Improving responses to pull production dynamics by the use of real time information

(1)

Master thesis technology and operations management

Rijksuniversiteit Groningen

Improving responses to pull production

dynamics by the use of real time

(2)

Improving pull production dynamic responses by the use

of real time information

University of Groningen

Faculty of Economics and Business

Course: Master thesis - Technology and operations management

Code: EBM766B20

Period: 2016-2A & 2B

Accessors: D. J. Powell & Dr. J. Riezebos

Assignment: Thesis proposal

Due date: June 20, 2016 before 9:00 am Hand in date: 20/06/2016

Student name: Student no.: Master degree program:

(3)

Preface

After six years of studying, including a bachelor, a pre-master and currently a master, I am proud to present to you my biggest achieved thus far, my master thesis. This report is my research for lean in a digital era, where I have researched pull production systems and tried to improve them by the use of real time information. Although I still need to finish my master after the completion of this report, this report feels as the end of my years as a student. Now, so close to the end of my years as a student, I have to admit that all this would not have been possible without the help of many different people. For this research I first thank the companies and company contacts of Variass, Altrex, Langehout & Cazemier and Propos for giving me the opportunity to use their companies and knowledge as cases during my research. I also my gratitude toward my supervisors Daryl Powell and Jan Riezebos for their guidance and feedback during my research, without which this research would not have been possible. Last but not least I thank my parents, my brother and sister, my brother-in-law and my sister-in-law for their support and encouragements during my whole study career. I give a special thanks to my brother for proof reading my thesis, which has improved the readability of this report.

(4)

systems behave according to stable state. The reality is however, that dynamics in the process do occur, for example because of demand variety and machine breakdown. Therefore the aim of this research is to determine how real time information can be used to control pull production dynamics.

Design/Method/approach – By conducting a case study, this research aims to give answer to

the research question giving practical insight and enlarging the theoretical body. A multiple case study is conducted, including two companies without a real time information systems and two company with a real time information systems. By conducting semi-structured interviews, receiving shop floor tours and through system demonstrations this research has generate rich full data in order to give answer to the research question.

Findings - When dynamics occur, the process needs to respond to keep the performance under

control. The possible responses in a pull production system are increasing capacity, increasing work in process or perform workload balancing. In current pull production systems it proves to be rather difficult to apply the proper response because of lack of information and the delay between the occurrence of a dynamic and the observation of the dynamic. In order to improve the response to dynamics real time information is needed. By implementing in check-out screens real time information is made possible in current pull production systems, removing the delay and creating a transparent process where responses to dynamic are based on real time data. The proper response to dynamics is distributed to the shop floor by the use of visual management.

Limitations/future research – Only three pull production systems and three response to

dynamics are included in this research. Future research could extent this paper by including additional pull production systems or responses.

Keyword - Real time information, pull production, dynamics, dynamic responses, IoT,

visualization.

(5)

Concept list

Concept Explanation

WIP Work In Process

POLCA Paired-cell Overlapping Loops of Cards with Authorization

ConWIP Constant Work In Process

IoT Internet of Things

OTIF On Time In Full

RTI Real Time Information

CT Cycle Time

TH Throughput

(6)

Johan Meijer -1-

1 Introduction

The principles of Lean production have enabled organisations to improve processes with various application of recognized tools and techniques (Powell, 2013). One of these techniques is the use of pull production systems which authorizes the release of jobs based on the system status (Hopp and Spearman 2000). Pull production systems aim to control throughput times by constraining the release of jobs onto the shop floor (Riezebos, 2010). However pull production systems often consider a stable state of the process and therefore do not take into consideration the effects of dynamics, for example, demand variety or machine breakdown.

In a pull system jobs are often released onto the shop floor by using physical entities such as cards, resulting in the name; based control systems. Through the years different card-based control systems have emerged, some examples are; Kanban (Al-Baik & Miller, 2014; Berkley, 1992), Constant work in process (ConWIP) (Spearman, Woodruff, & Hopp, 1990) and Paired-cell overlapping loops of cards with authorization (POLCA) (Riezebos, 2010). When cards are released, based on the systems status, they are worked on in the order of release or according to a sequencing rule. The cards are used to control material flow through the production, but they are also used to distribute information (Liu & Huang, 2016).

The control of the system is disrupted when a backlog is created. If several cards arrive at once, if a machine breaks down or another complication occurs in the process, there is a risk that some jobs will violate their due dates and that machines will become idle. When these dynamics occur, the process needs to respond to keep the performance under control. In order to determine the proper response to these dynamics, information about the process status is required. In current pull production systems there often exists a delay between the occurrence and the detection of dynamics in the system. To be able to respond more adequately this information has to become real time (Wallace J. Hopp & Spearman, 2011). Recent technological developments have made the constant availability of real time information possible, this contributed to the concept of internet of things (IoT).

(9)

By adding real time information to the current pull production systems a new situation is created. Occurrences of dynamics in a pull production system with real time information could be detected without delays and could lead to improved responses to these dynamics. This research aims to design and research this new situation by determining the effect of real time information on pull production systems, leading to the following research question:

“How can IoT and the real time information it provides be used to control pull

production dynamics?”

For answering the main research question this paper is structured in the following way. First the theoretical part is explained in the theoretical background. In the first part of the theoretical background the characteristics of current pull production are explained, including the characteristics of Kanban, ConWIP and POLCA. This is followed up by explaining what kind of dynamics occur in these systems and what kind of responses to these dynamics exists. Based on the explanation of the pull systems, the dynamics and the corresponding responses the first sub research questions can be answered:

(1) “What kind of pull production dynamics are there, what are the responses to these dynamics and what information is needed to determine them?”

If the necessary information can be gathered with the IoT, it could lead to an improved performance. But how this would improve current pull production systems, remains unclear, leading to the second sub research question:

(2) “How does real time information of process dynamics improve the responses to these dynamics?”

When this paper has determined what information should be gathered and how this could lead to an improved situation, then only the last part remains, which is how information obtained is shared between the management and the shop floor. Leading to the last sub research question.

