Introduction

Metals support advances in society by enabling modern technologies that depend on improvements in metal properties (Lu 2010) for such applications. These improvements can be facilitated by compositional designs that could involve varying the chemical composition of already existing metal grades, i.e., product specification, that result in novel alloys (Springer and Raabe 2012), or adding smaller amounts of elements that address certain trade-offs in metal production, within an existing product specification (Han et al. 2021). The resulting material is generally referred to as metal, but considering that their properties are a result of a combination of different elements, the term alloy is more apt. Steel is a prime example of this; from the basic iron and carbon combination, this metal has evolved to include a wide array of alloyed steels (Ashby and Jones 2013). It is an alloy system in a “state of fast development” (Deschamps et al. 2018), where stainless steels exhibit a higher growth rate compared to carbon steels (ISSF 2021). The variety of steel grades continue to expand, with one estimate stating that over 75% of the 3500 steel grades in currently in use did not exist 20 years prior (Worldsteel 2020). Steel maintains its position as an essential material under the sustainability agenda, due to its use in low carbon technologies (The World Bank 2017) as well as being fundamentally recyclable (ESTEP 2021).

This makes recycling very attractive for the steel industry because scrap is generally cheaper than ore-based primary production and requires a lower level of energy input that result in reductions in greenhouse gas emissions (Worldsteel 2020). It also lessens the burden on extracting ores for iron and alloying elements (Manabe et al. 2019). Overall, steel is the most recycled metal in the world (Worldsteel 2018; ISRI 2020).

However, as the newer grades continue to enter society, get used, and then, discarded, the recycling systems are unable to keep corresponding end-of-life streams unmixed. This is because the identity of the steel, more specifically the chemical or compositional information, diminishes as it moves along the value chain (Allwood et al. 2011). A reverse loop infrastructure that cannot manage the complexity of scrapped material contributes to an overall accumulation of contaminated scrap as observed by Dworak and Fellner (2021). It leads to nonfunctional recycling (Diener and Tillman 2015) where some alloying substances either accumulate in the wrong metal grade as tramp elements (Oda et al. 2010; Björkman and Samuelsson 2014) or become progressively diluted and lose their commercial viability for future recovery (Miranda et al. 2019).

This is a prevalent problem for steelmakers that use post-consumer scraps as production inputs. Obsolete scrap comes with inconsistencies (Pflaum 1989) that diminish the advantages of using scrap because there is insufficient knowledge of what exactly is in the scrap to be melted. Even with established scrap standards, there is a mismatch between the many steel grades as previously mentioned and the limited number of scrap categories the standards cover. For example, 11 categories are used within the EU (EFR 2007). Even a “more extensive” industry specification as the one being used in Sweden (JBF AB 2020) will still be insufficient to provide the required segmentation of scrap streams necessary to avoid losses brought about by this lack of information.

In order show that additional information can be beneficial, this study takes the concept of the expected value of perfect information (Szaniawski 1967) and reformulates it to answer the research question:

  • What possible value does perfect compositional information bring to scrap-based steel production?

The value of information analysis (Rothery et al. 2020) provides the means to “quantify the value of acquiring additional information to reduce uncertainty in decision-making” and is used in various fields, from agriculture to medical (Keisler et al. 2014). This paper sets out to associate value, primarily the economic value, with the effects of the lack of information about variation in scrap chemistry. It will methodically show how these variations in scrap chemistry, when neglected, can represent both real costs due to the need to add primary alloying elements, and additionally in terms of alloying excesses when exceeding product specifications.

Literature review

Lack of information and uncertainty in the recycling system

One restriction in increasing scrap usage in recycling is usually framed as an issue of tramp elements contained in scrap that could adversely affect the quality of the resulting steel (Spooner et al. 2020; Dworak et al. 2022). However, it is also a reasonable view that it is not the presence of these contaminants per se, but the uncertainty on the actual content of each element contained in the scrap portion of the feed to the furnace. Getting the blend of input materials correctly (Bernatzki et al. 1998) would mean that steelmakers can avoid using additional resources to deal with the discrepancy between what was fed and what comes out. The idea of the “perfect” scrap can be described as a quantity of scrap with all the desired content, accompanied by complete knowledge of its chemistry. In the context of scrap that is classified as home, prompt, and obsolete (Carlson and Gow 1936; OECD 2012; ECSIP Consortium 2013), home scrap is basically perfect for the steel mill to reuse because they have total information of its contents. The same could be said of prompt scrap, especially if there exists a mechanism for this material to return directly to the steel mill from their immediate customers (Sandvik 2018; Ovako 2019), whereas obsolete scrap is typically more heterogenous with more variety and uncertainty due to mixing of different material streams (e.g., Compañero et al. 2021).

