2013 Volume 53 Issue 5 Pages 848-853
The integrated scheduling for the continuous casting (CC) and hot rolling (HR) process remains challenging in the iron and steel production. Taking into account the technological and practical constraints, we establish a mathematical model with the objective of maximizing the number of slabs processed in hot charge rolling (HCR) or direct hot charge rolling (DHCR) mode. Using the decomposition strategy, a hybrid algorithm is proposed based on the combination of the mathematical programming and constraint programming (CP) methods. Computational results demonstrate that the hybrid algorithm is efficient and effective for solving the integrated scheduling problem (ISP) of steel plants.
Since the development of the automation and integration techniques in the iron and steel manufacture, the appropriate integrated production planning and scheduling for the continuous casting (CC) and hot rolling (HR) process has been of great significance to reduce the energy cost and improve the production efficiency. The integrated process requires that the production management should guarantee the time consistency, as well as the balance of material flow between two production stages. As the bottleneck process, CC can cast the high-temperature molten steel into slabs, which is a primary process to the upstream and downstream steps of the production line. In addition, the HR process substantially affects the quality of finished products, the due date of contract and the satisfaction degree of customers. Thus, the integrated scheduling for CC and HR process plays an important role in reducing the energy consumption, shortening the cycle life of products, enhancing the product quality, and controlling the inventory of intermediate products (slabs) and finished products (coils).
Currently, planning and scheduling decisions are made independently at different production stages. At each stage, the scheduling only considers the production constraints within the stage in question. Generally, the HR production plan is made firstly according to customer orders. Meanwhile, the required slabs are provided to the preceding production process. Thereafter, the steelmaking scheduling plan is made based on the HR production plan. However, this organization tends to induce substantial workload, thus delaying the production in the HR process. Therefore, it is essential and necessary to make an effective integrated scheduling plan.
In the iron and steel manufacture, production planning and scheduling were received considerable attention to study. In view of the importance to the theoretical researches and the practical applications, a large number of approaches1,2,3,4,5) were proposed to solve the scheduling problems. A comprehensive review of previous studies regarding the integrated steel production planning and scheduling could be found in Tang et al.6) Meanwhile, numerous approaches such as tabu search,7) genetic algorithm,8) and particle swarm optimization algorithm9) were developed to solve the HR batch scheduling problem over the past decade. For the integrated production plan, Tang et al.10) investigated two batching problems for the steelmaking and CC production, and proposed two heuristic algorithms for the corresponding problems. With respect to the HR scheduling problem, Zhao et al.11) formulated the batch planning of staple material as a vehicle routing problem with time windows (VRPTW), and optimized the batches of established units by adjusting the rolling sequences to reach a higher hot charge ratio.
In this study, we summarize the novelties of the integrated scheduling problem (ISP) as follows: (1) Unlike the problem for an independent stage, the integrated scheduling simultaneously considers the technological constraints within the CC and HR production process to guarantee the balance of material flow and the synchronization of scheduling time; (2) Our optimal objective is to maximize the total number of slabs processed in modes of hot charge rolling (HCR) or direct hot charge rolling (DHCR), further enhance the HCR and DHCR ratios to ensure reducing the energy consumption and shortening the preparation time before rolling, where HCR (DHCR) ratio is the percent of the amount for slabs processed in HCR (DHCR) mode with the total number of slabs. In the modern integrated production process, the HCR ratio and DHCR ratio are important indices for evaluating the performance of the iron and steel production.
In addition, a hybrid solution approach based on the decomposition strategy is developed to solve the complex ISP, which is motivated from the work of Jain and Grossmann.12) An integration algorithm for mixed-integer linear programming (MILP)/constraint programming (CP) was initially developed by Jain and Grossmann based on the Benders decomposition to solve a class of scheduling problems. Thereafter, various integration schemes13,14,15,16) with taking advantage of the complementary strengths of MILP and CP were proposed for solving the complicated combinatorial optimization problems.
The iron and steel production flow consists of three key stages, i.e., steelmaking, CC and HR. At the steelmaking stage, the mixture of molten iron and raw materials is reduced to desirable molten steel with the main alloy elements. At the immediate stage, CC casts the molten steel into slabs and provides slabs for the following production process. At the HR stage, the slabs are rolled into required steel plates or coils according to customer orders. The slab yard between the CC and HR stage includes a heat preservation pit and a cold slab yard, which are regarded as buffers. In the general production practice, the CC slabs may be transported to the customers as intermediate products, delivered to the slab yard to wait for rolling, transported to the heat furnace for reheating, or even directly rolled.
