An improvement plan of product quality at KROHNE Altometer

KROHNE

Towards improving the product quality at

KROHNE Altometer

Final report - graduation thesis project

S. Pu KROHNE Altometer Dordrecht

2016

Author: S. Pu

Student number: 69946 Study: Engineering

The place of publication: Vlissingen The year of publication: 2016

Publishing organization: HZ University of Applied Sciences Company: KROHNE Altometer

Supervisor: F. Wubben 2nd Examinator: J.Prins In-company mentor: C. Paul Version: 4.0

I

Summary

This report presents the result of the data analysis of calibration data in order to improve the product quality. KROHNE has built a large calibration database after years of calibration. The research considered analyzing these calibration data as a start point and tried to learn something new from the historical data. The main errors occurring during the calibration were abstracted through the database. Details about the calibration errors were discussed. This report dedicates to find out what was behind the errors at calibration and brings out an improvement plan based on the analysis of the calibration data. Therefore, it appears to create a proposal that should enable to optimize the calibration system and/or the production process. Application of the proposal should result in a better production quality.

The outcome of the data process was displayed by a step-by-step approach. The core error “Gk/Gkl deviation is out of range” was explained in detail. During the conclusion chapter, fish-bone diagrams were built up to show what factors influence the reliability of the calibration. More specific information was followed to describe what actions could be taken for increasing the efficiency of production and first pass rate. Besides, the SPC control charts were designed for KROHNE products to show the performance of system setting parameters. The control charts were realized in the software in close cooperation with another graduation trainee.

Abbreviation

Abbreviation Term

SPC Statistical process control

EMF Electromagnetic flowmeter

Gk/Gkl System setting parameter

UCL Upper control limit

LCL Lower control limit

CL Center line

USL Upper specific limit

II

Preface

This thesis is a final work for the bachelor degree of Engineering. The thesis project started in February and is ended in June. I really appreciate this opportunity that I was employed by KROHNE Altometer to do a project about improving the product quality or process based on analyzing the calibration data.

First, I would like to thank all people who were involved in this research. Their support and encouragement led me to the right way.

I would like to express my deepest gratitude to my supervisory teacher, Mr. Flip Wubben, who offered a lot of help and advices during the whole research.

I would also like to extend my sincere appreciation to Mr. Peter Maessen, who gave me such a great opportunity to work in KROHNE Altometer and always gave me a lot of suggestions.

My appreciation was also extended to Mr. Christian Paul. As my in-company mentor, he always guided and helped me a lot throughout the research.

Besides, I want to thank Mr. Max Walter to exchange different opinions and ideas about the assignment with me.

I want to give my special thanks to my family whose love and support enable me to finish my graduation project.

III

Contents

1.1. Background of the research ... 1

1.2. Problem statement ... 1

1.3. Problem analysis ... 2

1.4. Research Objective ... 2

1.5. Research Questions ... 2

1.5.1. Main research question ... 2

1.5.2. Sub-questions ... 3 1.6. Research Scope ... 5 1.6.1. Research focus ... 5 1.6.2. Research boundaries ... 5 2. Theoretical framework ... 6 2.1. Electromagnetic flowmeter (T1) ... 6 2.1.1. Signal converter ... 7

2.1.2. Working principle of the signal converter ... 7

2.2. Calibration for flowmeters (T2) ... 7

2.3. Statistical process control (SPC) (T3) ... 9

2.3.1. Control Charts ... 9

2.3.2. Pre-Control ... 9

3. Research method ... 11

4. Results ... 15

4.1. Errors at calibration (E1) ... 15

4.2. Analysis of error at calibration (E1 &E2) ... 16

4.2.1. Definition of Gk/Gkl ... 16

4.2.2. Flow deviation/zero flow deviation ... 23

4.2.3. Flow uncertainty/zero uncertainty ... 25

4.2.4. Other factors ... 25

4.3. Design of SPC system (E3) ... 25

IV

4.3.2. Specific components design & build ... 27

4.3.3. Control chart build & test ... 31

4.4. Test result ... 35

5. Discussion ... 36

6. Conclusion and recommendation ... 37

6.1. Conclusion for improvement (A2) ... 37

6.1.1. Summary for GK... 37

6.1.2. Flow deviation/zero flow deviation ... 38

6.1.3. Flow uncertainty/zero uncertainty ... 39

6.1.4. Other factors ... 40

6.2. Conclusion for SPC (A1, A2) ... 40

6.3. Recommendations ... 40

Bibliography ... i

Appendix ...ii

Appendix I. Errors at calibration for OPTIFLUX 4000 ... iii

Appendix II. Errors at calibration for WATERFLUX 3000 with IFC 100 ... v

Appendix III. Gk/Gkl distributions of OPTIFLUX 4000 ... vi

Appendix IV. Gk/ Gkl condition for OPTIFLUX 4000(detailed) ... vii

Appendix V. Gk condition for WATERFLUX 3000... ix

Appendix VI. Gk average for WATERFLUX 3000 IFC070(DN100) ... x

Appendix VII. SPC charts of Gk for WATERFLUX 3000 IFC 070 (DN 25) ... xii

1

1. Introduction

This report contains the background of the research, the problem statement, the theoretical information, the method applied on solving the problem, the result of the research and the conclusion for the outcome of the research. The chapter 1 introduces the necessary information of the research, which provides readers with clear and sufficient information to know the problem situation. The chapter 2 helps readers to know the theoretical knowledge to understand the problem statement. The chapter 3 states the method applied on the research for solving the problem. During the result part, chapter 4 shows the outcome of the research. The readers can see what was done and what was achieved for this research in this chapter. Conclusions for the whole research are described in the chapter 5, as well as answers to the main questions of the research.

1.1. Background of the research

Before each flowmeter leaves the factory, the calibration is a necessary part for the complete production. After finishing each calibration, the calibration data are all stored in the database of KROHNE. The purpose of the calibration is to ensure the reliability and accuracy of the product. (KRHONE, n.d.) If most of flowmeters can be calibrated successfully during the first calibration, it could save much time for the company.

The outcome (fail/ pass) of a calibration is influenced by the product quality, calibration procedure, the people who do the calibration and external circumstances. The research is based on analyzing the calibration data, which is defined as the start point for the research. The analysis result of the calibration data is researched for improving the quality of the product, calibration and/or production process.

The research environment appeared to be very complicated. The calibration parameters such as Gk value, Gkl value, zero point, etc. depend on the type of the device and the diameter of different flowmeters. On top of that, there are also differences between different calibration systems, as well as calibration process such as software versions and relevant technical information. All calibration information is all stored in a database. These data were till now not systematically analyzed statistically including calculating averages, ranges, significant level changes over time, total deviation, total spread, the effect of setting different calibration process, etc. After the data analyzing process, the result is researched for reducing the rejection and the failure rate of the calibration. The knowledge of Statistical Process Control (SPC) appeared to be an importance point, so it could be a good tool for improving the product, production or calibration.

1.2. Problem statement

Due to the non-predictability of the results at calibration process, the calibration-process must be repeated sometimes. This leads to a waste of effort in time and money. On top of that, it jeopardizes the promised delivery date for the client.