The level of what is known regarding the compositional information of scrap changes depending on which stage it is generated along the value chain. The outlook on steel production is that the availability of prompt, i.e., pre-consumer scrap whose chemical identity is rather intact when it returns to the steel mill, will decrease because companies may target waste reduction (Milford et al. 2011). That will then leave obsolete scrap (i.e., post-consumer) as the major scrap stock that can be bought externally. With obsolete scrap comes the challenge of utilization because of the need to deal with uncertainty in scrap composition. The information gap is a consequence of the considerable time that the steel products spends in use (Ruth 2004) and also the system of its collection and delivery back to the steelmaker. Scrap dealers (i.e., actors involved in scrap management that deliver to steelmakers) manage the complexity of having various sources, irregular timing, and limited capacity (Compañero et al. 2021). Characterizing and sorting a large quantity of material composed of numerous grades into a limited number of standard categories will result in scrap piles that convey only partial information regarding their true scrap chemistry.

In general, the current steel recycling system works with limited information about its most important input. The compositional information that reaches the steelmaker is lacking, but not entirely missing. The actual composition of what is loaded in the furnace is eventually revealed by analyzing a sample from the melt (Fergus 2000; De Saro et al. 2005). But any discrepancy in the resulting composition can only be dealt with via the use of additional resources; otherwise, the heat will fail to meet customer requirements with respect to the specific steel grade (Rong and Lahdelma 2008).

Scrap standards are used to facilitate transactions between scrap sellers and buyers (ISRI 2020) and elements that usually impact production are indicated as tolerance limits, as maximum allowed, or that an analysis should be secured prior to delivery (EFR 2007; JBF AB 2020). These standards are also seen as a response to supply-side inefficiencies related to a lack of information (Söderholm 2021; Tillväxtanalys 2020). Knowing what will end up in the melt helps in controlling its chemistry, which is important in obtaining a quality product (Brooks and Subagyo 2002). The higher accuracy regarding the real scrap composition, the lesser the uncertainty that needs to be accounted for in charging the furnace. Moreover, Matino et al. (2017) found that using a large quantity of low quality scrap (i.e., scrap with a high impurity content) can even affect the energy and environmental impact of the process. Therefore, more information can be mutually beneficial for scrap dealers and steelmakers (Panasiuk et al. 2022) and helps in increasing the amount of scrap that will be acceptable for production. The compositional information is an important feature of the physical materials that needs to flow at a high level between actors for a well-functioning circular economy (Nordic Innovation 2021).

Dealing with uncertainty as a decision to be made: the expected value of perfect information (EVPI)

Technical and systemic contamination (Baxter et al. 2017) creates a practical problem that manifests at the point of loading the furnace. Making an optimal furnace charge of scrap becomes challenging because of the uncertainty in composition and whether the components are available in the correct amounts and will lead to a successful heat (Rong and Lahdelma 2008). A decrease in this uncertainty leads to an improvement in furnace charging, i.e., achieving the benefits related to getting a correct recipe. Furthermore, it will be useful to compare the current state of operating, which only has access to partial information, with an idealized one where full or perfect information is available.

It must be recognized that the current scrap-based steelmaking practice can operate fairly well even with only access to partial information. If the outcome of the initial material loaded needs correction after melting, another step (i.e., refining unit or furnace) receives the melt and provides the pathway for adjusting the composition (Holappa 2014). The steelmaking scheme has operated well enough until now, which is also a possible reason why there is a general lack of investigations in attributing value to compositional information. However, these corrections require purer raw materials, mostly ore-based primary materials, and therefore cannot be considered resource efficient in some contexts (Huysman et al. 2015).

The question remains regarding what a recycling system could manage to do as a result of having access to more information. What is implicit in solutions that point to improved identification and scrap sorting (Bell et al. 2006; Graedel et al. 2022) is the recovery of material information to enable an allocation of raw materials to improve the production. Such an intervention requires investing in additional personnel and an upgrade of one, or a combination of, technologies. These types of investments then need proof of their ability to provide value (Li et al. 2011) to the totality of the recycler’s enterprise.

