Sie sind auf Seite 1von 6

More Than

by Matthew Barsalou

In 50 Words
Or Less
Root cause analysis
(RCA) should be driven
by facts and hard data.
To gather and interpret
these facts, the scien-
tific method should be
used together with the
plan-do-check-act cycle.
Most quality profes-
sionals should already
be familiar with the
scientific method, but it
is seldom presented in
an RCA-specific form.
ROOT CAUSE ANALYSIS

Just
OPINION
Ensuring root cause analysis
is driven by facts and data,
not unconfirmed observations
A ROOT CAUSE analysis (RCA) should be empiricalthat is,
based on data and facts. Although this should appear to be obvious, books
about RCA occasionally fail to mention the need to perform experiments,
test hypotheses or even look at a defective part being investigated.
Some experienced engineers have attempted to analyze failed parts by
looking at a failure mode and effects analysis (FMEA) to see what could
have happened, but they completely neglected to examine the failed part.
A brief glance at an FMEA is not an RCA. Ideas may be generated by
looking at an FMEA, but a root cause cannot be confirmed in a document,
and a new failure cause would not be listed in an FMEA. Looking at a docu-
ment may be one of the first steps when looking for the cause of a failure,
but it cant be the only step in an empirical RCA.
One consultant with several decades of quality experience claimed a
quality function deployment (QFD) is a good start for an RCA. After being
informed an empirical aspect was necessary, he replied, You should go
into production and look at the work instructions as part of a QFD.

March 2016 QP 39
Looking at work instructions can be helpful, but look- There is a feedback loop with one hypothesis leading to
ing at the part is even better. Work instructions wont tell anotherwith the original hypothesis leading to results that
an investigator whether a part is simply designed poorly. can be compared to data.3 The differences between expect-
Dorian Shainin reminded us to talk to the parts,1 a re- ed and actual results can lead to modifying the hypothesis.
minder that, unfortunately, is necessary for some quality Sometimes, failures must be investigated under condi-
professionals. tions with little initial information available, and the scien-
tific method can be particularly helpful here. To determine
The scientific method and RCA the root cause of a defect, it may be necessary to recreate
The scientific method should be an important part of any the failure to investigate the situation:4 Do so by guessing
quality professionals toolbox. It can be used to generate the most likely cause and attempting to simulate it. Such
sufficient data to begin using traditional quality tools such experiments make it possible to find the problems cause
as Ishikawa diagrams and statistical methods. The scien- by gaining new insights under experimental conditions.
tific method is well known, and its basics may already be Next, the investigator must find a hypothesis that could
understood by most engineers. Depending on the source, account for the failure. Studying the part or system in
however, there is some variation in its depiction. question can be a useful way of generating an initial hy-
The scientific method consists of recognizing the prob- pothesis to test.
lem, collecting data, performing experiments, generating a Depending on the situation, experiments may be sim-
hypothesis that fits the data based on methods or even sus- ple or complex. A quality engineer investigating the fail-
picions, making predictions based on the hypothesis, and ure of small electromechanical components, for example,
comparing experimental results to the predicted results.2 suspected a small mounting screw was denting the casing.

Flow chart of immediate Root cause analysis / FIGURE 2


actions / FIGURE 1

Problem discovered or an 1
improvement is needed.

Plan:
Describe and quantify the issue,
gather data and form
Plan: tentative hypothesis.
Approach to problem
determined.
Do:
Evaluate the hypothesis
empirically.
Do:
Identify needed support and
form team.
Check:
Observe and Use knowledge gained
interpret result. No
from previous test and
Did hypothesis stand form a new hypothesis.
Check: Act: up to testing?
Yes
Containment Sorting, inspections
needed? or rework. Yes

Root cause
No Act: not verified
Verify root cause
is true root cause. Begin Dig deeper.
improvement Not true root
Act:
actions. cause
Determine root cause.

