributed centers (Figure 6). Job orders arrive at the sales center and are selected by their marginal profit (Matsui 1985). The production center processes the job orders, minimizing its operating cost (Tijms 1977).

Coordination Cost

Two basic coordination configurations are analyzed: (1) a distributed coordination model in which an optimization module at either of the two centers coordinates the optimization process (Figure 7) and (2) a centralized coordination model where a module apart from both centers optimizes all operational parameters (Figure 8).

The distributed model requires the centers to exchange data in parallel with the optimization module. The centralized model provides an independent optimization module.

Coordination cost is determined by evaluating (1) the communication overhead per data trans- mission and (2) the transmission frequency over the optimization period. This method follows the concepts for integration of parallel servers developed in Ceroni and Nof (1999). Communication overhead is evaluated based on the message-passing protocol for transmitting data from a sender to one or more receptors (Lin and Prassana 1995). The parameters of this model are exchange rate of messages from / to the communication channel (td), transmission startup time (ts), data packing / un- packing time from / to the channel (tl), and number of senders / receptors ( p).

Results

The coordination of distributed sales and production centers is modeled for investigating the benefits of different coordination modes. Results show that the coordination cost and the number of negoti- ation iterations should be considered in the decision on how to operate the system. The numerical results indicate:

1. Same break-even point for the distributed and centralized modes at 400 iterations and A 2 jobs per time unit.

2. The lowest break-even point for the centralized mode at 90 iterations and A 5 jobs per time unit.

3. Consistently better profitability for the centralized mode. This effect is explained by lower communication requirements and competitive hardware investment in the centralized mode.

4. The distributed mode with consistently better profitability than the centralized mode at a higher hardware cost. This shows that distributed coordination should be preferred at a hardware cost less than half of that required by the centralized coordination mode.

From this analysis, the limiting factor in selecting the coordination mode is given by the hardware cost, with the distributed and centralized modes becoming comparable for a lower hardware invest- ment in the centralized case. Coordination of distributed parties interacting for attaining a common goal is also demonstrated to be significant by Ceroni and Nof (1999) with the inclusion of parallel servers in the system. This model of collaborative manufacturing is discussed next.

Collaboration in Distributed Manufacturing

Manaco S.A. is a Bolivian subsidiary of Bata International, an Italian-based shoemaker company with subsidiaries in most South American countries as well as Canada and Spain. The company has several plants in Bolivia, and for this illustration the plants located in the cities of La Paz and Cochabamba (about 250 miles apart) are considered. The design process at Manaco is performed by developing prototypes of products for testing in the local market. The prototypes are developed at the La Paz plant and then the production is released to the Cochabamba plant. This case study analyzes the integration of the prototype production and the production-planning operations being performed at distributed locations by applying the distributed parallel integration evaluation model (Ceroni 1999).

The production-planning operation (operation A) consists of six tasks, with some of them being executed in parallel. The prototype development operation (operation B) consists of four tasks, all of them sequential. Table 3 and Figure 9 show the description and organization of the tasks in each operation.

Means and standard deviations for the task duration are assumed. The time units of these values are work days (8 hours).

To contrast the situations with and without integration, two alternative solutions were developed. The first solution considers the sequential processing of the operations: the prototype was developed at La Paz and then the results were sent to Cochabamba for performance of the production planning. Parallelism is included at each operation for reducing the individual execution cycles. The second solution considers the integration and inclusion of parallelism in both operations simultaneously.

Assumptions were made for generating an integration model for applying the parallelism optimiza- tion.

Integrated Optimization

The integration process in both operations is simplified by assuming relationships between tasks pertaining to different operations. A relationship denotes an association of the tasks based on the similarities observed in the utilization of information, resources, personnel, or the pursuing of similar objectives. Integration then is performed by considering the following relationships:

The task relationships allow the construction of an integrated model of the operations. This integrated model preserves the execution order of the tasks as per their local model (Figure 9). Figure 10 shows the integrated model for operations A and B.

Once the integrated schema was generated, the parallelism analysis was performed. In order to evaluate the parallelism in the system, the time of communication and congestion delays needed to be estimated. The estimation of these delays was performed using the software TIE 1.4 (Khanna and Nof 1994; Huang and Nof 1998). TIE 1.4 allows the simulation of a network of distributed tasks with an Intel Paragon Supercomputer, up to a maximum of 132 parallel processors. TIE 1.4 uses a message-passing mechanism for communicating data among the computer nodes simulating the tasks. The data transmission can take place either synchronously or asynchronously. In synchronic data transmission the activity of the sending processor is stalled while waiting for confirmation from the receiver processor. In asynchronic data transmission, the sending processor does not wait for confir- mation from receiving nodes and continue with their activity.

