The design of a decision framework for selecting a shop floor control system for manufacturing plants.

(1)

- 1 -

The design of a decision framework for

selecting a shop floor control system for

manufacturing plants.

MSC THESIS

Double Degree in Operations Management

University of Groningen, Faculty of Economics and Business University of Newcastle Upon Tyne, department of Operations

2013-2014 Version 5.2

Student: P. (Patrick)Hoendervoogt

Email: patrickhoendervoogt@hotmail.com Student number: 1815156

Supervisor RUG: Professor dr. ir. (Hans) J.C. Wortmann Supervisor NUBS: Professor C. (Christian) Hicks

Coordinator: dr. J.A.C. (Jos)Bokhorst

(2)

- 2 -

Summary

Purpose: The aim of this thesis is to design of a decision framework for selecting a shop floor control system for manufacturing plants.

Design/methodology/approach: A case-study has been performed at an agricultural company with 120 employees and a turnover of 25mln€.

Findings & Practical implications: The framework provided not just information about the feasibility of the different SFC systems, but delivered also accurate information about the current state of the organization.

Research limitations: The framework could not be tested on its validity; this has to be done by another person. Also with one case study this is not possible. Level of insight, speed and effectives are measures used in this research.

Originality/value: This framework will be valuable for companies to reveal which SFC system is a feasible match to their manufacturing characteristics. The assessment of company characteristics matched to a SFC based on advanced planning & scheduling & control in contrast to WIP controlled SFC has not been done before.

(3)

- 3 -

Preface

This dissertation is submitted for two degrees of Master of Science; “Supply Chain Management” at the University of Newcastle Upon Tyne and for “Technology and Operations Management” at the University of Groningen. The research described herein was conducted under the supervision of Dr. Ir. Warse Klingenberg and Professor Dr. Ir. J.C. Hans Wortmann in the department of Operations, University of Groningen and Professor Christian Hicks in the School of Mechanical and Systems Engineering, University of Newcastle upon Tyne, over the period from June 2013 to January 2014.

This work is to the best of my knowledge original, except where acknowledgements, notes and references are made to previous work. Neither this, nor any substantially similar dissertation has been or is submitted for any other degree, diploma or other qualification at any other university.

It has been an honour and joy to work with many individuals, both inside and outside the university, who made this dissertation possible. I am grateful to my supervisors at both universities, for their careful, insightful and efficient supervision during my program in both countries. Without their generous and indispensable help neither the conduct of the study nor the accomplishment of this thesis would have been possible. The design and plan of the work were wholly due to Warse Klingenberg. I have been deeply grateful that, during the rough time he had, he still managed to read through my thesis and offered me detailed feedback and comments. Unfortunately, Warse passed away without seeing the completion of my dissertation in which he had invested much time and energy.

(4)

- 4 -

1. Introduction

Today companies are faced with several new challenges such as globalization, increasing competitive pressure, shorter product life cycles, and shorter technology innovation cycles. Customers are demanding more variety and increased quality, delivered in shorter lead times and at a lower price. Never before companies faced a greater need to become more agile and responsive (Goeldner

et.al.2011). In order to survive, companies have had to develop strategies to deal with these challenges.

Two popular ways are the implementation of lean practices and the integration of information technology (IT). Lean1 production is a strategy that aims for the production of high quality products at the pace of customer demand and with little waste (Ward and Zhou, 2006).

Since a good functioning of the production planning and control concept is crucial for the economic success of the enterprise, the selection of a fitting PPC concept is an important decision process (Henrich

et.al.2004). PPC is concerned with reconciling the demand and the supply of products and materials in

terms of volume, timing and quality (Meyer et.al.2011 p.1). The activities required to achieve this are typically clustered into four broad functions: (1) loading, (2) sequencing, (3) scheduling and (4)

monitoring and control (Slack et.al.2004). The first three collectively constitute the production planning

function; the fourth the production control function. Production control is an optimization problem that typically addresses the question of when and how much to produce in order to achieve a satisfactory customer service level (measured by how quickly customer demands are satisfied), while keeping low in-process inventories. Difficulties in production control arise because of queuing delays due to variability in production capacity (e.g., due to the failure or maintenance of a machine) and demand for final or intermediate products (Liberopoulos &Dallery,2000). This Thesis will focus on scheduling & monitoring and controlling production, or SHOP FLOOR CONTROL (SFC). Typical functions of a SFC system include reducing WIP, minimizing shop floor throughput times and lead times, lowering stockholding costs, improving responsiveness to changes in demand, and improving delivery date (DD) adherence (Stevenson et.al,2005). There are important objectives, and choosing the right SFC system is hence a crucial strategic decision. The characteristics of the production environment determine the degree of variety the SFC approach should be able to cope with. Tools of SFC directly address control by short term decisions, by moving material and workers, adjusting processes and equipment and taking actions required to ensure the system continues to function towards its goal (Hopp&Spearman, 2008).

The scope of control lies within manufacturing plants and is focused on local control of capacity resources within production departments and on material coordination between those departments. Within the recent literature on operations management, a debate is going on whether production control should be executed with modern IT systems for SFC or not. This seems to have led to the development of two divergent paths; one for those who have applied computer-aided production management in the form of material requirements planning (MRP), Manufacturing Resource Planning

(7)

- 7 -

(MRP 2), or scheduling software packages; and the other for the “lean purists who have neglected to see the value in the application of computer systems”2 (Strandhagen et.al.2012).

The last mentioned path is will be elaborated now first. It is part of a more general approach to tackling the production control problem, viz. to restrict the search for a production control policy to a class of simple, sub-optimal policies that are easy to implement and try to determine the optimal policy within this class (Liberopoulos &Dallery, 2000). Much of the research effort in this area has focused on developing and evaluating simple production control policies that depend on a small number of parameters and have often emerged from actual industrial practice. In many of these policies production is triggered by actual demands for final products or based on the status of the production system. Such policies are often referred to as WIP production control systems. An important feature of these systems, which is common to many other production control systems, is the use of tokens, which usually consist of cards that authorize certain production tasks to be performed (González et.al.2012). This thesis also deals with variety oriented SFC approaches that control order release by limiting the amount of orders in the process. Each approach anticipates on different types of variability, but there is a limited amount of theory available which addresses this issue (Dogger et.al,2010).