(3) “How should the information of the proper response to dynamics be distributed to the shop floor?”

(10)

2 Theoretical background

In this chapter the theoretical background of the research is explained, starting with the explanation of the characteristic of pull production and current pull production systems in paragraph 2.1. This is followed by the explanation of the dynamics that occur in these system and the corresponding responses to these dynamics in paragraph 2.2. After this the input of the new variable, real time information (RTI) is explained in paragraph 2.3.

2.1 Pull production

The way production system produce products can be divided into two types of systems: push and pull. In a push system, for example MRP, work releases are scheduled. In a pull system order release are authorized (Spearman et al., 1990). One of the reason for using a pull system is that push system do not always create feasible plans. Push systems use fixed lead times which are not based on the actual job utilizations. Resulting in a high variation in the process throughput times and cycle times (CT). Pull systems are designed with more flexibility by only authorizing releases of jobs and limiting the amount of jobs released. By constraining the amount of jobs released one can keep the Work in Progress (WIP) at a constant level, thereby controlling the throughput times (Riezebos, 2010). A high variety in pull systems exists, each containing its own advantages towards the other (Gonzalez-R, Pedro L., Framinan, Jose M. Pierreval, 2012). In sections 2.1.1 to 2.1.3 the three most often discussed and applied pull systems are described.

2.1.1 Kanban

(11)

Figure 1 - Basic Kanban

One of the characteristics of a Kanban system is “blocking”. Blocking occurs when the total number of jobs in the following buffer is equal to the buffers capacity (Berkley, 1992). When this occurs the station must remain idle until the station downstream creates an “empty slot”. A short coming of the Kanban is that it requires the process to behave nearly perfect, meaning no scrap loss, low set-up times and low fluctuations in demand (Spearman et al., 1990). A pull systems that is better in dealing with this is the Constant work in process (CONWIP) system or the paired-cell overlapping loops of cards with authorization (POLCA) system.

2.1.2 Constant work in process

The CONWIP system also releases a limited amount of jobs into the process. However the authorization signal for a job release is not done from station to station, but from the end of the process to the beginning of the process (Spearman et al., 1990). Meaning that when a job has finished the entire process, it signals the beginning of the process to release a new job into the system, see Figure 2 - CONWIP. The CONWIP system does so in order to enable more variety in the process while maintaining the limited amount of WIP and the control of throughput times.

The performance of a CONWIP system is based on the amount of cards released (Framinan, González, & Ruiz-Usano, 2003; Spearman et al., 1990). The job sequencing is determined before release and is based on all the jobs on the shop floor and in the backlog list (list containing all the jobs that still have to enter the shop floor). However, once released, the sequence of jobs cannot be changed and are worked on in the release sequence. Therefore when complications occur in on the shop floor the sequencing cannot be altered forcing the process to run its course. Making decisions on the right order release sequencing of the backlog is very important for the process performance (Leu, 2000; Spearman et al., 1990).

(12)

2.1.3 Paired-cell Overlapping Loops of Cards with Authorization

The POLCA system is designed to operate in environments with high demand variability (Gonzalez-R, Pedro L., Framinan, Jose M. Pierreval, 2012; Riezebos, 2010). A POLCA system is a material control system that regulates the releases of jobs on the shop floor in a cellular manufacturing system (Riezebos, 2010). Authorization signal for job release are done by the use of cards. Cards are not product specific as with the Kanban system, but are paired between cells. Therefor giving signals from one cell to another (through the use of cards) is not an inventory replenishment signal, but a capacity signal which signals available capacity (see Figure 3 - Polca). Cards are fixed between two cells, staying with the job on the journey between them (Stevenson, Hendry, & Kingsman, 2005). The POLCA system aims to increase the speed of job transfers between cells and to reduce unbalances between cells (Riezebos, 2010). POLCA not only controls the release between cells, but also signals the release of jobs onto the shop floor.

Figure 3 - Polca

(13)

2.2 Pull production dynamics and corresponding responses

Often in literature when explaining pull production systems it is assumed that the process behaves according to a stable state, this however is not the case since customers change their mind, ask for favours, machines breakdown, release signals do happen at the same time, etc. The reality of almost every shop floor is that complications in the process happen and therefore some jobs do need special treatment in order for the process to perform (Wallace J. Hopp & Spearman, 2011). These kind of complications happen at all types of pull system, for which examples were given for Kanban, CONWIP and POLCA. When these dynamics occur the system needs to respond to keep the process under control. The possible responses to system dynamics that are addressed in this paper are: card number adjustment, workload balancing or capacity adjustment (Belisário & Pierreval, 2015; Germs & Riezebos, 2010; W J Hopp & Roof, 1998; Renna, Magrino, & Zaffina, 2013). The responses addressed are limited to these three due to the boundaries of the research.

2.2.1 Card number adjustment and capacity adjustment

Due to variations in demand and or system performance the system has to respond in order to maintain its performance target. The first two responses are based on little’s law.

 Little’s law: Work In Process = Throughput x Cycle time

Pull production systems aim to control throughput times by constraining the release of jobs onto the shop floor (Riezebos, 2010). But Little’s law tells us that the control of the throughput can be achieved by either changing the cycle times or by changing the work in process. In a pull production system by decreasing the cycle time and keeping the WIP the same, the throughput times would decrease. CT’s are decreased by increasing the capacity of the system. Work in process in a pull system is increased by increasing the number of cards that are released into the system.

(14)

A decrease of the CT is achieved by increasing the capacity. This is often done by increasing the number of machines or by increasing the number of employees. The CT of the system is however always depend on the bottleneck in the system. Therefor the maximum capacity of the system is equal to the maximum capacity of the bottleneck.

