Keywords

1 Introduction

There is evidence that the competitive environment of process industries is becoming more challenging as firms in a sector which is already cost focused are forced to control their costs while pursuing new markets, or improve their flexibility and responsiveness to defend existing markets (Yoho and Rappold 2011). The impact of this includes increasing product variety, reduced order size and reduced lead time, leading to increased manufacturing complexity, which when the fixed and inflexible nature of process manufacturing capacity is considered, results in significant challenges in the areas of planning and production control (PPC).

Since the 1980s, discrete manufacturers have benefitted from lean manufacturing techniques and pull production, to better align demand and capacity (Lyons et al. 2013) and create flow, delivering improvements in lead-time and cost. The process industries however have been slow to follow the discrete sector in the use of alternate PPC approaches to improve competitiveness (Dennis and Meredith 2000b). It has been suggested (Abdulmalek et al. 2006; Pool et al. 2011) that this is due to the unsuitability of process industry product/process characteristics, which may hinder the use of such mechanisms.

The authors of this paper would argue that the unsuitability lies with the mechanisms themselves, which are discrete industry founded, and so may not necessarily apply to the process environment, or help process firms compete in the same way they have discrete manufacturers (Crama et al. 2001). Fundamentally, there would appear to be a lack of evidence to support the application of pull in such environments (Belvedere and Grando 2005) and this paper is intended to address that deficiency.

2 Literature Review

2.1 The Process Industries

It is argued that the competitive environment of process manufacturing firms is becoming more difficult (Yoho and Rappold 2011), forcing changes in fulfilment strategy, which brings challenges for PPC and issues of ‘fit’ between fulfilment and manufacturing (Karmarkar and Rajaram 2012, p. 680). Numerous frameworks have been developed (Lyons et al. 2013) that classify products against processes to understand different operations and their manufacturing strategies (Stavrulaki and Davis 2010).

2.1.1 Published Industry Taxonomies

Hayes and Wheelwright (1979) built on the work of Skinner (1969) and Abernathy and Townsend (1975) cited in Lummus et al. (2006) introducing the product-process framework for aligning products and their life cycles with their corresponding production process life cycle. Traditionally the process industries have been clustered in the bottom right hand zone of the matrix. Crama et al. (2001) and McDermott et al. (1997) claim that firms should rarely exist off this diagonal due to the impact on cost/profit.

However, with changing markets, hybrids which compromise traditionally accepted manufacturing strategies are becoming increasingly common (Crama et al. 2001). Safizadeh et al. (1996) observe that most of these off-diagonal companies belong to the process industries where Kemppainen et al. (2008) suggest a “misfit” between capital-intensive equipment and expensive product changes.

Numerous authors have discussed the validity of this product-process approach (Kemppainen et al. 2008; McDermott et al. 1997; Stavrulaki and Davis 2010) proposing additional dimensions including batch consistency, material and product variety. Dennis and Meredith (2000a) argue that as Hayes and Wheelwright’s (1979) model was originally developed for the discrete industries, the assumption is that process firms will face the same challenges and can use the same solutions as a discrete firm based on their position on the diagonal. They (ibid) propose that in contrast, manufacturing systems are actually organised by how products are made rather than by the actual end product itself. Abdulmalek et al. (2006) propose the additional dimension of discretisation, which, consistent with the process sector’s ‘off-diagonal’ movement, can facilitate fulfilment flexibility (Pool et al. 2011). Due to fundamental differences (Crama et al. 2001) however process industries may not be comparable with discrete industries.

To distinguish between process and discrete manufacturing (Taylor et al. 1981) as two mutually exclusive strategies is inconsistent with manufacturing reality (Abdulmalek et al. 2006). Most process plants are actually hybrids because their non-discrete products become discrete at some point (Billesbach 1994). Pool et al. (2011) summarise production characteristics on either side of this point, which they claim can support different PPC approaches. Puttman (1991) and Abdulmalek et al. (2006) suggest however that in some systems, characteristics might actually be shared. The literature discusses characteristics typical of the process industries as shown in Table 1.

Table 1 Nature and impact of process industry characteristics
Full size table

In more recent years differences within the process industries have received more attention. Fransoo and Rutten (1994) distinguish between process/flow and more flexible batch/mix industries but their model is arguably a constituent of those presented a decade earlier, implying that firms within each group have similar processing patterns (Dennis and Meredith 2000a). While Fransoo and Rutten (1994) lacked an “empirical approach” (Van Donk and Fransoo 2006, p. 211), Dennis and Meredith (2000a) addressed this with their study of the differences within process industry firms proposing that there is considerably more complexity in the process industries than previous research suggests.