González et.al.(2012) explain that this approach became popular since the emergence of the lean philosophy in the late seventies, most firms changed the way they control production and inventory. One of the major issues they have to face is the reduction of work in process (WIP), without deteriorating the customers’ satisfaction. Production control by reducing WIP while meeting the target production rate, improves the ability of identifying quality problems in the system earlier (Colledani&Tolio, 2011). In this thesis; production control based on regulating maximum set WIP levels through controlling order releases and work flows is further referred as “WIP-control”.

The other path is part of a second more general approach to tackling the production control by formulating it as a nearly optimal control problem and then try to determine an optimal control policy for this problem with regulating WIP levels, “scheduling based production control”. The scheduling system should only suggest the release of workable work to the plant. It is counterproductive to release scheduled work until there is available capacity, where equipment and tooling are performing within specifications, and sufficient materials are available and of acceptable quality (Bell, 2005).

Ensuring the release of workable work requires interaction between the schedule, material planning and control, and capacity planning. A capacity planning system considers resource availability and detailed routing information when calculating the schedule, looking for gaps in the schedule where jobs (or portions of jobs) may be inserted. The more detailed and accurate information available to the scheduling system on capacity, routing steps, and start times, the more precisely it can calculate expected due dates, maintaining a valid release schedule (Bell, 2005).

(8)

- 8 -

Scheduling, monitoring and control of manufacturing equipment and automated control of manufacturing steps have made great progress in recent decades (Meyer et.al. 2011). Advanced automated scheduling, monitoring and control can help make production systems leaner, and even more significantly, make the entire supply chain leaner (Bruun&Mefford, 2004). Recent developments in wireless technologies have created opportunities for developing next-generation manufacturing systems with real-time traceability, visibility and interoperability in shop floor planning, execution and control (Zhang et.al.2011). In this thesis, the use and possible integration between ERP, Manufacturing Execution System (MES) &Advanced Planning and Scheduling system (APS) shall be used as an example of “Scheduling based production control” from this point referred to as “Scheduling”.

Both methods are proven to increase productivity, but within the literature there is still not a procedure or a model that could tell when a company should choose for production control done by scheduling without WIP-control, or to choose to implement WIP production control systems.

Within literature, researchers focussed on the comparison of Kanban WIP-control and MRP based scheduling concepts for the control of repetitive manufacturing systems (Rice&Yoshiwaka, 2003). Or description of different card-based WIP-control systems (Gonzales et al. 2012). And Stevenson (2005) wrote a review of production planning and control: the applicability of key concepts to the make-to-order industry, which included MRP as well. A gap within literature is the analysis of which manufacturing characteristics determine the applicability of more advanced automated scheduling systems compared to the applicability of WIP-controlled systems for SFC.

While selecting and implementing a suitable SFC concept, different stages3 can be distinguished (Henrich

et.al. 2004).

Preliminary Study & Evaluation;

Detailed Investigation & Final Selection; Implementation.

Henrich et.al.(2004) continues to say that, in practice mostly external consultants support companies in selecting a suitable concept in the ‘Preliminary Research & Evaluation’ stage. This decision-making process is frequently based on intuitive reasoning rather than on an objective evaluation of the company characteristics and the considered production planning and control concepts. Moreover the selection is based on the experience of the advisor, collected in prior projects. There is a big need to make this initial selection procedure more transparent.

Therefore, the main research question of this thesis is:

(9)

- 9 -

How can a decision framework be designed which helps a company to determine, within the

Preliminary Study & Evaluation stage, which system should be implemented for shop floor control?

The decision framework that will be developed in this thesis will include different already existing methods of production control. The aim of this thesis is to assist companies in the decision to implement a specific token-based WIP-control system, or to implement automated scheduling, monitoring and control. Therefore a number of sub research questions have to be answered first. These are presented as follows:

1. How to conceptualise IT solutions for SFC, which choices are faced? 2. Which different token-based SFC systems do exist for WIP-control?

Sub question of 1 & 2: What are the individual characteristics of these control systems? 3. Which characteristics of production are relevant to determine a SFC method?

The questions above are addressed to gain information which is used to design the framework. For the evaluation of the framework sub questions are presented as follows:

4. How useful is the framework to give relevant information of the current state of the company? 5. How effective could the decision framework be used by the company in advising an appropriate

method for SFC?

(10)

- 10 -

2. Methodology

2.1 Case study

In this study a qualitative approach has been adopted, as the aim is to provide an in-depth analysis on which considerations will determine whether a token-based WIP-control or a Scheduling system or a combination should be implemented for SFC. The designed framework will be used within a case study and tested on its effectiveness.

Silverman (2001) argues that qualitative data has the ability to provide a deeper understanding of certain phenomena than quantitative; therefore this approach is selected by applying an explorative case study research methodology. Yin (2009) suggests that case study research is the most appropriate overall research methodology if how or why questions are being posed, if the researcher has little control over events, and if the focus is on contemporary phenomenon within a real life context. As the research question is a how-type question and as the focus of the project is on contemporary phenomenon within a real life context (to or not to implement WIP-control for SFC), a case-study research is chosen to carry out. The goal in case study research is analytic generalisation and not statistical generalisation (Yin, 2009). One drawback of this methodology is however its time-consuming nature, which makes it necessary to limit the number of studies to a maximum of one.

2.2 Data Collection

The data for the decision framework is gathered from scientific journals. To test the framework, semi structured interviews are conducted over an 1,5-month period with members involved with production control. 4The Case study is done at an agricultural equipment factory, which for privacy reasons is called AGRIFAC. In terms of data collection for the case study involved a semi structured interview with a primary on-site contact, which was the production manager. Consultants and project manager involved in the ERP or pull system implementation were also present. Participant observation, and document reviews shall be used to collect data from the organizations. Digital field notes and digital recorded audiotaped discussions shall be saved in a case study database. This shall provide multiple perspectives, strengthening objectivity and pattern recognition in data gathering and triangulation and leading to synergy (Chen et.al.2008). In turn, this will help to establish external construct, internal validity, and reliability of the data which is supported by Eisenhardt (1989).

2.3 Data analysis

The data analysis starts with a case analysis in order to determine the pattern of data. This is done first to ensure familiarity with the complexity AGRIFAC faces, to allow the unique patterns of the case to emerge (Eisenhardt, 1989).