Again, the circumstances of making recipes, or a blend, of input materials to be loaded in the furnace, with partial or full information can be considered as deciding if additional information is worth it. It is a decision that clearly impacts scrap usage in recycling. This falls under the domain of decision analysis, which is according to Koenig (1998):

“a methodology for calculating the tradeoff between making a decision with partial information versus expending resources to obtain more information and thereby making a decision based on more complete information”

Decision analysis is useful for evaluating complex decisions that involve “multiple (and usually conflicting) objectives and uncertainty,” with the resulting benefit sometimes referred to as the expected value of perfect information (EVPI) (Parnell 2009). Szaniawski (1967), who is attributed with the first conceptualization of EVPI, wrote that information “provides a basis for better solutions of decision problems” and in relation to this work, the value of information is the “highest price (in utility) to be paid for the information, compatible with the condition that the best use made of perfect information is at least as good as any action previously available.” Further developments in the study of perfect information were made based on this work (Boncompte 2018). EVPI is also seen directly as the expected increase in profitability (Avriel and Williams 1970).

Using the concept of EVPI, the research question can then be framed this way: is there a possible improvement in the outcome if the level of compositional information is full or perfect (i.e., uncertainty on what is being loaded in the furnace is lower)? The resulting EVPI is then an estimate of how much it is worth investing in more information.

It is interesting to approach the uncertainty problem in steel recycling, and possibly of other materials, as a decision that is taken by recycling actors. They also behave typically of profit-making firms and would like to be price/cost efficient (Carlsson 1972). Looking at activities associated with recycling (see Table 1), recovering material information is already embedded because these actors need it to do business effectively. This means that there is a prevailing and acceptable degree of information that these actors are obtaining now. For example, analysis-secured scrap categories are present in some industry standards (e.g., Baillet 1995; JBF AB 2020) to certify scrap deliveries. The willingness to do more depends on understanding what the benefits are. This is rarely studied in literature. However, what is seen is a focus on using programming or the application of mathematical models for selecting appropriate ingredients (Aggarwal and Gupta 2014) and/or blending of raw materials (Sakallı and Baykoç 2011; Gyllenram and Westerberg 2016).

Table 1 Information relating to the scrap and what it might contain is recovered in end-of-life management of scrap

Current approach on scrap mix optimization

The two levels of information already described are available to steelmakers. First, partial information on steel scrap is captured in industry-established documents. Regional (EFR 2007; ISRI 2018) and country-specific industry standards such as Sweden’s skrotboken (JBF AB 2020) are examples of these types of documents. Additionally, characterization technologies (equipment) and standards (protocols) are used. Second, perfect information is also available but only after melting when all the input materials are in a homogenous state. At this stage, the actions available to the steelmaker will be limited; primarily in terms of how much material can be added without exceeding limitations in the production system, i.e., heat weight and lot sizes.

Aggarwal and Gupta (2014) discussed the use of mathematical tools to aid production managers to optimize their resource allocations under various constraints. “If wrong decisions are taken, it will result in losses,” according to them. Predictive models make use of mixing rules that define combinations of scrap feedstocks to satisfy customer requirements at the lowest production cost (Miletic et al. 2008). However, all models need reliable data to produce reliable results.

Studies on blend optimization point to the need for knowledge of scrap chemistry (Miletic et al. 2008; Mombelli et al. 2021), although the overall idea is to improve production performance rather than addressing the issue of valuable elements dissipating. Preventing these from becoming tramp elements should also be prioritized. This makes the approach used in this study different. While it still covers the uncertainty in scrap as input, the operational concern is not how to plan around it. Rather, to show that operating with perfect information can fulfill both economic and environmental incentives that at times are thought of as conflicting (Andersen 2007) and a barrier to Circular Economy (CE).

Methodology

To achieve the comparison between the two levels of information, the steel recycling process is scaled back as primarily an operation of melting and refining of existing material. This allowed for the premise of the study to take form as a decision on how steelmakers deal with uncertainties in terms of scrap compositions. The uncertainty in the present work is an interpretation of the compositional tolerance limits published in material standards. Comparing partial and full information scenarios gives an insight to what value access to additional information can bring. In this work’s context, partial or imperfect information refers to only knowing that the scrap is within the tolerance limits that are indicated in the scrap standard. On the other hand, full or perfect information refers to knowing the exact composition before it is processed in the furnace. While not meant to be a comprehensive decision analysis, the model and resulting data nevertheless increases the understanding with respect to EVPI, or simply acquiring additional information, in the context of recycling.

