cated the use of MAS technology as an integral part of this thesis. Finally, we discussed the purpose of using a proper MAS design methodology. We concluded that Prometheus is a practical, compre- hensive and easy to understand methodology that will be used in this thesis.
3.4
Operations Research Applications in the Primary Aluminium
Industry
Production planning in the primary aluminium industry is a challenging task, due to imbalances among production and fabrication steps comprised. Only a minor mistake, like inferior anode place- ment and postponed metal tapping, may negatively impact the performance of a potroom. Unex- pected variations in the process could result in temporary shutdowns of (parts of) the system (e.g.,
3.4. Operations Research Applications in the Primary Aluminium Industry 55
FIGURE3.6: Prometheus MAS Design Methodology (Padgham and Winikoff, 2005)
disruptive electrolytic-cells) and ultimately negatively impact the throughput to the casthouse. Con- sequently, it may take a considerable amount of time for a cell to recover from such a disruption and get back to its most productive stage again (Meijer, 2015).
An adequate logistics planning for distributing anodes is essential for stable production in pri- mary aluminium plants. Meijer (2011) highlights that anode changing is an important parameter that influences the stability of a cell and that optimizing this material flow can create a more efficient overall performance and higher productivity of the smelter. Anodes that arrive too early at their destination could lead to stacks blocking the passage for other equipment or bottlenecks, while a de- layed anode delivery could affect the electrolytic process that takes place in the cell. Moreover, with the even growing number of cells used in modern smelters (see Figure 3.7), the need for sophisticated applications of analytic methods to help make better decisions regarding logistics planning of anode, is of substantial value. Furthermore, as the aluminium industry is a capital intensive industry, only a marginal improvement with the application of analytic approaches can lead to significant reduction in costs realized by aluminium companies. Thus, improving the decision-making of smelter operations, with a focus on logistics involved in anode changing activities, represents a promising opportunity in reducing process variation and operational expenses.
There is a significant stream of studies taking analytical methods for aluminium smelters into con- sideration. The Light Metals series (e.g., see Grandfield, 2014; Hyland, 2015; Williams, 2016) contain a collection of articles covering the value chain from bauxite to final products and alloys. Technological advances have contributed to various alternative production and manufacturing processes. Research findings and reviews over the last five decades, in the field of aluminium production and related light metals technologies, are bundled as well (see The Minerals, Metals & Materials Society, 2013). Dutta, Apujani, and Gupta (2016) performed a concise survey of operations research (OR) approaches ap- plied to the aluminium industry. Similar techniques have been studied in other industries such as the steel industry (e.g., see Dutta and Fourer, 2001; Ouelhadj and Petrovic, 2009).
Winkelmann et al. (2016) developed a dynamic discrete-event simulation model to support plan- ning processes in downstream manufacturing processes. They mapped the material flow of rolling mills and extrusion plants by the configuration of generic simulation models. In their models, they
1960 1970* 1980 1990 2000* 2010 100 200 300 400 500 600 700 120 188 256 365 543 720 Year Number of Electr olytic-cells
FIGURE3.7: Typical Number of Electrolytic-cells in Modern Aluminium Smelter Facil- ities. Adopted from (Simulating Pot Room Logistics in Aluminium Smelters). The *-sign indicates that linear interpolation of the depicted direct neighbor years is used to esti-
mate the number.
included handling equipment and production facilities under consideration of process models. Their work mainly differs from ours because we do not focus on operations in the secondary aluminium process (e.g., rolling mills, extrusions plants, etc.). We focus on the primary aluminium process and the anode transportations that is thus limited covered by Winkelmann et al. (2016).
Pablo, Racca, and Bustelo (2012) consider a simulation model to identify the main bottlenecks in the process of liquid aluminium transportation. They analyzed dynamic system behavior, apply heuristics balancing algorithms and validate possible solutions for increasing casting capacity. The focus of their study is investigating the impact of the logistics of liquid metal transportation taking into account a desired production of rods. The difference with our study is that we primarily consider anode transportation instead of liquid aluminium. Eick, Vogelsang, and Behrens (2001) developed a discrete-event simulation model of a smelter based on high-level Petri nets. Their model incorporates features such as operation scheduling, collision detection, and material handling processes. Also, they explicitly modeled the anode and butt transport as part of their Petri net. Although both models presented in Eick, Vogelsang, and Behrens (2001) and Pablo, Racca, and Bustelo (2012) can be used to gather insights into the logistical performance in aluminium smelters, both publications do not focus on the impact of tactical and operational decisions on their anode transportation approach.
Steinrücke (2015) considers a Mixed-Integer Linear Programming (MILP) model of a so-called multi-stage production, shipping, and distribution scheduling problem in the aluminium industry. The used relax-and-fix decomposition method aims to connect the material flows of the adjacent subsystems including bauxites mines, aluminium oxide refineries, and the aluminium smelters. The main distinction with our study is that we provide an operational strategy for determining how to manage the internal logistics in the aluminium smelter and thereby support the material flows on a network-wide scale. Steinrücke (2015) simplified operational decisions made on sites and only considered the planning of production and shipping quantities.
An interesting and closely related line of research comes from Nicholls and co-authors. Nicholls and Hedditch (1993) present an integrated mathematical model of an aluminium smelter incorporat- ing the raw materials feed, carbon bake, rodding, potrooms and ingot mill-areas, including non-linear feedback loops. Besides evaluating the impact of technological, organizational and financial changes, such as capacity variations and electric current variations, on the strategic planning level within an aluminium manufacturing facility, their model takes into account the anode setting cycle. Optimizing the throughput of an ingot mill in an aluminium smelter is discussed by Nicholls (1994). Their MILP model optimizes the throughput under varying percentage time availability of furnaces, troughs, and casters. Other examples can be found in Nicholls (1995a), Nicholls (1995b), Nicholls (1997), Nicholls
3.5. Integration of Solution Techniques 57