2

1.3. Problem analysis

On basis of the calibration data analysis, the research is concentrated on finding out what exactly causes the problems and what actions might be done to solve these problems. When a flowmeter is not qualified after calibration, the factor that causes this problem may be unknown. The problem could be originally in the product, calibration rig or the combination of both. Usually, this flowmeter would be calibrated for several times until it is qualified, which leads to a waste of time and cost. Besides, there is also a risk that the settings of the flowmeters are incorrect, which reduces the quality of the product.

This research was dedicated to finding ways to improve the product system or production process quality. In order to increase the first pass rate, it is possible to improve the meter design, meter assemblies, calibration rigs, calibration process, physical production process, production environment (dust and pollution), etc. No matter what can be done for the product design improvement, the research will focus on how it can be improved, such as how the calibration rigs can be improved. The most important thing is to obtain some useful information from the calibration data analysis and optimize the calibration system or production. This is what this research wants to reach to at least. The flowmeter needs to be calibrated to complete the production process and data obtained during the calibration are analyzed as a start point of the graduation thesis.

The research assignment was focusing on the feasibility of an application of SPC for KROHNE. If possible, a specific SPC system should be defined. The best case is a high level first pass such as 95% during the calibration, which also meets requirements on the total deviation and spread. Besides, the time of calibration, different calibration factors and noise were also taken into account. Therefore, an improvement plan for reducing the production problems is delivered to the company.

1.4. Research Objective

The main purpose of this research is to increase the first-pass calibration rate, the reliability of product and the production efficiency as well as decrease the problems in the future production. Therefore, the results will be focused on finding out the best way to increase the reliability of the products and production processes of KROHNE.

1.5. Research Questions

In order to achieve the research objective, main and sub research questions are formulated.

1.5.1. Main research question

The main research question is formulated as follows:

What factors influence the reliability of the calibration and what actions might be taken for improving the product quality of the electromagnetic flowmeter on basis of processing the calibration data with a statistic tool?

3 1.5.2. Sub-questions

In order to find an answer on the main research question, the research is split up into three parts. 1. Gathering basic theoretical information about the project, resulting in theoretical questions. [T] 2. Exploring the current situation at KROHNE by raising empirical questions. [E]

3. Finding an improvement by raising analytical questions by answering the question for KROHNE by how to design an improvement plan for the product and production quality of a KROHNE Flowmeter. [A]

Finally, the following sub-questions must be answered: T1. What’s the principle of electromagnetic flowmeters? T2. What’s the general working principle of calibration? T3. What are the capabilities of SPC?

E1. What’s the relation between the calibration data and calibration? E2. What will cause the miss-calibration?

E3. What will the result of calibration data interpret?

A1. Can statistical process control (SPC) help to improve the product of KROHNE?

A2. What’s the way to reduce the problems and increase the first pass rate of the product of KROHNE?

4

The sub-questions are derived from the main question (see Figure 1).

What factors influence the reliability of the calibration and what actions might be taken for improving the product quality of the electromagnetic flowmeter on basis of

processing the

calibration data with a statistic tool? T. Basic theoretical information about the project

T1. What’s the principle of electromagnetic flowmeters?

T2. What’s the general working principle of calibration?

T3. What are the capabilities of SPC?

E. Current situation with in the company

E1. What’s the relation between the calibration data and calibration?

E2. What will cause the miss-calibration?

E3. What will the result of calibration data interpret?

A. Feasible point of improvement

A2. What’s the way to reduce the problems and increase the first pass rate of the product of KROHNE?

A1. Can statistical process control (SPC) help to improve the product of KROHNE?

5

1.6. Research Scope

1.6.1. Research focus

Due to the complexity of the assignment, it has been decided that the research will focused on KROHNE Altometer in Dordrecht. The research is scoped down into the topics calibration and statistical process control. Technical design aspects of the flowmeter are excluded. Depending on the results, a further application within the other KROHNE subsidiaries might be considered in the near future.

1.6.2. Research boundaries

In order to fit the research within the available time period of the graduation internship, the research boundaries are defined as followed:

1. Methodological limitation. There is a possibility that the interviews and questionnaires contain several sources of bias. It is possible that the interviewees are biased and that their mood influences the nature of their answers.

2. This is a complex research with multiple aspects, which have to be researched. However, as there are only 20 weeks to finish this research, it will not be possible to research every topic in depth. Because of the time restriction, it is decided to only perform a quantitive research with a qualitative research.

3. The researcher did not have any prior knowledge about the calibration process. Gradually, during the internship period the researcher became more knowledgeable in the matter. However, at times, decisions had to be taken in which the researcher was not knowledgeable yet. Therefore, further research might be required, in order to fill in the gaps.

4. The research needs to be conducted within a limited time scale of five months, starting February 8th 2016. Because it is a complex research project which addresses multiple subjects, it is expected there will be a need and a recommendation for further research.

6

2. Theoretical framework

In this chapter, the basic theoretical information related to the problem statement is described by answering the theoretical questions.

T1. What’s the principle of electromagnetic flowmeters? T2. What’s the general working principle of calibration? T3. What are the capabilities of SPC?

2.1. Electromagnetic flowmeter (T1)

Electromagnetic flowmeters (EMF) can measure the volume flow of liquid, slurries, sludge and pastes in almost any industry. The only requirement to use EMF is that the measured process liquid must be at least electrically conductive. Figure 2 shows an example of the complete EMF of KROHNE, which is mainly made up of the primary head and signal converter.

The principle of EMF is based on Faraday’s Law of Induction. It states that when a conductor is moving through a magnetic field (see equation 1), the induced voltage will be created across the conductor. The functional principle of electromagnetic measuring devices relies on this law of nature.

𝑈⃗⃗ = (𝑣 × 𝐵⃗ ) ∙ 𝐿 (1) Where

𝑈⃗⃗ Induced voltage (vector) 𝐵⃗ Magnetic field strength (vector) L Length of the conductor moved

𝑣 Velocity of the conductor moved (vector)

The signal converter generates the current to the field coils. Then, the field coils are going to create a magnetic field. When the process liquid flows through a magnetic field, an electrical voltage is induced in the process liquid due to the movement of the liquid. The induced voltage U is proportional to the velocity v and also the volume throughput.

The following applies to a circular tube cross section: 𝑞 = 𝑣 ∙ 𝜋 ∙ 𝐷2⁄ (2) ₄

𝑞 = 𝑈 ∙

Figure 2 structure of an EMF

① Field coils

② Measuring tube

③ Electrode

7

The volume flow is expressed as q (see equation 2, 3). The induced voltage signal is then picked up via two electrodes in conducting contact with the process liquid and sent to a signal converter. (Hofmann, 2011)

2.1.1. Signal converter

The signal converter is a quite important component of the EMF. It has different functions to perform in the EMF. One of the tasks is to supply the current required by the primary head field coils to generate the magnetic field, which is for EMFs with pulsed dc field. The main function is to process the signal voltages. The task is to amplify the signal voltage without feedback. For this, the impedance of the input amplifier of the signal converter must be high enough so that there is no effect on the measuring accuracy from the internal resistance of the electrode circuit. The amplified electrode voltage is then converted into digital values. Besides, complex digital filter techniques are used to free the signal voltage of any superimposed interference (such as noise and other interference signals). The signal converter then scales the digital values in accordance with the specified operating parameters (such as full scale range, span of the mA output, etc.) and converts the scaled digital values into suitable standard signals for the process, which can be shown on the local display. The flow velocity 𝑣 and the volume flow 𝑞 are calculated using scaling of the calibration constant and the nominal size of the primary head. In addition, the signal converter also provides a variety of diagnostic data including the indication of flow profile, the conductivity of the process liquid, etc. with an internal device bus. The measured data of output can be displayed visually.