Figure 1 lays out the conceptual flow in detail. If the compositional information about the input scrap for production is perfect (full information), meaning that the exact amounts of the elements are already known prior to melting, an opportunity arises to directly combine input materials to match the target product composition. This was compared to a situation of furnace loading with a degree of uncertainty (partial information). In the latter, one standard recipe would have to be used to deal with the range of all possible values of each element contained in the scrap. Subsequently, the composition is adjusted to meet the target composition in the refining unit, by adding pure Fe, Cr, and Ni. The following subsections explain important components of the model implementation.

Fig. 1
figure 1

A flowchart that shows how the idea of partial versus full information was operationalized in the study

Scrap, product, and alloying elements

A scrap grade that is the end-of-life counterpart of a common steel grade was used to ensure that the necessary data for both the scrap input and target product, such as specifications and raw material prices are available. The stainless steel grade AISI 304 was selected as the target steel product, and its corresponding chemistry was sourced from Euro Inox (2007). Choosing a stainless grade to serve as a model material instead of an unalloyed one was made for this study with the purpose of including valuable, and thus costly, elements such as Cr and especially Ni. These two main alloying elements in stainless steel represent substantial values in use and any excess of either element would represent excess costs. This will emphasize the effects of even small percentage differences for these two components in the results. Furthermore, all other elements except for Fe and C were excluded in this model. The scrap chemistry was obtained from the Swedish scrap book (JBF AB 2020) (i.e., 951 styckeskrot is the scrap category of stainless steel AISI 304).

The compositional limits of elements for both scrap and product are summarized at the top of Table 2. A spreadsheet program was used to first simultaneously generate random Cr, Ni, and C data, assuming a normal distribution within the specified boundaries. Then, the ten scrap deliveries were randomly selected from this lot via the same spreadsheet. The resulting variations were intended to exemplify deliveries from different scrap supply streams (e.g., producer, origin, and date of manufacture), but of the same category. See Table 2 for the composition of each delivery.

Table 2 Generated data for scrap were labeled as deliveries and imported to the software. The values were randomly generated and only the main elements of interest were included

For the target composition of the steel melt, the setting in the simulation software was aimed at the lower limit (i.e., 17.5 wt% Cr–8 wt% Ni). This aligns with the economic view of a steelmaker of a favorable outcome in that no excess of expensive alloying elements will be used up in the final product. As for the composition of the pure raw materials, Fe, Cr, and Ni were assigned 100 wt%. The raw material prices were based on discussions with an experienced industry practitioner during the first quarter of 2022. This gives a snapshot of prices that are congruent with the time period the simulations were performed.

Steelmaking simulations

Both the product and scrap data as described from the previous section were imported into the Rawmatmix® (RMM) software to make simulations of an idealized steelmaking operation as outlined in Fig. 1. RMM is a web-based, multi-purpose optimization tool using linear programming methods and process models for cost minimization and greenhouse gas emission estimations as the major objectives. It can be used for charge optimization, value in use calculations, and to assess the impact of changes in raw material properties and prices (Gyllenram and Westerberg 2016; Gyllenram et al. 2021). The execution of the melting-refining portion of the model was implemented in this software environment.

The steelmaking model was comprised of two stages. It starts with the melting unit followed by a refining unit. The former corresponds to an electric arc furnace (EAF) or an induction furnace while the latter stood for an argon-oxygen decarburization (AOD) vessel and/or ladle furnace in actual operations. The latter will function as the correction step to reach a desired melt composition. With a focus set on raw materials, the simulations were made without particular attention to process parameters, meaning that the settings of the furnaces in the software were unchanged. However, the distribution factors were modified such that all the elements report to the steel melt and nothing was lost to the slag or gas phase. This simplified the stainless steelmaking route comprised of the EAF, AOD, and ladle furnaces into the two-stage melting and refining process as described earlier. As a consequence, the calculated production costs consist of raw material costs and energy costs. Other details related to changing the composition of the melt, meaning aspects such as slag removal and decarburization, were deferred to a possible future extension of this work.