An often considered method for determining the amount of cards for a ConWIP system is called statistical throughput control (STC). This method controls throughput by dynamically changing the number of cards (W. J. Hopp & Roof, 1998). It measures the average throughput time, the standard deviation and the average CT after the completion of every job. This output is then compared with a target throughput level and the maximum CT. If the target CT is not achieved, then either the throughput target is reduced or capacity is added. For the average throughput value it is determined whether it is below or above three times the standard deviation. When it’s below three times the standard deviation a card is added to the system. When it’s above a three times the standard deviation a card is removed from the system and when it neither above nor below three times the standard deviation the system does not change. When either capacity is added or the amount of cards in the systems is changed, a warmup period of n products is observed before further decisions are made regarding CT or WIP level (W J Hopp & Roof, 1998). The amount of products is dependent on the process, for example the process length and number of production steps.

Another approach similar to that of Hopp & Roof (1998) is that of difference throughput rate control (DTC) (Liu & Huang, 2009b). This method measures the throughput time achieved by a system, the throughput time achieved is called real time throughput (RTH). This RTH is then compared to the target throughput value (TTH). If TTH is equal to RTH the response is to do nothing. If TTH is smaller than RTH a card is removed from the system. If TTH is bigger than RTH, first the CT achieved is compared to the maximum CT. The minimum CT is based on the bottleneck of the system, since the CT can never be lower than the CT of the bottleneck. If the CT achieved is higher than the minimum CT, capacity is added to the system. If the CT achieved is equal to that of the bottleneck, then a card is added to the system. After a change in the process has been applied, then a similar warmup period as that of Hopp & Roof (1998) is used before determining if additional responses have to be taken.

(15)

significantly. The decision to adjust the number of cards is based on the evaluation of two moving average computations of the customer demand. One moving average is computed on a greater horizon time and one on a lower horizon time. These two are compared to determine whether customer demand is increasing, decreasing or stable. If demand change is significant cards are added, removed or stay the same.

2.2.2 Workload balancing

Workload balancing (WLB) is another method to control the performance of a pull production system. WLB is the capability of a system to balance the total workload among the workstations on the shop floor (Germs & Riezebos, 2010b). A balanced system results in better control of arrival times of job at workstations on the shop floor, leading to shorter average que length to achieve a certain utilisation level and can lead to a reduction of time between arrival and completion of an order. An effective system of WLB reduces both the shop floor throughput times and the total throughput times. The shop floor throughput time is the time from the point an order is released onto the shop floor until its completion. The total throughput time is the time between the arrival and the completion of an order. (Germs & Riezebos, 2010). WLB reduces the total throughput time by decrease variability of the workload at each workstation. Balance of workload on the shop floor can be achieved by restricting the release of orders to workstations that are busy and by releasing orders to workstations that are waiting for work. Although the cellular lay-out of a POLCA system does bring forward some WLB capabilities, it does not perfectly detect and signal imbalance in workload (Germs & Riezebos, 2010).

2.3 Internet of things and real time information

This paragraph explains the new variable in this research, namely RTI. This paragraph starts with discussing IoT and how this can lead to RTI. This is followed by explaining what information IoT should gather in order for pull production system to improve. This paragraph finishes by proposing a method for distributing the gathered information.

2.3.1 Internet of things

(16)

measurements. This data can be used to answer “what-if” questions, e.g. what-if we change the sequence of production or what-if we increase capacity. The answer of these questions should lead to the proper response for the occurring complications.

Gathering and distributing RTI in the current pull production systems is not feasible, because the gathering and distribution of information is done physically, e.g. by the use of cards. However the digital era has enabled RTI by the introduction of the internet of things (IoT). IoT is the use of sensors and network in order to connect “things” together to create an intelligent network that can exchange information among things and with humans (Li et al., 2014). In the early stages IoT started with the development of radio frequency identification (RFID) tags. RFID is used to identify product wirelessly through the use of radio waves (Hozak & Collier, 2008; Kumar, Kadow, & Lamkin, 2011; Powell, 2012). Nowadays IoT in the form of sensors can be imbedded in any object in a process in order to keep real time track of e.g. machineries status, vehicles locations, robots or any other object in the process (Lu & Cecil, 2015; Saint-Exupery, 2014). This creates a machine-to-machine environment enabling machine-to-machine interactions and improving human-to-machine interactions.

(17)

2.3.2 Internet of things and response to dynamics

Based on the description of the different pull production systems in chapter 2.1 and on the description of the possible responses to dynamics in a pull production environments in chapter 2.2, we now describe what IoT should measure in order to create the right RTI.

The first proposed response to dynamics is that of WIP adjustment or capacity adjustment, which are based on little’s law. WIP is the total workload that is present on the shop floor at a given time. The amount of WIP on the shop floor is kept at a certain level by constraining the amount of jobs released on the shop floor (Germs & Riezebos, 2010b). Therefor it should be measured what is on the shop floor, what is leaving the shop floor and what is released onto the shop floor in order to get RT WIP information.

The other measurement is the capacity of each station individually and the capacity of the system in total. The capacity of the total system is limited to the maximum capacity of the bottleneck of the system, if the system has a bottleneck. By increasing the capacity the CT of each product decreases. This report uses the following definition for CT: “Is the average time from release of a job at the beginning of the routing until it researches an inventory point at the end of the routing” (Wallace J. Hopp & Spearman, 2011). When capacity is added the average time form the beginning until the end decreases.

The third measurement proposed is that of the workload. The workload at each station is the number of jobs at each station times the planned CT of each job. The planned CT is the processing time plus the setup time of each job or a set of jobs. In order to determine an imbalance of workload the position of all the individual jobs should be measured at all times. When an imbalance of workload occurs in the system, management or the shop floor should change the sequence of jobs or the release of jobs in order to balance out the jobs in the system. By doing so the utilisation of each station increases and therefore increase the throughput of the system

(18)

might not be achieved. When this occurs the underlying reason should be determined, in order to do so RTI about utilisation is necessary. An easy way to determine underlying reasons is by measuring machine or station status. The management should be able to see whether a station or machine is in production, has a breakdown, is being repaired or is idle.