(11)

- 11 -

After the in-depth understanding is shaped, an analysis and comparison between theory and practise findings at AGRIFAC will be performed, which will enhance generalizability of the results for identical companies. By constructing a table, the similarities and differences can be easily visualized between the theoretical framework and findings out the case study analysis.

(12)

- 12 -

3. Background theory

There are three focal areas to this thesis. The first is SFC, which sets the boundaries for the research, the second is token-based WIP control which could be supported by IT or not and the final area is automated advanced scheduling, monitoring and control. By conducting a literature review, this section explores all three concepts in order to identify useful insights from theory.

The distinction in this thesis between WIP control and Scheduling is related to the distinction known in literature as push and pull SFC systems. The two have emerged from fundamentally different approaches to production management: WIP control out of Just-in-time, which is related to pull production from Japan (Sugimori et al., 1977); and push scheduling from America and the West (Wight, 1984).

3.1 PUSH vs. PULL

(13)

- 13 -

Pull production control can be divided into three main levels: job entry, job release, and priority dispatching/WIP-control (Powell et.al,2013). The job entry level considers demand and capacity management, for example the decision as to whether to accept or reject orders, the setting of due dates, and the management of capacity. Job release determines the time of release for accepted orders, and short-term adjustments in capacity. Finally, priority dispatching/WIP-control manages the flow of orders on the shop floor (Powell et.al,2013).

Examples of both push and pull systems will be given next, where characteristics and functionalities in addition to the advantages and disadvantages will be covered. The summary of this analysis will be covered in a table at the end of this chapter. Based on the following analysis a decision framework will be presented (at pg. 42) which matches the functionalities of the different methods with company characteristics.

3.2 Push systems

MRP 1 is a set of techniques that uses bill of material data, inventory data, and the master production schedule to calculate requirements for materials. MRP makes recommendations to release replenishment orders for material. The need for an integrated or unified infrastructure has driven the development of MRP I, a tool for generating planned order releases, to MRP 2, the overarching planning system incorporating MRP 1. Finally to ERP systems which extend the MRP 2 hierarchy to multiple facility systems that supports a large number of core business processes and represent the closest commercially available single unified system (Hicks, 2007).

3.2.1 Traditional MRP II systems

According to Liu et.al.(2002) most ERP systems (based on MRP 2 logic) have Rough Cut Capacity Planning (RCCP) feature. RCCP looks into the capacity requirements for key resources and compares them with the available or demonstrated capacity for the key resources. This comparison assists the master scheduler in establishing a feasible master production schedule. Another capacity planning tool in ERP systems is Capacity Requirements Planning (CRP). Working together with the MRP 1 or 2 modules, CRP determines in detail the mount of labour and machine resources required accomplishing the tasks of production.

3.2.1.1 MRP Advantages

There is much that is positive about the recent evolution of ERP5 systems. The integration and connectivity they provide make more data available to decision makers in a more timely fashion than ever before (Hopp&Spearman, 2008). Recently research is done on lean production combined with ERP systems, which showed that ERP systems could support pull production; one of the basic principles of lean production (Womack and Jones, 1996). To which extent the usage of a company’s ERP system supports pull production practices could be measured by a capability maturity model developed by Powell et.al.(2011). At first, this combination looks contradictive, but as Powell et.al.(2013) explain, the

(14)

- 14 -

effectively usage of ERP systems really could contribute to lean production where both fundamental different methods originally support each other. In combination with card based WIP control system, MRP systems could support pull production.

A disciplined approach to standardized business processes is the key. There will, in any case, be a clear decoupling of the two concepts, and a push–pull point will be identifiable (Powell et.al. 2013)– this is the point at which the push planning of the ERP system ceases, and the control of the pull system takes over on the shop floor. At the job entry level, this means the demand and forecast data is available for the pull system in order to allow for manual demand smoothing. At the job release level, the ERP system should support both orderless and order-based production6, as this is a requirement for firms who are just starting lean implementations. At the order control level, ERP should have functionality for manual printing of cards and colour-coded release lists (Powell et.al. 2013).

3.2.1.2 MRP Disadvantages

A pure push system can release work to a very congested line, only to have the work get stuck somewhere in the middle. The result will be a loss of flexibility in several ways (Hopp&Spearman, 2008). First, parts that have been partially completed cannot easily incorporate engineering (e.g. design) changes. Second, high WIP levels impede priority or scheduling changes, as parts may have to be moved out of the line to make way for a high-priority part. And finally, if WIP levels are high, parts must be released to the floor well in advance of their due dates.

Using traditional push ERP systems based on MRP 1- or 2- logic, without a combination of pull card based systems to control the WIP level on the shop floor often results in problems. It means that there is no feedback loop to shop floor production scheduling inventory and quality control data (Bruunet.al.2004). This type of information is essential to collaborative production planning and inventory management throughout the supply chain as well as making build-to order production feasible. This control system dictates the workstations at each moment what to do. However, by assuming infinite capacity, the dynamics of limited resource availability is completely ignored and therefore stresses the planning of resources too much and often results in unfeasible plans for companies where machines are always overloaded (Liu et.al2002; Vandaele et.al. 2005). Another disadvantage is that MRP uses a fixed lead time approach to control work releases, regardless off the level of effective utilization. Consequently, lead times are often dramatically underestimated (De Boeck et.al2007). Patches such as MRP 2 and CRP may improve the system in small ways, but they cannot rectify this basic problem. MRP 2 is still rather oriented on material problems rather than capacity aspects and the control of throughput time can still be a lot improved (Bertrand et.al. 1990). This fundamental flaw of MRP has been carried over into traditional ERP (Hopp&Spearman, 2008). To summarize, Stevenson et.al.(2005) argue that the customer enquiry stage is not specifically addressed, nor the capacity decisions at the point of job entry and job release are considered. So these traditional push systems are not aligned with the principles of lean and

(15)

- 15 -

therefore should this IT solution not used as a stand-alone solution. The next section will cover more advanced ICT solutions.

3.3 Advanced scheduling based production control

There exists software to provide the links between planning and shop floor activities, for closed loop SFC which is not provided by the push system described previously7. For example the integration between MES and APS- Scheduling System (APS) as additions to the existing ERP system.