In the software, the same price was assigned to the scrap deliveries, as these were supposed to be of the same category and bought without accounting for the ingrained compositional variation. The recipe development is described in more detail in the “Creating scenarios based on the conceptual model” section.

Creating scenarios based on the conceptual model

Three production scenarios were tested to investigate the potential value of perfect compositional information about the scrap. The crucial aspect in creating these scenarios is to simulate the decision-making at the melt shop. In particular, the consequences are brought about by the uncertainties in the scrap composition during furnace loading. This will be reflected in the recipes made when having access to partial or full information for the production heats. The term heat is used to indicate the simulated production of a single batch of at least one-hundred-ton steel melt.

In total, three scenarios with increasing complexities were designed to investigate the value of having access to perfect information. The first scenario considers only target composition. In the second, volume constraints are added, and in the final scenario, scrap availability is considered over 10 consecutive heats, instead of individual heats.

Scenario 1: constraint with respect to the target composition

The first scenario began with only a constraint that the steel melt should meet the desired product specification. The standard recipe based on partial information was straightforward. Each delivery as received was assumed to be of the right composition (i.e., steelmaker receives AISI 304 scrap and uses it directly to produce AISI 304 steel), and the simulated steel production was implemented by loading the melting unit with one hundred tons of a single scrap delivery to produce the steel melt. The program was restricted to adding pure alloys, without any limit on quantity, only to the refining unit to correct for any compositional deviation. Ten heats were made to correspond to each of the ten deliveries.

Next, the whole production schedule was repeated for another set of heats in order to simulate production based on full information. Pure alloy additions were now already allowed at the melting unit and the simulation software found and used an optimized recipe for each heat.

Scenario 2: constraint with respect to the target composition and volume

In this scenario, the amount of material that can be added to the refining unit was limited to a maximum of 5 tons. This is more in line with reality where steelmakers normally have limits on the amount of additional material that can be added, especially in the refining unit.

For recipe development, one can perceive from the compositional limits of both product and scrap (see Table 3) that a problematic situation will arise with max Cr and max Ni content in terms of getting the melt composition within the tolerance limits of the steel grade intended to be produced.

Table 3 Possible analyses of scrap deliveries at the specification limits

A scenario was then conceived where the steel mill strives to limit the amount of alloy additions after the melting unit. For a standard recipe created with access to partial information, the goal is to preemptively address the possibility that the delivered scrap containing excess Ni is at its maximum limit by diluting it with Fe to reach the product’s upper limit of 10.5 wt%. This was achieved by combining 81 tons of scrap and 19 tons of pure Fe as input to the melting unit. However, this also resulted in Cr being diluted below the product lower limit of 17.5 wt%, which in turn required the addition of more than 5 tons of material to the refining unit. This situation particularly affects deliveries having relatively low-Cr contents. Thus, the final recipe required a portion of the 19-ton Fe addition to be changed to Cr. The final standard recipe consisted of 81 tons of scrap, 16 tons of Fe, and 3 tons of Cr loaded to the melting unit, to fulfill both the Cr and Ni content requirements in the steel melt.

Again, by having access to full information, more precise adjustments to the recipe for each heat can be made. Pure raw material additions will be added in the melting unit instead. Therefore, the simulated, full information results in this scenario should give the same results as in scenario 1, full information.

Scenario 3: constraint on target composition, volume, and scrap availability

The final scenario adds scrap availability as a constraint. In a production setting, this can be thought of as the situation wherein material can be drawn from different scrap piles to balance the composition. This meant that any quantity of material used in one heat becomes unavailable for the other heats. Therefore, this final scenario was treated as a production schedule.

The standard recipe for partial information is the same as in scenario 2. As a consequence of this, it should be noted that simulation results with partial information in this scenario will give the same results as scenario 2, partial information.

Results

The simulation output from RMM is summarized in Table 4 to give a quick overview of the results before addressing the specifics of scenarios 1–3 that can be found in the “Scenario 1,” “Scenario 2,” and “Scenario 3” sections. Individual scenario results will also be presented in more detail in the following subsections. The full information results of scenarios 1 and 2 lead to the same results as the program was allowed, without a limit on availability, to use as much pure material in the melting unit as possible. For scenario 3, combining different scrap deliveries in order to maximize scrap usage was assumed to be possible only with perfect information.