2.1.2. Working principle of the signal converter

The central communication of the signal converter is the internal device bus, which connects functional blocks, data setting, diagnostic values, etc. All the data can be called up, recorded, set through the device bus interfaces.

The microprocessor (𝜇𝑃) of the primary signal processing system controls the switching of the field current supply. The active, periodically pulsed dc current is supplied to the field coils of the primary head. The field current creates the magnetic field in the process liquid which moves through the measuring tube. According to the Faraday’s Law of induction, there is a voltage induced in this case. The signal voltage induced from the process liquid is transferred via a shielded signal cable from the electrodes of the primary head to the input amplifier in the signal process system. The amplifier with extremely high impedance amplifies the electrode voltage and then supplies it to the A/D converter. The A/D converter samples the electrode voltage synchronously to the field cycle and converts it into digital values. The 𝜇𝑃 filters the diagnostic values and measuring data and scales them based on the calibration data, which will be stored in a backplane. Through the device bus, all the measurement and diagnostic values will be sent to the I/O unit and display and operating unit.

2.2. Calibration for flowmeters (T2)

Calibration is a necessary step for each flowmeter before leaving the factory. Calibration is a comparison between two measurements. One of them is proven to be right. The magnitude of this

8

measurement value is known, which is also called set value. And another measurement is made in a similar way on the other device (Hofmann, 2011). The result of the comparison when calibrating is expressed as a difference. This difference can be expressed as a deviation in the unit of the quantity measured or as a percentage (KROHNE, 2016). It is important to measure the mutual deviation. There are mainly two calibrations, zero calibration and Gk calibration, which are most important. Gk is a system setting parameter, which makes the device measure more accurately. Zero calibration is for checking the zero point (offset) for the flowmeter. The other calibration issues are not very important, which is introduced in the research proposal.

GK calibration

The flow velocity, v cannot be measured directly. The flow velocity is derived from the electrode voltage. The voltage will be sent to the signal converter to process. The flow velocity has a relation with the definitional value of GK, the measured value of GK, the definitional voltage and the measured voltage. Normally, the equation is listed in the following text. Then, the flow volume or flow rate can be gained through the flow velocity. (KROHNE, 2016)

𝑣 = (𝐺𝐾𝑑𝑒 𝐺𝐾_𝑚𝑒)

−1 ×𝑈𝑚𝑒

𝑈_𝑑𝑒 𝐺𝐾_𝑑𝑒, the definition value of GK

𝐺𝐾_𝑚𝑒, the measuring value of GK 𝑈_𝑚𝑒, the measuring value of the voltage 𝑈𝑑𝑒, the definition value of the voltage

Zero calibration

zero_new = zero_old +zero_reading

Typically, the zero_old is set to zero. The zero_new will be stored as new calibration value after calculation. The zero_ reading is the average flow reading of the cyclic object.

Relation between zero calibration and GK calibration

The zero calibration enables the flowmeter to find the correct zero point, while the GK calibration makes the correct slope for the measurement so that the measure flow velocity can get as close as possible to the set flow velocity. (See fig 3)

9

2.3. Statistical process control (SPC) (T3)

Statistical process control (SPC) was used to be considered to solve chronic quality problems. (Wheeler, 2000) There is an important use for SPC, which is not used as a tool to solve the problem. The main function of it is to monitor an improved product or process so that the problems can be reduced or not appeared. There are mainly two branches of SPC: Control Charts and Pre-Control. Now, the ratio of control chart and pre-control is 4: 1.

2.3.1. Control Charts

The control chart mainly consists of a central line for the average, an upper line of the upper control limit (UCL) and a low line of the low control limit (LCL). All these lines are determined by historical data. (Keki R. Bhote, 2000) Nowadays, the most popular SPC chart is 𝑥 control chart with R- control chart. Figure 4 show the basic control chart. If one point is out of the line of control, this means that the process may be out of control. (Salacinski, 2015) (Oakland, 2002)

2.3.2. Pre-Control

Pre-Control is new, simple, user friendly and it is not as popular as the control chart in the SPC world. V set V measured Without calibration Ideal condition 45⁰ offset zero_new

Figure 3 Graph for calibration

V set, the set velocity of flowmeter

V measured, the actual velocity of flowmeter

10

In pre-control, the drawing tolerance is divided into three zones (see fig 5). These three zones are color coded, which are green, yellow and red.

The middle 50% of the chart is the green zone. And there is a 50% yellow zone or two 25% yellow zones next to the green zone. The red zone is outside of the specification limits. The line between the green zone and yellow zone is called Pre-control line (PC line), and then the specification limits are between red zone and yellow zone. (Keki R. Bhote, 2000)

Figure 5 example of a pre-control chart

The pre-control offers a shortcut in which a sample of only five consecutive units are taken. The unit is the variable that wants to be monitored. Only when all five continuous units are in the green zone in a row, the pre-control can get started. If one of the continuous five units is outside of the green zone, the pre-control is not allowed to start.

Pre-Control in production

- If two units fall in the green zone, the production continues to move on.

- If one unit is in the green zone and the other one is in the yellow zone, the production will continue. - The process is still in control, but the courting restarts.

- If both units fall in the yellow zone, the production will be stopped for adjusting the process and conduct an investigation into the cause of variation.

- If one unit is in the red zone, there must be a rejection on the project. The production will be stopped.

- Once the production is stopped, the process will restart after being corrected.

For this assignment, the specific data set in the SPC depends on the result of the statistical analysis. The specific SPC control charts will be defined for KROHNE product, so it is possible that the SPC for KROHNE product is not designed based on the strict SPC rules. The variation and deviation that lead to miss-calibration will be analyzed according to the basic definition of the control charts of SPC. More theoretical information can be found in the research proposal.

11

3. Research method

The calibration data analysis is considered as the start for this project.

The further research is considered as two separate parts: The conceptual design and the research technical design ( (Verschuren, 2015). In the conceptual design, the researcher has settled the objective, the research questions and model and finally the research definitions and boundaries. In the Research Technical Design, the research strategy is determined. Based on that, the research material is explored and defined. Finally, the research is planned.

During the conceptual design the research questions are formulated. According Verschure 2015, theoretical, empirical and analytical questions are formulated.

The conceptual design is partly described in chapter 1. In chapter 2 the theoretical questions are answered. In chapter 4, the results of the empirical research are given and the research is concluded by answering the analytical questions.