APS-scheduling systems could be used for production scheduling within the PPC framework and MES Systems are used for monitoring and control. Production control by IT, only works well in such a way, that these different systems should work together and therefore it is important to explain how these systems work on planning and scheduling as well in addition to SFC. The real time feedback loop is what these systems make effective.

Figure 1 Example of integrated IT (Liu et.al2002).

3.3.1 MES systems for monitoring & control

In order scheduling software to make accurate recommendations it must know the current status of every work-centre and resource, every job, and all materials available for production. MES mainly concentrates on managing shop-floor operations such as monitoring as well as its execution and control, timely informing shop-floor supervisors in terms of equipment status, material delivery and consumption as well as manufacturing progress in real time (Zhong et.al.2013). It often is combined with (electronic) visual dashboard where planning, routing and performance related information is made available to assist shop floor workers to the process. This function can also be used to visualise where operators are required on the shop floor and to make adjustments in capacity thus supporting line balancing.

(16)

- 16 -

Figure 2 MES Functionality8

Huang et.al.(2008) explain that WIP inventories are traced and tracked throughout the manufacturing processes on a real-time basis. The real-time traceability and visibility of WIP materials and information facilitate the identification of the shop-floor bottleneck and improve shop-floor performance. Capacity data are essential to several planning decisions, which mostly rely on estimated data. A widely used practice for estimating capacity of currently installed equipment is to start with the rated capacity (e.g. parts per hour) and reduce this number according to various detractors (machine downtime, operator unavailability, setups, etc.). Since each detractor is subject to speculation, such estimates can be seriously in error. For this reason, it makes sense to use the production tracking module to collect and update capacity data used by other planning modules (Hopp&Spearman, 2008).Such up-to-date shop-floor information is then send back to ERP and MES for better planning, scheduling, and control decisions. (Huang et.al.2008).

Wireless manufacturing can replace the paper-based manual WIP tracing and tracking system by RFID-based smart devices and systems, and ultimately improves and optimizes the shop-floor operational efficiency and effectiveness. The system closes the loop of SFC by providing real-time feedback of such sensory data for adaptive decision making (Huang et.al.2008).

8

(17)

- 17 -

Figure 3 MES implication 9

3.3.2 APS-scheduling systems

A second tool that often is used for production control, in specific production planning and scheduling is APS-based scheduling systems. Gruat-La-Forme et.al.(2005, p.4) use the definition out of the APICS Dictionary, which defines APS-based scheduling system as follows: “APS describes any computer program that used advanced mathematical algorithms or logic to perform optimization or simulation on finite capacity scheduling, sourcing, capital planning, forecasting, demand management. These techniques simultaneously consider a range of constraints and business rules to provide real planning and scheduling, decision support, available-to-promise capabilities. APS often generates and evaluates multiple scenarios. Management then select one scenario to use as the official plan”.

The issues and problems that are addressed by APS range from the longer term capacity requirements to the short-term day-to-day management of the production floor explains APS patent maker Chua

et.al.(2011). At the higher level, it examines the customer forecast/demand against the existing capacity

constraints, in order to facilitate strategic decision as to whether a major purchase of capacity should be made or should sub-contract to third-party be considered. On a daily basis, it determines what should be the product-mix to be released in a particular day or shift, so that there is continuity between successive product loadings while observing the capacity constraints. At the same time, detailed production schedule is also being generated to guide the production supervisor on the actual execution of the customer orders on the production floors, considering the various production goals of minimisation of WIP, maximisation of machine utilisation and meeting the customer order due date (Chua et.al., 2011). The heart of the APS is its Finite Capacity Scheduling (FCS) capability; FCS ensures the jobs are loaded into work-centres such that no work-centre capacity requirement exceeds the capacity available for that work-centre. FCS outputs a listing of manufacturing orders in priority sequence down to the machine level. This dispatch list is sent to the shop floor MES for production control and execution (Liu et.al2002).

9

(18)

- 18 -

Figure 4 closed loop system integration (Liu et.al2002).

3.3.3 MES & APS combination

(19)

- 19 -

taking emergency actions can prevent causing serious disruption of the factory for little gain (Hopp&Spearman, 2008).

In order for APS to work, sufficient and up-to-date data must be communicated from ERP systems. This is not sufficient and dynamic real-time shop floor data is captured by MES. Hence one of the challenges for a successful implementation is to ensure the timely tight integration between not only APS and ERP, but also APS and MES (Liu et.al2002).

Besides its detailed planning and scheduling functionality, APS also plays an important pivotal role to achieve manufacturing synchronization. The bi-directional relations among APS, ERP and MES reflect the innate control and feedback nature between each two of them. For example, APS eventually controls the production since the MES production schedule is provided by APS. But control is only feasible provided that MES can give timely feedback of the real shop floor status. Hence the control and feedback between systems are closely related, with data flow in both directions (Liu et.al2002).

(20)

- 20 - 3.3.4 Advantages scheduling

According to Bruun&Mefford (2004) intelligent resources can help make production systems leaner, and even more significantly, make the entire supply chain leaner. The same authors argue that as lot and order sizes come down due to the closer coordination of production schedules, firms will be forced to develop faster and more efficient ways of setting up runs of products and order delivery to customers. With advanced planning systems closer coordination of production schedules and faster adjustment to changes in demand are possible, this will facilitate information transmittal internally within the firm. De Boeck et.al.(2007) say that for companies working in a make-to-order environment, it is vital that the lead time quoted to the customers is reliable. The functionality of the APS module allows to dynamically adjust lead time quotations to changes in demand or shop floor conditions, such that the sales department’s promises are in sync with the manufacturing department’s capabilities and due date performance is secured. This function is done by a “What if” simulation (Hopp&Spearman, 2008). If output times of specific jobs could be accurately estimated, and the module could project how the system will evolve (i.e. what jobs will be in the line and what jobs will be waiting in queue) over time, the performance of the line could be simulated. This would provide the basis for a “what if” tool for analysing the effects of different priority rules or capacity decisions on outputs. Capacity requirements planning (CRP), a module of MRP 2 software often used in ERP systems, attempts such an analysis. However, as Hopp and Spearman (2008) pointed out, CRP uses an infinite-capacity model that invalidates predictions beyond any point where a resource becomes fully loaded. More sophisticated, finite capacity models for making such predictions are available.