In conclusion there remains relatively little research on the nature of the process industries (Crama et al. 2001; Dennis and Meredith 2000a). There has been much analysis and validation of models but the fundamentals of many such models remain founded in discrete manufacture and so there is limited consensus as to their applicability in the process industries. This suggests that due to the diverse nature (Lyons et al. 2013) and increasingly ‘off-diagonal’ behaviour of process firms (Crama et al. 2001) the current literature frameworks are inadequate and may lead to selecting the wrong PPC approach.

2.2 Planning and Production Control (PPC)

Taylor and Bolander (1994) state that a firm’s planning and scheduling must be tailored to its manufacturing environment and competitive strategy, which in turn has implications for off-diagonal process manufacture. In order to discuss the applicability of production control systems, Kingman’s equation is used to explain the impact of variation and the interplay between variation and utilisation. There are three factors within Kingman: arrival variation; process variation and utilisation, all of which influence waiting times or cycle time. Expanding on Kingman we can determine that variation is buffered by Inventory, Capacity or Time (Bicheno 2011) and that the mechanisms available stem from these three fundamentals (Fig. 1).

Fig. 1
figure 1

Expanding on Kingman’s equation (Bicheno 2011, reprinted with permission)

Full size image

De Treville and Antonakis (2006) agree, suggesting that the lean paradigm can be defined as a system which maximises capacity utilisation and minimises buffer inventories through minimising variability. In some environments capacity can be increased or utilisation reduced by adding labour whereas in others the only option is capacity in the form of plant/equipment. As this normally comes at significant cost, manufacturers normally focus on reducing variation using the mechanisms discussed above. The use of an appropriate pull system can also serve to reduce variability (Schonberger 1983) and stabilise flow.

2.2.1 Production Control Systems

Pull versus push is the control of WIP versus the control of throughput, typically controlled in relation to capacity which must be estimated and is subject to variation (Hopp and Spearman 2004). Pull systems are more responsive than push systems (Cheng and Podolsky 1993 cited in Hopp and Spearman 2004). By accounting for system status, pull improves manufacturing cycle times to achieve a lead time shorter than the expected delivery time. Spearman et al. (1990, p. 880) explain however that push and pull are not mutually exclusive concepts and that most systems are hybrids of the two “containing make to order (MTO) and make to stock (MTS) elements”.

Whilst kanban was the first production control system to be termed pull (Hopp and Spearman 2004) its limited applicability (Bicheno and Holweg 2009; Hall 1983; Liker 2004) has motivated the generation of alternatives (Gaury et al. 2000) such as CONWIP (Spearman et al. 1990) and DBR (Goldratt and Fox 1986) in addition to various hybrids. However, whilst discrete manufacturers have benefitted from such pull production control systems (Lyons et al. 2013), the literature is “devoid of examples” which address process manufacturing (Yoho and Rappold 2011, p. 61).

Dennis and Meredith (2000a) explain how process industries have had mixed success with ERP systems with Van Donk and Fransoo (2006) suggesting that one of the main issues has been the inadequacy of MRP to plan process industry capacity. MRP logic tends to assume infinite capacity, fixed lead times, (Yoho and Rappold 2011) fixed batch sizes (Darlington and Schmidt 2013) regardless of plant loading and product mix (Karkarmar 1991) and when used for shop-floor scheduling ERP is effectively a push system (Powell and Strandhagen 2011). Schedule feasibility/adherence issues are exacerbated in process industries where materials, BOMs, yield and changeovers are all sources of variation and where there is the need to sequence production based on shared capacity and constraints (Schuster et al. 2000). Consequently other methods to guide and execute the schedule must be found (Schuster et al. 2000).

Based on the practices of process firms, Taylor and Bolander (1994) propose PFS a general approach to scheduling. In contrast to MRP (which uses product structure), PFS uses process structure to find a feasible schedule where capacity, due date, lot sizing and sequence dependency can be accounted for. However infrequent schedule violations are an assumption of PFS (Schragenheim et al. 1994) and as such it may not be suitable in higher product variety, contingent capacity process environments. Hubbard et al. (1992) incorporate the group technology (GT) philosophy into process scheduling. GT improves efficiencies by exploiting similarities and has been successful in cell manufacture where it normally involves equipment dedication (Shahin and Janatyan 2010). However, the philosophy can also be employed in constructing shared capacity schedules (Jamshidi 2009) by grouping products into families (Shahin and Janatyan 2010; Soman et al. 2004).

