Vous êtes sur la page 1sur 5

The influence of machine settings on solder faults by design of

experiments
Anubhav Tewari, Madhu Dason (Senior Member IEEE)

ABSTRACT — In automated selective soldering processes degree of importance of control parameters on the
unoptimized machine settings are believed to play a crucial response parameter with a minimum number of runs, and
role in contributing to solder faults. This study investigates hence obtain the optimal settings which would minimize
the effectiveness of a statistical design of experiment the number of solder faults.
approach to determine the optimal settings of a selective
soldering machine, and hence reduce solder faults. The The interaction of certain short listed explanatory
experiment was run on a selective soldering machine used to variables turned out to be a well documented fact. A
selectively solder connectors and varistors on a printed
quadratic response surface design was chosen to analyze
circuit board. It helped determine the optimal settings for
the machine and hence reduce solder faults by a considerable the interaction and second order effects. Industrial
margin. Overall, the design of experiment approach proved processes that require third and higher order models are
an effective tool to reduce defects and characterize the highly uncommon, hence third and higher order models
selecting soldering setup. were ignored. Quadratic response surface design provided
advantages over other designs like ease of finding optimal
Index Terms — Optimization methods, characterization,
Selective Soldering, Quality.
machine settings, making the design more robust against
external and non-controllable influences and fewer
experimental runs. The model turned out to be quadratic
I. INTRODUCTION as assumed and gave the desired optimal settings, which
were successfully implemented.
For a quality conscious company into manufacturing
electronic circuit boards, fixing solder faults takes a fair
bit of time and effort given a high fault rate, resulting in II. DESCRIPTION OF PROCESS
inconsistent quality standards, increased costs and low
Two main printed circuit board (PCB) configurations
efficiency. It is believed there are various factors
were being run through the selective soldering setup. The
contributing to solder faults which can be classified into
setup enabled through-hole devices on a PCB ‘selectively’
four broad categories namely machine based, material
soldered from below. The machine utilized robotics to
based, management based and human errors [1].
individually move each board over stationary soldering
Unoptimized machine settings out of these four are stamps.
believed to be a major contributor to solder faults.
The process firstly required the through-hole devices to be
There are various approaches to finding the optimal placed manually on the PCB’s, which were then sent to
machine settings for a soldering machine ranging from the fluxing station via a conveyor. Flux is the first process
naïve techniques like trial and error to more sophisticated of soldering, flux plays an important role in soldering. It
techniques like Failure mode and effect analysis (FMEA), prevents oxidation of the board during heating, and at the
Fishbone analysis etc. Design of experiment approach was end of the process the flux still needs to have enough
chosen because it takes into consideration unpredictable activity to prevent bridging when the assembly leaves the
interactions that might occur between variables and also solder bath.
because it allows several variables to be analyzed
simultaneously, which perfectly suited the problem in The next step in the soldering process is pre-heating,
question. Statistical formulas determine the optimal which is not intended to reduce thermal shock, but rather
conditions for the variables and indicate if interactions to dry the flux, evaporate the solvents in the flux and
exist between variables which result in positive or bring enough energy to the assembly to achieve good,
negative effects [2]. Practical considerations like cost, filled solder joints, this is achieved by allowing the PCB
time and the ease to conduct the experiment too were to remain over the heater for the desired time. Preheat
being met by the design of experiment approach. temperature should be within a target range between, on
the lower end, the solvent evaporation temperature and,
The precise objective of this DOE was to determine the
on the upper end, less than the temperature where the flux avoid any variation from them, the standardization itself
begins to decompose before soldering begins. helped reduce the solder faults by a small percent. The
process also provided a new outlook to the operators who