The last measurement proposed is that of finished goods inventory. Which is especially important in a Kanban systems, since the finished goods inventory (FGI) determines whether a job is released or not.

2.3.3 Collecting and distributing real time information

Adding RTI to the pull production environment creates new information that needs to be managed. Information management is described as ensuring that valuable information that is obtained is used to its fullest extent (Hicks, 2007). It involves activities such as “the creation, representation organisation, maintenance, visualization, reuse, sharing, communication and disposal of information” (Bevilacqua et al., 2016; Hicks, 2007). Improving information management can add value to the organisation by improving the overall efficiency, competitiveness and responsiveness. This directly relates to the lean since good information management could lead to an improvement in all five lean principles stated by Womack and Jones (1996).

A tool often used in lean for distributing information and reducing information wasted is visual management. Visualized information is easier to understand because it becomes tangible in the mind (Hall & Obregon, 2002). The goal of visualization is taking massive amounts of data and converting it to visual shapes and colours in order to make it easily interpretable. This data in its original form is either too small or too big and would therefor take some time for people to understand. In a lean organisation, by making this information visual, people are able to understand the status of the process just my looking around (Tezel, Koskela, & Tzortzopoulos, 2004). Lean visual management is an approach that communicates with the “doers”, in order for parts of the organisation to become self-employed. According to Tezel et al. (2004) there are several functions of visual management: Transparency, discipline, continuous improvement, job facilitation, on-the-job training, creating shared ownership, management by facts, simplification and unification.

(19)

to the shop floor is managed by the use of “entities”, most often cards. The information is kept to a minimum in order for the process to be easy to control.

(20)

3 Methodology

In this chapter the methodology of the research is presented. First the method chosen and the reasons for choosing this method are explained. This is followed by the case selection procedure and this chapter finishes with explaining the data collection and analysis method.

3.1 Case study

The research questions stated in the introduction is a so called “how” questions, which in this case is about a proposed relation between RTI and pull production control mechanism. Case study research has been recognised as being good in answering these kind of questions and has also been found particularly suitable for developing new theory (Voss, Tsikriktsis, & Frohlich, 2002). To the best of my knowledge the proposed relation has been not been established in literature yet, therefore this case study can be described as theory building. In the previous chapter it has already been explained why pull production environments need to be controlled, how they can be controlled and how RTI could have a positive effect on this. Although in theory this could lead to an improved situation in practice this might be unnecessary or infeasible. Therefore this research has selected cases to research this relationship and establish an overview of the advantages and disadvantages. Multiple case study research allows studying a real-life phenomenon, such as this relationship, in depth (Voss et al., 2002; Yin, 2009). Moreover, studying multiple cases allows comparison of the capability development patterns across multiple cases, which increases generalizability and external validity. Although this research focusses on pull production control mechanism the unit of analysis goes beyond that. There are multiple reason for doing so, for example, the complexity of the orders have effect on the pull production systems chosen. Therefore the unit of analysis of this multiple case study research is the organisation. Starting from the order arriving in the organisation up to the order shipment.

3.2 Case selection

(21)

frame to help uncover, confirm, or qualify the basic processes or constructs that underpin the study. When building theory from case studies, each case should be selected so that it either: (1) predicts similar results; or (2) produces contrary results but for predictable reasons (Voss et al., 2002).

In order to find cases that could lead to similar results, first “action one” of Miles and Huberman (2004) was applied; setting boundaries. Although not expected, RTI systems for shop floor control do already exists. This made it possible to research the effect of RTI on shop floor control. But since this paper specifically researches the effects of RTI on pull production environments, the RTI system(s) should be applicable to pull production environments. Therefor the first boundary of the case selection is that the case should have experience with, or expert knowledge of pull production environment and controlling dynamics in these systems.

Secondly, “action two” of Miles and Huberman (2004) was applied; creating a sample frame. Since the research aim is theory building, this research studies the relationship between RTI and pull production control from multiple perspectives. Therefor not only cases with a RTI system where added to the sample frame, but also cases without a RTI system. Since multiple pull production systems exists, of which three are explained in this paper, also multiple different pull production systems have been researched. Resulting in the following cases:

Table 1 - Case selection overview

Since this research has many variables, namely; different pull systems, different RTI systems and different companies and products, the outcome of the research could be different when replicated with different companies. The possible variation in outcome is due to the fact that different companies have different needs and therefore might require different RTI.

3.3 Data collection and analysis

The data collection is has been done in three steps; pre-visit preparation, on-site data collection and post-sit data collection. In order to increase reliability and valid a research protocol has been designed, with as core a list of interview questions. (Voss et al., 2002; Yin, 2009). The

Company name Case criteria

Variass POLCA

Altrex Kanban/ConWIP

(22)

full protocol can be found in appendix B. Since this research uses two type of cases, the research protocol was developed to be suited for both types. The first case type are companies with a pull production system without RTI. The second case type are companies that possess a system that results in RTI.

The person with which the interview was held was a person with expert knowledge of the pull production system or the RTI system at the company. This could be a consultant, a production manager, or any other person that could give detailed information about the system from a shop floor perspective as well as a management perspective. The interview questions of both kind of companies will be formed according to the so called “funnel model”, where broad questions are asked first and more detailed questions follow as the interview progresses. The interview questions will be send upfront in order for the contact person to prepare for the interview. For the pull production companies the visit should include an interview and a guided tour throughout the facility. It may also include a demonstration of systems used to control the pull production system, for example ERP, which will increase triangulation and thereby the reliability (Voss et al., 2002; Yin, 2009). For the companies with a RTI system the visit should include an interview and a demonstration of the RTI. For the participating companies it was determined beforehand what their role is in the total research. Each participating company will be the object of a case study involving multiple contact points which includes at least one company visit.

(23)

4 Analysis

The analysis of the collected data is explained and discussed in this chapter. First the individual cases are discussed separately in paragraph 4.1, in order to discuss them with each other in paragraph 4.2. An overview of the case information can be found in appendix C.

