• No results found

6.2 Model Content

6.2.3 Demand Management

The demand management controls ingoing and outgoing anodes and anode pallets. It keeps track of anode pallets arriving in the rodding shop and pallets ultimately again arriving at the rodding shop. To this end, it communicates relevant data to the other agents in the system. As addressed before, information regarding the arrival and departure of anode pallets in practice is done through the Manufacturing Execution System (MES).

We decide to include a comprehensive approach by which the entire process from the moment an anode is needed until the transport job is released and dropped-off, is included. For that reason, we first introduce some definitions and characteristics in Subsection 6.2.3.1. Subsection 6.2.3.2 outlines the demand flow methodology embodied in the simulation model. Finally, Subsection 6.2.5 discusses some heuristics for initializing the demand schema’s.

6.2.3.1 Demand Characteristics and Definitions

This subsection clarifies some terminology and further defines the scope of the evaluation model. The focus is first on addressing how shift & sections are defined. After which load characteristics are discussed.

We define a shift as a time-window in which all activities of a certain activity type (i.e., anode changing, metal tapping or pot tending) are scheduled. In a section, only one activity is performed

6.2. Model Content 97 during a shift. The activities anode changing, metal tapping, and pot tending follow each other in succession. The length of a shift is usually eight hours for each activity type, but this is modifiable in the model. Both the shift duration as well as the duration per activity type are adjustable. For exam- ple, one could schedule the anode changing activity with a duration of eight hours, metal tapping of six hours, and pot tending of ten hours. Once the shift length is set, we consider this length as fixed during the entire systems’ life.

Recall that in a shift & section way of working, cells are clustered in sections and the working schedule is fitted to the activities carried out in these sections. We define a section as a cluster of adjoining electrolytic-cells that may share anode pallet demand and in which common activities are carried out. Two cells directly placed opposite to each other on the main aisle are also considered in the same section. In theory, a section can comprise all cells in the potroom, but we limit ourselves to the extreme lower limit of one section comprising two opposite neighbor cells in the center aisle of a segment and the extreme upper limit of one section covering an entire potroom. The minimum size of a section is, therefore, two and the maximum size comprises an entire potroom.

In a transportation control system, the desire to have the load on-time at the destination is often modeled by time-window restrictions. Each transportation request is then characterized by, for ex- ample, an earliest- and latest delivery time. We characterize the transport jobs by time-windows as well. To this end, we define the following job characteristics:

• W ESTp,s,a, Workforce Earliest Start Time: start time when the crane operators may start their

anode changing shift in potroompsectionsanode shifta;

• W LSTp,s,a,Workforce Latest Start Time: latest time when the crane operators can start swapping

anodes on a pallet;

• W STp,s,a, Workforce Start Time: actual start time when the crane operators start their anode

swapping activities;

• F ERTp,s,a,Fresh (anode pallet) Earliest Release Time: release time at the rodding shop from whereon

the fresh pallet can be picked-up. Set by the user (default:4hours, alternatively one can use the formula as discussed in Appendix J);

• F LRTp,s,a,Fresh (anode pallet) Latest Release Time: latest time when the fresh anode pallet can be

picked-up at the rodding shop. Set by the user (default:W LST);

• F P Tp,s,a,Fresh (anode pallet) Pick-up Time: fresh pallet pick-up time at the rodding shop;

• F DTp,s,a,Fresh (anode pallet) Drop-off Time: fresh pallet drop-off time at the cell segment;

• BERTp,s,a,Butt (anode pallet) Earliest Release Time: release time at the P/D-point at which the

butt anode pallet can be picked-up;

• BLRTp,s,a,Butt (anode pallet) Latest Release Time: latest release time when the butt anode pallet

can be picked-up. Set by the user (default:4hours after being dropped-off);

• BP Tp,s,a,Butt (anode pallet) Pick-up Time: butt pallet pick-up time at the cell segment;

• BDTp,s,a,Butt (anode pallet) Drop-off Time: butt pallet drop-off time at the rodding shop.

A transport picked-up later than the latest release times (F LRT andBLRT) would not yield a desirable result. A fresh anode pallet picked-up after this time would namely be too late at the des- tination and result in excessive waiting times for the workforce that handles the crane operations. Likewise, we have to limit the latest release time of the butt anode pallets because they may other- wise burden efficient vehicle movements. However, it is unavoidable that in some cases these latest release times will be exceeded. Therefore, we include so-called "emergency" shipments for handling those late anode pallets. An emergency shipment includes the entire process from picking up the pallet until dropping of the pallet without any interference with the AGV system. We assume these emergency activities are carried out immediately. These assumptions can be advocated in practice by acting adequately when these outliers are expected to happen. For instance, a human-driven forklift can be used to maneuver almost freely through the system as a substitute for an emergency team. However, remark that we keep track of these emergency shipments and ideally one would eliminate them completely. We expect this can be done, for example, by examining the system performance and then fine-tune parameters or by considering different experimental configurations. We leave this up for further research.

