Вы находитесь на странице: 1из 11

SPC IN PAINT APPLICATION: MISSION IMPOSSIBLE?

SUMMARY

This paper describes the introduction of process control methods based on facts and figures in a complex

process (paint application in the automotive industry) that traditionally has been operated based on experience

and visual evaluation. An introduction to the process is followed by a description of the method we developed

to make the outcome of the process measurable. Once this method was established several designed

experiments were used to get to know the influence of application parameters. As a next step statistical process

control was introduced. A lot of attention is given to the way we convinced people of the validity and

importance of this - for them - new way of running a process, as this has been and still is the most difficult

aspect of the project. This is an ongoing project, so we do not only conclude with the lessons learned but also

with on overview of steps to be taken in the future.

INTRODUCTION

Volvo Cars Europe Industry (VCEI - Ghent / Belgium) invests a lot in reducing or eliminating negative

environmental impacts. One of their biggest investments has been the installation of a new paint shop that uses

water based paints in stead of solvent based paints. In comparison to the existing paintshop this new shop is

heavily automated and almost all manual spraying has been eliminated.

Contrary to what one would think process control for this automated plant is more difficult than for a

manually operated plant. The operator is extremely flexible and capable of handling small process changes, by

adapting the spraying conditions immediately (slightly changing distance, amount of paint, …). His closed

control loop is based on the visual evaluation of the result of the spraying process. A robot does not have this

visual evaluation and just follows a pre determined program.

With this in mind the management of the paint shop decided that more preventive control systems were

needed and that Statistical Process Control (SPC) could become a valuable tool. A special task force (see next

paragraph) was put in charge of the introduction of process control systems. The long term goal is to obtain a

capable paint process.

1
A TEAM EFFORT

An automotive paint shop is a continuous line where the car bodies undergo a series of treatments as can be

seen in figure 1.

Internal color External color External color


Preparation
coating coating layer 1 coating layer 2

Drying and Clear coat Clear coat Drying and


baking external internal baking

Figure 1: schematic presentation of the paint process

In order to study this process and introduce process control systems, knowledge from different areas is

needed. In the SPC task force several competence’s were brought together. The group includes people from

pre-production that had gained experience with water based paints in laboratory tests and have a lot of

experience on solvent paint application from the old shop. VCEI works with its paint suppliers in a partnership

relation so technical support people from the paint suppliers are also part of the team. The production area

manager, an operator and a quality control responsible complete the team. An external consultant was brought

in to guide the team and to bring the necessary statistical knowledge into the group.

The original intent was to actively involve maintenance people in this work. However, the introduction of

control systems coincided with the installation and start-up of the factory and all maintenance capacity was

needed to get the installation running. In view of the conclusions of various elements of our work (see further)

the active involvement of maintenance will be a necessity in the future.

In fact, the almost inevitable start-up problems have seriously added to the complexity of our work. There

were numerous disturbances (mechanical, software,…) and a lot of changes, not only to the installation but also

to the materials. Water based paint is new for the organization and it became clear that not all of the experience

gained over the years on solvent based paints was still valuable.

2
The purpose was to improve the quality of the process, not to introduce statistics! Classroom teaching on

statistical techniques has not been part of the project. Processes were studied and if statistical techniques could

help they were introduced. All knowledge transfer was done by real time application of techniques, helping

people solve problems and better understand their own processes. Basic principles were explained with special

attention to statistical thinking, rather than statistical formulas. Now, after about one year of work, some people

want to know more about the theoretical background because they saw the power of these techniques and want

to use them in other applications.

This has been a deliberate approach. As a consultant I have seen many cases where a lot of money was

invested in training but the net result was extremely poor. You can only convince people of the value of using

new techniques if you can prove it works for them. And even then it is difficult to introduce new ideas. Don’t

forget the golden rule (Shuster, 1994): “No one can change another person’s mind, but one can place others in

an environment that enables them to choose to change their own minds”.

MAKING THE OUTCOME MEASURABLE AND GETTING THE FIRST RESULTS

As can be seen in figure 1 a car body entering a paint shop is subjected to a lot of processes. After dust

removal, the interior is color sprayed with robots, then the exterior gets its color batch (for metallic paints this

is done in two consecutive steps) and then the paint is dried and baked. After that clear coat is applied both

internally and externally and this varnish is dried and baked. After additional cleaning the car is inspected for

paint quality. It is clear that in the entire cycle a huge number of parameters are involved. When problems are

detected in inspection, experience is called in to try and locate the cause of the problem as quickly as possible.

This is obviously an “after-the-facts” type of control. Because of the length of the line and the production speed

a lot of cars have been painted by the time a defect is detected.

In most SPC applications it is at least clear what has to measured and what the tolerances are. This is not the

case in this paint application. It is a practical impossibility to evaluate the quality of the various steps on the car

body. Paint tolerances (thickness, visual aspects) only exist for the overall layer on the finished car.