3.3.5 Disadvantages scheduling

Scheduling problems are notoriously difficult and Information technology without a suitable model of the flow process is fundamentally flawed (Hopp&Spearman, 2008). While more accurate then CRP, finite loading models frequently have massive data needs and complex computations. Finite scheduling uses heuristics and simulation approaches which is based on deterministic calculations. This causes a solution never to be optimal but instead gives a possible good feasible solution. However, the deterministic calculations do not take into account variability. Therefore this approach could only work well if the schedule gets updates frequently or has a real time feedback loop.

Another important issue that should be considered choosing intelligent resources combined with APS is that the throughput time is not reliable. The main factor contribution to this is the possible wide variation in workload to the different capacity resources in combination with the possible variation in processing times and the big diversity in types of capacity. The software fills gaps within the schedule with tasks that do fit without deteriorating the constraints.

(21)

- 21 -

planning and control decisions on them (Hopp&Spearman, 2008). The next paragraph will give an introduction to SFC which regulates workloads in a plant by controlling WIP levels.

3.4 Pull - WIP control card based systems

In the last 30 year of Ops management research a vastly amount focused on how and when to authorize jobs to be completed, so as to maintain a reduced amount of WIP while meeting customers’ demand (Hopp&Spearman, 2008). In token-based systems, the processing of jobs on a given workstation is authorized if the corresponding tokens or different release mechanisms are available in the control panel.

As jit literature correctly points out, if one wants to maintain high levels of throughput with low WIP levels (and short cycle times), one must reduce these disruptive sources of variability (failures, setups, recycle, etc.) Hopp and Spearman (2008) note that, the source of this pressure is the limited WIP level, not the mechanism of pulling at each station. Pulling at each station controls the WIP level at every point in the process, which would not necessarily be the case with a general WIP cap. However, reducing overall WIP level via a WIP cap will reduce the WIP between various workstations on average and thereby will apply the pressure that promotes continual improvement. Deeper analysis of different card based systems is performed in chapter four of this thesis.

3.4.1 Pull advantages

Compared to basic MRP push systems, Spearman and Zananis (1992) conclude that: there is less congestion in pull systems, and are therefore inherently easier to control then push systems. The benefit of a pull environment owes more to the fact that WIP is bounded than to the practice of pulling everywhere. Riezebos et.al(2009) state that pull systems are transparent and easy to understand, and do not rely on computer technology to control workflow. In the design of these systems, special attention has to be paid to the release of work to the production system and to ensure that the estimated throughput times are accurate. These accurate estimates play a key role if other production units in a supply chain or production process are demanding stable throughput times. Also, card-based pull systems are cheaper to implement then complete IT solutions.

(22)

- 22 - 3.4.2 Pull disadvantages

An important disadvantage of these SFC approach is that card based systems depend on (additional external/ not integrated) sequencing and scheduling modules to determine which orders will be released in the system and what the planned release date for the order will be (Riezebos et.al2009).

Another important point is that the average lead time will impact the average number of cards that need to be present in the system while information on the lead time percentiles is useful when determining the number of safety cards necessary to protect the target throughput rate against variability. In the absence of adequate lead time information, card levels are set rather ad hoc or based upon experience; this may either lead to missed throughput rates (when card levels are set too low) or to an unnecessarily high level of WIP on the shop floor (in case card levels are set too high) (De Boeck et.al,2007).

According to Lidell, (2008) and Liu et.al,(2002) paper card based systems are too manual; too slow to react to change, too time consuming in case of resizing the cards or replenish lost cards. Furthermore, manual card-based control systems are not able to accurately prioritize upstream work and are not able to intelligently sequence upstream work, as intelligent IT systems mentioned later on can do.

3.5 Functionalities assessment

(23)

- 23 -

3.6 Typology of production situations

(24)

- 24 -

Figure 6 Typology of customer order decoupling point adapted from Wortmann (1992)

(25)

- 25 -

Figure 7 Decision pointanalysis for Push/Pull adapted from Bell (2005)

Bell (2005) gives a guideline to determine the optimal push/pull positioning of a particular value stream within the overall supply chain at each CODP, which is showed in Fig. 7. At the far upper left corner of the diagonal is Engineer to Order, where each order is unique, resulting in a variety of disturbances such as emergency orders and frequent engineering changes. In most cases demand cannot be planned in advance, so most materials must be ordered from suppliers and all production is push scheduled, usually in a discontinuous job shop or on-site assembly project environment.

Down the diagonal is MTO, where similar products are fabricated to customer order. In this case some basic inventories may be planned in advance, and some processes may be grouped (product flow analysis, group technology) to achieve some degree of flow and demand pull (Bell,2005).

Because a considerable amount of parts and components used in ETO/MTO products are one of a kind, the involved setup times and processing times are difficult to estimate (Zhong et.al.2013).Such uncertain disturbances will affect normal production-plans and -schedules. Furthermore, these disturbances cause snow-ball effects such as delays on customer orders, logistics errors and high level of WIP inventories

(Zhong et.al.2013).

Further down the diagonal is ATO, where core assemblies may be purchased and produced in advance, stored in a final assembly supermarket, and pulled by the customer order (or a level schedule where each unit is configured separately) into a rapid final assembly process (Bell,2005).

(26)

- 26 -

A method for negotiating or setting job due dates, as the characteristics of jobs can vary greatly and due dates (DDs) must be determined individually.

A job entry/release stage and a method to determine the appropriate job entry or release time. A control scheme to manage the intersecting and conflicting routes of highly customized

products with variable shop routings.

These requirements are taken into account assessing the different SFC approaches later on.

Until this point in the graph, the production planning focus is on order execution and the performance measures are order focussed e.g. average response time, average order delay. The competitive priority is shorter delivery lead-time (Van Donk&Gaalman, 2004). However, customization invariably leads to non-standard product routings on the shop floor, and without the insurance of a FGI, lead times are naturally longer than for MTS companies (Kingsman et.al,1996). As a result, the ability to accommodate customization is a key test of the applicability of the SFC approaches reviewed in this thesis to the different industries.