4.1 Individual case analysis

In this paragraph the cases are discussed individually, leading to a cross case in the second paragraph of this chapter.

4.1.1 Variass

4.1.1.1 General information

Variass, located at Veendam and Drachten, is a high tech system supplier and EMS specialist of electronic and mechanical products and systems. With a turnover of approximately 25 million and 130 employees it can be considered as a medium sized company. The product produced at Variass can be characterized as low volume, high variety. Variass was selected as a case study because of their POLCA (non-RTI) system.

4.1.1.2 Process

The products on the shop floor move from production cell to production cell, making the products and the process ideal for a POLCA system. The first step in the system is combining a POLCA card with an order or batch. Before the job is processed, first a check is performed whether capacity is available in the following cell. POLCA cards at Variass are not route dependent, as with traditional POLCA system. A POLCA card at Variass is linked to a single cell. When a job or order has been processed the single cell POLCA card is put back on the POLCA board in the centre of the shop floor visualizing available capacity to the other cells. The reason for having single cell dependent POLCA cards, is that otherwise an excessive amount of cards would be necessary which could lead to confusion and information overflow. 4.1.1.3 Dynamics and responses

(24)

can be present on the shop floor. Changing capacity or balancing workload is determined by the foreman on the shop floor and is done with a capacity tool in combination with experience. When certain products become “hot” they are able to skip the queue by the use of a priority card. The amount of priority cards on the shop floor is limited by three, otherwise the flow of the process could become compromised or other products could become “hot”. When a customer changes its mind during the process or complications occur, the job can be put on hold by using an on hold card. When this occurs the POLCA card is detached from the job and put back on the POLCA board in the centre of the shop floor.

4.1.1.4 Performance indicators

The key performance indicators (KPI’s) for Variass are customer delivery reliability, quality achieved and efficiency achieved. The first is determined by measuring the amount of orders that are delivered on time in full, the quality by the measuring the amount of customer complaints and the efficiency by comparing estimated calculation with the actual performance. If the actual performance deviates significantly from the calculation performance, then either the process is improved or calculation is adjusted. Information available on the shop floor are weekly achieved efficiency per cell, capacity overviews and failure costs.

4.1.1.5 Real time information opportunities

The POLCA system at Variass has some opportunities for improvement and one of them can be the use of RTI. First the available capacity on the shop floor is based on the number of jobs on the shop floor. When a job enters a cell the capacity is only made available again after the entire job is finished, resulting in unnecessary blocking.

The second opportunity is due to the delay between efficiency achieved and efficiency planned. When a job takes more time to produce than originally planned, efficiency is not met. The difference could have multiple causes for example, employee needs additional training, complication during processing of the job but it is not put on hold, etc. Determining the actual cause proves to be difficult due to a delay of a week between occurrence and measurement/ feedback.

(25)

4.1.2 Altrex

Altrex, located at Zwolle, is a producer of climbing equipment, ladders and scaffolds for almost 65 years. With an employee count between 160 and 200, Altrex can still be called a medium sized company. The main process of Altrex consists of assembling parts into finished goods. Throughout the years batch sizes at Altrex have become smaller and product variety has increased. We chose Altrex as a case study company as they have implemented a Kanban (non-RTI) based system.

4.1.2.2 Pull system

Before implementing the Kanban system Altrex had difficulties with the variety in customer demand. Altrex was not able to respond fast enough to the dynamic customer demand, because they used a push system. In order to increase customer delivery reliability and reduce throughput times Altrex decided to implement a Kanban system. Although traditional Kanban systems have minimum and maximum buffer levels, Altrex only focuses on finished goods inventory (FGI). When a slot become available in the FGI a new order can be released onto the shop floor, which is worked on according to a first-in first-out policy. Since the end of the process (the FGI) signals the beginning of the process to release a job onto the shop floor and no additional in process signals (buffers) are used as with a Kanban system, we can conclude that the shop floor functions according to the ConWIP system mechanism. The implementation of the Kanban/ConWIP system also resulted in a more predictable process, which resulted in a decrease of raw materials in the front end of the process.

4.1.2.3 Dynamics and responses

(26)

The decision to add capacity is currently based on the planning and forecasted planning. By comparing the planning with the available capacity they can easily predict whether additional capacity is acquired.

4.1.2.4 Key performance indicators

The KPI’s of Altrex are quality, efficiency, FGI level and customer delivery reliability. The quality KPI is a list of situations of what went wrong and caused quality lose. The quality list is updated every week and is used for continuous improvement. The continuous improvement of quality is performed by the foreman of the shop floor, who checks the list every week and gives feedback when necessary. The efficiency is determined for each production line, which gives limited insight. The FGI level is necessary for controlling customer demand dynamics and is controlled by determining the average FGI capital. The customer delivery reliability is the most important KPI and is measured by the amount of on-time-in-full (OTIF) delivery level achieved. A good OTIF level is achieved by controlling the efficiency levels and the FGI levels. 4.1.2.5 Real time information opportunities

Currently, the efficiency is measured for each production line, giving only insight into the performance at the beginning and the end of the process. The measurement is performed on a daily basis and therefore a delay exists between the moment that the efficiency is not achieved and the responses to get the performance under control again.

The current measurement of efficiency for each production line does not give sufficient insight for control. The operations manager of Altrex therefore proposes to implement a check-in check-out system for each person and each job, in order to determine efficiency achieved for each job and for each person. The result will be a more controlled system, since there is more accurate and reliable information and there are no delays between occurrences and information obtained.

(27)

4.1.3 Langehout & Cazemier

Langehout & Cazemier, office located at Drachten, is a consultancy company which provides two services. The first is consulting companies about facility, design and planning and the second is selling and implementing their RTI system, called ElancE. The development of the ElancE system was due to customer demand to increase control of their lay-out or shop floor. In order to satisfy the constant rising customer demand for more shop floor control, Langehout & Cazemier developed their ElancE “hours control based system”.