2.2.2 Planning and Production Control in the Process Industries

Process industries “lag” behind discrete manufacturers in the effective use of PPC (Dennis and Meredith 2000b, p. 68) despite more complex decisions regarding product replenishment (Yoho and Rappold 2011). An increasingly competitive environment for process industries has ushered a shift from MTS to hybrid MTO/MTS, (Crama et al. 2001; Fransoo 1992) bringing with it additional PPC complexity. Combined MTO/MTS is “neglected in the literature” and much of what does exist has limited applicability in process industries assuming for example no setups and batch sizes of one (Soman et al. 2004). The relationships between setups, process yield and differences in run productivity cause issues with estimating process industry capacity and plan feasibility. Bottlenecks can move (King 2009) and schedules initially found to be reasonable can become invalid (Fransoo 1992) because the impact of product mix and sequence are not considered (Dennis and Meredith 2000b), rendering the critical piece of processing equipment a constraint by the number of setups (Schragenheim et al. 1994).

This relationship between demand, capacity and inventory in the process industries creates real cost trade-offs (Fransoo 1993 cited in Crama et al. 2001), which must be managed. Combined MTO/MTS creates issues for shared capacity production control. Stock production is sometimes manufactured in the queue ahead of real demand, creating trade-offs between due-date performance, inventory, flexibility and capacity. The key issue in planning this capacity therefore is to determine what level of inventory is appropriate, how to make that decision and which products to produce at which time, to meet demand in the most cost effective manner (Cooke and Rohleder 2006). This creates further trade-offs between setups which result in waste/lost capacity and larger lot sizes which reduce waste/increase capacity but impact on service (ten Kate 1994). Toelle (1996) cited in Schuster et al. (2000) suggests that the lot sizing trade-off should be viewed as one between set-up costs and capacity constraints. In the context of the process industries waste, contamination and quality costs should be added (Cooke and Rohleder 2006).

In process manufacture, production sequence which is often determined by setups (Van Dam et al. 1993) is of particular importance. In light of the competitive developments in process industry markets, in addition to capacity and waste/cost, trade-offs arguably include service levels all of which can be mitigated by production sequence (Clark et al. 2010; Cooke and Rohleder 2006; Soman et al. 2004) which the authors conclude to be of fundamental importance to PPC in the process industries. This is well supported in the operational literature.

2.3 Flow and Pull

Bonney et al. (1999) suggest that in practice most systems comprise elements of both push and pull. Pull however cannot be viewed in isolation and whilst it is outside the scope of this paper, the authors recognise that pull is effectively a constituent part of “the larger lean construct” (Hopp and Spearman 2004, p. 144).

Lean manufacturing (Krafcik 1988) emerged in post-war Japan (Womack et al. 1990) where observations of the work of Ford creating flow in a mass production environment and learning from other leading intellectuals were assimilated by Japanese industrialists. They combined them with their own ideas to create a hybrid, (Bicheno and Holweg 2009) holistic and sustainable system of management focused on reducing waste (Womack et al. 1990). This system compressed lead time (Schonberger 2015) leading ultimately to the creation of flow in a low volume, high variety environment with pull as a core concept (Liker 2004).

Pull and the achievement of flow are fundamental elements of the five lean principles (Womack and Jones 1996). For Ohno and Toyota the mantra was “flow where you can, pull where you must” (Rother and Shook cited in Liker 2004, p. 108) suggesting that even for Toyota, achieving flow was not straightforward.

There is considerable ambiguity (Bonney et al. 1999) regarding the definition of pull in the literature with JIT, kanban and pull often used interchangeably (Hopp and Spearman 2004). In the mid 1990s the definition of pull became distorted and synonymous with MTO. Hopp and Spearman (ibid) cite Womack and Jones’ (1996) introduction to pull as a key catalyst of this distortion:

Pull in simplest terms means that no one upstream should produce a good or service until the customer downstream asks for it…

Hopp and Spearman (2004) caution against distinguishing pull/push by either MTO or MTS, arguing that both can be either pull or push. King (2009, p. 241) argues that Ohno “didn’t explicitly define pull” with Hopp and Spearman (2004) describing Ohno’s (1988) picture as a high level, conceptual view of pull as opposed to the means to make it work. They (ibid) distinguish between “strategic” and “tactical” pull, or what Bicheno and Holweg (2009, p. 148) term “principle” and “mechanism” arguing that it is here that the ambiguity has arisen.

Hopp and Spearman (2004) suggest that the literature requires a definition of pull based on what is needed to obtain its benefits rather than on how it is executed, concluding that unlike push, pull explicitly limits system WIP. The over-riding distinction between the two then is that the pull system takes account of system status, preventing both the system from becoming overloaded and the queue from growing exponentially resulting in stable and minimal cycle and lead times.

Hopp and Spearman (2008) suggest that Toyota’s success was achieved as a result of a pull based system which improved material flow, lead time, quality, flexibility and hence stability and service. However, their conclusions are discrete industry founded with a lack of evidence regarding the applicability of discrete mechanisms (see Table 2) in process manufacture (Abdulmalek et al. 2006; Yoho and Rappold 2011).