At the far lower right corner of the diagonal is MTS which offer a low variety of producer-specified and typically, less expensive products. The focus is on anticipating the demand (forecasting), and planning to meet the demand (Kingsman et.al,1996).. The competitive priority is a higher fill rate. The main operations issues are inventory planning, lot size determination and demand forecasting. The performance measures are product focussed e.g. line item fill rate, average inventory levels (Vollman et al., 1997; Silver et al., 1998. Thus, production is repetitive, perhaps according to a level schedule or continuous flow, and all customer demand is pulled immediately from a finished goods supermarket (Bell,2005).

(27)

- 27 -

3.7 When do you need advanced scheduling?

As a manufacturing enterprise positions its products and processes along the diagonal, it must also determine the appropriate IT capabilities that add the most value (Bell, 2005). Generally speaking, as an operation tends toward the discontinuous end of the spectrum, complexity increases, requiring additional decision support and process control tools as suggested in Fig. 8. So naturally the degree of IT support required for planning, scheduling, and execution depends on the inherent complexity of the environment.

Figure 8 Bell (2005)

Unlike MRP systems which are often combined with card based systems, APS-schedulling simultaneously plans and schedules production based on available materials labour and plant capacity. APS& MES has commonly been applied where one or more of the following conditions are present (Lidell, 2008):

ETO/MTO (as distinct from make-to-stock) manufacturing;

Capital-intensive production processes, where plant capacity is constrained;

Products 'competing' for plant capacity: where many different products are produced in each facility;

Products that require a large number of components or manufacturing tasks;

(28)

- 28 - 3.7.1 Order-less vs. order-based manufacturing

In contrast to the order-based, time-phased requirements of push production, pull production uses a mechanism of order-less, rate-based planning. Production orders are no longer required in pull production, as requirements are communicated using tokens. However, the time buckets in the heijunka schedule are multiples of what is known as the takt time (Powell et.al.2013). As will be explained in 3.7.2, heijunka is not always possible. And in certain industries, e.g. medicine, aviation and aerospace industry governmental regulations oblige bookkeeping of all order specific manufacturing data related to each part/batch/ingredient its process and processor. Besides, scheduling and planning effort can increase as a result of converting from an ‘order-less’ production control system to one involving orders and tracking. Information needed to complete the order (components required, workcenters and operations required, tooling required, etc.) may be printed on paper or tickets, often called shop orders or work orders, which are distributed to production personnel. However, this could also be done digital. The information already exists in the databases (in case of MES implementation) so by integrating and executing this with APS& MES systems, efficiency can increase.

3.7.2 Scheduling and sequencing

Bell (2005) explains that scheduling in a discontinuous environment is particularly important because it is so unpredictable. In a repetitive environment with a level schedule, if you know the heijunka schedule at the beginning of the week you have a good idea of the expected rate of throughput. Combined with an understanding of the predefined product mix, backlog contents, and supermarket levels, you can make realistic promises to your customers. Not so with a discontinuous environment, where a single order can suddenly change the capabilities of the entire plant. So it is important to schedule a discontinuous environment as best you can, communicating the status clearly and regularly.

The goal is to provide people on the floor with enough information to enable them to make reasonable control choices, but not so much as to overly restrict their options or make the schedule unwieldy. What this means in practice is that different plants will require different scheduling approaches. In a simple flow line with no significant setup times, a simple sequence of orders, possibly arranged according to earliest due date (EDD), may be sufficient (Hopp&Spearman, 2008). Maintaining a first in first out (FIFO) system ordering of jobs at the other stations will yield a highly predictable and easily manageable output stream for this situation.

However, in a highly complex job shop, with many routings, machine setups and assemblies of subcomponents, a simple sequence is not even well defined, let alone useful10 (Hopp&Spearman, 2008). In the more complex situation, it will not be clear that the MPS is feasible. Consequently iteration between the MPS module and the advanced sequencing/scheduling module will be necessary. In complex situation such as this, a fairly detailed schedule may be needed with specific release times for jobs and materials and predicted arrival times of jobs at workstations. The data requirements and

(29)

- 29 -

maintenance overhead of the system required to generate such a schedule may be substantial but this is the price to pay for complexity.

To make appropriate sequencing decisions at the time of release, and as a downstream WorkCentre chooses jobs from an upstream queue, the software requires accurate information on capacity, cycle times, alternate routings, and constraints. Machine capacities and configurations must be known, including specific rules governing the sequencing and batching of work11 (Bell, 2005).

3.7.3 Multiple bottlenecks

In some systems, this can simply be a count of the units in the system. But in systems where different part types require vastly different process times, the same number of units does not represent the same use of resources (Hopp&Spearman, 2008). Hence, in order to maintain a level loading on the system, it makes sense to measure the WIP in terms of time required at the bottleneck. If the bottleneck is stable, this method can work well. If the bottleneck is different for different products, bottleneck-time is still a reasonable measure of WIP provided that the product mix does not vary too widely. If this is the case, more advanced computing systems are needed.

3.7.4 Shared resources

Shared resources complicate control of card based systems12. Control is complicated at a shared resource because a job must be picked to work on from multiple incoming routings. Shared resources also complicate prediction. Simple methods could be used for estimating the exit times of jobs with a product or line specific production line; it is not nearly as accurate for a line with resources shared by other lines. The reason is that the outputs from one line can strongly depend on what is in the other lines (Hopp&Spearman, 2008). A simple MRP module is not able to do this.

11

This is illustrated in appendix 1 note 5.

(30)

- 30 -

4. Deeper analysis of pull card based systems

This chapter gives a deeper analysis of different WIP control systems. After this chapter, the decision framework is presented. Within this thesis 4 card based approaches are presented: Kanban (González

et.al.2011;Stevenson et.al. 2005; Riezebos et.al2009), CONWIP (González et.al 2011; Stevenson et.al.