According to the temperature of slabs and the way through which the CC provides slabs for the HR, the linkage mode between the CC and HR process defines the following production processes, cold charge rolling (CCR), HCR, DHCR, and hot direct rolling (HDR), as illustrated in Fig. 1.
Linkage modes between CC and HR process.
In the traditional CCR process, CC slabs are transported firstly by transfer rails to the slab yard. Thereafter, the slabs are loaded into the heat furnace according to the rolling plan. The charge temperature is generally below 400°C.
In the HCR process, tepid slabs are sent to the heat preservation pit after the CC process. Then, the slabs are lifted from the pit and charged into the heat furnace for reheating according to the rolling requirements. The temperature of initial slabs ranges from 400 to 700°C.
In the DHCR process, hot and immaculate slabs are directly charged into the heat furnace through transfer rails, and then the HR process is undertaken. The charge temperature varies from 700 to 1000°C.
The main features of the HDR process are as follows: without being heated in the heat furnace, hot and immaculate slabs are directly sent to the HR machine after the slab edges have been heated by the heater. In general, the slabs temperature is kept over 1100°C before rolling.
As the process objects (casts, rolling units) are cast or rolled continuously at a high temperature, more direct linkages between the CC and HR stage indicate less energy consumption and shorter production lead time. It is obvious that the HDR technique can achieve the highest energy savings, followed by DHCR, HCR and CCR. Similar trends exist in the preparation time before rolling. Due to the necessity for a special configuration of the production line and the high organization level, the HDR technique is seldom implemented in the modern iron and steel manufacture.
The steel strips in which rolling unit and the sequence of strips are defined at the planning level with considering multiple factors, such as the orders and products delivery date. In the steelmaking plan, the slabs included in which casts and processed sequence in the corresponding cast are also determined with considering the HR plan and the limitations of the CC stage.
As specifically defined here, there is only one machine at the HR stage. In addition, the following assumptions are made for the ISP.
(1) ACC (HR) machine can process at most one cast (rolling unit) at one time. A cast (rolling unit) can only be processed on one CC (HR) machine at any time.
(2) The set-up time is required between two continuous casts on the same CC machine, which is independent with the casting sequence and casts’ properties.
(3) For two consecutive rolling units, only when the former has been finished and after a relative set-up time of the rolling machine, can the immediate next one be started.
(4) Slabs included in the same cast (rolling unit) are continuously processed on the CC (HR) machine.
(5) For the required rolling slabs from the CC machines, only when the casting process has been finished and transported to the rolling machine, can the rolling operation be started.
(6) The cold slab yard can infinitely provide slabs for the HR production process.
(7) The required slabs in the rolling process can be completed punctually by the heat furnace.
Our integrated scheduling aims to provide an effective scheduling plan within a given time horizon and the objective is to maximize the number of slabs processed in the HCR or DHCR mode. The tasks of the ISP are as follows: (1) Determine the processing equipment for each cast, as well as the processing sequence of casts and rolling units; (2) Decide the start time for casts and rolling units, from which the start time of slabs in casts and rolling units can be derived. Furthermore, the linkage mode of slabs can be fixed based on the interval time, which eventually help to determine the destination (cold slab yard, heat preservation pit, heat furnace, and HR machine) of the CC slabs.