Table 4 Relevant production results for all simulations. The refining unit is not used if full information is available. Therefore, the raw melt and alloyed melt cots will have the same value

For each additional constraint in the scenarios, and in line with the level of information, the production costs and material usage characteristics evolved. Production costs, excess costs, and excess melt are consistently lower when full information is available. Detailed information for all heats in all three scenarios is covered in the “Scenario 1,” “Scenario 2,” and “Scenario 3” sections.

In addition to the actual costs attributed to material, energy, and production costs, a recurring item was introduced in the simulation software that was designated as an excess cost in the reported results. For each heat, it was calculated as the monetary cost equivalent of Cr and Ni that end up in the final melt, but already in excess of the minimum product requirement. The steel product does not become any better due to these excess amounts, and the potential loss is that the surplus amount of these alloying elements could have been used in other products instead.

Utilizing the refining unit in the model became necessary only in the recipes made with partial information for attaining the desired steel melt composition. Consequently, the results show that with full information, there is almost no excess cost. The main reason behind this is because the Cr and Ni contents which are already present in the scrap are maximized. Thus, exactly the right amount of pure alloying components is added to fulfill the lower end of the product requirement. This situation is certainly possible with partial information, but it can be considered a lucky outcome rather than a deliberate result of action.

The results are clearly influenced by the current situation of raw material pricing. Notice that the price (0.45 €/kg) for “pure” Fe is only 20% of the high-alloy AISI 304 scrap priced at 2.18 €/kg. The scrap’s value can be attributed to its Cr and Ni contents. The effect of this price differential is that the program, having an available low-cost diluting material as “pure” Fe, favored its use as it optimizes the cost. Thus, even if the results seem to show that the scrap usage is less when having access to full information, it will shift depending on the prevailing scrap price in the market.

Scenario 1

The results of the simulation which assumed that only partial information is available are shown in Table 5. The raw melt costs are consistent across all heats since the same quantity of scrap was loaded in the melting unit. Without any limitation on the quantity of material that can be added in the refining unit, almost half of the simulated melts (1, 5, 8, and 9) used significant amounts of pure materials, requiring between 19 to 30 tons to acquire a correct composition. As for heats 3 and 7, these only needed scrap since the scrap chemistries of the deliveries are already a close match to that of the product. However, the overshot of Cr and Ni contents still resulted in excess costs.

Table 5 Simulation results for scenario 1 with partial information

The melt program used a recipe created with partial information for the rest of the heats and the following trends were observed:

  • There were excess costs for each heat. Most of the cost is related to the use of excess Ni content due to its price; around 95% on average can be attributed to it. The excess Cr content also contributed in some heats, but the average excess cost was 19 €/ton. It must be noted that the assumption of a normal distribution within the tolerance limits given by the scrap specification gave the deliveries an average Ni content of 10 wt%. This value already exceeded the minimum product requirement of 8 wt%. Therefore, no Ni addition was required for any of these heats. Nonetheless, it calls attention to the effect of disregarding the unique compositional identity of deliveries as it will be materially costly.

  • Correcting composition requires additional resources. The refining unit in the melting program was mainly used to achieve a final melt composition that fulfills the product requirement. It functions as the last step for correcting the melt chemistry via alloying with pure materials. Operating the refining unit will also require energy input.

  • Pure iron price has a large effect on the costs. When the price of pure iron is much less than the scrap and there is no limit with respect to the amount of material additions in the refining unit, the alloyed melt unit cost ends up being cheaper than the raw melt unit cost. This unusual effect is because of the increase in the final melt quantity due to the pure material additions resulting in total, the use of more material. In a real operating meltshop, there would be restrictions on the total melt weight. Combine that with the relatively low corresponding cost of the added material and the immediate effect as seen here is that the alloyed melt cost per ton will seem cheaper than just the melted 100 tons of scrap. This is seen in all heats where additional material was added in the refining unit. This “side effect” especially affected heats 1, 5, 8, and 9 where between 19 and up to 30 tons of total material were added; majority (20 tons) of that is pure iron, and approximately 5 tons more of Cr. The same phenomenon can be seen in heats 2, 4, 6, and 10, although to a lesser degree.

