Introduction

On the surface, many production processes can be represented as concise and apparently deterministic mathematical models. Consequently, as deterministic and structured problems, they can normally be readily solved using an appropriate algorithm that results in a precise and hopefully optimal solution.

In many circumstances, these 'hard' models contain parameters that appear well defined and quantifiable but are in reality at best 'guestimates', subject to complex sub-processes that determine their values. Often, these sub-processes are not fully understood despite the fact that they form a part of the 'known' production process. Theoretically, these parameters can be specified, but unfortunately, it is not always possible to measure or quantify with any accuracy their component values due to the nature of the production process (ie, physically 'hostile'). Additionally, due to the lack of 'real-time' 'online' understanding of how some of the sub-processes actually work, the rectification of day-to-day production problems by factory floor workers is undertaken using 'experience' (a kind of alchemy). These two aspects of sub-processes based substantially on tacit knowledge (hereinafter referred to as 'the tacit knowledge problem') are typified by the aluminium smelting process. The resolution (or at least better management) of the tacit knowledge problem has considerable rewards for smelter owners.

The reader is invited to imagine a scenario that will later be laid out and discussed more thoroughly: it is the scene of a pot room in an aluminium smelter, slightly misty, very warm near the pot with a slightly acrid smell and extremely dangerous with 'white' molten aluminium lying beneath the often red hot crust in the pot. There are massive electrical currents in play (enough to vaporize a person) and at each end of one particular pot, molten aluminium begins to spill out on the pot room floor (due to a wave effect that has developed in the pot). The molten aluminium on the pot room floor is also 'live' with the electrical current. The pot room operators, suited and totally masked against the pot room gases, can be seen placing long poles of pine into the pot to burn off gases; others are adding soda ash and yet others increasing the raw material input into the pot. The scene is not unlike that of medieval alchemists at the furnace.

This paper explores this scenario branding the expertise of the operators as 'alchemy'. Even today, much is not understood about some aspects of the operation of the smelting process. Knowing how and when to stabilize the fierce forces of the pot and to make it behave for best aluminium production is the challenge. Also, knowing how to include this knowledge into a full production model becomes the problem this paper seeks to address.

Understandably, the problems of incorporating the tacit knowledge (discussed above) are glossed over or ignored in much of the literature. The impact of these problems is twofold. Firstly, the nature of the critical parameters that originate from a sub-process that precludes their reliable measurement (as discussed above) together with the fact that they can vary in value quite significantly depending on the time frame of the model can result in vastly differing (and 'not too robust') solutions being obtained. Secondly, the not well-understood sub-processes are (as previously mentioned) 'managed' in the main by human operators on the 'factory floor' who use an 'alchemy'-based approach to dealing with them, that is, they rely on their own intuitive experience to deal with matters rather than any precise science. Consequently, depending on the operator's own highly individual actions, varying effects will be wrought on the sub-processes output (which in the aluminium smelting industry cannot be practically measured with any real degree of accuracy). This situation can result in the destabilization of the sub-process and result in its highly variable and unpredictable 'bad behaviour' over an extended time frame. There are additional problems also associated with the sub-process selected for examination in this paper, including the fact that accurate measurement of the output is not possible! These and other problems are dealt with later in the paper.

Recognizing these problems is part of their resolution. However, effort should (and often is) be directed to both obtaining greater understanding of the sub-process at a technical level (knowledge enhancement) and harnessing the explicit and implicit knowledge of the operators (knowledge perpetuation).

As a means of achieving the objectives mentioned above, it is suggested that the concept of a 'Community of Practice' within the sub-process provides an ideal approach. In the aluminium smelting industry dealt with in this paper, Communities of Practice (at the most informal and almost intangible levels) exist as 'brotherhoods' of alchemists in an almost unaware way, discussing, learning and debating their 'alchemy-based' approaches to managing the sub-processes. Communities of Practice are discussed in detail later in this paper. The level of awareness of the Communities of Practice is (at least from anecdotal observation associated with the smelter used in this paper) very small, but the impact of any greater collective understanding can have a significant impact on output and thus profit. It is suggested that by increasing the level of awareness of the existence of Communities of Practice within the workplace and encouraging their development, that the greatest resolution to the problems outlined above can be obtained. Some suggestions are made as to how this might be achieved.

In order to obtain best practice production solutions for the smelter, while recognizing the difficulties associated with the troublesome sub-process discussed above and also working towards a perpetuation, transfer and enhancement of the tacit knowledge of the operators, a mixed-mode modelling approach to obtaining best practice is employed. Figure 1 highlights the structure of the problem, while Figure 3 illustrates the essential nature of the mixed-mode modelling approach used for obtaining a continuously improving best practice solution for the operation of the overall plant. Mixed-model modelling (see Lehaney and Clarke, 1997; Nicholls et al, 2001 for details of this approach) is essentially the combination of hard and soft models of processes that are then solved in a macro sense using a soft heuristic to obtain best practice. The hard model in Figure 1 clearly pertains to the production or manufacturing process (part 1 of Figure 1), while the soft 'model' is the Community of Practice (part 2 of Figure 1) relating to the tacit knowledge problem. An overall mixed-mode modelling 'soft' heuristic arrives at 'best practice' solution by incorporating the previous two models (and their solutions (Part 3 of Figure 1)). The solution to the tacit knowledge problem (represented by the Community of Practice 'model' including the knowledge perpetuation processes) operates continuously throughout the year and has a significant input into the monthly or yearly model of the overall production process through improved production efficiencies and more accurate estimates of parameters. Much of this input is incorporated into the mathematical model of the smelter prior to determination of best practice. This best practice solution of the plant would be done on a monthly and/or yearly basis, and each time it is undertaken, it would (having incorporated the knowledge and process improvements) yield a better solution than otherwise would have resulted.

Figure 1.

Interaction of the Community of Practice with the production process and sub-process—a mixed-mode modelling approach.

Full figure and legend (95K)

Figure 3.