The ElancE system is developed as an add-on for ERP systems. The implementation and training are included in the purchase price of the system and is designed for medium size companies. The goals of the system are; increased control of the process and the processing times, increased accuracy of the information obtained and removing of the delays between the shop floor occurrences and the information available for the management.

4.1.3.2 Real time information system

The ElancE system works in the following way. First a customer order arrives and is put into the ERP system, depending on the company the order is with or without a pre-determined workload. The order goes through a planning process in order to determine a detailed planning. The detailed planning is input for the ElancE system, which corresponds the detailed planning with the respective departments. The order is immediately made visual for the entire process after completion of the detailed planning. The departments can see what the total workload is over a longer period of time, could be weeks or months. The workload can also be seen from different perspectives, namely: (1) based on department workload, (2) total shop floor workload and (3) product based workload. Moreover, the available capacity for each department is adjusted and made visual after each order has entered the ElancE system. Once an order is on the shop floor the details of the order is send to a station. The employee at that station can see on a screen what product needs to be produced. On the screen comments about the product can be seen or added, the detailed drawing of the product are available, as well as the pre-calculated production time, the due date and additional information if required. 4.1.3.3 Responses to dynamics

(28)

bar, the bar turns red. At this point management can take action or the employee can add a comment to why it has exceeded the pre-determined production time.

4.1.4 Propos

Propos, office located at Doetinchem, develops and sells a planning and monitoring software under the same name. The Propos system is based on the POLCA mechanisms and is developed to fully automate pull planning, automate prioritization and serve as communication tool between management and shop floor. The system is suited for every production system that has a low volume high variety production mix, with multiple routings and therefore multiple production steps.

4.1.4.2 Real time information system

Propos states that traditional systems have too many details planned which are never accurate and also causes a lot of waiting time for jobs. Therefore, Propos proposes to create production cells which would lead to an improved overview of the shop floor. The Propos system monitors the jobs based on the main production steps and should lead to autonomous teams. The sequence of production steps are visualized by the use of colours. Each colour represents a cell or department. For example green is followed by blue then the employee knows immediately where the product should go for the next product step, as with traditional POLCA system. The Propos system starts with the input of an order into the companies ERP system. The order should include a due date, production routing, order description and production times. Then the order stands in a queue waiting to be released onto the shop floor. To minimize the amount of time jobs are on the shop floor, the orders are kept in the queue as long as possible In addition, jobs on the shop floor should be finished before another order is released, resulting in a constant work in process on the shop floor. Buffers between cells are actively controlled with RTI, in order to prevent idle cells. Orders in queue and on the shop floor are constantly prioritized, based on due dates, throughput times, preventing cells from becoming idle, batching and production peaks.

(29)

4.1.4.3 Responses to dynamics

The special message is prioritization or on hold. Prioritization is used to increase the performance of the shop floor. Prioritization is used

Prioritization could be due to order batching, behind schedule, in order to control capacity, prevent cells from becoming idle, maximize throughput or reduce setup times. On hold could be due to batching, complications or missing parts or tools.

Management information includes proposed capacity for each cell, WIP analysis, and buffer time analysis. Since Propos is a shop floor control system, it states that the biggest amount of control should also be done on the shop floor and not by management.

4.2 Cross case analysis

4.2.1.1 Pull production dynamics and responses

When looking at the pull production systems without RTI (Altrex and Variass), already a lot of variation exists. Altrex has several production lines while Variass has multiple cells, therefore Altrex has a Kanban/ConWIP system and Variass a POLCA system. Altrex adjusts capacity and WIP to keep the process performance under control, while Variass also performs WLB. The biggest dynamic for both cases is the customer demand, which both handle in different manners. Besides adjusting capacity and WIP Altrex also uses forecasting and FGI in order to coop with the dynamics demand.

At Altrex and Variass the WIP level changes when additional employees are hired. Hiring additional employees is done in order to increase the capacity of the system and therefore reducing the CT. At the same time Altrex allows the employees to pull more jobs onto the shop floor, in order to prevent employee form becoming idle. By doing so, Altrex increases its WIP level. Variass changes the WIP level of the shop floor in order to cells from becoming idle. 4.2.1.2 Key performance indicators

(30)

4.2.1.3 Collection and distribution of process information

At Variass the efficiency achieved is calculated every week per cell, while at Altrex the efficiency is calculated every week per production line. When efficiency is below the pre-determined efficiency there exists a delay in both cases, making it hard to determine the actual cause.

Variass still uses cards for the distribution of information, but agrees with Altrex that this is not an ideal mechanism since cards can be misplaced or lost. Therefore, cards are no longer the leading source of information. Key information comes from the work orders attached to the job. Variass does still use special cards to deal with “on-hold” and “prioritization” situations. Altrex and Variass both agree they could benefit from RTI to be able to more actively control the process and obtain more accurate and reliable information. They both also agree that the removal of the delay between occurrences and availability of this information for management could lead to improved responses.

4.2.1.4 Real time information systems

Two systems who could accomplish the improvement are the ElancE and the Propos system. ElancE was developed based on customer demand while the Propos system is based on the theory of POLCA.

The main goal of systems is achieving the customer delivery reliability. The Propos system aims to achieve this by focussing on controlling the performance of the process, not focusing on the achievement of the planning. In other words when the pre-determined production time is not achieved, action is taken. While the ElancE system focuses on the pre-determined planning and the planning achieved, signalling management and shop floor when the expected efficiency deviates too much. Propos states that planning is not feasible in the first place and therefore does not use the planning.

4.2.1.5 Distribution and collection of process information

The systems on the shop floor for are similar, the ElancE system has one or two computer screens per station or department and the Propos system always has one touch screen per cell. The information distributed on the screens difference significantly. While ElancE has many options for management and shop floor, the Propos systems tries to minimize the information distributed.