The research strategy is determined by a case study because the research is covering the specific situation within KROHNE Altometer, Dordrecht. According (Verhoeven, Doing Research, 2015) a case study involves one organization of group. Based on Verschure, the case study at KROHNE as characterized as follows:

1. Small domain

2. Labor-intensive approach 3. More in depth than wide

4. A selective strategical sample size from the huge database of calibration data 5. Open retrieval of all data without limitations

The research starts with increasing the knowledge of the researcher. According to processing the calibration data from historical records, the assignment aims to use these data to know some interesting things which can help to improve the production, product system or calibration. All the calibration data are stored in the database after each product is calibrated, so the calibration data already exist. The information data are not gathered directly and personally. In addition, these data are used to analyze for a totally new topic as described in this report. Thus, the secondary analysis method is applied on this project. (Verhoeven, 2015) In order to obtain some useful information from the outcome of the analysis, the statistical approach is used for this part, which includes calculating the average, the range, first pass rate, etc. as well as finding some cross-sections with different kinds of information. The purpose of the calibration data analysis is to know the variations or deviation of the product, calibration system or process that result in the miss-calibration and low first pass rate so that some improvements can be done. According to the result of the data analysis, the project will continue to research the feasibility of defining an SPC system for KROHNE. Therefore, it is a combination of both qualitative and quantitative analysis. Clearly, other data collection methods are not suitable for this project.

In order to answer the research questions, questions are answered by the following way according to the table 1.

12

Research Question

Research Method

T1. What’s the principle of electromagnetic flowmeters?

Literature research Interview with experts

T2. What’s the general working principle of calibration?

Literature research Interview with experts

T3. What are the capabilities of SPC? Literature research Interview with experts

E1. What’s the relation between the calibration data and calibration?

Analyzing calibration data Literature research Interviewing experts

E2. What will cause the miss-calibration? Analyzing calibration data Experimental set-up

Consulting experts inside & outside KROHNE

E3. What will the result of calibration data interpret?

Analyzing calibration data

A1. Can statistical process control (SPC) help to improve the product of KROHNE?

Consulting experts inside & outside KROHNE Confirming results by internal interview

A2. What’s the way to reduce the

problems and increase the first pass rate of the product of KROHNE?

Data-analyzing

Consulting experts inside & outside KROHNE Confirming results by internal interview Table 1 Research questions related to the research method

After finishing the first part of the research, applying the V-model is then considered as the method for the next part of the project. (See fig 6 and 7).

13

For defining a SPC control system, V model is a better methodology than Delft Design Method. Delft Design Method focuses more on creativity. It would be more suitable, if the project is about designing a totally new product (TU Delft cooperates with many other educational and research institutions, 2014). At the same time, during each phase in the Delft Design Method, more details and options need to be compared and decided for each part of the product. While, V model is more suitable and more widely applied if some improvements can be done on the existing product. (SDLC-V-Model, n.d) Most of all, V-model is more flexible compared with other methods.

The application of V-model is done as follows:

Step 1 Study the basic knowledge

First, basic knowledge of the electromagnetic flowmeters and the specific calibration procedures are necessary to learn. Basic knowledge usually lays a foundation for further study. This step is also called as literature study. This step is related to the theoretical framework in both research proposal and final report.

Step 2 Analyze the calibration data

After making clear about most of the technical information, calibration data will be analyzed. During this step, some calculations are done such as calculating the mean, the standard deviation, spread, etc. Besides, some cross-sections are also considered. This is an important step for the whole project. This step is related to chapter 4.1 and 4.2.

Step 3 Do a research about the result

When finishing the data analysis, a research based on the result is going to be done. Errors at calibration will be discussed. Also, actions and improvements for these problems are put forward. The feasibility of the application of SPC is discussed. Step 3 refers to the result part 4.1 and 4.2.

Step 4 List requirements for SPC

• Study the basic knowledge

Step1

• Analyze the calibration data

Step 2

• Do a research about the result of the analysis

Step 3

• List requirements for the SPC system for KROHNE

Step 4

• Design a basic sructure of SPC system

Step5

• Design specifc parts for SPC system

Step6

• Build and test

Step 7

• Set up an improvement plan

Step 8

Figure 6 Flowchart of method

14

Additionally, the requirements for SPC system are listed after analyzing the calibration data, which is used for set up a basic structure of SPC system for KROHNE. Chapter 4.3 is followed for this step.

Step 5 Design a basic structure of SPC system

According to the requirements, a basic monitor of SPC system is discussed. During this step, the basic information of SPC chart are confirmed such as choosing what kind of control charts, what kind of variables to be displayed, etc., which is described in chapter 4.3.1.

Step 6 Design more specific parts for SPC

Once the basic structure of SPC control charts are decided, more specific parts are discussed such as determining the sample size, sample frequency, control limits, center lines, etc., which is showed in chapter 4.3.2.

Step 7 build and test

All specific designs are combined in this phase. Probably, the result of SPC charts will be checked compared with the realistic situation. (See chapter 4.3.3 and 4.4)

Step 8 set up an improvement plan

15

4. Results

In this chapter, the results of the research are displayed, which are mainly made up of two branches: the result of calibration data analysis (4.1 & 4.2) and the design part of SPC (4.3).

E1. What’s the relation between the calibration data and calibration? E2. What will cause the miss-calibration?

E3. What will the result of calibration data interpret?

4.1. Errors at calibration (E1)

For different flowmeter, the rank of error occurred at calibration was different. Two Pareto charts were made after filtering information from the calibration database. Two typical types of products were selected. First,

OPTIFLUX 4000 was selected because it is one of the popular products with enough system setting parameters that can be analyzed. Fig 8 clearly shows the rank of error at calibration for DN10-DN150 OPTIFLUX 4000. The errors at calibration for other diameters of OPTIFLUX 4000 are listed in the appendix I.

The WATERFLUX 3000 with IFC 070 was selected to analyze because this type of flowmeter had some problems according to the feedbacks from the clients. Fig 9 is the proportion of error for WATERFLUX 3000 with IFC 070 DN25-DN150. The research focused on the errors that were important for the production in this report.

Figure 9 Error at calibration for OPTIFLUX 4000 DN10-DN150

16

4.2. Analysis of error at calibration (E1 &E2)

During this part, the main errors at calibration were discussed. The root causes behind them were also described as well as relevant actions for these errors. There were mainly six parts for analyzing the errors at calibration. (See fig 10) The core error at calibration that was analyzed in this report was the Gk/Gkl deviation out of range. Gk/Gkl would be explained in the following text.

4.2.1. Definition of Gk/Gkl

“Gkl/Gk deviation is out of range” was first analyzed in the following text. It was considered as an important point for this project. Both Gk and Gkl are system setting parameters and there is no big difference from each other. The Gk/Gkl was analyzed on basis of the historical calibration data. Before that, it is necessary to know about the definition about this kind of system setting parameter.