Table 2 Literature summary of pull production characteristics
Full size table

There remain gaps in the literature as the vast majority of research is focused on the implementation of pull in discrete manufacturing environments (Belvedere and Grando 2005; Crama et al. 2001; Yoho and Rappold 2011). Furthermore there is a lack of evidence but the literature suggests that discrete pull mechanisms do not directly transfer to process manufacture (Abdulmalek et al. 2006; Pool et al. 2011). However, with a contingent approach, opportunities to create flow may exist in terms of aligning demand and capacity and minimising interruptions (Lyons et al. 2013).

It is interesting to note discussion regarding the lack of process industry lean/pull literature and case implementation in the 1990s (Billesbach 1994; Schragenheim et al. 1994) and the fact that over a decade later the literature is still making the same point (Belvedere and Grando 2005; Pool et al. 2011; Yoho and Rappold 2011).

2.4 Conclusions on the Literature

In conclusion, further examination of flow and pull based PPC models in the process industries would add to the current body of knowledge.

The bulk of the literature is focused on MTS environments and research on MTO is scarce (Germs and Riezebos 2010). The fact that most process businesses are a combination of both carries questions with regard to how pull should be implemented, when capacity is shared and consumed by competing value streams and production for both real orders and stock. This forces the issue of prioritisation (Germs and Riezebos 2010), which in process manufacture brings trade-offs and decisions regarding production sequence where there is a deficiency in the lean literature.

This is particularly striking when the “unparalleled” (Kouvelis et al. 2005, p. 462) extent of the traditional operational literature on scheduling and sequence is considered.

3 Methodology

A single case study of a process industry that supplies commercial fish food has been selected with triangulation achieved through the existing body of knowledge on the process sector, PPC approaches and pull, with case company data comparison and discussion, accompanied by simulation to combine theory and case environment.

The research was conducted at Skretting UK’s operation in Longridge, England.

Nutreco is a global animal nutrition company with approximately 10,000 employees in 30 countries of which Skretting is a global subsidiary specialising in the manufacture of high energy pelleted fish feed for commercial fish farms, being part of the aquaculture value chain. Skretting has operations in 14 countries, selling approximately 1.9 million tonnes of feed for over 60 species of farmed fish and shrimp.

Skretting Longridge is the UK’s speciality feed plant manufacturing in excess of 200 different products, for 7 species of fish, on a single extrusion line for more than 400 customers UK wide. An increasingly diverse product portfolio accompanied by pressure on customer lead time and shared capacity has led to an increased focus on PPC and flow as a means to improve responsiveness (Schonberger 2015), which provides a suitable environment for the research.

The case company will be compared with the literature characteristics and taxonomies of process industries and the case environment considered in terms of the fundamentals of demand, capacity and variation, the PPC literature and suitability for pull. A sequencing simulation will be carried out to compare typical pull scheduling methodologies with that of the case company’s pull system.

4 Case Study Analysis and Discussion

Comparing the case company to the literature on process and discrete industries highlights some differences (see Table 3).

Table 3 Characteristics of Skretting as a process business
Full size table

This analysis confirms both the inadequacy of comparing process and discrete industries and the generalisation that process industries consistently show the same characteristics. Skretting is not an exact fit with either classification and in addition to hybrid and certain unique characteristics, clearly displays aspects of both process and discrete manufacture.

Skretting is typical of the literature which considers that many process industry products become discrete late in the transformation. However, in contrast to Pool et al. (2011) Skretting demonstrates significant complexity and commonality in characteristics on both sides of this point (see Fig. 2) suggesting that the additional complexity of batch/mix process businesses brings elements of both process and discrete manufacture.

Fig. 2
figure 2

Skretting process versus discrete manufacture (adapted from Pool et al. 2011)

Full size image

Despite fitting the APICS (2013) definition of batch/mix production, when Skretting is considered in the context of Fransoo and Rutten’s (1994) model of process industries it is found to be atypical of batch/mix businesses and as such its place within process industry taxonomy remains unclear. This is confirmed using a qualitative view (supported by Lyons et al.’s (2013) descriptions) of Skretting’s position on Dennis and Meredith’s (2000a) four criteria model where Skretting displays two significant differences which bring trade-offs between equipment flexibility and variety, impacting on PPC:

  • Equipment v Materials Diversity: Skretting had a significantly higher number (×10–20) of raw materials and finished goods than those businesses which had similar equipment characteristics (and displayed the lowest material diversity). This raises PPC issues in terms of flexibility and responsiveness, cost, fulfilment strategy and service level;

  • Equipment v Run Time: Average run time at Skretting was 10 times shorter than the lowest run time average and more than 50 times shorter than those plants with similar equipment characteristics. In the context of process manufacture this brings PPC issues in terms of run length, changeover, waste and capacity.