A mixed-mode modelling solution methodology for the smelter problem.

Full figure and legend (140K)

In order to show how the problems and their solutions are affected, this paper initially describes the operations of the smelter, outlines a mathematical model, identifies and discusses the difficulties associated with the tacit knowledge problem (and the implications for the hard model investigated), suggests a 'model' that encapsulates the tacit knowledge problem and then suggests how the objective of knowledge enhancement and perpetuation might be achieved. Finally, the overall problem is represented as a mixed-mode model and a solution heuristic suggested that leads to the determination of best practice.

Modelling an aluminium smelter—the 'hard' model

In this section, an outline of the operation of an aluminium smelter is provided and a simple model that represents the monthly macro operations of the smelter is given. This mathematical model represents the traditional 'hard' model usually employed to solve such problems. The difficulties associated with one of the sub-processes are then discussed.

In order to understand and appreciate the nature of the difficulties associated with the tacit knowledge problem already discussed, it is necessary to appreciate what the modelling per se of a smelter entails. The development of the mathematical model of the smelter has been extensively documented elsewhere. For an electrochemical model of the operations of the aluminium reduction cell and other processes associated with aluminium smelting, see Grjotheim et al (1977). For models of a smelter as a whole, see Nicholls and Hedditch (1993), Nicholls (1995). For the reader wanting a 'lay' representation, see Habashi (2002). The aluminium smelter discussed in this paper required around A$2 billion to construct and was commissioned in 1986 on a green-field site of some 250 acres. The mathematical model was initially developed to provide management and smelter supervisors with the ability to determine what the optimal operating values of the key variables of the plant should be, to evaluate capital expansion plans and alternative technologies, etc. The model was a monthly model that could be linked together to provide a yearly model.

The smelter is a value-added plant rather than a profit maximizing one, which together with the fact that the various business entities of the plant, known as Areas, are autonomous creates the need for a bi-level programming approach. The complicated production process associated with aluminium means also that a number of non-linearities are encountered. The production process per se comprises each of the Areas of the plant interacting in specific ways. The Anode Area manufactures the anodes that go into the reduction cells (pots) in the Pot Rooms Area (where the aluminium is produced), and is also responsible for the ordering and delivery of all raw materials. The Anode Area takes coke, pitch and small amounts of recycled anodes from the pot rooms (plus scrap from the Anode Area and the Rodding Area) and forms a 'green' paste which is then moulded into 'green' anodes (weighing a little over one tonne). The green anodes are then placed into a special furnace that over a period of two weeks bakes the anodes (thus forming the 'baked anode') ready to be passed on to the Rodding Area and eventually passed on to the pot rooms.

The Rodding Area takes the baked anodes and inserts a metal rod into the anode to allow it to be suspended in the pot (reduction cell in the pot rooms) where a current is passed through it. This is done by pouring molten 'pig iron' into the holes in the anode into which the rod is inserted. They are also responsible for the gathering of spent anodes (butts) and their reprocessing (ie, their grinding and transfer to the Anode Area for reuse). In the Pot Rooms Area (using the Hall–Heroult process), a current is passed through the suspended anodes (which have been passed on to the pot rooms from the Rodding Area) in the pots containing reacted alumina (alumina through which has been passed aluminium fluoride gas and a 'bath' comprising cryolite and soda ash) and the resultant aluminium is siphoned off into 10 tonne crucibles on a daily basis and despatched to the Ingot Mill. There are two pot rooms with in excess of 400 pots. Each pot holds approximately 10 tonnes of aluminium. There are 4 km of pots in the smelter (four 'rows' each of 1 km, two per pot room). Each pot has 32 one tonne anodes suspended in it and is approximately 10 m long and 4 m wide. Replenishment of the raw materials is done on a needs basis (mostly automatically) and the replacement of anodes is done on a regular basis. This is discussed below. The Ingot Mill receives the crucibles of molten aluminium from the pot rooms and tips their contents into holding furnaces where the aluminium is cooled from around 1000^compfnC to approximately 750^compfnC. The molten aluminium from the cooling furnaces is then tipped into 'casting' machines that take the liquid metal and make ingots which are then shipped around the world.

It should be noted that the modelling of a smelter and its constituent processes can be undertaken using a variety of time fames. The Pot Room Area especially can be viewed on a 'real time' 'online' basis through to a daily or monthly time frame. Very little work has been done on modelling pots on a 'real time' 'online' time frame with the main need, stemming from a planning perspective, being a monthly one.

The objectives of the smelter are to ensure that the maximum aluminium is produced and that the production process costs are at a minimum. The costs associated with the pot room are linear with respect to use of electric current (kA) averaged across all the pots, while those associated with the manufacture of anodes is inversely proportional to the length of time that anodes are left in pots (which is the same for all pots in the smelter). Figure 2 summarizes the aluminium smelting process. In terms of capital equipment and cost, the pot rooms are the most expensive production area followed by the Anode Manufacturing Area. The cost of electricity is the single most expensive aspect to smelting after the initial establishment of the capital equipment. In the Anode Area, the greatest costs are raw materials and gas (used in the baking of the green anodes).

Figure 2.

Production activity and process flows for the aluminium smelter.

Full figure and legend (72K)

A careful analysis of the operations of the plant revealed that the two main production activities or 'Areas of commonality of purpose' of the smelter are only indirectly linked by a material feedback loop and the necessity to produce at least the required amount of intermediate product (anodes). The Pot Rooms Area (aluminium production Area) has direct control over the quantity of aluminium produced while the Anode and Rodding Areas are in direct control of the quantity of Anodes produced (but must meet the Pot Room Area's minimum anode requirements). The Pot Rooms Area (incorporating the Ingot Mill which is directly controlled by the electric current also), the 'leader' Area, has no logical process control over the anode production per se other than indirectly specifying a minimum required number of anodes. The Anode and Rodding Areas are the 'followers' and are bound together by anode production. The setting cycle (SC) is the number of days that the anodes are left in (all) the pots before being replaced. The Anode and Rodding Areas determine this length of time. The Ingot Mill is directly driven by the output from the Pot Rooms Area. The Pot Rooms Area controls the output of aluminium by the amount of current it uses (kilo Amperes (kA) across all the pots). The Anode and Rodding Areas determine their production activities by the SC they choose. The longer the SC, the lower the production rate of anodes and the lesser the Anode and Rodding Areas' activities (hence the lower the cost). The output of aluminium is directly proportional to the current used (kA), while the production (and therefore cost) of producing anodes is inversely proportional to the SC used. The material feedback loop is comprised of the recycling of spent anodes from the Pot Room Area. The new anodes must contain spent anodes (within an upper and lower limit).