  • Excess costs are mostly due to Ni. The high price of Ni was expected to impact this value depending on how much extra Ni that is added to the heat. The influence of Ni is higher than that of Fe or Cr. This can also be observed in Table 5. Heat 4 has the lowest excess cost owing to its already low Ni content. The other heats with relatively lower excess costs compared to the rest are heats 5 and 9, which both used scrap with Ni contents greater than 11.5 wt%. However, due to a high level of dilution, the excess costs did not exceed 200 €/ton.

The reader should be reminded that the results are also a consequence of how these specific scenarios were set up and that the interpretation of the results should be done in a careful manner. For example, exceeding 100 tons in the final melt is acceptable in some cases but reaching up to almost 30 tons additional molten metal, as in heat 5, is unlikely to happen during actual production conditions.

For the results of the full information approach in this scenario, the details can be seen in Table 6 where the recipe is now adapted specifically depending on the composition of the delivery. The raw melt and alloyed melt costs would have the same value in this case.

Table 6 Simulation results for scenario 1 with full information

It is noticeable that, taken on average, the raw melt cost for the heats was 15% cheaper compared to using only scrap (partial information). Again, this is due to the raw material prices as the model simply dilutes each heat to reach the bare minimum requirements for the Cr and Ni alloy contents in the final melt. The scrap usage average is around 73 tons. However, it can be as low as 63 tons as in heat 1, which had the highest Ni content. No Ni addition was required in any recipe because all deliveries already met the minimum threshold values for the product.

Scenario 2

The previous section showed simulated production results, where only compositional constraints were considered. However, volume constraints (i.e., quantity that can be added with respect to the furnace capacities) are also highly important. Notably, some heats in scenario 1 (heats 1, 4, 5, and 9) showed vast material additions to the refining unit just to meet product specifications. In practice, this results in a product surplus without any customer demand and no consideration of physical limitations to the heat size. The recipe that was developed (81:16:3 proportions of scrap-pure Fe-pure Cr) should also be able to address the identified worst-case scenario as described in the methodology.

The results with partial information summarized in Table 7 showed two important things. Firstly, there was a substantial decrease in the amount of pure alloying additions to the refining unit compared to the partial information state in scenario 1. Secondly, the production quantities look more regulated. In fact, only four heats (1, 4, 5, and 9) required adjustments in the refining unit. Alloyed melt unit cost for these four heats is, of course, higher and quite notable for heat 4 due to the use of more than a ton of Ni. This underlines the issue again when operating a process based on partial information; a generalized recipe is supposed to address the probability of having max Ni in the scrap. But in one heat (using scrap delivery 4), additional Ni was still required after melting because the actual content in delivery, 8.53 wt%, is much closer to the product minimum of 8 wt%. As for the rest of the heats, the use of the 81–16-3 material blending scheme did achieve the required chemistry specifications of the final melt.

Table 7 Simulation results for scenario 2 with partial information

It should be noted that simulation results with full information in this scenario will give the same results as scenario 1, full information. In this scenario, costs are reduced by 10% on average compared to having partial information. Nevertheless, the blending of scrap with a proportion of 19 tons of pure material resulted in a decrease in excess costs, up to 40% lower, compared to that of the partial information case in scenario 1. These positive effects from adding a constraint comes from the relatively low cost of pure Fe and high Ni content in the scrap. Compared to heat 4, where Ni was low, the cost reduction with full information was as much as 20%.

The raw melt costs in this partial information scenario are lower compared to those of scenario 1, due to the relatively cheap Fe and Cr units loaded at the melting unit compared to the price of the scrap. Consider that the material cost for 100 tons of entirely scrap loading (218 k€/heat) compared to 100 tons of blended loading (194 k€/ heat). In addition, less material was also needed in the refining unit. Indeed, there was a slight increase when looking at the average alloyed melt cost, but the numbers are coherent when comparing the results to the partial information in scenario 1.

Here, a trend can be seen that the scrap usage decreases as the scenarios develop. This is a result of how the simulation objective of minimizing the cost, targeting the bare minimum requirement for Cr and Ni content in the final product and the availability of low cost “pure” Fe for dilution.

Scenario 3

In this final scenario, all the scrap deliveries corresponding to 1 000 tons were simultaneously made available for use in each heat. For partial information, the results are the same as in scenario 2, partial information. But with access to full information, there is a higher degree of freedom to systematically choose and combine proportions of different scrap deliveries together with pure alloying elements to round off 100 tons. The target was still to produce ten individual heats of 100 tons each in stainless steel grade 304 that uses as much scrap as possible, but subject to the underlying intent to minimize costs in each heat. Table 8 shows the completed melt program.