2005; Spearman et.al.1990), and POLCA or “Paired-cell Overlapping Loops of Cards with Authorization” (González et.al 2011 Stevenson et.al 2005; Suri,1998, Riezebos et.al.2009) and COBACABANA “control of balance by card-based navigation” (González et.al.2011; Land,2009). Each of these is a token system of cards for enforcing limits on work-in-process (WIP) in a shop. Together, these systems address four fundamental levels of WIP-control (Harrod&Kanet,2013):

Control of WIP associated with a single work-centre (Kanban); Cumulative control of WIP across the whole shop (CONWIP);

o

Cumulative control of total load of the system of WIP across the whole shop (CONLOAD); o Cumulative control of WIP across each set control loop (M-CONWIP);

Control of WIP associated with a given route between two work-centres (POLCA);

o Control of workload associated with a given route between two work-centres (LB- POLCA);

o Control of WIP, associated with availability of capacity of all workstations in the routing of an order at the time of release (G-POLCA);

Control of WIP by controlling the queues/workload of critical workstations (COBACABANA). Within literature there are more approaches described which are designed to cope with variety aspects, for high variety production environments (Krishnamurthy&Suri, 2009). However, the four described approaches with four additional variations have successful implementations described within literature. For the other more advanced approaches this is not yet the case (Stevenson et.al,2011) and it is common that these approaches have implementation issues (Hendry et.al,2008).

4.1. Kanban

(31)

- 31 -

Harrod&Kanet(2013) further explain that Kanban was originally conceived for systems with little product variety, i.e., product is homogeneous or commodity-like (e.g. MTS environment). Production processes in these systems are dedicated to repeatedly performing the same operation to the same production stock keeping unit. Rather than dealing with delivery dates for individual customer orders, it is the production rate or throughput rate that is the key concern. The focus then was traditionally on watching for and eliminating bottlenecks in the production process, with the assumption that any increase in production capacity automatically generates more throughput (and revenue). On the other hand, for ATO / MTO systems this assumption that capacity automatically generates more throughput does not hold. For those systems, it is the satisfaction of externally generated individual customer orders that drives revenue (Harrod&Kanet,2013).

For MTO companies, a direct relation between signal and product type is not useful. MTO companies face a much higher product variety, which would lead to a very large number of different bins or card loops. Next, the repetition of identical orders is not that frequent, which would lead to long waiting times of the intermediate stock in a bin (Riezebos,2010). In isolation, the Kanban is a de-centralized shop floor signalling system and lacks control at the customer enquiry stage, job entry and job release stages (Stevenson et.al.2005). Despite this, it may be possible to use Kanban on the shop floor in conjunction with a higher level-planning tool such as WLC; however this would still need a way to accommodate product variation, while the WLC job release function means that the shop floor can be controlled through simple priority dispatching without the need for Kanban signals. Although Kanban can only be implemented for standard items, aspects of the JIT philosophy and lean thinking approach, such as attitude towards waste and stockholding, could be adopted also in ATO & MTO environments (Stevenson

et.al.2005). For a more detailed explanation on the topic, see Sugimori.et.al.(1977) or Stevenson et.al.(2005).

4.2 CONWIP

(32)

- 32 -

parameter concerning this system is NC, i.e. the number of cards (González et.al.2011).

Figure 9 Conwip line (Hopp & Spearman, 1991)

CONWIP uses a combination of physical and virtual authorisation mechanisms (Riezebos et.al.2009). The physical mechanism, which may use either cards or containers, provides authority to the operators for new order releases. The virtual mechanism is needed to provide guidelines on which order to release. Hopp and Spearman (2008.p.469) denote this virtual mechanism as a “sequencing and scheduling module”, which might be similar to a dispatching rule such as earliest due date first, or a more complicated heuristic. Riezebos et.al.(2009) continue to say that in a CONWIP system the choice of items to produce is determined by the virtual mechanism. The physical system only indicates that a new order may be released, but does not limit the set of orders from which to choose. The sequencing and scheduling module determines which orders will be released in the system. The quality of this element within a CONWIP system may significantly affect the timing and balancing capability of the whole system. Hence, the role of IT support in achieving a good performance of the CONWIP ‘pull’ system should not be underestimated.

There are two aspects under discussion with respect to the operation of a CONWIP system (González

et.al.2011). The first is whether it can be seen as a push or pull system. Although classically considered a

pull system, some authors consider that has characteristics from both types of systems (see e.g.Pyke&Cohen.1990). The second is whether it performs better than other systems. Since usually, the performance of a system is dependent on the type of environment (scenario) where it is implemented, it is not possible to obtain general results (González et.al.2011). Within MTS environment Fowler

et.al.(2002) consider Kanban to be throughput control oriented while CONWIP is naturally more

focussed on WIP. Nevertheless in this environment, CONWIP can provide a greater throughput than Kanban (Spearman&Zazanis.1992).

(33)

- 33 -

conditions, it may no longer be reasonable to fix the WIP level in a CONWIP loop by holding the number of units in the line constant (Hopp&Spearman,2008). Since CONWIP only controls the number of orders and not the work content (processing times) in the system, this can lead to a fluctuating workload in the system. Furthermore the use of just one parameter at the system level can lead to an increase of the variability of the workload at each workstation and, thereby, decreases the workload balance.

CONLOAD is developed to consider the processing time variability (Dogger et.al,2010). As Conwip, it can be integrated with a high-level planning system. With CONLOAD each order that enters the system increases the total load of the system by the sum of bottleneck processing times divided by the average throughput time of this order type. The CONLOAD control has the advantage that the parameter setting is more intuitive than for the classical CONWIP-control, CONLOAD takes into consideration how much load is added to a single machine or a group of machines by a particular lot to decide on releasing this lot into the fab or not (Rose,1999). The only CONLOAD parameters that have to be determined in advance by simulation or queueing analysis are the average cycle times for each product (Rose,1999).

A Problem with both CONWIP and CONLOAD is that they control the workload at an aggregate level and therefore do not control the direct loads in front of each workstation. A decreased WIP level cause an average decrease of queue lengths, but only restricting average queue lengths is not sufficient. If average queue lengths decrease but variances do not, the idle time at workstations will increase. Therefore, the systems WIP level does not consider e.g. routing variability. It is possible that only orders with one type of routing are in the production system, while orders from a different routing wait outside the production system. This can lead to idle time at workstations although orders for these workstations are present, which means that these SFC approaches lack workload balancing capabilities. A way to deal with routing variability is to use a control loop for every possible routing in the production system instead of using just one control loop for the whole system. Dogger (2010) refers to three other sources; Germs and Riezebos (2009), Khojasteh-Ghamari (2009) and Hopp&Spearman (2008) who describe the M-CONWIP approach for a process with variable routings. Out of these sources is to conclude that the processes with variable routings within an M-CONWIP system have a better workload balancing capability than CONWIP.