The model developed to represent the Portland Aluminium Smelter was as follows:

s.t.,

In model (1), the two variables are kA and the SC. As mentioned, the power input into the smelting process (kA) is proportionally related to the output of aluminium, and thus maximizing kA maximizes the output of aluminium. The kA must be greater than or equal to kA_l to prevent the pots from freezing. The aluminium production per se is independent from the production of anodes for use in the reduction cells except for some specific technical conditions and the recycling of used anodes. The larger the SC (ie the longer the anodes are left in all the pots), the fewer raw materials are consumed in making and dealing them. Thus, the objective is to maximize the SC (ie the length of time that anodes are left in all the pots) thereby minimizing the cost associated with this activity. The cost of making anodes is considerable and the process, apart from a technical connection, is separate from the pot rooms. In this formulation, the leader variable is kA and the follower variable SC. The SC must not exceed SC_u as determined by the kA_l. Note that all coefficients, vectors and variables are monthly averages across the entire smelter.

The first constraint group (1.1) is associated with the consumption of reduction cell-based resources (power, etc) and are linear. Here, the a₁ is a vector of reduction cell related and Ingot Mill coefficients relating to primarily the consumption of raw materials. These have been considerably simplified here as in reality they comprise in many instances complex algebraic expressions with up to 40 parameters. The vector b₁ is related to the availability of the Ingot Mill and Pot Room Area's raw materials. The second constraint group (1.2) is related to resource consumption (excluding recycled proportions of anodes from the reduction cells) associated with anode production and is reciprocally related to the SC. Here, a₂ is a vector of coefficients. The vector b₂ represents the availability of raw materials and capacities associated with the Anode and Rodding Areas. The third constraint sets an upper limit to the SC given the value of the kA currently in use and is a reciprocal constraint with respect to kA. This constraint owes its existence to the fact that a minimum proportion of the anode is required to be left at the time or replacement for technical and recycling reasons.

The fourth constraint is related to the consumption of petroleum coke in the production of anodes and is determined by the amount of recycled anodes available and the fixed proportion of pitch (the other ingredient required to make the anodes), and as such is a 'filler' raw material whose quantity is influenced by the kA and SC. If the kA is increased, the consumption of anodes increases also. Consequently, for a given period of time there will be less remaining of the anode to recycle. The longer the SC, the less anode is left at the time of replacement and therefore the more coke is needed; however, offsetting this is the fact that if the SC is large, then less anodes will need to be produced per month. This constraint may be treated as a concave quadratic. The fifth and six constraints related to the lower and upper limits permitted for recycled anode material in the new anodes. The actual proportion is influenced by both the SC and kA variables in a reciprocal and linear way, respectively). Both these constraints are convex quadratic. To this point, the model looks quite straightforward and deterministic.

The problems associated with a substantially tacit knowledge-based sub-process and its impact on the hard smelter model

In (1), every equation type except (1.2), involves the parameter 'current efficiency' (CE). CE represents the percentage of current (kA) consumed by a 'pot' that is actually utilized to produce aluminium. The higher the CE, the more efficient the smelting process is and thus the greater the output of aluminium (using a given amount of current). An increase of 1% in the CE could mean a multi-million dollars increase in profit (or in this smelter's case, reduction in cost) per annum.

Theoretically, the production of aluminium is calculated thus:

Estimated output is arrived as follows:

where:

In (2), (3) and (4) above, the terms are defined thus:

The difficulties associated with the current efficiency parameter (CE) can be classified as follows:

1. Lagged estimation. The CE that is used in the model representing the current period, is the one estimated from previous period's production operations. Thus at best, the CE is one period out of date. The estimation of CE is normally done over at least a month in order to avoid the considerable variations that can occur over shorter time frames. This is explained further in (ii).
2. Variability and unpredictability—operator knowledge The CE is an average concept over all pots. However, an individual pot's operations will be potentially highly variable and influenced for considerable time periods by various operator and non-operator factors that are themselves not predictable nor fully understood. Effectively, if the operators do not keep the pots under control as far as possible, the output for that time period will be lower than could be reasonably expected and consequently the CE will also be lower.
3. Non-measurability of actual aluminium production. The estimation of AS_t-1 and AS_t is very approximate, as it is not possible to know how much aluminium has been produced during each shift. This is due to the fact that the sides of the pot are not at a constant temperature, but vary considerably. This leads to a 'build up' of 'bath' (cryolite) around some of the sides of the pot and thus reduces the capacity for aluminium to form. Aluminium is produced and sits at the bottom of the pots. When the pots are tapped at the end of each shift, a given tonnage is removed, rather than the day's production. This is because there is no way of knowing from the height (ie depth) of the aluminium in the pot, whether the aluminium occupies the space or a combination of aluminium and side build-up that was formed during the shift.
4. Non-measurability of other useful parameters. Because of the hostile and destructive environment of a pot, many of the mechanical devices that are needed to measure temperatures and also ratios of various substances cannot exist for very long. This unfortunately makes the measurement of potentially useful indicators (such as electrolyte alumina content) impractical thereby reducing the information that might allow operators to more effectively control the pot. Urpani (1996, p 45) notes that there are many other important parameters whose measurement frame of reference varies from minutes to days. Operators really need to be able to use these to control the pot in a more effective way. Additionally, many of the observable parameters (such as pot temperature, voltage, etc are often affected by many other interdependent factors other than the ones that lead to the problems (such as anode effect) that the operators are attempting to control.