2.2. Mathematical Programming FormulationThe following notations are employed to describe the ISP:
Indices and sets:
nC: total number of casts;
nR: total number of rolling units;
MC: total number of CC machines;
i, j: index for casts, i, j = 0, …, nC+1, where 0 and nC+1 denote the fictitious cast;
q, g: index for rolling units, q, g = 0,…, nR+1, where 0 and nR+1 denote the virtual rolling unit;
m: index for CC machines, m = 1, …, MC;
NC: set of slabs in all casts,
l: index for slabs, l ∈ NC, l ∈ NR;
DC: set-up time of CC machines between two consecutive casts;
DR: set-up time of the HR machine between two adjacent rolling units;
tC: processing time for each slab on CC machines;
tR: processing time for each slab on the HR machine;
Tr: transportation time between the CC and HR stage;
v: identifier of linkage mode between the CC and HR stage, v = 1denotes the DHCR process, v = 2 expresses the HCR process, and v = 3 expresses the CCR process;
H: time horizon for the production plan;
M: a sufficient large number;
Decision variables:
xijm: binary variable, 1 if cast i and j are assigned to the same CC machine m, and cast j is processed directly after i, 0 otherwise;
yim: binary variable, 1 if cast i is processed by the CC machine m, 0 otherwise;
zqg: binary variable, 1 if rolling unit q is the direct predecessor for g, 0 otherwise;
Accordingly, the mathematical model ISP_M1for ISP is presented as follows:
(1) |
(2) |
(3) |
(4) |
(5) |
(6) |
(7) |
(8) |
(9) |
(10) |
(11) |
Comparing to the traditional scheduling problems, the technological constraints related to the CC and HR production process are considered in the ISP, as well as the consistency between two stages. The main decisions involved in the ISP include the assignment of casts to CC machines, the processing sequence and the start time for casts and rolling units. Thus, the difficulties led us to develop a hybrid algorithm based on the decomposition strategy for solving the ISP.
In the present work, the original ISP is split into a master problem and a CP subproblem in a natural way using the special features of the integrated production. To obtain potentially good solutions, the master problem is used to find the partial solutions, and the CP subproblem is to check the feasibility of the obtained solutions and generate a complete schedule plan. Specially, we propose an iterative scheme where iterate between a master problem and a CP subproblem, in a similar fashion as in Jain and Grossmann.12) The start time of the casts and rolling units, the sequence of the rolling units are determined in the master problem, while the CP subproblem is used to derive a feasible complete solution. It means that the CP subproblem should identify the processing machine for casts and the corresponding casting sequence for casts processed by the same CC machine with considering the obtained start time. If a feasible solution is obtained, it becomes the final solution. Otherwise, the causes of infeasibilities infer cuts, which are then added to the master problem to perform the next iteration. A simplified flow diagram of the proposed algorithm for the ISP is shown in Fig. 2.
Schematic diagram of the proposed hybrid algorithm.
For the master problem, the continuous variables (SiC, SqR) are considered, which are important to specify the processing mode for slabs. Since the production practice that there is only one machine at the HR stage, the sequence of the rolling units is also considered in the master problem to optimize the high level solutions. Especially, the sequence of the rolling units is important to derive the feasible casting sequence for casts. Otherwise, more infeasible solutions would be generated to satisfy the technology constraints of the DHCR or HCR process excluding the resource (machines) constraint. Thus, it is necessary to consider the sequence of the rolling units in the master problem. Then, the model ISP_M2 for the master problem can be written as follows:
obj: (1)
s.t. (4), (6) — (11)
Since the slabs are processed continuously at the CC and HR stage, the master problem should determine the start time of the batches (casts, rolling units), and then the start time of the slabs can be derived. The interval time of the slabs between two production stages simultaneously exists in constraint (9) and the objective function (1), where constraint (9) gives a lower bound of the interval time, whereas each linkage mode gives the domain of this time. This implies that the interval time should be short enough to ensure a large number of slabs charged in the DHCR or HCR mode.
3.2. CP SubproblemAfter the master problem is solved, the CP subproblem decides if this partial solution can be extended to a complete solution by fixing the assignment variables (yim) and the sequencing variables (xijm). In this paper, we model the CP subproblem using the modeling language of ILOG’s OPL, which has a number of global constraints and special constructs for the scheduling problems. In the CP subproblem, the parallel CC machines are assumed to be unary resources in OPL. The defining attribute of a unary resource is that the CC machine can process only one cast at a time. Using this basic framework of OPL the CP model ISP_M3 can be written as follows.
(12) |
(13) |
(14) |
(15) |
The CP subproblem is used to determine the processing machine and the feasible casting sequence for casts; therefore, no objective function is included in the ISP_M3. Constraint (12) establishes the relationship between the master problem and the CP subproblem according to the start time of the casts. Constraint (13) expresses that the duration of a cast is independent with the CC machine. Constraint (14) plays the same role as constraint (3), where assign [ci, m] is used to denote the binary assignment variable. Similarly, constraint (15) is equivalent to constraint (5), and seqC denotes the binary sequencing variables.