𝑄 =𝜋 ∙ 𝐷 2 4 ∙ 𝑣 (4) 𝑈 = 𝑘 ∙ 𝐵 ∙ 𝑣 ∙ 𝐷 (5) U: induced voltage [V] K: device constant

B: magnetic field strength [T] V: mean flow velocity [m/s] D: tube diameter [m]

When the diameter of the flowmeter and the magnetic field keep unchanged, the induced voltage is in proportion to its flow velocity (See equation 4, 5). However, for the same velocity with different diameters, flow signals from the electrodes are not the same. Therefore, Gk is a system setting parameter that shows the flow signal characteristic of a flowmeter during the measurement. The

Errors at

caibration

GK/GKL

deviation out

of range

flow deviation

out of range

zero flow

uncertainty

out of range

flow

measurement

uncertainty

out of range

coil resistance

out of range

others

17

name of this kind of system setting parameter can be expressed different such as Gkl, Gk, which depends on the applied field current.

𝐺𝐾 =840𝜇𝑉

𝐸1𝑚/𝑠 (6)

The Gk is defined as the ratio of 840μV and the voltage signal when the velocity is 1m/s. (See

equation 6) (KRHONE, n.d.)

4.2.1.1. OPTIFLUX 4000

The first research product was OPTIFLUX 4000. First, it was interesting to see if there was any trend that the system setting values Gkl or Gk could be determined by the diameter of the flowmeters. To make clear whether the diameter could lead to different system parameters setting, Fig 11 was set up to show the average values for various diameters. The chart was based on the relation only between Gkl/Gk and the diameters. From the fig 11, Gkl/Gk was different for various diameters, so the diameter of EMFs does influence the setting values.

However, the discrete levels of both Gkl and Gk were not desirable. Fig 11 also shows the spread of measured Gkl values of OPTIFLUX 4000 stored in the database after successful calibration. For

diameter

18

example, the average Gkl of DN 25 was 6.232, but the standard deviation of it was nearly 200%, which means that the range of Gkl for OPTIFLUX DN25 was very large. Because of that, it is not possible to determine one certain value for a certain diameter. Other factors also influence the value of Gkl. Since the chart of Gk was similar with Gkl according to the definition of this system setting parameter, the spread of GK was not displayed in the main text but showed in the appendix III. Next, more factors were considered. A new chart combined diameters, applied converters and lining materials was built for further research. Fig 12 showed Gkl values applied on the OPTIFLUX 4000 from DN25- DN150. The typical lining material was PFA from DN25-DN150. For the same lining material, Gkl values for a certain diameter applied on different converters were very close to each other, except for the converter IFC 040. The Gkl applied on the IFC 040 was always larger than Gkl applied on the other converters for the same lining material and same diameter. Thus, lining material, diameter and applied converter can effect on the setting value of Gkl. The charts with other diameter sizes were displayed in the appendix IV.

To go further research, DN150 was analyzed for getting more information, as DN 150 was a popular and typical one. Fig 13 was based on the calculated average Gkl value from the database for 12 months in 2015. The upper part shows the average Gkl values of typical material PFA applied on the converters except for IFC 040. Clearly, the spread of Gkl was acceptable. This means that the Gkl values are close to each other for the specific diameter, the specific converter and the same lining material.

The lower part of fig 13 shows the Gkl in special conditions for DN150. The blue vertical axis was the values of Gkl, while the red vertical axis shows the spread of the Gkl in each month. For instance, the Gkl value for special lining materials was 7.2429 in January, while the average Gkl for typical lining materials PFA was 5.8329. If the pre-call value were not set properly in this situation, it can cause the error at calibration for “Gkl deviation is out of range.” In October, the Gkl for the lining material ETFE instead of PFA was 6.6265, which was also far away from the average value 5.835 for PFA lining material for the same diameter and converter. Besides, the Gkl applied on IFC 040 was always larger than the Gkl for other converters.

Diameter

19

Fig 14 is a gauss curve for DN150 OPTIFLUX 4000 with PFA lining material, which was built in EXCEL after collecting enough data from calibration database. Most of Gkl values were statistical and gathered in the average area within 6 SIGMA, which showed a nice performance. But there were two unsatisfactory points that were located near 5.5 and 6.2. They could be exceptions, but there’s also a possibility that the setting of these two flowmeters were not correct. The gauss curve can clearly show whether the data were statistical, which was quite useful. The statistical data can be input to SPC system.

month month Gkl

Gkl

Figure 13 Gkl condition for OPTIFLUX 4000 DN150

20

4.2.1.2. WATERFLUX 3000 DN25-DN150 with IFC 070

Fig 15 was the average Gk for WATERFLUX 3000 with IFC 070 from DN25-DN150, which was built after filtering necessary information. For this kind of flowmeters, the lining material was the same. The applied converter was IFC 070. From the figure, it is further verified that the Gk is influenced by the lining material, applied converter and diameter.

The spread of Gk was a little bit large for DN100, so a gauss curve was made for showing the spread of the Gkl (See fig 16). Although a few points gathered at the area that was a bit far from the average, it was statistical basically.

Normally, for the same converter, same lining material and same diameter, Gk/Gkl values of them are statistical. Thus, it is possible to define an SPC system for KROHNE product based on the Gk values.

Gk

Diameter

Figure 15 Gk conditions for WATERFLUX 3000 with IC 070

21

4.2.1.3. Detailed analysis

The volume of the flow is got from the liquid velocity (see equation 5), while the velocity is obtained by the following equation 8. Through adjusting the measured GK value, the measured velocity gets as close as possible to the set velocity, so the deviation between the measured velocity and set velocity will be as small as possible. The deviation is under control, which ensures the accuracy and reliability of the product. 𝑣 = (𝐺𝐾𝑑𝑒 𝐺𝐾_𝑚𝑒) −1 ×𝑈𝑚𝑒 𝑈_𝑑𝑒 (8) GKde, the definition value of GK

GK_me, the measuring value of GK

U_me, the measuring value of the voltage [V]

U_de, the definition value of the voltage [V] when the flow velocity is 1 m/s.

When the flow velocity is 1m/s, the measured GK can be calculated by equation 9. The measured voltage Ume, is obtained by a multimeter through the electrodes.

𝐺𝐾

= (

𝑈

𝑈

) × 𝐺𝐾

Since the volume is in proportion with induced voltage in the case of that both the diameter and induction don’t change, the formula can also be expressed as equation 10, which easier to understand. For an accurate and reliable flowmeter, the actual volume should be extremely close to the measured volume. The deviation between the measured GK and definition value GK is large, which means there’s a difference between the measured volume and actual volume (see equation 10). Therefore, the “Gk/Gkl” deviation is out of range is an important factor for the quality of the products.