Of the four difficulties specified above, the most significant are (ii) and (iii). This is because in the case of (ii), operators can increase the output of aluminium by their actions and reduce the instability of the pots. Significant cost and profit implications reside here. In (iii), to some extent, the operators can influence the side build-up with their treatment of the pots. These difficulties are (as has already been mentioned) referred to as the tacit knowledge problem associated with sub-systems of a production process and are separate from the difficulties associated with the inability to know exactly what production is per time period. In the case of (iv), until technology advances sufficiently to provide appropriate monitoring devices (something that is gradually happening within the industry), nothing can be done.

The sub-process of operating pots and their efficient running and the resultant CE is not a small area of concern or importance. This is a sub-process at the heart of the smelter and is of major significance. It is in fact where the greatest secrecy lies within various smelting companies, as high CEs are their competitive edge and are worth millions of dollars.

In the literature, there are many articles concentrating on the objective of maximizing CE (usually theoretically based or using a laboratory pot). See, for example, Sterten (1988), Langon and Peyneau (1990), indirectly Desclaux (2001) and Tarcy (1995). These attempts at understanding the complex electrochemical reactions within a pot still do not yield solutions to the reality of managing pots in real-time and online on a day-to-day basis. There has been no real advance in the ability to accurately measure daily aluminium production. Additionally, where research has been conducted into the way in which operators could better manage pots, this is not generally published in detail as there are considerable commercial benefits to the company undertaking the research that gives them a competitive edge.

Research by Urpani (1996) using an object-oriented approach to determine if best practice by operators (with respect to managing pots) could be uncovered from operational data, revealed no immediately usable results. Rodrigo (1998) using a multidisciplinary modelling approach (incorporating Petri nets and Markov chains, etc) found a method of determining CE of a pot; however, the day-to-day use of this methodology across say 500 pots in a smelter, was not developed. In fact, as far back as 1947 due to Pearson and Waddington (1947), it was known that the CE could be estimated by using a ratio involving carbon monoxide and carbon dioxide. However, as always, outside the laboratory pot, the impracticality of this meant it was not a viable method (eg holes would have to be drilled in the anodes (32 per pot by say 400 plus pots) and the gases sampled).

Thus, even to this day, the real world 'smelter floor' management of pots is still to a significant extent 'alchemy', despite the amount of knowledge that has been assembled about the Hall–Heroult smelting process. These smelter floor (operator oriented) practices and problems are discussed briefly below.

The operators' role and complexities they face

From time to time, various indicators will alert the operator that a serious occurrence is about to happen within the pot. When the aluminium oxide content of the bath drops below a critical level (and it is not always easy to know when this has occurred), gas forms underneath the anode and an 'anode effect' commences. This can be stopped, reduced or alternatively exaggerated by the operator overriding the automatic feed of reacted alumina to the pot and/or the addition of soda ash and aluminium fluoride. Additionally, the gas build up can be burned-off by placing a wooden stake in the pot at the appropriate place. An increase in the pot's voltage and/or the appearance of the aluminium moving en masse as a wave in the pot (often spilling onto the pot-room floor) heralds the commencement of an anode effect. The CE will be significantly reduced by the anode effect and this could last for many days. Additionally, the CE can be unfavourably affected by power outages (ie interruptions to the supply of power from the supplier) as well as numerous other environmental factors. The operator's reaction to these occurrences is critical and will determine the performance of the pot for some time. There is still no agreement as to what the 'right' thing to do is (even by the experts) or when to do it for all the occurrences. So, the 'experience', intuitive judgement and anticipation by the operators is critical. Also note that the extent to which operators can actually 'do something' is quite limited, that is, they have only a relatively few actions that they might take depending on the 'trigger' (sudden high temperature of the pot, anode effect, voltage rise, etc). In summary, these actions are, overriding the automatic feeding of reacted alumina to the pot, adding soda ash and burning off gases under the anode. The effects of these actions can be either a worsening of the situation or the converse, its rectification.

'Modelling' the tacit knowledge problem using a Community of Practice

This tacit knowledge problem associated with the sub-process (pot operation) was (anecdotally) observed to be accompanied at the Portland Aluminium Smelter by an embryonic (ie existing) Community of Practice. Earlier in this paper, the role of Communities of Practice in knowledge perpetuation and enhancement was briefly discussed and it could be seen that a Community of Practice would potentially form the basis of providing the smelter with a 'framework' (ie a 'model' or paradigm) to manage the problems of a substantially tacit knowledge-based sub-process and its impact on the smelter model.

Communities of practice in Pot Rooms Area

As already indicated, operators in the Pot Rooms Area have a variety of experiences in dealing with the complexities of their job. Due to the (in part) alchemic nature of their job, a natural incentive exists to have a loose and very informal Community of Practice with respect to pot management. On the other hand, there are the 'technicians' (chemists and engineers) who look at the problem of pot management from a scientific and (so-called) objective perspective. Evidence of their attempts at maximizing CE has been dealt with previously. Because of the differences of approach and perspective, there is little incentive for these two groups to mix, and indeed, there is little incentive for a stronger Community of Practice to occur among the operators other than through natural curiosity and a sense of being correct in their individual approaches to pot management. The research undertaken by Urpani (1996) uncovered the fact that when there was some interaction between operators regarding their management of pots, they were actually astounded to find that generally, there were vast differences in the way each handled various situations. They had presumed that practice was 'standard'. This revelation (anecdotally) was seen as the catalyst for the subsequent greater degree of interaction between operators on this issue. However, there are excellent grounds for the smelter management to encourage the intermixing of the technicians and operators as well as the operators per se to expand their Communities of Practice activities. The technical analysis of pots is all too often confined to macro analysis of hundreds of pots as well as laboratory pots that are very small and operate under a very controlled and unrealistic environment. In this case, a weak Community of Practice had emerged almost accidentally, as is often the case. This paper proposes that the nurturing of such Communities of Practice more intentionally by management may hold great potential for more complete modelling and optimal performance management of processes in that part of the smelter. Wenger (2003, p 84) suggests that the natural boundaries of a group need to be observed to some extent, and in this case, the pot room operators had a natural boundary with technicians, and to some extent with each other. However, the fact that a spontaneous Community of Practice of sorts emerged also suggests that careful cultivation of a more systematic Community would have been very possible. Wenger also notes that the boundaries between the Community of Practice and, say, the technicians, can be bridged rather than remaining a sealed and excluding border by careful nurturing from an organization. Without this bridging, the Community of Practice can simply serve to '..reflect relations of power among practices' (Wenger, 2003, pp 84–86).

