(749b) Integration of Automation Logic and Scheduling for Real-Time Batch Chemical Plant Optimization

In this work, we present a novel approach to incorporating a model of the low-level automation logic in a chemical plant into the online scheduling problem. We address general scheduling problems in network production environments with the following characteristics:

• Tasks can consume and produce multiple materials, batches can be mixed or split, and recycle streams may be present.
• The resolution of the scheduling problem is on the order of a single task (e.g., batch reaction).
• The plant's automation system operates at a much finer resolution, so that a single reaction may consist of 10 or more steps.
• The automation logic may restrict the dynamics of the plant in ways that are not accounted for in the scheduling problem; for example, a delay due to automation logic (not accounted for in scheduling) may lead to delays in the completion of a task.

Two of the challenges that arise from the mismatch between the scheduling model and the actual plant dynamics are that (1) during execution of a schedule, disturbances or delays that render the current schedule infeasible, but which can only be detected using the detailed model of the automation system, may occur; and (2) the scheduling solution, obtained using an optimization model, may lead to behavior that is prevented by the automation system. To address these challenges, we develop a methodology that integrates automation-derived information with the scheduling model.

First, building upon previous work (Gupta et al., 2016; Subramanian et al., 2012), we develop an iterative procedure that contains (1) a generalized state-space discrete-time resource task network (RTN) model that can be implemented online, and (2) the update equations that carry information from one online iteration to the next. The model includes features that have not been examined before, such as considering delays in processing time of tasks as optimization decisions, which may be triggered due to delays in other tasks and/or limited resource availability. In addition, we generalize the definition of tasks to account for the multiple steps involved. Furthermore, we develop a systematic way to introduce disturbances into the scheduling model based on measurements from the plant. The model accounts for new types of disturbances, including (1) processing delays detected before the completion of a task, (2) disruptions from nominal pump flow rates, (3) yield changes during the execution of a task, and (4) other disturbances during the execution of tasks (e.g., resource requirements) that can lead to task delays.

Second, to detect disturbances during the execution of a schedule, we construct a finite transition system that models the plant's automation logic (Rawlings et al., 2017), and augment the model with historic data on the amount of time the system spends in each step. With this detailed model of the plant's dynamics, we treat the current schedule as a specification of the desired system behavior, and compute the set of states from which the plant could complete the schedule. If the plant ever reaches a state that is not included in this set (due to a disturbance), then we conclude that the schedule needs to be recomputed. Additional information can be extracted from the model to explain why the schedule was no longer feasible, and to add constraints to the scheduling problem when resolving to account for the detected disturbance. The same approach also applies to the case when infeasibility of the scheduling solution will be detected (and corrected), even in the absence of any disturbance.

To evaluate our methodology, we use SIMIT, a simulation platform developed by Siemens, to build a simulation model that mimics the process dynamics and automation logic of a real plant. We develop a Delay Prediction Module (DPM) that analyzes automation logic to determine step transitions. Delays are detected by DPM and the schedule is recomputed to account for these disturbances. Other disturbances such as yield changes or disruptions from nominal pump flow rates are communicated to the scheduling model from SIMIT. The three components (SIMIT, DPM, and RTN model) are interfaced through the “executive program”, which ensures proper data flow between these components.

We illustrate the applicability of our methods through a case study that involves both batch and continuous processes, multiple final products and tasks with various steps (including material filling and withdrawing, reaction, maintenance, and quality control steps). We study how the integration of scheduling and automation may lead to better quality closed-loop solutions, with higher throughput than in solutions obtained through iterative rescheduling.

References

1. Gupta, D., Maravelias, C. T., and Wassick, J.M. (2016) From rescheduling to online scheduling. Chemical Engineering Research and Design, 116, 83-97.
2. Subramanian, K., Maravelias, C. T., and Rawlings, J.B. (2012) A state-space model for chemical production scheduling. Computers & Chemical Engineering, 47, 97-110.
3. Rawlings, B.C., Wassick, J.M., and Ydstie,B.E. (2017) Application of formal verification and falsification to large-scale chemical plant automation systems. In CPC 2017.

This work is sponsored in part by the Department of the Army through the Digital Manufacturing and Design Innovation Institute. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Department of the Army.