(31)

addition the screens are used for distribution of work orders, which results in a paperless shop floor and preventing work orders from becoming lost or damaged on the shop floor. The Propos system distributes for every job the due dates, the routing and the order description. Propos still uses work orders to go with the job throughout the process. Both systems also produce a backlog list of jobs that are to be processed. For the ElancE system the jobs are worked on from top to bottom, automatically putting the most urgent job on the top of the list. This is based on the due date of the jobs. The Propos systems gives the employees the choice of what to produce, unless it signals that a job has prioritization over the other job. When a certain job has priority a button with “backlog” lights up, visualizing the employee which job to process first.

4.2.1.6 Responses to dynamics

The ElancE system controls the process by signalling to the shop floor when the pre-determined processing time or the due date is exceeded significantly. To signal to both management and shop floor when these times are exceeded the system uses colours. The colour orange is used when a job is close to exceeding either the processing time or due date and red when it has exceeded the time significantly. When this occurs either the shop floor or management can respond to the out of control situation.. Since the systems is based on the planning it also visualizes the effect which the out of control state has on the other jobs in the planning. In some cases management can decide to adjust the planning to prevent any further delays.

At propos the shop floor is controlled by the Propos system and the employees on the shop floor and the employees on the shop floor. The input for the system is the due date, the routing, the description and the production times. The check-in check-out screens result in RT process information, with this process information the system knows when respond in order for the process to stay under control. The possible responses are: WIP adjustment, prioritization, WLB and batching. Prioritization is performed in order to prevent jobs from exceeding their due dates, prevent job from waiting for batching and prevent cells from becoming idle or imbalanced. WIP is adjusted when the system foresees cell from becoming idle or when releasing of the job results in long waiting times. Batching is performed in order to reduce setup times. WLB is performed when an imbalance in workload is experienced which could lead to cells from becoming idle. The capacity adjustment is still under control of the management and will only change when customer demands increases for a long period of time.

(32)

5 Discussion

The discussion chapter is used to make a connection with the theoretical background and the analysis results of the cases studies. In order to build a connection between theory and practise we follow the same structure as in the theoretical background chapter. Starting with the different pull production systems, then the different pull production dynamics and corresponding responses, followed by and finishing with how RTI could improve responses to dynamics.

5.1 Pull production systems

In the theoretical background this paper has discussed three pull production systems; Kanban, ConWIP and POLCA. In order to determine the opportunities for RTI in pull production this paper first has researched two pull production systems without RTI. In the cases of Altrex and Variass this research has found that the distribution of information through the use of cards is no longer the leading system. Most information in the cases is distributed by work orders and weekly updates on the KPI’s.

Spearman (1992) stated that in order for Kanban systems to perform the system needs to behave in an almost stable state. Altrex experienced this to be true, since the FGI alone was not sufficient to deal with the demand fluctuations. In order for Altrex to coop with the demand fluctuations Altrex designed a hybrid system. The hybrid system of Altrex combined Kanban with ConWIP and added forecasting to the mix in order to completely deal with the dynamic customer demand.

(33)

For the POLCA system safety cards are used in order to deal with complications in the process (Riezebos, 2010), Variass uses special cards in order to put certain jobs on hold or give them priority in the queue. Altrex made use of special cards in the past in order to increase FGI level to deal with customer order peaks. However, these cards are no longer in use since they were easily lost or misplaced resulting in an unnecessary high FGI level.

5.2 Pull production dynamics and corresponding responses

Wallace J. Hopp & Spearman (2011) state that pull production systems do not behave according to a stable state, since customers change their mind, ask for favours and machines breakdown. When these so called dynamics occur some jobs need special treatment in order for the system to perform. Looking at the different cases, this is indeed true. The biggest dynamic comes from the dynamic customer demand. Altrex and Variass have designed their production process in order to deal with these dynamics. However, when the design alone is no longer sufficient responses are required in order for the system to perform.

By the use of literature this research has described three possible responses: Increase the WIP, increase capacity or perform WLB. All these responses are based on not achieving the target throughput value. For the cases of Altrex and Variass adding capacity is based on the customer demand. When planning signals that demand exceeds the capacity both cases are able to hire additional employees to increase capacity and therefore reduce the CT. By decreasing the CT both cases are able to increase the throughput, which makes it possible for the cases to meet their due dates.

For the cases of Altrex and Variass the most important KPI is the delivery service level, mostly described as on-time-in-full (OTIF). When the OTIF and corresponding due dates are at risk the cases respond in order to get the process under control. In the case of Altrex and Variass the most often applied response is the increase of capacity.

(34)

At Variass also WLB is performed between the cells. Riezebos and Germ (2010) state that it can be difficult for the POLCA system to detect and signal unbalanced workload between cells in convergent routings. Difficulty with workload balancing is experienced by Variass, since WLB is based on daily updates and not on the real time status of the shop floor.

In order to balance workload on the shop floor Propos uses prioritization, on hold and work order release signals. Although this has been described in literature, in the propos system no additional information, e.g. release list (Riezebos, 2010), is necessary since the information is visualized on the screens when necessary.

5.3 Real time information

This paper introduces a new variable in current pull production systems, namely the IoT. IoT in the form of RFID tags and sensors should be able to realise RTI. With RTI manufacturing companies should be able to improve process control. By the use of RTI complications can be detected in a timely fashion, giving management the possibility and the time to formulate a proper response (Wallace J. Hopp & Spearman, 2011). In practise two RTI cases are used; Elance and Propos. They do not make use of RFID tags or sensors, but have implemented

Table 2 - IoT method applied

Although this does not reduce human labour on the shop floor, as proposed by Paul & Rho (2015), it can reduce human labour by decreasing the amount of planning required as with the Propos system. However, in the ElancE case the control is based on planning activities and therefore does not decrease the amount of planning required. Both systems do make the information gathered and distributed more real time, more reliable and more accurate as described in literature.In theory six real time measurements are proposed in order to control the shop floor performance. The measurements are: WIP, CT/capacity, workload throughout the system, throughput achieved, utilisation per cell/station, machines status and FGI levels (only Kanban). In practise other measurements are used in order to control the shop floor performance. The measurements of the different cases are: efficiency (pre-determined processing time compared to achieve processing time), WIP level, workload at every cell, due dates and FGI levels. See table 3 – measurements theory and practise overview.