The potential gains that can result from the existence of a Community of Practice in the Pot Rooms Area are significant. The resultant sharing of the tacit knowledge among the operators and the gains to knowledge per se could potentially:

1. reduce the instability of the pots (through better handling of anode effects and other day-to-day pot irregularities) thus increasing their efficiency (and thus the CE);
2. increase the accuracy of production measurement through better pot handling leading to the pot temperature being more even, thus leading to lesser bath build up in the pots (thus also impacting on the CE);
3. increase the perpetuation of knowledge and also lead to a potential sharing of knowledge with others (including technicians).

These consequences have significant immediate effects on the output and efficiency of individual pots, that is, their CE. Thus, over all the pots in the smelter, the CE will improve. This improvement will be reflected in the overall CE measured at the end of the month and input into the mathematical model (1) (the exact extent cannot be calculated as noted earlier). It should be remembered that small percentage increases in the CE lead to large decreases in cost. Additionally, in the longer term, the knowledge perpetuation and enhancement will also be favourably affected.

The nature of a Community of Practice

In order to better understand what a Community of Practice 'model' is and how it might operate and/or be encouraged to do so (ie, the 'solution' to this 'model'), it is necessary to understand a little more about them. Communities of Practice are essentially informal groups of people in a single organization (as in the case of 'commercial in confidence' operations of smelters and other manufacturing entities) or in more 'common knowledge' areas, across organizations as well. As Louis (2005), Wenger et al (2002, p 13), Davies et al (2003) and Burk (http://www.tfhrc.gov/pubrds/mayjun00/commprac.htm) accessed 20 September 2005) indicate, Communities of Practice are founded on common knowledge or common work tasks where the coming together of people to share stories and discuss work practices (ie sharing explicit and tacit knowledge) gives support, knowledge and a sense of belonging to people. These interactions can occur in a real or virtual sense depending on the nature of the organization and whether the Community of Practice exists beyond one such work site. In fact, the Community of Practice is one of the 'basic building blocks of a social learning system' as long as it revolves around a joint enterprise, mutuality and a shared repertoire (Wenger, 2003, pp 79–80).

It is worth noting that the name of Communities of Practice was applied initially (see Orr, 1996; Louis, 2005) to a group of copy machine technicians who met in a social setting (around the water cooler). Here they shared common experiences and methods of doing their job that were not a part of the 'technicians' manual'. Membership of a particular Community of Practice is not necessarily fixed but may vary considerably (as will its leadership). In this paper, the nature or classification that could be assigned to the Communities of Practice (see Vestal, 2003) is a hybrid of the 'best-practice' and 'knowledge stewarding' categories. It would certainly embody the three minimum elements of being a Community of Practice, namely having a domain of knowledge, a community of people who care about this domain, and a sharing of practice with a view to being more effective (Wenger et al, 2002, p 27). The aluminium smelting industry is ideally suited to the informal emergence of Communities of Practice and for their encouragement in a formal way by the organization. Anecdotally, the existence of at least one Community of Practice within the Portland smelter was noted (as found by Urpani, 1996) but it was at an embryonic state and very confined (effectively to a small group of process operators on one shift). Other Communities may exist, but have not been actively sought at this stage. By contrast, in the float glass manufacturing industry, because of the multinational nature of glass companies and their worldwide manufacturing operations, the Communities of Practice often exist on a much larger and more formal basis but are still very much confined to the one business entity rather than across manufacturers.

The value of Communities of Practice can be considerable. Communities of Practice can systematize the exchange of knowledge as well as encourage the formalization and interchange of tacit knowledge. In some ways they can also take on the informal role of 'teacher', and mentor for members of the Communities of Practice. With greater sharing of knowledge and experiences, a greater understanding of the production process (or sub-process) ensues and the consequence of this is enhanced performance. It must be acknowledged that they can also be dysfunctionally narrow, cliquish and exclusive, although careful cultivation and stewarding of the community can positively steer the group away from these negative facets (Wenger et al, 2002, p 139; Wenger, 2003, p 80).

When the tacit knowledge problem exists for a production process (in this case a sub-process), a traditional 'hard model' normally cannot be constructed in order to determine the 'best practice' operation of the process. Consequently, a combined approach of 'hard' and 'soft' modelling is required. In this case, when a hard model (the mathematical model) is coupled with a soft model, the framework (or 'model') of a Community of Practice associated with the sub-process, facilitates arrival at a solution. The solution process per se will provide a focus to achieve future knowledge improvement and long-term better understanding of the operations. This synergy is illustrated in Figure 1, where it will also be noticed that if the Community of Practice does not exist, the knowledge flow and consequent sub-process handling by operators does not flow back into the production process.