6.2.3.2 Demand Flow Methodology

The pallet’s capacity (in most cases) exceeds the demand from one individual cell, therefore, individ- ual anode demand from cells is combined into pallets and those pallets are transported to dedicated

places in the section. Anode changing operations are covered on a high-level only. That is, we do not model operators and crane movements explicitly, but we model the anode changing process as a time-consuming activity only with the inclusion of AGV path blocking restrictions. Ideally, one would have the anode pallets as close as possible to the corresponding cells because the required crane movements are then limited. This is, however, not always possible, because of, for example:

• pallet placement restrictions;

• anodes located on the back aisles need to be moved by crane via the center aisle; • pallet orientation should be taken into account;

• conflicting pallet placement interests (e.g., where to position one pallet if two neighboring cells require one anode each and another cell a couple of meters further requires one anode?); • possible workforce flow through different sections.

Based on this analysis and our already made design choices, we could deduce at least three design elements that should be examined in more detail: (1) the transition from anode demand to anode pallet demand, (2) the interaction with the crane, (3) and the pick-up and delivery sequence of pallets. Below, we address the methodology followed to cover these design considerations.

To model the anode pallet demand properly, we propose the material flow determination ap- proach as shown in Figure 6.2 of which Appendix K presents the corresponding entity relationship diagram. TheDemand Managementfirst establishes an anode demand scheme. This scheme repre- sents fresh anodes that need to be changed in a shift. Currently, the demand for anodes follows a cyclic and predictable pattern. That is, anodes need to be changed when the setting cycle elapsed. We assume this setting cycle or at least its distribution type is known upfront. In the case of equal set- ting cycles for all anodes, the anode setting cycle scheme repeats itself each setting cycle period. The scheme should be obtained from customers and is expected to be unique per customer. Alternatively, Appendix L provides a set of five heuristics that can be used to mimic the anode demand normally arising from the MES.

FIGURE6.2: Material flow determination approach representing information flows be- tween the Demand Management, Section Management, and Vehicle Scheduling.

Once the anode demand is known, theDemand Managementgenerates the anode pallet demand. The resulting anode pallet demand schema contains a schedule of pallets that need to be delivered to a section for anode changing shifts. Appendix N presents three heuristics as alternatives to a manually provided schedule.

TheSection Managementthen obtains information regarding the pick-up and drop-off locations of pallets. We consider a static pallet position of which the location cannot be changed anymore once

6.2. Model Content 99 they are allocated to a position. However, the pallet may be a different one than initially scheduled. For example, when the first planned AGV A has a delay and another AGV B that is earlier at the section has to wait before AGV A arrives, then AGV B can already drop-off a pallet at the place orig- inally assigned to the delayed pallet A because otherwise the initial planned pallet position of AGV B may block the passage for the pallet from AGV A. The main goal of determining the pallet drop-off locations is to deliver them as close as possible to the corresponding cells to avoid unnecessary crane movements. Parallel to modeling the crane behavior, the allocation of pallets thus requires attention. The proposed pallet allocation heuristics are presented in Appendix O.

In addition to determining proper P/D-locations, the sequence in which the jobs are handled is of importance. To this end, theSection Managementuses the anode pallet demand schema and P/D- locations, and determines the sequence in which the anodes are interchanged. The sequence relies on how the crane operators plan their anode changing activities. We consider having one workforce team per shift per section. Usually, the teams follow a certain pattern like starting changing anodes in the western end and gradually work through the eastern end. To adequately model the behavior of cranes, we propose some heuristics to simulate the crane operation "flow" through the potrooms in Appendix P. These heuristics are limited to handling the anode demand per pallet.

Although the way the crane operates affects how and where the pallets will be positioned and visa versa, the approach we propose considers the crane operations mainly as leading followed by the pallet position allocation. Our approach focuses on the supportive role of pallet transport which will marginally affect the way crane operators continue their activities. Furthermore, our approach allows the model to be flexible for evaluating different work systems and it is a straightforward approach that is easy to understand. An approach that, for example, simultaneously address crane operations and pallet allocations might be superior to our approach, however, we suggest to investigate this potential in future research. So, first, the pallets are assigned statically, after which the workforce flow through the section is modelled.

One remark regarding the maximum number of possible pallet placements per segment on each side of the center aisle. This number equals the number of electrolytic-cells minus the two outer cells (on each side) because on these outer ends their pallet drop-off and pick-up is not permitted. When the possible assignment positions are not sufficient to cover the anode demand in a section, we consider that the leftover pallets (following from the heuristics) are delivered to the section by "emer- gency" shipments. In further research, one could, for example, include a dynamic pallet assignment approach in which multiple pallets are assigned to one position with different reserved time slots. The considered crane behavior heuristics affect how the butt pallets transport requests are generated. If all anodes in the pallet are interchanged, we decide to release the subsequent pallet transport request. Lastly, we include three variants of rodding shop pick-up and drop-off rules. As evaluation rules we examine (1) the selection of a random rodding shop, (2) the selection of the nearest rodding shop, and (3) the selection of the farthest rodding shop. The rodding shops are selected once the transport request is generated and do not change intermediately.

To summarize, the demand generation heuristics as discussed above consists of: • Generating fresh anode pallet demand (4 heuristics, Appendix L).

• Transition to fresh anode pallet demand (3 heuristics, Appendix N). • Assignment of pallet locations (5 heuristics, Appendix O).

• Modelling crane behavior (3 heuristics, Appendix P). • Selection of rodding shop as pick-up locations (3 heuristics). • Selection of rodding shop as drop-off locations (3 heuristics).