So in order to be able to study and control the processes of paint application a simulation of the process was

developed. In this simulation standardized plates are sprayed in the same conditions as the car body. These

plates are subsequently baked and evaluated for layer thickness. This gives us an idea of the state of the specific

spraying apparatus under study without interference of the other processes. Figure 2 gives a schematic view of

this simulation. This simulation is called a “paintbrush” and the final result is a paint distribution as illustrated

3
in figure 3.

Paintspray

Obtained pattern
Figure 2: Schematic presentation of the paintbrush process

15.0
Thickness (µ)

10.0

5.0

0.0
-40.0 -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0 40.0

Figure 3: Good paint distribution pattern from paintbrush test


Distance from center (cm)

In order to evaluate the value of the paintbrush simulation we compared the values obtained on the brush

with the values found on a car body. To do this we took a classical paintbrush and then fixed a plate on a scrap

4
body and painted this scrap body under production conditions. The result of the comparison for external robot

spraying is given in table 1, showing an extremely good correlation between car body and paintbrush.

Table 1: Comparison in thickness between paintbrush and car body

(external robots 3 and 4)

Thickness on Thickness on

paintbrush (µ) car body (µ)

4.60 4.61

4.56 4.54

3.51 3.52

2.85 2.86

3.42 3.58

2.79 2.58

From the paint distribution of the paintbrush a large number of numerical characteristics can be derived. They

are related to the amount of paint (total surface under the curve), the general pattern of the paint (thickness and

width) and the shape of the pattern (skewness and peaks). The paintbrush shown in figure 3 gives an idea of

how a good pattern should look like.

The group was aware of the fact that this simulation would not allow us to predict all possible defects.

However, we still considered the method to be worth while to use as a way to study the processes and maybe to

keep them under control. For each application process (internal, external, color, clearcoat) a specific way to

take a paintbrush had to be developed, because there are large differences between them (layer thickness,

spraying positions, using overlap spraying or not, …). The preliminary tests performed to develop these

methods gave us a first useful application of the paintbrush concept.

During the tests we regularly observed patterns that deviated considerably from the optimal form. Two

examples are given in figure 4.

When examining the robots producing these deviating patterns, we could link the pattern to a defect on a

specific robot component. Worn out or slightly damaged components would cause specific deviations to the

pattern. Also errors in reassembling robots after cleaning immediately show up. We are currently investigating

the relationship between the shape of the pattern and the robot components in more detail.

5
These first results gave us the necessary confidence to use the paintbrush to study the various application

processes.

Thickness (µ)
15.0

10.0

5.0

0.0
-40.0 -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0 40.0
Distance from center (cm)

15.0
Thickness (micron)

10.0

5.0

0.0
-40.0 -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0 40.0
Distance from center (cm)

Figure 4: two examples of clearly abnormal patterns (peaked and asymmetric)

STUDYING THE APPLICATION PROCESSES

As indicated previously there is a lot of experience in-house on the influence of parameters on the spraying

process for solvent based paints. In addition a lot of pre-production laboratory tests had been done on the

application of water based paints. We still felt it was necessary to evaluate the influence of the major

application parameters in production conditions. By performing these tests on paintbrushes they could be done

quickly and at low cost.

Tests in the past had always been of the one-factor-at-a-time type. Introducing factorial and fractional

factorial designs required some explanation. This is an example of knowledge transfer while working on the

process. The important factor was to explain the basic principles of designed experiments (blocking,

6
randomization, interactions, effects) and to apply the technique. Tests were performed on various processes and

with the knowledge derived from them and proven in production the confidence in the method grew. People

were impressed how very few tests could give them a lot of information on the behavior of the process.

As an example of the use that was made of designed experiments table 2 gives the results of a test on

internal spraying of color paint, consisting of a simple 2³ factorial design with a 4 times repeated center point.

Most designs were of this simple nature.

Table 2: Example of results from a 2³ experiment on internal robot nr. 3 paint color 421

FACTOR 1 FACTOR 2 FACTOR 3

Flow (cc/min) Atomising Fan air (Nl/min) STOTAL (µcm) THICKNESS (µ) WIDTH (cm)

(Nl/min)

390 270 180 114.6 7.1 16.1

390 270 260 67.3 3.8 17.7

390 390 180 146.3 14.5 10.1

390 390 260 101.2 9.9 10.2

490 270 180 139.9 8.5 16.5

490 270 260 85.6 4.8 17.8

490 390 180 168.5 15.3 11

490 390 260 141.8 10.6 13.4

440 330 220 120.8 7.2 16.8

440 330 220 118.8 7.7 15.4

440 330 220 122.0 7.8 15.6

440 330 220 120.7 7.7 15.7

These tests did not only show us the influence of application parameters, but also gave us an idea of the

stability of the paintbrush. From these test results a standardized method for setting the parameters for new

paints was developed. This is illustrated in figure 5. Similar tests were made for the various other processes.