The final step in the soldering process is the soldering of were getting more quality conscious as a result of their
the components. The right temperature and contact time involvement in the experiment [1].
are required to allow for good wetting enabling the molten
solder to penetrate through the hole and create a good It was decided to vary four factors for the DOE namely,
fillet without withdrawing the solder from the assembly. ‘Pre-heat time’ (The time spent by the printed circuit
board on the heating station), ‘Solder time’ (time spent by
The board then comes out of the setup and is manually printed circuit board in contact with molten solder),
inspected by an operator. Both boards were generating ‘Solder temperature’ (The temperature of the molten
large quantities of errors that needed rework therefore not solder) and ‘Maintenance’ performed on the machine
being as efficient as desired. The main errors for rework (Cleaning the solder pot of oxides).
that occurred regularly were; non wetting, solder splashes,
solder spike, solder bridging, and not enough solder on The first three variables are continuous in nature and were
the top side of the hole. Other errors such as excessive varied at 3 levels each. The fourth variable was blocked to
solder, no solder, wrong orientation of connector, study its influence on the response. Levels for the factors
component tilt, connector pin not through were not were chosen based on a series of small experiments,
considered as potential problems since they occurred very engineering intuition, customer specifications and industry
rarely or if they did occur they could be classified under standards. The ranges were kept broad to enable
one of the main identified errors. optimization in case the optima happened to be outside the
specified ranges. The range for Pre-heat time was decided
III. EXPERIMENTAL PROCEDURE to be 48 - 72 seconds, 1- 4 seconds for solder time, and
290 - 300 degree C for solder temperature. The variable
The methodology mainly consisted of the process of short maintenance was blocked at two levels namely, ‘before’
listing the factors that would be included in the DOE and and ‘after maintenance’ (blocks 1 and 2, respectively).
choosing an appropriate design that would best deal with
‘After maintenance’ was defined as the runs done within
these factors and give the desired results.
two hours of machine clean-up.
Selecting factors and levels
The response variables were numerous; some of the more
frequently occurring ones being solder bridges, solder rise
Past data was investigated and brainstorming sessions
and non-wetting. Based on past data it was evident that
were conducted with engineers and operators to come up
solder rise was the major contributor out of these, hence it
with an initial list of potential explanatory variables. A
was decided to treat all faults under one heading called
small team went through the list of potential factors,
total defects, as the optimization would not be achievable
trying to classify the candidate factors into four categories
for each response variables individually, given the
[1].
infrequent occurrence of some of them.
- Factors that can be readily adjusted and which we would
Experimental Design
like to vary in the experiment.
- Factors which are important but which cannot be easily
It was decided to use the statistical package ECHIP
set to predetermined levels. Such factors are likely to be
because it has the advantage of providing designs that
good explanatory variables, but be excluded from the
require fewer experimental runs than its counterparts. This
experiment in favor of factors which can be readily
is because the designs are intended to allow estimation of
adjusted.
the response surfaces and hence the optimal regions for
- Factors which will be standardized in order to reduce
the variables of interest, instead of focusing on
variation from that source.
determining the level of significance of the design
- Things that will be neglected at least for the first
variables in effecting changes in the response variables. A
experiment.
response surface quadratic design best suited the needs of
this experiment as it provided advantages like ease of
The above process proved fruitful by helping in detecting
finding optimal process settings, making the design more
certain anomalies in the setup like warped carriers,
robust against external and non-controllable influences
damaged flux brushes etc. These were standardized to
and fewer experimental runs. The design given by ECHIP Table 2: Orthogonality check
is given in table 1.
Pre-Heat
The trials were randomized to avoid variation due to Solder Temp Solder Time Maintenance
Time
unanticipated sources. Some trials were replicated to
capture the variation in the machine setup.
Solder Temp 1
Table 1: DOE Based Experimental Runs