In conclusion, analysis suggests that while clearly a process manufacturer, Skretting arguably displays complexity not described by the literature and exhibits characteristics of process/flow, batch/mix and discrete industries.

4.1 Skretting Manufacturing, PPC and Fulfilment

Typical of the process industries (Crama et al. 2001), the Skretting facility is V plant oriented (Cox and Spencer 1998). Skretting has three points of differentiation which typically provide opportunity to alter fulfilment strategy by relocating the decoupling point (Naylor et al. 1999) but this is practically and economically unviable. This combination of inflexible plant and limited decoupling means that Skretting fulfilment is a combination of MTS decoupled at finished goods inventory and MTO decoupled at raw material inventory.

Skretting’s continuous layout means that intermediate process WIP is both constrained and almost non-existent. It could be argued at a mechanistic level that as a result of this natural bounding of WIP, the Skretting system cannot be overloaded and is already pull based. However, demand and capacity can still be misaligned and the wrong product/quantity manufactured, resulting in over-production waste, slow moving inventory and poor flow to the customer despite this apparent bounding of WIP. As such it could be argued that the mechanistic definition of discrete industry pull in terms of bounded WIP does not apply to process manufacture.

Skretting’s V plant configuration includes shared capacity common routing which, consistent with the literature, brings trade-offs between cost/waste, capacity and service which render scheduling critical. In V plants, scheduling is typically focused at constraints and points of differentiation where traditionally opportunities for decoupling exist. Due to decoupling limitations, Skretting scheduling focuses on the former, where due to the combination of early product commitment and resulting setups, extrusion is the dominating process constraint. Historically, Skretting scheduling has been MRP push (Hopp and Spearman 2004). This push approach has resulted in misalignment of demand and capacity with off-peak over-production, and a finished goods inventory curve which doesn’t reflect seasonality. As a result Skretting has implemented a pull based system which is discussed in Sect. 4.2 below.

Skretting’s hybrid MTO/MTS fulfilment is an industrial reality, but is not well described in the literature creating issues for PPC frameworks which regard MTO and MTS as mutually exclusive fulfilment strategies requiring disparate manufacturing characteristics. In the case of Skretting, this is not practically possible and so a solution must be found in terms of PPC, however the literature PPC frameworks do not adequately describe hybrid fulfilment. As such in the context of this paper, it is considered necessary to examine PPC fundamentals to determine opportunities for hybrid fulfilment solutions and the implementation of pull.

Annual demand is seasonal fluctuating significantly with customer growing strategies, harvest plans, water temperatures and oxygen levels. Product demand follows typical Pareto behaviour. Order demand follows a similar pattern with the majority of customer orders smaller than the minimum run length. This ‘long tail’ (Anderson 2009) provides further PPC challenges where early committed products are manufactured on inflexible process industry equipment. This supports both the need for a MTS/MTO fulfilment strategy with finished goods inventory buffering based on robust demand analysis (King 2009) and the equipment/variety conflict highlighted by the analysis of Skretting using Dennis and Meredith’s (2000b) model.

Arrival variation is atypical of the process industry literature and confirms Skretting’s off-diagonal position. Several decision points/handovers are present in the Skretting supply chain and manufacturing is decoupled from real demand resulting in amplification (Lee 1997). Numerous countermeasures have been implemented but short term demand is still subject to significant biological and environmental variation. This variation is confirmed using demand class analysis (Boylan et al. 2008; Syntetos et al. 2005) where over 50 % of Skretting SKU’s are higher variation ‘lumpy’ or ‘management control’ with significant variation within these demand classes. Furthermore, only 3 % of volume has a low coefficient of variation (CV).

Analysis demonstrates that Skretting process variation is consistent with the process industry literature:

  • Material Variation—processing functionality variation (not explicit until processing has started) due to natural sourcing and seasonal availability;

  • BOM Variation—variability of material quality, availability and price results in BOM variation which can cause significant change in processing functionality impacting product quality, yield and capacity due to changing proportions of variable materials;

  • Machine Variation—equipment is large scale, capital intensive and processing is inexact cooking/chemical reaction giving rise to significant process variation of which changeovers are a significant proportion impacting on time, capacity, material waste and quality;

  • Yield/Quality—The combination of material, BOM and process variation results in variation in both yield and right first time (RFT) quality, impacting flow, system predictability and service levels.

In conclusion the Skretting case demonstrates a significant degree of variation in processing, materials, BOM and yield/quality that is consistent with the process industry literature providing less opportunity for reduction than that within discrete manufacturing and so any buffering strategy should take this into account.

This, coupled with long tail demand, arrival variation and off-diagonal fulfilment not typical of the process industry literature results in a particularly high variation environment for Skretting as a process manufacturer (see Fig. 3).