3.3. Generating CutsThe cut is used to forbid the infeasible values and tighten the domain of variables, which is important for the master problem. Jain and Grossmann’s cut-generating scheme relies largely on the fact that the resource units (machines) are independent. Therefore, the cuts can be separately generated for each individual machine. In the present study, it is challenging to design the effective cuts generation method due to the interaction between the rolling sequence and the casting sequence.
The following notations are adopted to introduce the cuts generation method:
im: the index for the cast with the maximal starting time in
Then, the following observation concerning the feasibility of the schedule plan is introduced:
Observation 1. The solution is infeasible if, for some t, (1) the duration of cast i overlap with that of some cast j, or (2) the interval time between cast i and cast j is less than the set-up time, where
In an infeasible solution, it is assumed that at least one of the conditions in Observation 1 is satisfied for some cast j and
Since there are no benchmarks test sets available for the ISP, several instances are generated according to the collective production data from an iron and steel plant in China. The solution methodology is implemented in C++, and performed on a PC with Intel Core(TM) 2, 2.66 GHz using the Windows XP operating system. The MIP optimizer ILOG CPLEX 11.2 and CP 2.1 are used for obtaining the optimal solutions.
The parameter settings used in our experiments are as follows: MC = 3, Tr = 8 min, DC = 20 min, DR = 10 min, tC = 5 min, tR = 2 min, [
The sizes of the experimental instances are illustrated in Table 1. Since the 24 h operation period of the steel plant is divided into three shifts, 3 groups of experimental instances are generated accordingly. In Table 1, NRC denotes the number of required slabs at the rolling stage from the CC machines.
inst | H(m) | nC | |NC| | nR | |NR| | NRC |
---|---|---|---|---|---|---|
11 | 480 | 5 | 143 | 3 | 104 | 87 |
12 | 8 | 199 | 5 | 168 | 147 | |
13 | 9 | 217 | 6 | 185 | 158 | |
21 | 960 | 9 | 267 | 5 | 216 | 193 |
22 | 10 | 308 | 6 | 258 | 236 | |
23 | 11 | 348 | 7 | 281 | 252 | |
24 | 12 | 387 | 8 | 343 | 303 | |
25 | 13 | 423 | 9 | 389 | 335 | |
31 | 1440 | 11 | 457 | 5 | 349 | 324 |
32 | 12 | 497 | 6 | 406 | 382 | |
33 | 13 | 546 | 7 | 457 | 419 | |
34 | 14 | 594 | 8 | 514 | 466 |
In order to evaluate the performance of the proposed algorithm, the optimizer CPLEX is used to solve the model ISP_M1. The computational results for solving the 3 group sizes of instances by CPLEX and the decomposition algorithm are respectively shown in Tables 2 and 3. Of note, the model sizes and the search process of the master problem, and the CP subproblem in the last iteration are shown in Table 3. “rDHCR” and “rHCR ” respectively present the DHCR ratio and HCR ratio. “cuts” represents the number of cuts inferred by the CP subproblems. “obj” represents the objective value.
inst | nodes | iteration | variables | constraints | time (s) | rDHCR (%) | rHCR (%) | obj |
---|---|---|---|---|---|---|---|---|
11 | 107 | 1672 | 1 51 | 315 | 3.656 | 20.69 | 74.71 | 83 |
12 | 3826 | 125080 | 320 | 593 | 5.812 | 24.49 | 72.11 | 142 |
13 | 49624 | 2048867 | 395 | 686 | 64.875 | 57.59 | 36.71 | 149 |
21 | 5712 | 69286 | 381 | 743 | 6.735 | 34.2 | 62.69 | 187 |
22 | 413591 | 6885765 | 462 | 906 | 308.657 | 36.02 | 57.63 | 221 |
23 | 1603828 | 71136199 | 551 | 1023 | 2809.45 | 38.89 | 59.52 | 248 |
24 | 4103061 | 176432119 | 648 | 1218 | 6901.11b | 31.68 | 54.46 | 261a |
25 | 1968637 | 74570569 | 753 | 1383 | 3701.3b | 46.57 | 34.93 | 273a |
31 | 1475784 | 667999044 | 521 | 1139 | 3071 | 35.19 | 54.32 | 290 |
32 | 1689931 | 74279046 | 614 | 1344 | 4628.44 | 34.29 | 53.93 | 337 |