7
Low flow
High atomising
Low fanair

T >>> T<<<
and or and
W>>> W<<<

Adjust FLOW

T<<< and W>>> T>>> and W<<<

Atom (+10;T +.25µ Atom (-10;T -.25µ


W -.5 cm) W +.5 cm)
Fan (-10; T +.25µ Fan (+10; T -.25µ
W -.2 cm) W +.2 cm)
Figure 5: Strategy for setting the parameters to obtain the desired paintbrush in

internal color spraying (derived from the evaluation of the results of table 2)

8
INTRODUCING STATISTICAL PROCESS CONTROL ON EXTERNAL ROBOTS

The work done so far was performed within the SPC group. The obtained results had their effect on the

operation of the line, but in an indirect way. Detection of defective components was done using paintbrushes

and the results of the designed experiments were used for original parameter settings. In addition the

information from these tests could be used to solve problems as we were having better knowledge of the

influence of the parameters.

The original goal was to develop a preventive control system, not a tool to help us solve problems once they

occur. We were now faced with the evaluation of the possibility of using the paintbrush simulation as a method

to continuously control the performance of the spraying process. In addition this method would have to be

introduced in production and people had to be convinced of the advantages of using it.

The group decided to start the SPC on the spraying of colors on the external robots. This process involves

four robots spraying specific parts of the exterior of the car body. As each robot is a unique element in the

process and taking and evaluating a paintbrush requires some time, we decided to use a moving range chart and

take one measurement a day. Figure 6 gives an example of a control chart on the amount of paint on the plate

for one of the robots. Obviously, control charts were set up for all characteristics that can be derived from the

paintbrush evaluation.

The first results showed that the moving range charts could be used as a means of continuous control, and

we started to measure the paintbrush every day to see if we could detect trends and other out-of-control

conditions. We also explained to the production and maintenance people what the charts meant and how we

were planning to use them as a warning system for the detection of process changes. We thought that everything

was properly explained and that people were convinced of the validity of the method.

However, until now it has been a continuous struggle to get operators and maintenance to react to out-of-

control signals from the control chart! The paint process has traditionally been run on the motto: “If it works,

leave it alone!”. Deviations that do not lead to rejects on final inspection are not considered meaningful. As the

external robot spraying only partly contributes to the final result it is likely that changes in this process will not

immediately lead to rejections.

9
STOTAL (µcm)
360
UCL
350

340

330
Average
320

310

300
LCL
290

280
27/01/97

27/01/97

28/01/97

28/01/97

29/01/97

29/01/97

30/01/97

30/01/97

31/01/97

31/01/97

3/02/97

3/02/97

4/02/97

4/02/97

5/02/97

5/02/97

6/02/97

6/02/97
Moving Range (µcm)
40
35
UCL
30
25
20
15
10 Average
5
0
27/01/97

Figure 6: X-moving range chart for STOTAL - External robot 2

It has become clear that preventive thinking is an extremely difficult concept to introduce. We are also paying a

price for not involving maintenance from the start, because they are very reluctant to react to signals from a

chart.

As indicated earlier, people can only be convinced by facts, so we continue to gather information from the

control charts. We have been able to show that reported problems can be attributed to trends observed on the

paintbrush charts. No reaction was taken on the first signals but later on defects started to emerge. These

examples are very valuable aids to convince the people of the fact that they can prevent problems if they are

willing to listen to the charts.

10
CONCLUSIONS AND FUTURE ACTIONS

During this project we learned the following lessons:

* Don’t think statistical techniques cannot be applied to your complex process. We found them to be

applicable to processes that were not even measured and where tolerances are unknown.

* Not involving maintenance in the group from the start was a mistake. We now have problems in

getting them involved in process control.

* Never underestimate the difficulty to get people to think preventive. People are used to react

to problems, not to prevent them.

* Introduce techniques while using them, so you can prove their value.

In this project we still have a lot of work to do. First of all we need to convince the organization of the

necessity to react to out-of-control signals from the control chart. We need to introduce control charts on the

other paint application processes. In order to do this in a practical way automation of paintbrush taking and

evaluating is foreseen. We are already introducing actions to reduce variation on the paintbrush, so we have

started on the never ending journey of capability increase. Finally, we are evaluating the possibility of using the

paintbrush for other characteristics (color, appearance,…).

REFERENCE LIST

Shuster, D.H. 1994. Making Cultural Change Happen: Putting Management Transformation Theory to Use.

Quality Management Journal Volume 1 issue 2 (January 1994): 7-17

11

Вам также может понравиться