Further, it should be noted that a critical value of the Communities of Practice is the learning loop for the perpetuation of knowledge in the operators over time. Unless there is some process in place that nurtures and cultivates the Communities of Practice and also attempts to bring the tacit knowledge into more explicit forms, there is a risk to the enterprise of knowledge loss over time. Each generation of skilled operators with high levels of accumulated tacit knowledge needs to systematically pass this knowledge to the newcomer operators, or the organization has failed to manage its knowledge base. Knowledge management is a source of competitive advantage and a key to a firm's longevity (see Saint-Onge and Wallace, 2003; Wenger, 2003; Lehaney et al, 2004). A haphazard approach to the knowledge transfer from experienced hands to newcomers may be sufficient, but also leaves a large amount of knowledge preservation and transfer to chance. It might not happen in a comprehensive way. If there were to be a sudden and unexpected turnover of personnel, the knowledge base is deeply at risk, and so is the competitive advantage. A Community of Practice approach can therefore more consciously capture, preserve and perpetuate crucial knowledge of various processes which are otherwise difficult to codify into conventional training courses, even on-the-job ones. With this understanding, a 'solution' to the Community of Practice 'model' will be suggested later in the paper.

The mixed-mode modelling solution of the smelter problem

In suggesting the mixed-model modelling solution below that involves the hard mathematical model (1) and the soft Community of Practice 'model', it should be noted that there are two distinct time frames involved. The 'macro' model of the smelter (1) is the main planning model and operates on a monthly (or indeed a yearly) time frame. However, the sub-process of the pots operates on a much shorter time frame (typically an online real-time basis) which is under the control of the pot room operators. The Community of Practice 'model' employed to represent the tacit knowledge problem operates in continuous time. These two time fames are easily dealt with when it comes to obtaining an 'overall' solution to the smelter problem (the mixed-mode model) since the results of the Community of Practice model 'model' will provide input into the monthly (or yearly) model of the smelter (1) which will be incorporated into the mathematical relationships there and thus impact on the solution obtained. This is illustrated in Figure 3 and discussed later in this paper. Below are described the solutions to the smelter model and the Community of Practice 'model' and then their interconnectivity via the mixed-mode modelling approach, arriving at a best practice solution.

The solution of the mathematical model

A solution to the mathematical model of the smelter developed in this paper (and for more complex versions) has been extensively covered in past literature. For the one month model with no stocks of raw or intermediate products (as shown in this paper), the solution is almost trivial and is affected by a simple solution algorithm (see Nicholls, 1995; Falkowski and Nicholls, 1999). For solution of the problem over multiple months and also where inter-temporal transfer of capacity associated with some of the processes is involved, the solution is more complex requiring a 'cascade' solution heuristic and a Lagrangian approach, respectively. The objective of the model is to determine best practice within the smelter with respect to the leader and follower business units by determining the maximum kA and the maximum SC (given the kA selected). This will then ensure that the smelter will not violate resource and capacity constraints and operate in a minimum cost and maximum output manner (or at least on a 'best practice' basis).

Community of Practice 'solution'

Communities of Practice at their most effective have a fundamentally egalitarian, informal existence that is by definition not a mirror of the formal organizational reporting lines and structures (Saint-Onge and Wallace, 2003; Ackerman et al, 2003). They therefore need to side-step both organizational power and politics and sit outside that sphere of influence, owned and clearly driven by the community of operators, and simultaneously be respected and acknowledged by the power and management group. Organizations more accustomed to patterns of conventional line authority that give great power to managers may therefore have some difficulty with the existence of a collective activity which management does not own or drive or control. Yet this is the essence of the more open, egalitarian culture associated in modern management practice with the 'learning organisation' (see Senge, 1990; Wenger, 2003). In encouraging Communities of Practice, it is therefore wise for management to firstly face the fact that they can neither control nor overly influence the Communities of Practice, but their support and cultivation in terms of providing time, space and patronage/legitimacy to the activity is vital.

While Communities of Practice do not sit in the conventional authority structure of the firm, they do need to reflect and sit comfortably with its organizational culture. Where there is, say, a strong history of antipathy between management and operatives, it would become very important that management be seen to be entirely unconnected with the desire for a Community of Practice to be established, since cynicism and some paranoia in the operators that management 'wants to get something extra, to exploit' would spell the death of the Community of Practice before it ever formed. Where technical and professional expertise has tended to be seen as belonging to a group of unrealistic, somewhat irrelevant idealists who do not really know anything of value, then they must also be seen as quite separate from the process at least in the first instance. Respect cannot be forced in settings like this, and it is therefore wise to note local company conventions and traditions and to work with them in encouraging Communities of Practice, at least at the outset. Over time, various careful bridging strategies can be put in place by skilled leaders to open the natural boundaries to other relevant groups with expertise that is pivotal to the Community of Practice (Wenger, 2003).

In this smelter case, the existing culture presented a distant but mutually respectful relationship between senior management and the pot room operatives. Certainly the distance was generally sufficient that it would not be appropriate for management to direct that Communities of Practice be established, nor to be involved in them unless by specific invitation. The pot room operators (in the main) regarded the technical and professional employees with some degree of cynicism, seeing them as having 'no clue' about the real world at the 'furnace' front. It would therefore not be appropriate for these professionals to be directly or immediately involved in Community of Practice discussions unless later invited willingly by the pot room operators at a point where some input from the white-collar specialists might be welcomed.

In a context where a union is both powerful and has high potential to be obstructive, it would be unwise to proceed without their direct blessing, since there is a risk that any initiative which seems related to securing greater productivity or company advantage is just another form of potential exploitation. In this aluminium smelter case, union relationships with management were reasonably co-operative. While the union was strong, there was no immediate history of extreme use/abuse of that power, and in the rural setting of the plant, it was always a concern of the union that employment remain high. Job loss would be a serious matter and whatever cooperation with management was needed to ensure the success of the smelter overall, would be given reasonable consideration. It would therefore be possible for a Community of Practice approach to be instituted and driven by local union leadership, and that would only have occurred if those leaders had been thoroughly educated and inspired to pick up this role. At least one of the strategies for establishing Communities of Practice is therefore to work closely with local union leaders to gain understanding of the nature and value of the Communities of Practice, and to then support the union's efforts in establishing the community meetings by validating the necessary time, providing a 'reliable budget' (Wenger, 2001, p 41) and providing simple facilitation in the form of some secretarial and organizational help if requested, and perhaps some refreshments or the like to enhance the experience of operators' meetings. Even union-driven establishment of Communities of Practice relies upon operators continuing to feel that their participation is at all times voluntary, and that the sharing of knowledge remains something which is neither required nor carries any direct consequences if avoided.