33 | 3873950 | 88383317 | 715 | 1515 | 6864.69 | 34.12 | 54.89 | 373 |
34 | 1641603 | 42469713 | 824 | 1714 | 3601.47b | 27.47 | 54.29 | 381a |
inst | ISP_M2 | ISP_M3 | cuts | rDHCR (%) | rHCR (%) | obj | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
nodes | iteration | variables | constraints | time (s) | branch | Choice point | variable | constraints | time (s) | |||||
11 | 38 | 741 | 25 | 194 | 0.891 | 3 | 5 | 23 | 8 | 0.047 | 1 | 40.22 | 55.17 | 83 |
12 | 66 | 1518 | 50 | 335 | 3.344 | 4 | 6 | 35 | 8 | 0.047 | 2 | 36.05 | 60.54 | 142 |
13 | 327 | 5908 | 65 | 373 | 9.813 | 3 | 5 | 39 | 8 | 0.188 | 5 | 53.16 | 39.24 | 146 |
21 | 194 | 2996 | 51 | 430 | 5.766 | 3 | 5 | 39 | 8 | 0.109 | 3 | 36.79 | 60.11 | 187 |
22 | 488 | 7242 | 66 | 531 | 32.828 | 4 | 6 | 43 | 8 | 0.203 | 7 | 44.92 | 48.73 | 221 |
23 | 206 | 4305 | 83 | 577 | 25.438 | 5 | 7 | 47 | 8 | 0.234 | 6 | 48.8 | 49.6 | 248 |
24 | 345152 | 5807110 | 102 | 691 | 495.39 | 5 | 7 | 51 | 8 | 0.047 | 1 | 32.01 | 60.4 | 280 |
25 | 18545 | 183989 | 123 | 777 | 354.718 | 5 | 7 | 55 | 8 | 0.141 | 4 | 45.07 | 48.96 | 315 |
31 | 638 | 19306 | 53 | 690 | 22.719 | 4 | 6 | 47 | 8 | 0.063 | 3 | 43.52 | 45.99 | 290 |
32 | 36459 | 383073 | 68 | 818 | 153.828 | 5 | 7 | 51 | 8 | 0.109 | 2 | 39.53 | 48.69 | 337 |
33 | 13610 | 293395 | 85 | 911 | 281.86 | 5 | 7 | 55 | 8 | 0.188 | 6 | 36.28 | 52.74 | 373 |
34 | 7214 | 83635 | 104 | 1019 | 950.484 | 5 | 7 | 59 | 8 | 0.156 | 3 | 33.05 | 58.58 | 427 |
It can be clearly seen that all instances are successfully solved by the decomposition algorithm, while CPLEX fail to obtain optimal solutions for 3 instances as the memory space is occasionally exceeded. Thus, the suboptimal solutions are obtained within the maximum allowable time for these instances. In all problem sizes, the first data set are relative easy to solve for both the hybrid algorithm and CPLEX. The sizes of the model ISP_M1 and ISP_M2 increase much rapidly with the size of instances. To a certain extent, the optimal solutions are obtained by the master problem with the cost of long computational time for the big M constraint (8). Therefore, the computational time of the model ISP_M1 and ISP_M2 increase with the size of instances and the time horizon. For all the experimental instances, the CP time is less than the model ISP_M2 in terms of the determined start time for casts. Furthermore, comparing to the model ISP_M2, the CP time varies slightly with the size of instances.
The HCR ratio obtained by the hybrid algorithm is higher than the DHCR ratio for most instances. Although the effective integrated scheduling plan relies largely on the production plan, the constraints of the HCR process are relatively easy to meet in comparison to the DHCR technology. For some instances with the optimal solutions obtained by CPLEX, it can be seen that the DHCR ratio obtained by the hybrid algorithm is relatively higher than the model ISP_M1. It means more energy saving, as well as the shorter lead time before the HR process. According to the computational results, it is concluded that the hybrid algorithm is more effective and shows a great superiority in efficiency for solving the ISP.
This study focused on the ISP for the CC and HR production process in the iron and steel production. Our aim is to generate an effective scheduling plan within the given time horizon in order to ensure more slabs charged in the HCR or DHCR mode. For the established mathematical model, a hybrid algorithm was proposed based on the decomposition strategy. Cuts generation scheme based on the temporal precedence relationship was devised for the master problem. Computational results indicate that the hybrid algorithm is capable of solving the ISP efficiently. Compared to CPLEX, the proposed algorithm is more effective.