𝐺𝐾

= (

𝑄

𝑄

) × 𝐺𝐾

Q_actual, the actual volume obtained from the primary head

Qmeasure, the measured volume after combining the sensor and the applied converter

Therefore, if some problems happen during the process of measuring the voltage, it can influence the measured GK values. (KROHNE, 2016)

22

Since the field current is supplied to the coil resistance from the converter, the problem could also be caused by the field coil which is used to create the magnetic field. (See fig 17)

𝐻 ∙ 𝑙 = 𝑁 ∙ 𝐼 When H is constant 𝐵 = 𝜇0∙ 𝜇𝑟∙ 𝐻 When H with iron

𝐵 ∙ 𝑙

𝜇0∙ 𝜇𝑟 = 𝑁 ∙ 𝐼 (11) H, magnetic field [A/m] B, magnetic induction [T] N, numbers of turns in the coil I, magnetic field current [A] 𝑙 , magnetic path length [m] 𝜇0, magnetic constant

𝜇𝑟, relative permeability of the material

For different converters, the supplied field current may be different, which causes the difference of magnetic induction (see equation 11). In this case, the differences of current from converters lead to the differences of GK. (Hofmann, 2011)

𝑈 = 𝑘 ∙ 𝐵 ∙ 𝑣 ∙ 𝐷 𝐺𝐾 =840𝜇𝑉_𝐸

1𝑚/𝑠

Further, 𝑅𝑀 is called kind of resistance for magnetic field and is expressed as equation 16. Normally, 𝑅𝑀 is expressed as 𝑅𝑀(𝑖𝑟𝑜𝑛) (see equation 14). If there’s air gap between the iron core, the 𝑅_𝑀 will change (see equation 16). When the current remains unchanged, the induction B will change, which also lead to the possibility that spread of measured GK is big.

𝜑 = 𝐵 ∙ 𝐴 (12) 𝑁 ∙ 𝐼 = 𝐵 ∙ 𝑙 𝜇0∙ 𝜇𝑟 = 𝜑 ∙ 𝑙 𝜇0∙ 𝜇𝑟∙ 𝐴 (13) 𝑅_{𝑀(𝑖𝑟𝑜𝑛)} = 𝑙 𝜇₀∙ 𝜇_𝑟∙ 𝐴 (14) 𝑅_{𝑀(𝑎𝑖𝑟)}= 𝑙 𝜇₀∙ 𝐴 (15) 𝑅_𝑀= 𝑅_{𝑀(𝑖𝑟𝑜𝑛)}+ 𝑅_{𝑀(𝑎𝑖𝑟)} = 𝑙 𝜇₀∙ 𝜇_𝑟∙ 𝐴+ 𝑙 𝜇₀∙ 𝐴 (16) 𝜑 , magnetic flux [Wb]

A, area of the cross section [m^2]

Figure 18 showed the components of core part. The core needed to be fastened with two plates tightly to keep the reluctance to perform well.

23

If the connection between the core and the plate was not close enough, it is possible that there’s air gap existing that influenced the GK value (see fig 19). When the current kept unchanged, the increasing reluctance would make the magnetic induction decrease (see equation 13). So, the induced voltage also reduced which would result in the larger GK compared with the original GK. Under this situation, the manufacturing procedures during the construction need to be take care. Action can be taken such as fasten the connection between the frame, plate and the core part.

4.2.2. Flow deviation/zero flow deviation

“The flow deviation out of range” mainly derived from the nonlinearity of the flow speed. Two points are measured during the calibration. When the line passing through the zero point and two measured point is not linear and its slope is not desirable (see fig 20), this means that the flow deviation is out of the acceptable range. It will lead to reducing the accuracy. The nonlinearity derived from that flow profile is different for low and high flow speed. Specific information was not able to be obtained through calibration database. For further research, an expert who is specialized at calibration process Figure 18 Core components for WATERFLUX 3000

24

was asked for help. The main reasons behind this error were the leakage of the flow and improper design of the position of the coil. If the coil is not positioned well, the induced magnetic field cannot cover the whole cross section (See sketch 1 (b)). That’s not a good magnetic field for the flowmeter.

The cross-talk will result in the zero offset. The cross-talk occurs when the layout of the electrode, coil and cables cable is not designed as well, especially when the cables were too close. When the coil and electrodes were too close, the extra mutual capacitance and mutual inductance would interference the main signal. Sketch 1 displayed three structures of core position. Sketch (a) was the most suitable one. The magnetic field was not good in Sketch (b), despite there was no cross-talk occurring, which will result in the nonlinearity of the flow curve. The cross-talk could happen in sketch (c) because of the short distance between each other. Besides, the leakage also influences the zero point.

Figure 21 was the most suitable layout. The brown things are the combination of the coil and cores. The white cable and purple cable were connected with the signal converter and the electrode. The blue, green and yellow cables were also connected to the converter. The yellow cables were used to check the performance of the coil and other connections.

(a) (b) (c)

Sketch 1 Layout of the core V set

V measured 45⁰

zero_new

Figure 20 Nonlinear flow speed

Figure 21 Structure of electrode and core

25 4.2.3. Flow uncertainty/zero uncertainty

The flow/zero uncertainty out of range mean that the noise of the flow or zero flow is not ideal. The main reason behind them was that the limitation of time. The flow may take a long time to be stable during the calibration, especially when there is dirt or air between the process water and the electrode. It is not wise to spend too much time on calibrating one flowmeter. The only thing that could be done is to make sure that the electrodes are clean enough before calibration, which saves the time effectively.

4.2.4. Other factors

There are also other errors occurring at calibrations such as “coil resistance is out of range”, “hardware time is zero”, etc. If the coil resistance is out of range, this means that the measurement of the coil resistance may be not correct or the surrounding temperature influences the calibration process. “Hardware time is zero” is the problem from the software. In this report, those errors were not discussed in this report, because they are not very important compared with other errors.

4.3. Design of SPC system (E3)

Because the Gk values are statistical under certain conditions, SPC is a good tool to monitor the performance of them. The design of both requirements and structure of statistical process control applied on KROHNE’S product was described in the following text.

4.3.1. SPC design

After processing the calibration data, it is obvious that the GK is of great importance for the product. It must be useful if GK value of each product can be monitored by some tools. SPC is a great and priority monitor tool to monitor variables for seeking to improve the process performance. It aims at reducing the variation in key parameters and building the quality into the product (Montgomery, 2009).

There are mainly two branches of SPC: Control Charts and Pre-Control. For this research, the control chart was applied. Reasons are listed as follow:

- The control chart is highly useful in process improvement. - The variation inflation risk of the control chart is minimal. - The control chart can be accepted broadly.

- The control chart is statistically valid. - The control chart is easy to understand.

- The control chart is also the foundation for improving the product which is convenient to find some problem occurring.

GK value is a core point for an electromagnetic flowmeter. When the control chart was used to track the GK values, the performance of GK values can be monitored, which is a good way to improve the product or production. Based on the previous result of the GK analysis, it is clear that Gk has a

26

relation with the applied converter, the lining material and diameter. To achieve a perfect effect, the control charts must be able to filter the GK by lining material, applied converter and diameter. For one electromagnetic flowmeter with the same applied converter, same lining material and same diameter, both the average GK trend and the spread of GK were displayed on the monitor. Fig 22 is a table list of requirements that the SPC control charts need to meet.

Requirements

The SPC chart is able to show the average GK trend clearly. The SPC chart is able to show the spread of GK clearly.

The SPC chart is able to select the variable that has to be displayed. The SPC chart is able to select different features (including diameter, lining material and converter).

The SPC chart should be suitable for KROHNE.

The SPC chart is able to show the performance of GK for the specific product. The design of SPC chart is able to be realized in software.

The design of SPC chart can provide a clear diagram for the company. The design of SPC chart is easy to understand.