Pre- Total Solder Time 0.118 1


Trial Solder Solder Maintenance
Heat Defects
No. Temp Time
Time (response)
3 300 2.5 1 48 7 Maintenance -0.024 -0.089 1
9 295 1 1 48 0

9 295 1 1 48 1 Pre-Heat
0.148 0.101 -0.029 1
13 300 2.5 2 72 20
Time

11 290 2.5 2 48 11

1 290 1 1 48 45 Cell Contents: Pearson correlation (should be close to zero for


orthogonality)
2 300 1 1 72 101

15 295 1 2 72 22
IV. RESULTS
16 300 4 2 48 18

14 290 4 2 72 106
The data collected from the experimental runs is shown in
table 1 under the heading total defects.
5 300 4 1 72 33

8 290 4 1 60 23 From the experimental analysis, response surface for data


7 290 1 1 48 11 is best shown by means of 3-D plots. The 3-D plot is
12 300 1 2 48 7
shown in figure 1.
4 295 2.5 1 60 0
The plot shows that the properties are non-linear and
6 290 1 1 72 10 optimum lies within the graph.
10 290 1 2 60 2

1 290 4 1 48 11
One of the uses of the statistical analysis is to predict the
value of the response variable as a function of the control
parameters. The predictions include the p-value for each
factor indicating the statistical significance of the factor.
An orthogonality check was performed on the
The model was found to have an R-square value of 0.681.
experimental runs to avoid the problem of confounding of
variables, since this could lead to erroneous conclusion
A Pareto chart (Effects graph) derived from the regression
about the interaction of variables. The Pearson
model is shown in figure 2. Strong interaction between
correlations were found to be approximately orthogonal,
solder temperature and solder time can be observed from
due to the fewer experimental runs. The results of the
the graph. The interaction plot between solder temperature
orthogonality test are attached in table 2.
and solder time is shown in figure 3.
A data collection protocol was decided upon to
Since only 3 levels of the control parameters were chosen,
standardize the proceedings.
the optimization capabilities were limited. In addition, the
scatter in the results further limited the optimization of the
5 boards were inspected for each setting to get a better
results. However, even with these limitations, it was
picture of solder fault occurrences. Boards soldered under
possible to analyze the response surface for optimizations
experimental condition were visually inspected, but
of control parameters to obtain zero defects. The
otherwise treated just the same way as other boards. In
following results were obtained:
addition the overall quality of the printed circuit boards
(PCB’s) was monitored for any unusual effect caused by
the experimental runs.
Figure 1 Figure 3

Total_Defects Interaction - Total_Defects


Solder_Temp = 299.0
Maintenance = 2 Maintenance = 1
100 Pre_Heat_Time = 60.0

80

30
60

40

0
HIP
20
EC

E C H IP
1
0
2 70
So 65
lde 60
290 291 292 293 294 295 296 297 298 299 300
r _T 3 ime
im 55 at_T
e
4 50 re _He Solder_Temp
P

Low Solder_Time = 1.0


Middle Solder_Time = 2.5
High Solder_Time = 4.0

The model predicts that least defects would be observed


for Pre-heat time of 52.80 seconds, Solder temperature of
Figure 2
299 Degree C, Solder time of 1.30 seconds and with a
clean solder pot (Maintenance block 2).

Pareto Effects for Total_Defects A check on the residuals using a probability plot and a
scatter plot was done for the response variable ‘Total
-
4 Solder_Temp*Solder_Time defects’. The results showed residuals followed a normal
+
9 Pre_Heat_Time^2
distribution and there was no unusual deviation.
+
3 Pre_Heat_Time
+
8 Solder_Time^2 V. DISCUSSION
+
Based on the statistical analysis the following
Term

7 Solder_Temp^2
+
2 Solder_Time observations can be made: The control parameters can be
+
5 Solder_Temp*Pre_Heat_Time significantly correlated to total defects and hence the
-
1 Solder_Temp
response can be empirically predicted. The results show
- that all the control parameters were essential for
6 Solder_Time*Pre_Heat_Time
- determining the solder faults, since they appear in either
10 Maintenance[2]
the linear/ non-linear terms. This means that based on this
ECHIP

none of the variables can be discarded. It should be noted


that if a parameter is not significant statistically does not
0 10 20 30 40 50 60 70 80 90
mean it is physically insignificant, since there may too
Effect much scatter in the data to get significance.

The properties show a curvature and thus show a non-


linear behavior that justifies the choice of a quadratic
design of experiments. There are some strong interactions
between control variables.
The fitted model is explaining 68.1 % of the variation.
Indicating that though the major factors were included in
the study, there are other factors that were left out and
account for the unexplained variation
Finally, optimization shows the power of the DOE. It
shows that there were enough data within an accuracy to
permit optimization to be reached. Thus, the minimum
number of runs needed to determine optimum properties
was successful.

VI. CONCLUSION
The DOE is adequate (for a first pass) to determine the
key parameters and optimize them (which is a key
objective of this project). Minimization of the solder faults
can be achieved by setting of the control variables to the
optimum values obtained. The study brings out the power
of a statistical DOE as a tool to characterize the soldering
setup and reduce solder faults by optimization of settings.

ACKNOWLEDGEMENT
The author wishes to acknowledge Madhu Dason for
providing the opportunity to work on this project and for
his constant support and encouragement. Professor Kaye
Marion and Panlop Zeephongsekul at RMIT, for their
valuable guidance. And Chris Kelly for his assistance
throughout the experiment.

REFERENCES
[1] G.K.Robinson “Practical strategies for
experimenting,” pp 81-87, 2000 Edition.
[2] Kert Jensen and Prasanna Paralkar “Analysis of
design of experiments used as a tool to implement
environmentally safe cleaning process in
manufacturing”

Vous aimerez peut-être aussi