The simulation with TIE 1.4 is modeled based on two types of programs: a controller node and a task node. The controller assigns each of the programs to the available computer nodes and starts the execution of the first task in the execution sequence. The implementation in TIE 1.4 will require as many computer nodes as there are tasks in the operation plus the controller node. For example, operation A has 6 tasks, requiring a partition of 7 nodes for its execution on the parallel computer.

Simulation Results

Three models were simulated: operation A, operation B, and integrated operation. A total of 10 simulation runs were performed for each model, registering in each case the production (II), inter- action (T), and total (<I) times, and the degree of parallelism ('¥), which is a concurrency measurement for the system. The results obtained for each case are presented in Tables 4 to 6. The simulation of the individual operations allows us to generate an estimate of the delay times due to communication and congestion, both required to optimize the operations locally.

The results obtained from the simulation of the integrated operation were utilized for determining the parallelism of the tasks. The parallelism optimization assumes the tasks’ duration and commu- nication times as per those generated by the TIE 1.4 simulation (Table 5) and congestion time as 0.02e0.05*'¥. The results obtained are presented in Table 7 and Figures 11 and 12.

The solution generated includes the number of parallel servers for the tasks shown in Figure 12.

Local Optimization

For generating the local optimal configurations of the tasks, the PIEM model was applied to both cases with the congestion delays computed according to the expression 0.02e0.05*'¥. The results ob- tained for each operation are presented in Figures 13 and 14.

The numerical results obtained from the local and integrated scenarios are presented in Table 8.

The results in Table 8 show a slight difference in the total production time, which seems to contradict the hypothesis that the integrated scenario will benefit from a reduction of the cycle time.

However, it must be noted that the number of subtasks required by the local scenario for achieving a comparable total production time is double that required by the integrated scenario. This situation is the result of no constraint being imposed on the maximum number of subtasks for each task in each scenario (infinite division of tasks). In particular for operation B, the number of subtasks in which each task is divided is nine, which can be considered excessive given the total number of four tasks in the operation.

For evaluating the comparative performance of the integrated and local scenarios, the local scenario for each operation was chosen according to the final number of subtasks in the integrated

scenario. Therefore, the number of subtasks was set at 15 for operation A and 11 for operation B. This makes a total of 26 subtasks in both local operations, which equals the number of subtasks for the integrated scenario. The values for the performance parameters show a total production time (II + <I) of 2.7769 shifts, direct-production time (II) of 2.3750, and interaction time (<I) of 0.4019 shifts. These values represent an increment of 54% for the total production time, an increment of 104% for the production time, and a decrement of 37% for the interaction time with respect to the integrated solution.

The results obtained from this case study reinforce the hypothesis that exploiting potential benefits is feasible when optimizing the parallelism of integrated distributed operations. Key issues in PIEM are the communication and congestion modeling. The modeling of time required by tasks for data transmission relates to the problem of coordination of cooperating servers. On the other hand, the congestion modeling relates to the delays resulting from the task granularity (number of activities being executed concurrently).

Variable Production Networks

The trend for companies to focus on core competencies has forced enterprises to collaborate closely with their suppliers as well as with their customers to improve business performance (Lutz et al. 1999). The next step in the supply chain concept is the production or supply networks (Figure 15), which are characterized by intensive communication between the partners. The aim of the system is to allocate among the collaborating partners the excess in production demand that could not be faced by one of them alone. This capability provides the entire network with the necessary flexibility to respond quickly to peaks in demand for the products. A tool developed at the Institute of Production Systems at Hanover University, the FAS / net, employs basic methods of production logistics to pro- vide procedures for the efficient use of capacity redundancies in a production network. The tool satisfies the following requirements derived from the capacity subcontracting process:

• Monitoring of resource availability and order status throughout the network

• Monitoring should be internal and between partners

• Support of different network partner perspectives (supplier and producer) for data encapsulation

• Detection of the logistics bottlenecks and capacity problems

A key aspect of the system is the identification of orders the partner will not be able to produce. This is accomplished by detecting the bottlenecks through the concept of degree of demand (a comparison of the capacity needed and available in the future, expressed as the ratio between the planned input and the capacity). All the systems with potential to generate bottlenecks are identified by the FAS / net system and ranked by their degree of demand. The subcontracting of the orders can be performed by alternative criteria such as history, production costs, and throughput time.

The system relies on the confidence between partners and the availability of communication channels among them. Carefully planned agreements among the partners concerning the legal aspects, duties, responsibilities, and liability of the exchanged information are the main obstacles to imple- menting production networks.