Table 8 Simulation results for scenario 3 with full information

It is noticeable that deliveries 3, 4, and 7 were not consumed due to their respective Ni contents, which were the lowest among all the deliveries. The RMM program proceeded to use those with high Ni contents first, because of the availability of low-cost Fe for dilution. This is also a consequence of the model implementation with one kind of scrap together with one type of steel product. It sequentially diluted those with higher Ni contents, bringing its level in the melt composition to the minimum values and reserving the ones with similar compositions to the product for future heats. By using cheaper Fe and Cr in the recipe, as already seen in the “Scenario 2” section, the average unit melt cost went down to 1 867 €/ton. This is the lowest among all studied scenarios. In addition, having an excess use of valuable alloys in the heat was avoided. Thus, the excess costs were reduced to almost zero.

It must be pointed out that if a similar product but with a higher Ni requirement was part of the simulation, then the program would distribute the scrap accordingly. The results show that having access to the full information on the composition of scrap will enable production planning in not only targeting the lowest costs but also ensuring the suitability of the scrap for melting into products where their individual alloying components are functionally retained.

There was no additional material added in the refining unit. However, the overall scrap usage was the lowest among all studied scenarios. This is again due to the high Ni content of the deliveries coupled with the cheap price of alloying Fe which is readily available.

Discussion

Making the most of the available raw materials is essential for steel producers, especially since material inputs can comprise as much as 70% of the cost structure in an EAF production (Renda et al. 2013; Medarac et al. 2020). The results from all three scenarios show a significant potential for savings when information is fully available.

The concept of perfect information and the outcome of the comparison gave a new perspective when studying steel recycling and highlighted the importance of information in recycling. The use of EVPI gives an indication of how much additional resources is worth putting into obtaining more information through, e.g., improved sorting and characterization. As shown by Compañero et al. (2021), improvements in sorting and characterization are limited by the willingness to pay for such activities. The simulations in this paper show that the EVPI estimates can be useful in this setting, and it also provides rough estimates in a simplified environment. Future models should include more materials and products. Scenario 3 has shown that the value of information increases with increased model complexity, rather than the opposite.

The software used in this study was set to minimize costs. This is in line with current industrial praxis. If the relative cost of primary raw materials increases or if the software is set to maximize scrap usage instead, the results will change in favor of higher scrap usage. For scenario 3 in this paper, 100% scrap utilization is possible if perfect information is available. With only partial information, a maximum of 81% was possible. These results are in line with, e.g., Koenig (1998).

Suggestions for future work and concluding remarks

This work was an initial undertaking to look at the value of information in scrap-based steelmaking. A reasonable progression in future investigations would be to increase the number of scrap classes and stainless steel grades. For example, the AISI 316 stainless steel grade would be a good candidate to investigate along with the AISI 304 grade. It also has its own scrap category in Sweden (962 styckeskrot), similar in chemistry to 951 but with the additional specification on molybdenum, another alloying element that is even more expensive than Ni. Some of the simplifications made in the model set up should be revised, such as changing the distribution factors for metals of interest to simulate its behavior in the melt (e.g., Cr also reports to the slag phase) and including tramp elements (e.g., Cu). There are also factors that affect raw material costs that were excluded for simplification purpose such as that different supplier deliveries having differences in pricing depending on the amount of alloying content.

The requirements and methods to establish the exactness of the bulk chemistry of scrap loaded into a furnace is perhaps a subject for a different type of study.

This study concludes with an emphasis on its contributions:

  1. i.

    A model was created that appraises the effects of the level of what is known about the composition of the scrap. Variations are inherent in obsolete scrap even if they are classified under the same category. Uncertainty affects scrap usage and inhibits the presumed economic and environmental advantages for the scrap-based steelmaker.

  2. ii.

    The results emphasized the need for upgrading the characterization and sorting infrastructure. Stating this need is nothing new but using the concept of EVPI to indicate economic potential is one distinct contribution to substantiate this need. The study has clearly shown that both material and the corresponding information flows influence scrap-based production of steel.

  3. iii.

    An interdisciplinary view of material production (steelmaking) systems is advantageous in that the solution space that is explored is reasonably constrained but can still provide relevant information for decision-making in different units of the firm other than production. For example, in the purchasing department and raw material handling.