(530f) Batch Scheduling with Quality-Based Changeovers

Batch scheduling is a frequent and complex operation performed in all process industries. The combinatorial nature of the problem difficults finding non intuitive solutions to capture additional value. Because of the high frequency of these operations, even small improvements can quickly translate into millions of dollars in savings. The planning horizon spans from two days to a few weeks. In the case of discrete-time models, the horizon is divided in hourly periods. The resulting models are difficult to solve, thereby motivating intensive research in this area[1]. One of the features that adds more complexity to the problem is the presence of sequence-dependent changeovers (SDC)[2], in which the length of the cleaning task performed between two successive processing tasks depends on the sequence in which the processing tasks are performed. For example, if in a given reactor, task B is performed after task A, then a cleaning of 2 hours might be required between the tasks. However, if task A is performed after task B, maybe the cleaning time could be longer, not required, or the transition from B to A might be even forbidden. The cleaning is frequently required to prevent cross contamination between products.

In many applications, like the production of chemicals of the same product family, the cleaning can be avoided by making sure enough batches of the second product are performed in a row, and any impurities from the previous product are diluted enough such that the resulting product meet quality standards. We refer to this type of transitions as quality-based changeovers (QBC). This kind of transitions have not been addressed in the Process Systems Engineering (PSE) literature. The required models need to consider the option of performing either a quality-based changeover or a traditional cleaning changeover. We implement and test quality-based changeovers in three of the major general purpose scheduling frameworks in PSE, namely State-Task-Network (STN)[3], Resource-Task-Network (RTN)[4], and Unit-Operation-Port-State-Superstructure (UOPSS)[5]. Implementing in the three frameworks allows to compare their computational efficiency, and main features, including extensibility to accommodate novel features such as quality based changeovers. The implementations are tested in a small process network, and in an industrial-sized network. For the second case, a richer objective function is proposed, including minimization of priority-dependent backlog, changeover time, and deviation from a previously determined plan.

The results obtained give valuable insights on the three frameworks. Some highlights of the comparisons performed are:

QBC can seamlessly integrated in the STN and UOPSS frameworks by extension of their SDC constraints. On the other hand, RTN requires the addition of many auxiliary tasks and resources.

QBC reduces computational burden to the problem by tightening the SDC constraints.

Since few SDC constraints are active at the optimum, they can be effectively handled with use of lazy constraints.

UOPSS requires the addition of less SDC constraints, which makes it more efficient for large problems.

The modeling extensions allow to better represent industrial problems, while the insights obtained yield guidelines for model selection. These two aspects combined represent an important contribution to the adoption optimization of scheduling operations.

References

[1] Harjunkoski, I., Maravelias, C.T., Bongers, P., Castro, P.M., Engell, S., Grossmann, I.E., Hooker, J., Méndez, C., Sand, G. and Wassick, J., 2014. Scope for industrial applications of production scheduling models and solution methods. Computers & Chemical Engineering, 62, pp.161-193.

[2] Kim, S.C. and Bobrowski, P.M., 1994. Impact of sequence-dependent setup time on job shop scheduling performance. The International Journal of Production Research, 32(7), pp.1503-1520.

[3] Kondili, E., Pantelides, C.C. and Sargent, R.W.H., 1993. A general algorithm for short-term scheduling of batch operations—I. MILP formulation. Computers & Chemical Engineering, 17(2), pp.211-227.

[4] Pantelides, C.C., 1994, July. Unified frameworks for optimal process planning and scheduling. In Proceedings on the second conference on foundations of computer aided operations (pp. 253-274). New York: Cache Publications.

[5] Zyngier, D. and Kelly, J.D., 2012, June. UOPSS: a new paradigm for modeling production planning & scheduling systems. In Symposium on Computer Aided Process Engineering (Vol. 17, p. 20).