Fig. 3
figure 3

Euler diagram illustrating Skretting variation complexity

Full size image

Skretting setups/changeovers are frequent and constrain capacity (Schragenheim et al. 1994). Changeovers are also sequence dependent (Yoho and Rappold 2011) consuming different amounts of time and so capacity varies with product mix, and the number (lot size) and type of changeovers. In addition to lost capacity, setups/changeovers also incur energy cost and generate material waste.

The significant impact of setups/changeovers are consistent with the literature and indicate that the typical discrete-pull focus on setup/batch size reduction (Thun et al. 2010) is less suitable for the process manufacturing environment at Skretting, due to increased waste and potential capacity constraint (Kim and Tang 1997; Schragenheim et al. 1994).

Consequently, typical discrete industry workload levelling would also be unsuitable in the Skretting process environment (Bicheno and Holweg 2009) due to the increased number of changeovers (Powell et al. 2009).

Analysis shows that Skretting Capacity is contingent on:

  • Product Mix—due to the processing speeds of different sizes and product types which require varying difficulty of production effort (Seidman and Holloway 2002);

  • Constituent Materials—due to their processing functionality which can cause the bottleneck to shift;

  • Bill of Materials—due to changing proportions of variable raw materials;

  • Yield/Quality—due to large scale, inflexible inexact processing and variable BOMs/Materials.

Figure 4 illustrates the cumulative capacity impact of each variation source contrasted with the capacity contained within MRP.

Fig. 4
figure 4

Cumulative impact of variation on capacity of a specific SKU

Full size image

This contingent capacity has implications for PPC and establishing flow, while in the context of pull mechanisms, the variable product/process characteristics and cycle times, also confirm the unsuitability of discrete industry takt for the Skretting environment.

4.2 Pull in the Skretting Process Environment

Analysis of the Skretting production environment suggests that many of the traditional mechanisms of discrete pull may not be applicable to process manufacturing at Skretting (see Table 4).

Table 4 Summary of implications of Skretting manufacturing environment for pull
Full size table

However, a pull system has been introduced at Skretting, which is consistent with the principles of flow and pull and the fundamentals of demand, capacity and variation (see Table 5).

Table 5 Elements of Skretting pull system and their consistency with the literature
Full size table

In summary, Skretting now employ hybrid MTO and demand based pull replenishment MTS using a combination of time and inventory buffers and Advanced Demand Information (ADI) (Claudio and Krishnamurthy 2009) to reduce changeovers and enable schedule smoothing (see Fig. 5) on a product group basis to deliver high service levels while mitigating the trade-off impact of changeovers to reduce waste and improve flow.

Fig. 5
figure 5

Buffering of arrival variation to smooth the schedule at Skretting

Full size image

As a result of this demand based pull, run lengths and inventory levels for higher demand, more stable products have been extended reducing material waste and shortened for lower demand, more variable products improving obsolescence. The periodic review system has also stabilised manufacturing cycles (Yoho and Rappold 2011) reducing the standard deviation of time between cycles. However, Skretting did not experience an improvement in quality/yield (ibid) as a result of more stable production cycles. As a result of this improved alignment of demand and capacity, system flow is improved resulting in a more stable inventory profile and an upward trend in OTIF.

Since the original research, analysis by a third party working capital consultancy one year after the pull system was implemented has confirmed the reduction of weighted average maximum and minimum system lead time by over 50 %.

4.2.1 Conclusion on Pull in the Skretting Process Environment

From the analysis above, the authors conclude the potential for and benefits of pull production control in a process environment as demonstrated by the Skretting pull system case, further proposing that sequence is an important element of the Skretting pull system and a critical element of achieving flow and pull in a process environment which will be examined next.

4.2.2 Importance of Sequencing in the Skretting Pull System

In this section simulation will be used to evaluate the importance of sequence as a constituent element of pull in the Skretting case.

To simulate sequencing the production demands generated by the Skretting pull system for 15 separate weeks in 2013 were re-sequenced using the following scheduling methodologies associated with lean/pull production (Bicheno and Holweg 2009; Hopp and Spearman 2008):

  1. 1.

    Shortest Processing Time (SPT)

  2. 2.

    Earliest Due Date (EDD)

  3. 3.

    Level Scheduling/Mixed Model (Level)

The results of the three lean scheduling methodologies are compared, (Table 6) with that of the actual Skretting sequence (Skr Seq).

Table 6 Summary of sequencing simulation results
Full size table

The SPT sequence gave the worst performance and in contrast to the literature (Hopp and Spearman 2008) did not decrease average manufacturing times. Rather than consolidating short runs (GT) to reduce setups and waste, SPT separates them to complete them first, exacerbating the number and severity of changeovers.

The impact of EDD and Level Loading are similar, increasing changeovers to either prioritise due dates (EDD) or spread workload (Level Loading). In contrast to the literature, level loading did not deliver higher utilisation, or when feasibility was accounted for, responsiveness for the customer (Hüttmeir et al. 2009).

