【正文】 D projects, etc. The system39。 and(b) a new railway bridge over river West Morava.In the course of the control model39。 is difficult and costly, so periodic inspections are preferred. The actual output observed at a current inspection point is measured in percentages of the total project (target amount). When realizing a certain project in `Hidrogradnja39。 can work with one, two and three shifts, the number of possible speeds has been taken to be three for all experiments.Evaluating the project39。s good name and to pete with other panies, the projects39。 through a mountain, various power stations, etc. `Hidrogradnja39。_N the average number of inspection points (without t0)._J the average index of the speed to be introduced within one simulation run.8. Practical applicationsIn 19971998, extensive experimentations were undertaken in several industrial plants in Serbia to test the fitness of the developed costoptimization production control model. The best results were achieved in a Serbian pany`Hidrogradnja39。s actual output, can be carried out only via timely inspections at pregiven control points. At every inspection (control) point, the decisionmaker observes the amount produced and has to determine both, the proper speed and the next control point. Assume that it is prohibited to use unnecessarily high speeds (especially at the beginning of manufacturing the products), unless there is an emergency situation, . a tendency to deviate from the target which maycause delay of the pletion time. This is because lengthy work at higher speeds when utilizing restricted resources (. manpower working in two or three shifts, etc.) can prematurely wear out the system. Assume, further, that the inspection and the speedreset times are zero. The costs of all processing speeds per time unit, as well the cost of performing a single inspection at the control point are pregiven.5. The heuristic algorithmReferring to Refs. [3,10], the heuristic control algorithm at each routine control point, ti, enablesminimization of the manufacturing expenses (Eq. (2)) during the remaining time D 255。s good name.Thus, given: the target amount to be acplished on time, the due date,several production speeds defined by their probability density functions, the average processing cost of realizing the manufacturing process per time unit for each production speed separately,the average cost of carrying out an inspection at the control point, andthe chance constraint to meet the deadline on time,the problem is to determine both, control points and production speeds to be introduced at each control point, to minimize the average manufacturing expenses within the planning horizon subject to the chance constraint. This is a plicated stochastic optimization problem with a random number of decision variables,Since an optimal algorithm to solve the problem cannot be found, a heuristic one is suggested and developed. The algorithm is based on simulation and pares, one by one, sorted couples of production speeds in order to find an optimal couple that results in minimizing the average expenses. The algorithm has to be realized at any control point to choose both, the speed to be introduced and the next control point.Note that all previous publications [25,10] are based on the risk averse principle, which is very efficient for noncost objectives, but cannot be applied to the newly formulated costoptimization model. It is therefore substituted for another one, namely the chance constraint principle, which is embedded in the heuristic algorithm and fits the cost structure of the paper is as follows. In Section 2, the description of the production system is outlined. Section 3 considers Notations and presents the stochastic optimization problem. In Section 4, the chance constraint principle is described. In Section 5, the heuristic algorithm is outlined. Section 6describes an illustrative numerical example, while in Section 7 extensive experimentation to verify the efficiency of the heuristic algorithm is presented. Practical applications of the algorithm on a real manmachine plant are outlined in Section 8. Conclusions and future research are presented in Section 9.2. Description of the systemThe system under consideration produces a single product or a production program that can be measured by a single value, just like the system described in [2], . in percentages of the planned total volume. Such an approach is often used in Ramp。machine production systems with random parameters, . construction, metallurgy and mining,research and development projects, for developing puter software and information systems, etc. For these not fully automatic systems, the actual output can be measured only at preset inspection (control) points. Such systems may normally use several possible speeds, which can be introduced by the decisionmaker at control points. Given the target amount needed, the due date and the amountproduced up to a routine inspection point, the problem is to determine at that point the new production speed and the next inspection point. Two objectives have been usually embedded in the stochastic optimization model: (i) to minimize the number of inspection points, and (2) to maximize the probability of pleting the production program on the due date. However, the number of publications on costoptimization online control models under disturbancesremains very scanty (see, . Ref. [3]), especially for control models under a chance constraint which have not been published elsewhere. To fill the gap, we suggest a newly developed production control model which incorporates cost parameters. Two basic conceptions are embedded in the model:A. The objective is to minimize the manufacturing expenses of pleting the production programon the due date。 Cost objective。 Chance constraint。s expenses within the planning horizon. A stochastic optimization problem is formulated, followed by a he