Where a union is perhaps not best placed to be the driver of Communities of Practice initiation, shift supervisors and leading hands become the obvious point of ownership and leadership in the first instance. Virtually any and every processing or manufacturing plant will have persons designated as leading hands, or shift supervisors or suchlike, people who are clearly in among and of the operators. Again, this may require some dialogue to engage the interest and understanding of those leading hands, some awareness raising generally to issues of knowledge loss/retention, some non-blaming direct discussion about the effectiveness of the training processes which are taking place on the plant floor.

In this case, pot room operatives were trained on-the-job in a highly informal manner with the significant exception of training for occupational safety reasons. Health and safety training was accepted as a given priority in such a highly dangerous environment, and so taken most seriously by all. All other training to do with the management of the processes in the pots was an unstructured process of mentoring, of anecdotal 'indoctrination' of new operators by 'old hands' who believed that they knew how the processes could be best managed for best outcome. This is an oral tradition of 'folk learning', and while it may be highly effective at best, it might also be highly inaccurate and inefficient at worst, or possibly very fragmentary, leaving significant gaps in new operatives' knowledge. In fact, many of the experienced operatives in the pot room held great pride in the knowledge that they had managed to accumulate over the years, and were quietly interested to demonstrate what they knew and compare when the opportunity came as described by Urpani (1996) and mentioned earlier in this paper.

A culture of pride in knowledge suggests that pot room operators would not respond overly well to suggestions that they form Communities of Practice in order to help the smelter become more profitable per se, not unless there was some bonus or financial benefit to be gained for themselves. Neither would they have wanted to engage simply because management or the professional specialists thought it a good idea and wished them to do so. They may well have resisted any effort to describe or see the Communities of Practice initiative as 'training' since they had long associated training with occupational health and safety training, which they respected and took seriously.

In linking the creation of a Community of Practice with the existing cultural values, it is therefore likely that the pot room operators would have been interested if the initiative had been presented to them as a chance to demonstrate and compare their craft knowledge, a subtle opportunity to allow gentle competition perhaps between shifts, or to allow older operators to share what they had intuitively learned over many years for the benefit of younger operators. Leading hands or shift supervisors are normally the best placed people to find the right level of challenge and flattery to engage their colleagues in this way. Once underway, it appears highly likely that the operators themselves would have had enough quiet interest in the exchange of knowledge to keep the process going.

Communities of Practice need to meet regularly in order to avoid becoming seen as 'flash in the pan' erratic events, but not necessarily every week. They need to find their own rhythm (Endsley and Kirkegaard, 2005, p 29). They may be steered by a shift supervisor to consider a particular topic at each meeting, but most properly should develop their own agenda of topics of interest to themselves (eg, 'What do you do if the pots start doing X, and you can see the temperature rising by Y degrees? What's the best way you've found to stop process Z from escalating?'). Endsley and Kirkegaard (2005, p 29) suggest that one of the key facilitation roles is to draw out the watchful 'lurkers', those less naturally talkative or confident, yet who must be included in order for the Community to genuinely share the learning tasks and resources.

In summary, Communities of Practice need to sit comfortably with the culture and values of the organization, or part thereof, and should be clearly owned and facilitated by the community of operators whose experience and knowledge is to be shared. Low key, facilitative, minimal leadership is best and may in fact become a rotating or shared function as agreed by the operators. Communities of Practice are normally best presented to operators as opportunities to share, compare and learn for the benefit of all, especially newer operators, so that the craft (in this particular case, their alchemy) is not lost or diminished. Although not consciously and intentionally developed in this case, the potential for the Communities of Practice to be more actively developed as a means of process improvement is outlined and signals its potential part in a more holistic model, that is, a solution to the problem of optimizing all parts of the smelter process.

The mixed-mode modelling solution process

Having considered the two components of the overall model of the aluminium smelter, and seen to some extent the way in which they interact, it is timely to indicate how these two model components, the hard model (of the smelter) and the soft model (the Community of Practice) can be used to obtain best practice aluminium smelting.

The tacit knowledge problem, represented by the Community of Practice 'model' is 'solved' by the development and implementation of management strategies aimed at enhancing its existence and level of activity. The smelter problem, represented by the mathematical model, can be readily solved using a variety of heuristics. The interface between the two models is a time dependent one where the Community of Practice Model feeds more accurate and stable parametric estimations (including CE) to the mathematical model. The CE will also potentially be improved due to better pot operation methods thus increasing production efficiency. The feedback from the mathematical model to the Community of Practice will facilitate implicit benchmarking and subsequent enthusiasm and potential gains (if handled correctly). If the solution of both of these models is undertaken with the input of the other and recognition of the existence and importance of each other then, the smelter stands to gain significantly. Figure 3 summarizes this solution interaction.

As can clearly be seen from Figure 3, the two models feed into one another, with the solution to the mathematical model arrived at monthly (or yearly) and the solution heuristic of the Community of Practice Model continuously in operation (over the full year). It should be noted that the implementation of the solution to the Enhancement of the Community of Practice Model leads not only to the knowledge perpetuation, but also to knowledge enhancement, which will in turn potentially lead to the more stable and accurate parametric values and increased productivity (increased CE).

The solution approach is an ongoing one that relies on the feedback loops and the continued application of the two 'solutions' to the 'models'. While this approach appears to be logical and is indeed simple, anecdotal evidence does not suggest that it is widespread. The actual solutions to the two models have already been discussed with the emphasis on the Enhancement of the Community of Practice Model. The solution to the mathematical model is well documented elsewhere.