As would be expected, Skretting setup variation was highest due to the combination of small changeovers where feasible, accompanied by large setups where unavoidable.

However, significantly and somewhat unexpectedly, the Skretting sequence delivered the lowest capacity CV. Although this result could be a function of the higher SKR sequence capacity, this suggests a degree of what in this environment could be considered level loading achieved by the Skretting methodology, the performance of which compares favourably with the discrete founded level loading sequence.

The importance of sequence was particularly apparent at higher levels of utilisation, which sequence impacted further, where schedule feasibility was a greater issue for the lean scheduling methodologies (see Fig. 6).

Fig. 6
figure 6

Impact of sequence on utilisation

Full size image

In high utilisation weeks, the superiority of the SKR sequence in terms of both material, energy and time waste was most pronounced. When schedule feasibility was taken into account, at high utilisation the SKR sequence also delivered better OTIF. Unsurprisingly, sequence was of lesser importance when product mix complexity was low where the SKR sequence did not deliver significant advantages in either manufacturing lead time or capacity.

The sequence simulation also serves to confirm the contingent nature of Skretting capacity. When the run by run capacity of the (least variable) SKR sequence is plotted against the average capacity (Hopp 2008), the variable and contingent nature becomes evident (see Fig. 7) bringing into question the suitability of this discrete founded capacity definition for building and executing the schedule in process manufacture.

Fig. 7
figure 7

Actual Skretting sequence capacity against Hopp’s (2008) average capacity

Full size image

4.2.3 Conclusion on Sequence

In conclusion, the process industry trade-offs identified in the literature are present within Skretting. Consistent with the literature, these trade-offs can be mitigated by a production sequence appropriate for the environment.

In contrast to the sequencing methodologies founded in discrete industry pull, the Skretting sequence provided highest performance in mitigating the trade-offs of capacity (throughput, lead time and variation), waste (material and energy) AND service (OTIF) identified in the literature (see Fig. 8).

Fig. 8
figure 8

Literature derived model illustrating analysed Skretting process industry trade-offs and the importance of sequence

Full size image

In terms of the principles of flow and pull, the Skretting sequence better aligns capacity and demand, minimises waste and optimises flow to the customer and as such in the Skretting case is an essential element of pull in process manufacture.

5 Discussion

5.1 Pull in the Process Industries

The literature view of pull is both ambiguous and strongly founded in discrete manufacture, focusing on the mechanisms of pull as opposed to the general principles that inspired Ohno. This is particularly evident, for example in Hopp and Spearman (2004), where make to forecast is regarded as pull if it is executed using kanban and takt. Forecast-based overproduction using kanban and takt is nevertheless overproduction and here Hopp and Spearman (2004) are arguably losing sight of the fundamental principles of demand-based flow to the customer in favour of the manufacturing mechanisms of executing pull.

The literature suggests that the process industries have been slow to adopt pull production and this paper concludes that the limitations of discrete based mechanisms contribute significantly to this (Table 7).

Table 7 Conclusion on discrete lean imperatives and limitations (adapted from Yoho and Rappold 2011)
Full size table

The authors agree with the literature distinction between principle and mechanism and conclude that as a result, the principles of pull can be executed in different ways.

The alignment of demand and capacity is at the heart of lean thinking (Lyons et al. 2013) and as such by focusing on the principles of flow and reduction of waste and addressing the PPC fundamentals of demand, capacity and variation rather than prescriptively applying discrete founded mechanisms, pull can be achieved in a process environment.

If we try to define process industry pull purely in terms of mechanisms which constrain inventory, then without the capacity flexibility of discrete firms, the process industries will have issues managing both demand and inherent variation.

This unavoidable process industry variation results in a significantly more hostile environment requiring alternate buffering and a variation accommodation strategy (Frei 2006), as opposed to the variation reduction typically found in discrete manufacturing.

Table 8 concludes potential process industry pull mechanisms based on literature principles and the Skretting case environment where continuous and inflexible equipment forces a different solution to discrete industry flexible capacity buffered, intermediate WIP bounded, single piece flow.

Table 8 Pull in the process industries: potential process industry approaches based on principles of flowal process industry approaches based on principles of flow and pull (adapted from Yoho and Rappold 2011)
Full size table

The Skretting case approach to pull provides the same benefits of throughput (increased capacity), inventory (reducing age profiles), rework (reduced waste) and customer service (high OTIF) championed by Hopp and Spearman (2008) on behalf of WIP bounded discrete industry pull. In Goldratt and Fox’s (1986) terms, inventory is improved, operational expenditure (in the form of material, obsolescence and energy waste) is reduced and service is improved for existing throughput whilst increased capacity is provided for additional throughput.