4.3 POLCA

(34)

- 34 -

based on the production plan and does not apply a fixed limit to the amount of WIP. It enables the planner to control the progress of orders by having planned release dates for each order in one or more cells (Riezebos et.al.2009). Even if a cell has a card available that enables it to start producing an order, it is not allowed to start until after the planned release date for the order. Riezebos et.al.(2009) further explain that the decision on which items to produce depends on the second authorisation mechanism, i.e. on the list of orders with the planned release dates, and on the type of POLCA card that has become available. The card identifies the first two cells that an order has to visit, but not the type of product. The same card may be attached to totally different orders in the course of time, as long as these orders subsequently visit the same combination of cells. POLCA cards are therefore not product-specific but route-specific. The mix of orders balances WIP with respect to routings. The content of WIP changes over time if new orders are released. The orders do not need to be similar to orders that have been released before. POLCA will keep a balanced mix of the workload in terms of the routing within the first two cells (Riezebos et.al.2009).

However, as with other production signalling methods, POLCA relies on assistance at the higher planning levels to determine delivery dates based on workloads and capacities at the customer enquiry stage (Stevenson,2005). Similarly, it needs to be incorporated with other methods if it is to address job entry and job release. POLCA is a sophisticated prioritization and shop floor pull technique, regulating the movement of work between operations and cells. Although POLCA requires (APS) scheduling software to optimize the sequencing and releases in advance, once the job is released to the shop floor, POLCA pull mechanisms aid the flow without further software intervention (Bell,2005).

Using the POLCA philosophy, cards are placed in pairings where jobs travel in one-direction and information returns in the other. To be useful in a job shop, the system has to accommodate high routing variability, including upstream travel. To allow multi-directional flow and variable routing, it may be possible to place two cards between a pair of cells to allow work and feedback data to travel in both directions (Stevenson,2005). For this type of configuration, known pull systems such as CONWIP and Generic KANBAN are not well equipped (Krishnamurthy et.al.,2004; Fernandes&Carmo-Silva,2006), while at the same time use of detailed scheduling techniques may not be appropriate due to its complexity and reliance on the accuracy of the information systems (Riezebos,2010).

Several alternatives of POLCA have been developed, like load-based POLCA (LB-POLCA) (Vandaele et.al,2008) and Generic POLCA (G-POLCA) (Fernandes et.al,2006). LB-POLCA transforms the original POLCA system intro a load-based version such that the number of cards in each loop is replaced by an allowable workload for that loop. It provides a more adequate and robust representation of available capacity in settings in which the processing times of production orders vary significantly and product mix changes occur frequently (Dogger et.al,2010).

(35)

- 35 -

routing of an order are available at the time of release (Dogger et.al,2010). The system also considers a pool of orders to set the release date of the jobs in the line. This release date is computed using the due date and the expected lead time of the job (González et.al,2011). Still, the basic idea of POLCA to use card loops for each possible combination of successive work-centres will reduce its practical applicability in shops with high routing mix variability (Land,2009).

4.4 COBACABANA

Dogger et.al.(2010) conclude that the above-mentioned SFC approaches are all designed for processes with a dominant flow direction. If no dominant flow exist and routings are highly variable, and even the sequence of stations varies across orders, the approaches are not able to balance the workload at every workstation. M-CONWIP and POLCA will need many control loops and corresponding cards per loop, therefore causing a decline in the controllability of the process. POLCA will also need to allow multi-directional flow, but no research is yet known whether or not this is feasible (Dogger et.al,2010).

Cobacabana (control of balance by card-based navigation) system is a simple card-based system for job shop control and is based on the concept of workload control (WLC), which has already proven its value in job shops (Land,2009). WLC is built on release methods using workload norms for all critical workstations; i.e. it controls the queues of all the critical workstations. This approach proves to be able to cope with both high routing sequence variability and long directed routings (Dogger et.al,2010). The philosophy underlying WLC is based on creating predictable and short throughput times for each critical workstation (Land,2009). Particularly, predictable and short throughput times are lacking in most job shops because of all types of variability that characterize this environment. Nevertheless, predictable throughput times are important for a good timing of order releases, for quoting realistic delivery times and for a good timing of capacity adjustments. Short shop floor throughput times increase the flexibility to deal with possible customer order changes, changes that are not uncommon in most job shops. Besides, short throughput times encompass the direct advantages of a transparent shop floor with low WIP. A predictable short throughput time at a workstation is enabled by keeping its direct load at a constant level, the direct load being measured as the sum of the processing times of orders waiting or already being processed at the workstation (Land,2009). Before being released orders will wait in a centrally controlled order pool. The pool buffer can absorb both fluctuations in capacity requirements and capacity availability. It is obvious that constant direct loads can only be realized by releasing an appropriate set of orders to the shop floor. This function of order release has been indicated as its load-balancing function (Land, 2009). It should be recognized that order release is the last moment one can effectively contribute to a constant direct load. The effectiveness of load-balancing dispatching rules (e.g. Work-In-Next-Queue) after release is limited.

(36)

- 36 -

The mechanism proposed by Land uses cards to control the flow of jobs, but in a different way than in traditional card-based implementations. Their system consists of two distinct parts: (i) order release and SFC, and (ii) order acceptance and due date promising. Regarding the SFC (part i), loops are set between the release stations and all critical stations.

Figure 10 González et.al.(2011)

Figure 10, adapted from González et.al.(2011), show the route followed by a specific order (intermediate buffers have been omitted for clarity reasons). Each loop between each station and the first station contains a number of cards NCi. Unlike other card-based systems, this system does not assign one card to each job (container), but each card indicates a certain amount of workload of a particular station. Thus, a job can need more than one card. This novel way of assigning cards can easily control the flow of jobs of different types of orders that share the same routing. Cards return to the control panel (release) at the end of operations in a given station. The system balances the workload for all critical stations, allowing a fixed number of cards (totalling 100%) at each station. Thus, each card introduced in the system represents the percentage of capacity consumption. The system also includes a control panel, where the available cards represent the opportunity of releasing new jobs and the free space of jobs released represents the percentage of load at each station.

The design of a decision framework for selecting a shop floor control system for manufacturing plants.