Reactive Scheduling in Batch Plants

HENNING, GABRIELA

Angra dos Reis

Workshop; “Pan American Advanced Studies Institute Program on Process Systems Engineering – PASI 2011. Process Modeling and Optimization for Energy and Sustainability”; 2011

CEPAC, Chemical Engineering PanAmerican Collaboration

Industrial environments are dynamic in nature and unforeseen events frequently disrupt in-progress schedules. New order arrivals, orders’ cancellations/modifications, unit breakdowns, changes in batch processing/setup times, late arrivals of raw materials, etc. are some of the unexpected situations faced in industry on a daily basis. As a result, in most plants, scheduling is an ongoing reactive process where evolving and changing circumstances continually force reconsideration and revision of pre-established plans. In the first part of this seminar an effort is made to systematize knowledge and to analyze relevant contributions already made in this domain. Definitions and concepts appropriate for most applications of rescheduling manufacturing systems will be introduced, and a framework for understanding rescheduling strategies, policies and methods will be presented. ·         Regarding strategies four categories of problems will be analyzed: (i) Completely reactive scheduling, (ii) Predictive-reactive scheduling, (iii) Robust predictive-reactive scheduling, and (iv) Robust pro-active scheduling. ·         Concerning policies three categories will be distinguished: (i) Periodic, (ii) Event-driven, and (iii) Hybrid policies. ·         Regarding methods, schedule repair and complete rescheduling methodologies will be distinguished. For each of them, various heuristic, algorithmic and MILP approaches will be presented. Furthermore, different types of events and performance measures will be considered. On the topic of unforeseen events, three key dimensions will be evaluated: cause, context and impact. In relation to performance measures, on top of the regular assessment measures associated with efficiency (e.g.., makespan, tardiness, earliness, number of tardy orders, etc.) and cost, which are regularly employed in predictive scheduling activities, different appraisals of schedule change need to be taken into account.  In fact, stability measures, which are related to a smooth operation of the plant, become relevant in reactive scheduling. Therefore, different metrics assessing the departure of the revised from the initial schedule will be proposed. In the second part of the seminar advances in the development of a reactive support environment will be presented. When addressing a rescheduling situation most of the objectives and basic constraints that define the original problem still apply; however, the partially executed schedule and the perturbation or triggering event has also to be taken into account. The support framework is able to capture this context knowledge by means of an explicit domain representation and to use it in the generation of solutions to rescheduling problems. The current proposal is oriented towards multiproduct stage batch plants, working under a batch-based approach and under different inter-stage storage and operational policies. It is limited to a set of unforeseen events (unit failure and performance modification, batch cancellation/modification/ new arrival). However, the underlying rationale of the system would allow extending it in the future to consider multipurpose batch plants, and a wider range of disruptive events (e.g. modification of non-renewable resources availability). The framework has been envisioned to operate under an event-driven rescheduling policy. It explicitly captures the status of the in-progress schedule, and typifies the unexpected event in order to characterize its context. This allows making a proper specification of the rescheduling problem to be faced. The resulting specification is then translated into a Constraint Programming (CP) model and, finally, the resulting formulation is solved by means of a CP approach. The proposed solution methodology attempts to get together the benefits of a repair-based method (limited schedule disruption and low computational requirements), with the advantages a complete rescheduling approach (non-myopic, overall view of the rescheduling system). Furthermore, one of its goals is to develop several alternative solutions in very low CPU times and to select the preferred one with reference to a set of objectives that measure both schedule efficiency and stability.