In conclusion, the Skretting case provides a process industry pull solution that is supported by the literature, the specific mechanisms of which will not necessarily apply to every process business but the principles arguably will. The majority of the literature’s attempts to define discrete pull do so referring to the mechanisms rather than the concepts and principles of flow, in response to which the authors propose the following principle based definition of pull in the process industries:

“The alignment of demand and capacity to provide the optimal trade-off between capacity, waste and service that delivers stable and predictable flow of that demand to the customer”.

5.2 The Importance of Sequence

Expanding on Kingman and the opportunities it presents within PPC, the traditional operational literature suggests that there is an additional element at play within process manufacture—that of sequence which can help mitigate trade-offs. This was supported by the Skretting case analysis where the Skretting sequence not only improved capacity and waste, but less expectedly improved service and unexpectedly reduced capacity variation.

However, the influence of sequence on PPC and the fundamentals of Kingman in the process industries whilst also unexpected, is further concluded here.

Lead-time in a queue is a product of time, utilisation and variation all of which are impacted by sequence in a process industry environment as shown in Fig. 9. In the Skretting case analysis a difference of between 10 and 20 % was seen in total process time, utilisation and capacity variation depending on the sequence used.

Fig. 9
figure 9

Model showing proposed process industry impact of sequence on Kingman (adapted from Bicheno 2011)

Full size image

Expanding on Kingman, the Skretting case analysis demonstrated that during weeks where utilisation was highest, sequence was extremely important and actually determined schedule feasibility. Changeovers are a source of variation and accordingly in the process industries, sequence impacts capacity in terms of the sequence of work and the generation of waste and load through waste generated failure demand (Seddon 2003) (see Fig. 10).

Fig. 10
figure 10

Model showing proposed process industry impact of sequence on utilisation (adapted from Bicheno 2011)

Full size image

6 Conclusion

The increasingly competitive environment and resultant ‘off diagonal’ activity within process manufacture causes issues for traditional literature models founded on the linear product/process approach. Existing literature frameworks are both discrete founded and taxonomy focused, inadequately describing underlying process industry complexity which is both inherent (e.g. complex BOMs, material variation and inexact processing) and relative to the environment (e.g. inflexible capacity combined with significant variability in demand and fulfilment). This complexity brings practical challenges for PPC where literature models do not satisfactorily describe the realities of hybrid MTS/MTO fulfilment and so a contingent approach based on PPC fundamentals and the trade-offs that can be influenced is required.

Discrete manufacturers have benefitted from pull production control resulting in improvements in lead time, cost, inventory and service but the process industries have been slow to follow this approach typically advocated for more stable, predictable production environments. The literature definitions of pull are ambiguous and research is focused on discrete industry implementations and the mechanisms as opposed to the principles of pull and flow. Commonly accepted discrete industry pull mechanisms lack applicability in process manufacturing but the principles of pull and the fundamental alignment of demand and capacity can be used to derive environment appropriate mechanisms which accommodate variation and support flow.

Such mechanisms must consider the process manufacturing trade-offs between capacity, waste and service which can be mitigated by sequence, the significance of which is not explicit in the lean literature. Here it should be noted that the case company sequence outperformed traditional lean scheduling approaches, generating outcomes contrary to the literature.

In process manufacture Yoho and Rappold (2011, p. 60) propose the use of a “complementary inventory policy” with “finished goods inventories in the right product at the right time…” (ibid, p. 67) asking “in what quantities and in which specific products should inventory be carried?” (ibid, p 59). In implementing demand based pull in a process environment we should add: “…and in what order should they be manufactured?”

Merging the significant operational literature on scheduling with the principles of flow/pull and the Skretting case analysis demonstrating both the influence of sequence on the fundamentals of Kingman and the impact of sequence on flow and waste, the authors conclude that the critical element of pull in the process industries is that of sequence.

7 Limitations and Future Research

Whilst the research accessed a longitudinal study of flow within the case operation, the data collection period, being less than one year does not represent the full extent of case company seasonality. The majority of this data is secondary data, not collected for the purposes of the research and therefore subject to bias and the context in, or purpose for which it was collected.

The conclusion whilst founded in the literature is triangulated using a single case study and, as such, it is not possible to generalise. The process industries encompass a wide variety of manufacturing with differing points of product commitment, differentiation and decoupling (between push and pull) ranging for example from petrochemicals where the vast majority is process based, to food and other FMCG where batch processes exist and products becomes discrete at some point. As a result, further research is recommended to determine if the research conclusions and the criticality of sequence apply to the implementation of pull in other process industry environments.

Additional research including multi-variant analysis is recommended to understand the degree to which each variable impacts flow. This may (dependent on the dominant variable) enable some reduction of influence and consequent reduction of buffers.