Concluding remarks

This paper has identified a problem that exists in many industries. The problem is associated with the lack of understanding of some of the sub-processes together with an inability to accurately measure some of their key parameters that are necessary for the overall modelling of an otherwise identifiable mathematical model at the macro level. The aluminium industry has been used as an example here. The solution recommended for this problem (of establishing 'best practice' through a macro model) has been to incorporate the concept of Communities of Practice in conjunction with a Knowledge Perpetuation protocol. It is suggested that the combination of the 'soft' and 'hard' models constitutes a mixed-mode modelling approach that will constantly see improvement in production practices within the sub-process as well as an increase in the accuracy of the key parameters associated with it. In summary, for the aluminium smelting industry, the use of Communities of Practice and Knowledge Perpetuation protocols is seen as leading the way towards:

• a decrease in the variability of a pot's production,
• an increase in the quality (purity) the aluminium produced,
• better behaved pots and a consequential more even temperature leading towards less build up of bath around the sides of the pot leading to a more accurate estimation of a pots production,
• a reduction in pot disturbances leading to the greater reliability is any estimates of production obtained from it,
• consequential increases in CE (which is also a more reliable 'guestimated' figure),
• providing a more realistic basis for determining best practice from the solution of the macro model of the smelter.

While the Portland Aluminium smelter has been used as an example of the problems that exist with some sub-processes, and of what might be achieved by the suggested use of Communities of Practice, etc, many industries share this problem, including the manufacture of float glass. The application suggested by this paper therefore has a much wider applicability than just the aluminium smelting industry per se and might be applied to any process with a sub-process that exhibits the tacit knowledge problem (ie driven by intuitive wisdom/alchemy rather than hard values and scientific control).

References

1. Ackerman M, Pipek V and Wulf V (eds) (2003). Sharing Expertise: Beyond Knowledge Management. MIT Press: Cambridge, MA, London.
2. Davies J, Duke A and Sure Y (2003). OntoShare—a knowledge management environment for virtual communities of practice. Proceedings of the 2nd International Conference on Knowledge Capture 2003. Association for Computing Machinery: New York, pp 20–27.
3. Desclaux P (2001). Carbon consumption and current efficiency in Hall–Heroult cell. Metall Mater Trans B 32: 743–746. | Article | ISI |
4. Endsley S and Kirkegaard M (2005). Working together: Communities of practice in family medicine. Fam Pract Mngt 12: 28–35.
5. Falkowski J and Nicholls MG (1999). Increasing the efficiency of selected grid search procedures through the introduction of a best step mechanism. J Heuristics 5: 199–214. | Article | ISI |
6. Grjotheim K, Krohn C, Malinowsky M, Matiasovsky K and Thonstad J (1977). Aluminium Electrolysis: The Chemistry of the Hall-Heroult Process. Aluminium-Verlag GmbH, Düsseldorf: Germany.
7. Habashi F (2002). Hall, Heroult and the production of aluminium. CIM Bull 95: 109–113. | ISI | ChemPort |
8. Lehaney B and Clarke S (1997). Soft systems methodology (and simulation) for mixed-mode modelling. In: O'Brien F and Orman A (Eds). Operations Research Keynote Papers 1997. Operational Research Society: Birmingham, pp. 16–33.
9. Langon B and Peyneau J (1990). Current efficiency in modern point feeding industrial potlines. Light Metals(18–22 Feb): 267–274.
10. Lehaney B, Clarke S, Coakes E and Jack G (2004). Beyond Knowledge Management. Idea Group Publishing: Hershey, PA, USA.
11. Louis LR (2005). Technical issues facing work groups, teams and knowledge networks. In: Coakes E and Clarke S (Eds). Encyclopedia of Communities of Practice in Information and Knowledge Management, 2005. Idea Group Publishing: Hershey, PA, USA.
12. Nicholls MG (1995). Aluminium production modelling—a non-linear bi-level programming approach. Opns Res 43: 208–218.
13. Nicholls MG, Clarke S and Lehaney B (eds) (2001). Mixed-model Modelling: Mixing Methodologies for Organisational Intervention. Applied Optimisation, Vol. 58. Kluwer Academic Publishers: Dordrecht-Boston-London.
14. Nicholls MG and Hedditch DJ (1993). The development of an integrated model of an aluminium smelter. J Opl Res Soc 44: 225–235. | Article |
15. Orr JE (1996). Talking About Machine: An Ethnography of a Modern Job. Cornell University Press: Ithaca, New York.
16. Pearson TG and Waddington J (1947). Discussions of the Faraday Society. Faraday Society: London.
17. Rodrigo HDR (1998). The establishment of a predictive model for current efficiency in reduction cells. PhD thesis, Swinburne University of Technology.
18. Saint-Onge H and Wallace D (2003). Leveraging Communities of Practice for Strategic Advantage. Butterworth-Heinemann: Boston.
19. Senge P (1990). The Fifth Discipline: The Art and Practice of the Learning Organization. Century Business: London.
20. Sterten Å (1988). Current efficiency in aluminium reduction cells. J Appl Electrochem 18: 473–483. | Article |
21. Tarcy GP (1995). Strategies for maximizing current efficiency in commercial Hall–Heroult cells. Fifth Australian Aluminium Smelting Technology Workshop 1995, University of New South Wales, Kensington, pp 22–27.
22. Urpani D (1996). Knowledge acquisition from real world data. PhD thesis, Swinburne University of Technology.
23. Vestal W (2003). Ten traits for successful communities of practice. Knowl Mngt Rev 56: 6.
24. Wenger E (2001). Organically grown. Train Dev J 55: 40–42.
25. Wenger E (2003). Communities of Practice and Social Learning Systems. In: In: Nicolini D, Gherardi S and Yanow D. (eds). Knowing in Organizations—A Practice-Based Approach, Chapter 4. ME Sharpe Inc: Armonk, NY, pp 76–99.
26. Wenger E, McDermott R and Snyder W (2002). Cultivating Communities of Practice—a guide to managing knowledge. Harvard Business School Press: Boston.

Acknowledgements

We wish to express our appreciation to the anonymous referees whose comments and suggestions improved the paper considerably.