The design of SPC chart is reliable. Figure 22 Requirements for SPC

Normally, both the mean values of the quality characteristic and its variability are monitored on SPC control chart. The control of the average condition is usually done by the control chart for the mean which is called 𝑋 control chart. The variability can be monitored by either 𝑅 control chart or 𝑆 control chart. 𝑅 Chart shows the range between the maximum value and minimal value, while 𝑠 chart determines that if the distribution for the process characteristic is stable. S is calculated by the formula of the standard deviation. S represents the standard deviation of each group. (Champ, 1987) At first, 𝑋- R chart was considered, because it is more popular compared with other kinds of control charts. But, R chart show the differences between the maximum value and minimal value instead of the statistical situation directly. When there are various Gk/Gkl values in a group, the R chart is not suitable for calculating the different sample sizes. Besides, the calculation in R-chart is a bit more complicated than the calculation of standard deviation (spread) (see Fig 23).

Although 𝑋 and R charts are widely used, it is occasionally desirable to estimate the standard deviation directly instead of indirectly through the use of the range R. So, 𝑋 − 𝑠 chart was thought as the core for GK display. First, the 𝑋 control chart can clearly show the mean of the GK value stored in the calibration database after successful calibration for a specific product in a controlled condition. The distribution of GK for a specific product can be displayed in the S chart. It is very clear and sufficient for the people who want to see the trend of both GK and its spread. Besides, S chart can

GK trend

GK spread

System

Data

27

also be used on various sample sizes. Therefore, the 𝑋 − 𝑠 chart is the most suitable control chart for this project. (Freund, 1957) (Bryce, (1997-1998))

4.3.2. Specific components design & build

After designing the basic structure of the control charts, more detailed and specific things need to be designed. (See Fig 24) There are two branches: 𝑋 chart and S chart. All the data used for designing control charts derived from the database. WATERFLUX 3000 with IFC 070 was selected as a product to test the availability and accuracy of the SPC control charts in this report. The lining material for WATERFLUX 3000 with IFC 070 was all Rilsan○R standard.

4.3.2.1. Sample size & sample frequency

Normally, the SPC control charts need sample the variables every certain time. (Dodge, 1959) At first, the month was considered as the unit for the horizontal axis. However, it is hard to sample from each month. The production was different every month, which meant that there were various numbers of Gk values stored in the database after calibration each month. The WATERFLUX 3000 with IFC 070, DN 100 was selected to test. There were 27 GK values stored in September and 154 GK values stored in November. If only 20 Gk values were sampled, it was not accurate at all. This was not a suitable sample way. Thus, the average Gk values for each month with all values were computed instead of sampling a certain number of GK every month. Each blue point was the average Gk value for each month. The black line was the average ( 𝑥 ) of all Gk values for DN100 in 2015. The red lines were the control limits computed by equation 17, 18. The diagram looks nice, but it was not easy to see the abnormal situation. The diagram can’t display some non-ideal situations. (See fig 25) The Gk values stored in each month were various, so the chart was not very accurate and persuasive. Besides, it would be difficult to structure the s-chart. (Wetherill, 1991)

Upper Control Limit (UCL) = 𝑋 + 3𝜎 (17) Lower Control Limit (LCL) = 𝑋 − 3𝜎 (18) Where,

𝜎 , the standard deviation of Gk

Spread (deviation)

GK average

Control limits

Central lines

Specific

system

Data

28

For the further research, a new control chart based on the same product in August was made. (See fig

26) Every Gk value in August for WATERFLUX 3000 with IFC 070, DN100 was displayed. Obviously, it

was a huge diagram if every point was considered on the control chart. This kind of control chart was also not suitable, even if it can clearly monitor each Gk value. (Montgomery, 2009)

Therefore, in order to make the control chart more accurate and effective, all GK values are sampled after discussion with the company mentors. The SPC control charts were structured with all calibration data in 2015.

There were 836 Gk values of WATERGLUX 3000 with IFC 070, DN 100 stored in the calibration database in 2015. The Gk values for testing have been already deleted. 20 Gk values were labelled as one group. There were 42 groups totally and 16 Gk values were in the last group (See fig 27).

month

Figure 25 Gk for WATERFLUX 3000 with IFC 070 in 2015

29

4.3.2.2. Center line

Two central values were calculated in the following text. One was the average line for 𝑋 Control chart and the other one was for S- control chart.

𝑋 - Chart

The average GK of each group were computed by equation 19. The center line ( 𝒙 ) is based on the average of 𝑋 of each group. (See equation 20) After calculations, 3.9947 is the center GK value for the 𝑋 control chart. The average Gk values for each group were listed in the appendix VI. 𝒙 =∑ 𝒙𝒏 𝒏 𝒊=𝟏 𝒏 (19) 𝒙 =∑ 𝒏𝒊∙𝒙𝒊 𝒎 𝒊=𝟏 ∑𝒎𝒊=𝟏𝒏𝒊 = 20∗(∑41𝑖=1𝒙𝒊)+16∗𝒙𝟒𝟐 41∗20+16

There’s a GK spread (standard deviation) in each group,

which is calculated by equation 21. (Montgomery, 2009) The spreads (standard deviation) for each group were displayed in the appendix VI. The average spread can be computed by equation 22 after computing 42 spread values. (Bryce, (1997-1998))

𝑠 = √_𝑁1[(𝑥1− 𝑥)2+ (𝑥2− 𝑥)2+ ⋯ + (𝑥𝑛− 𝑥)2] (21)

N=20, for the spread of the first 41 groups N=16, for the spread of the last group

𝑆 = [∑ (𝑛𝑖− 1)𝑆𝑖 2 𝑚 𝑖=1 ∑𝑚 𝑛_𝑖 𝑖=1 − 𝑚 ] 1 2 = [(20 − 1) ∗ (∑ 𝑆𝑖 2 41 𝑖=1 ) + (16 − 1) ∗ (𝑆422) 20 ∗ 41 + 16 − 42 ] 1 2 (22) 𝑆 = 0,1141

Control limits

The upper control limit and lower control limit can be calculated by following equations. 𝑈𝐶𝐿 = 𝑥 + 𝐴3∙ 𝑠 (23) 𝐿𝐶𝐿 = 𝑥 − 𝐴₃∙ 𝑠 (24) Date GK Date GK 10-12-2015 4,087 18-12-2015 4,114 11-12-2015 3,758 21-12-2015 3,962 11-12-2015 3,995 21-12-2015 3,815 11-12-2015 3,824 21-12-2015 4,047 11-12-2015 3,805 22-12-2015 3,845 11-12-2015 3,875 22-12-2015 3,862 11-12-2015 3,785 22-12-2015 4,07 11-12-2015 3,788 22-12-2015 3,823 11-12-2015 4,451 22-12-2015 3,799 14-12-2015 3,79 22-12-2015 4,295 15-12-2015 3,83 22-12-2015 4,151 15-12-2015 3,72 22-12-2015 4,125 15-12-2015 3,728 23-12-2015 3,803 16-12-2015 3,954 23-12-2015 3,746 16-12-2015 4,011 23-12-2015 3,723 16-12-2015 3,932 23-12-2015 3,779 16-12-2015 4,027 16-12-2015 3,905 16-12-2015 4,271 17-12-2015 4,406 group 41 group 42