Theory Cases

(35)

Table 3 - Measurements theory and practise overview

Three of the four cases actively measure CT by determining efficiency achieved in order to respond to significant deviations from the planned efficiency. Two cases determine and balance workload, one of which is real time and the other based on daily numbers and experience. WLB is done in order to prevent cells or stations from becoming idle or enabling batching. None of the cases actively measure throughput or machine status. Due to variation the TH of the process varies significantly, therefore the TH value is not a useful measurement. Machine breakdown can lead to serious production lose in a Kanban or ConWIP system, but is less serious in a POLCA system since POLCA gives priority to jobs that can actually be finished in subsequent stages.

In all four cases the leading measurement is the due dates of the different job. When there is a risk that some due date is violated or is going to be violated the process should respond. It does so by: adding capacity and decreasing the CT of the process, by prioritizing the jobs of which the due date would be violated otherwise or by releasing the job of which the due date would otherwise be violated early, increasing and possibly exceeding WIP level. An overview can be found in Table 4 - Measurement and responses.

When a cell or station would become idle if no response is taken, then this would reduce the TH of the system. By WLB, prioritization or WIP adjustment this can be prevented. When customers change their mind or produce a backlog order, it is often possible to combine this order with a similar one. By batching this order into one order the setup times can be reduced improving the system performance.

Theory Cases

WIP WIP

CT Efficiency

Workload Workload

TH Due dates

Utilisation FGI level

(36)

Table 4 - Measurement and responses

The biggest difference between the two RTI systems is the manner in which they control the shop floor performance. In theory it is stated that due to the information delay and the lack of information it is difficult to apply the right response to certain process dynamics. The ElancE system improves this by removing the delay between dynamic occurrences and increasing available information. By removing the delays and increasing information availability the management can actively control the shop floor performance, by distributing feedback, feeding forward necessary additional information and by actively communicating with customers about the expected performance.

The Propos system aims to create an autonomous shop floor. The control is performed by the employees on the shop floor and by the system itself. The autonomous controlled system results in a decrease in the planning activities and therefore a decrease in human labour. The system also automatically releases products, automatically determines priority and automatically provides management and shop floor with the required information.

5.4 Distribution and visualization of information

In the researched cases current visualisation through the use of cards is either removed or set to a minimum. The RTI systems create new information that needs to be managed. Improving information management can add value to the organisation by improving the overall efficiency, competiveness and responsiveness (Bevilacqua et al., 2016; Hicks, 2007). By implementing a RTI system companies are able to improve their system performance and therefore improve overall efficiency and competiveness, because of the improved responsiveness. A method proposed in theory and used in the cases is that of visual management. According to Tezel et al. (2004) there are several functions of visual management: Transparency, discipline, continuous improvement, job facilitation, on-the-job training, creating shared ownership, management by facts, simplification and unification. When comparing non-RTI system with RTI system, it can be concluded that several functions do improve when implementing a RTI

Theory goal Theory response Case goal Case response

Control TH

Capacity adjustment, WIP

adjustment

Control due dates

Capacity adjustment, prioritization, WIP

adjustment Prevent idle systems WLB, WIP level

adjustment Prevent idle system

(37)

system. Because the shop floor performance is constantly measured and this measurement is constantly available for every employee, it creates a transparent process. This transparent process is easier to control because it is management by facts, namely RTI.

This transparent process has two sides, a management side and a shop floor side. The interface of the shop floor should visualize a list of job in process, a list of jobs in queue and a list of jobs on the way. For each job the interface should show the product specification (job code and product name), next production step, the processing time planned and the processing time passed, the next cell/station on the route (the routing) and the due date. When a product has priority or needs to be put on hold a special bottom should light up on the screen. This button should include text and a colour explaining what action should be taken.

(38)

6 Conclusions, limitations and future research

The conclusions chapter consists of two parts. The first part concludes the findings of the research by answering the research questions and the sub questions. The second parts consist of the limitations of this research and how future research could extent this research.

6.1 Conclusions

This thesis has researched the influence of RTI on pull production dynamics. Although in theory in pull production systems there is often assumed to be a stable state of the process, in practise this is not the case. In practise dynamics do occur, since customer ask for favours, change their orders and machines do breakdown. The goal of the research was answering the main research questions:

“How can IoT and the real time information it provides be used to control pull production

dynamics?”

Dynamics causes the system to move form a in control state to an out of control state. The first sub research question is:

“What kind of pull production dynamics are there, what are the responses to these dynamics and what information is needed to determine them?”

(39)

Applying these responses in current pull production systems is difficult since there exists a delay between occurrence of the dynamic and dynamic information being available for management or shop floor (Renna et al., 2013). Also when dynamics occur no information is available about the actual state of the process. In order to determine the right response more RTI is needed leading to the second sub research question:

“How does real time information of process dynamics improve the responses to these dynamics?”

In current pull production systems the shop floor functions as a “black box”, where the performance of the shop floor is only visible at the start and the end of the process. Although in theory RFID and sensors have been proposed to result in RTI, in practise the use of check-in check-out screens lead to RTI. By the use of check-check-in check-out screens real time process information can be gathered. The goals of the measurements are to control and achieved the predetermined due dates, to prevent cells or stations from becoming idle and decreasing setup times. The responses in order to achieve the goals are: Capacity adjustment, WIP adjustment, prioritization and batching.

The information gathered and distributed this way is more real time, more reliable and more accurate. This leads to a transparent process where responses to dynamics will be based on actual data. It also increases the responsiveness to these dynamics, since the effect of dynamics are immediately visible for the management and the shop floor, removing the delay. Leading to the last sub research question:

“How should the information of the proper response to dynamics be distributed to the shop floor?”

(40)

6.2 Limitations and future research

Improving responses to pull production dynamics by the use of real time information

Master thesis technology and operations management

Rijksuniversiteit Groningen