Figure 27 Last two groups of data for DN100, IFC 070, WATERFLUX

30

𝐴₃ is a constant value which depends on the sample size. 𝑐4=

4(𝑛 − 1)

4𝑛 − 3 (25) 𝐴₃= 3

𝑐₄∙ √𝑛 (26)

From group1 - group 41, n=20

𝑐4 = 4(20 − 1) 4 ∙ 20 − 3= 0.9870 𝐴₃= 3 𝑐4∙ √20 = 0.680 UCL= 3.9947 + 0.680 * 0.1141 = 4.0722 LCL = 3.9947 – 0.680 * 0.1141 = 3.9171 For group 42, n=16 𝑐₄ =4(16 − 1) 4 ∙ 16 − 3= 0.9836 𝐴3= 3 𝑐₄∙ √16= 0.763 UCL= 3.9947 + 0.763 * 0.1141 = 4.0818 LCL = 3.9947 – 0.763 * 0.1141 = 3.9076 S – Chart

The upper control limit and lower control limit can be calculated by following equations. 𝑈𝐶𝐿 = 𝐵4∙ 𝑆 (27)

𝐿𝐶𝐿 = 𝐵₃∙ 𝑆 (28) 𝐵₃, 𝐵₄ are constant values which depend on the sample size.

𝐵3= 1 −

3

𝑐₄∙ √2(𝑛 − 1) (29)

𝐵₄ = 1 + 3

𝑐₄∙ √2(𝑛 − 1) (30)

From group1 - group 41, n=20

𝑐₄ =4(20 − 1)

31 𝐵₃= 1 − 3 𝑐4∙ √2(𝑛 − 1) = 1 − 3 𝑐4∙ √2(20 − 1) = 0.507 𝐵₄ = 1 + 3 𝑐₄∙ √2(𝑛 − 1)= 1 + 3 𝑐₄∙ √2(20 − 1)= 1.493 UCL= 1.493 * 0.1141 = 0.1703 LCL = 0.507 * 0.1141 =0.0578 For group 42, n=16 𝑐₄ =4(16 − 1) 4 ∙ 16 − 3= 0.9836 𝐵₃= 1 − 3 𝑐4∙ √2(𝑛 − 1) = 1 − 3 𝑐4∙ √2(16 − 1) = 0.443 𝐵4 = 1 + 3 𝑐₄∙ √2(𝑛 − 1)= 1 + 3 𝑐₄∙ √2(16 − 1)= 1.557 UCL= 1.557 * 0.1141 = 0.1776 LCL = 0.443 * 0.1141 = 0.0505

4.3.3. Control chart build & test

𝑋 - Chart

After confirming the specific details of the control charts, both of two charts were set up with the necessary information and calculations. The black line was the center line for GK. Each blue point was the average of every 20 GK values. The red lines were expressed as control limits. The X – chart showed the average condition of GK for WATERFLUX 3000 with IFC 070 and DN 100 (see fig 28).

32

S – Chart

The S- chart mainly presents the spread of Gk for WATERFLUX 3000 with IFC 070 DN 100 (See fig 29). Normally, the Gk values for a certain diameter, converter and lining material are within an acceptable range. The spread of Gk was derived from the standard deviation of Gk. The upper control limit (UCL) for the S-chart was not necessary, as long as the spread is small enough. The less the spread is, the better the Gk/Gkl performs.

The combination of X – chart and s-chart displayed a nice diagram of Gk for WATERFLUX 3000 with IFC 070, DN100. Every Gk value was taken into account.

For WATERFLUX3000 with IFC 070, DN25, there were 104 Gk values stored in 2015. In this case, 5 Gk values were set into one group. So, there were 21 groups together. The control charts for this type were displayed in the appendix VII. Table 2 shows the definition of the group size.

Table 2 Definition for group size

Number of Gk values Group size Number of Group

0-200 5 0-40

200-400 10 20-40

400-1000 20 20-50

1000-2000 40 25-50

For the SPC control charts, the control limits and center lines were set on each diagram. This kind of controls charts were checked, but it is found that there was not enough information which could be obtained. After discussion with the company, three more lines were decided to be display on the X – chart. There’s an expected Gk which is considered the Gk set value. 10% deviation of the expected Gk is the acceptable range for a successful calibration for WATERFLUX 3000 with IFC 070. The upper specification limit (USL) is based on the expected Gk plus 10% deviation, while the lower specification limit (LSL) is based on the expected Gk minus 10% deviation. Once there is any point which is out of

33

the specification limits, the “Gk deviation is out of range” will result in the failure of the calibration. The interface of the SPC control chart was created then. (See fig 30) The company considered that for a nice interface, the control limits for the last group were the same as the previous values so that the control limits on the charts were a straight line. But people should be careful when checking the last group.

First, the s-chart would be checked at first. Once the spread in one group in the s-chart was too high and even out of the upper limit control, the group with high spread would be checked in the 𝑥- chart for checking the average condition of this group. The group with high spread will be zoomed up. WATERFLUX 3000 with IFC 070, DN100 was discussed here. There were 20 Gk values in one group. From the s-chart in figure 30, the spreads of group 4 and group41 were higher than the UCL. Then, the two average Gk values of these two groups were checked in 𝑥- chart. In group 4, the average of this group with 20 Gk values was also above the UCL of 𝑥- chart. For group 41, although the average Gk was in control, the spread of this group was big. Then, both groups were zoomed up (See fig 31).

Variables Gk Feature 1 Converter IFC 070

Flowmeter WATERFLUX 3000 Feature 2 _Rilsan○R

From 01-01-2015 Feature 3 Diameter DN100

To 31-12-2015 Plot diagram

34

In both group 4 and group 41, there were successful Gk values which were actually out of the 10% deviation of the expected Gk. Normally, they cannot be acceptable by the client.

Figure 31 Abnormal condition

Emphasis was then put to the 𝑥- chart, after checking abnormal point of the s-chart. From the 𝑥- chart in fig 30, there was an area (from group 23 to group 27) that the continuous average Gk values were above the UCL of 𝑥- chart. Then the researcher went back to check this area in the S- chart. The values in Group 24 and group 25 were large in both control charts, so group 24 and group 25 were also zoomed up. (See fig 31) It is clear that in these four groups, there were Gk values that were not within the required range. As group 24 and group 25 were successive, the groups near these two groups also needed to be checked. It is found that there were also some unexpected results in group 22 and group 23. (See fig 31)

When looking back to all original data of WATERFLUX 3000 with IFC 070 DN100, the unexpected Gk values which were out of 10% deviation were found via the SPC control charts. The largest acceptable Gk value was 4.4. However, there were still just a few Gk values that just exceeded the specification limits a little bit (from 4.40 to 4.49). Those values were also possible to be tracked, because the UCL in s-chart was nearly 17%, which was a bit high. In this condition, if the UCL in s-chart was reduced such as decreased to 15%, it would be better to find more Gk from 4.4.0 to 4.49. (See fig 32) The group 13 has a large spread too. When looking at the Gk values in group 13, there was one Gk value that exceeded 4.4.