Correct

1 2016 Matthew Barsalou


2 2016 Matthew Barsalou

40 QP www.qualityprogress.com
1
ROOT CAUSE ANALYSIS

To test this hypothesis, he attached functioning and data is exploratory data analysis (EDA), which was cre-
failed units to an oscilloscope and attempted to generate ated by John Tukey.
a signal. Only the functioning units generated a signal, so Tukey compares EDA to the work of a detective. Accord-
the quality engineer took a unit known to be working and ing to Tukey, analysis of data and detective work requires
dented it using a hammer to simulate potential denting tools and an understanding of where to seek evidence. The
from the mounting screw. Components with a large ham- detective seeks clues that would be presented to a judge,
mer dent also generated an oscilloscope symbol, so the and the analysis of data seeks evidence that later can be
hypothesis was rejected. subjected to confirmation testing. Tukeys EDA is used to
The experiment was not ideal for confirming the root graphically depict data to seek out features that dont fit
cause, but it worked well for rejecting an incorrect hy- with what is to be expected or that stand out as odd.8
pothesis, thereby leaving the quality engineer free to in- Performing EDA may be as simple as creating a run
vestigate other potential causes of the failure. chart to observe the occurrences of failures over time.
A hypothesis should be conservative, modest, simple, Measurement data from different production batches also
general and refutable. This means it should not make too
5
could be compared to one another using box plots to see
many assumptions. This can be summarized with the prin- whether the batches are comparable. This information
ciple known as Occams razor, which requires the selec- can be used for hypothesis generation.
tion of the simpler or less complex of two competing hy- The hypotheses generated after using EDA may not
potheses. It also must be possible to refute the hypothesis. lead directly to the root cause, but they can provide a ba-
This element of a hypothesis is based on Karl Poppers sis to begin an investigation. The knowledge gained while
falsification: A hypothesis must be falsifiable.6 A hypoth- evaluating these hypotheses also can lead an investigator
esis that cant be falsified is irrefutable and unscientific. closer to the correct hypothesis during the investigation.
The hypothesis cant be evaluated because it may appear After sufficient data are on hand, John Platts concept of
to be true, regardless of whether it is true. strong inference (SI) can be used to provide RCA investi-
The hypothesis in the electromechanical component gators with a solid scientific approach to RCA.
example was dents are causing failures, and a simple Platt recommends forming multiple hypotheses for
experiment made it possible to quickly refute this hypoth- evaluation. After forming an initial hypothesis to investi-
esis. To form a hypothesis, data are needed. gate, there are three steps in SI:
The data used for hypothesis-forming may be readily 1. Devise an alternative hypothesis or hypotheses.
available. Otherwise, the data may need to be collected 2. Devise a crucial experiment that excludes, if possible,
and analyzed to form the first tentative hypothesis. A root the other hypotheses.
cause investigator already may have a preconceived idea 3. Perform the experiment.
about what the root cause may be, but data are still need- The procedure is repeated, with each hypothesis be-
ed to evaluate the hypothesis. Otherwise, the hypothesis is ing modified. The initial hypothesis is used to predict an
little more than an unconfirmed opinion. experiments results. Platt explains something should be
The use of basic quality tools, such as Pareto and predictable if it is true, and there can be no knowledge
Ishikawa diagrams or run charts, may be helpful here. If without prediction.9 In addition to a lack of knowledge, a
there are insufficient data to form a plausible hypothesis, hypothesis that cant be used for prediction has no value
a root cause investigator could perform exploratory inves- because its accuracy cant be evaluated.
tigations to gather data. A quality engineer investigating the electromechanical
An exploratory investigation is less structured than the component failures hypothesized dents are causing the
formal scientific method,7 but it should not be confused failures. The alternative hypothesis was dents are not
with searching for data in a truly haphazard way. During causing the failures. The crucial experiment was inten-
an exploratory investigation, the root cause investigator tionally denting the components to determine whether
quickly forms hypotheses and seeks to quickly reject them they could still function when clearly dented. It is impor-
or hold them for later, more-structured testing. tant to remember that failing to reject a hypothesis does
Exploratory investigations should be considered a part not prove the hypothesis is true.
of data gathering and not part of a search for a definitive Popper warns that a hypothesis never can be proven
root cause. A useful method for examining preliminary true because it could always be rejected by new data at

March 2016 QP 41
some future time.10 For practical purposes, you should tested, results of the first test are compared to the hypoth-
consider a hypothesis to be potentially true when it has esized results, and conclusions are drawn. If the hypothesis
been ruggedly tested and survived the testing. This testing is refuted, induction is used to form a new hypothesis based
is in the form of an experiment. on the results. The iterative inductive-deductive process
Ideally, the results of each experiment should be used repeats, and repeated cycles of the process can lead the
to refine a previous hypothesis or form a new one that can experimenter to the true state of nature:12 the root cause.
be evaluated through experimentation. Experimentation An RCA can be performed using the scientific method,
should not be thought of as a linear process but as a cyclic including deduction and induction with plan-do-check-act
one that leads a root cause investigator closer to the un- (PDCA) cyclealso known as the Deming cycle, Shewhart
derlying truth with every cycle or iteration. cycle or plan-do-study-actas a basis for experimenting to
George E.P. Box, Stuart Hunter and William G. Hunter find a root cause.13 Using PDCA has advantages that are al-
recommend using the iterative inductive-deductive pro- ready well known in industry and are both structured and
cess.11 flexible.
Deduction uses a general premise to form a conclu-
sion. An example of this would be when an engineer no- Three phases of PDCA iterations
tices a machine is not functioning and concludes a system There are three phases with cycles of PDCA when per-
prevents the machine from functioning so there must be a forming RCA, and each phase contains steps that should
system failure somewhere. be taken. The first phase of a PDCA cycle is the immediate
Induction uses empirical facts to form a general but action phase. The second is the RCA phase, which may
tentative conclusion, such as when an engineer observes involve multiple cycles of PDCA. The third and final phase
threads are stripped from a bolt, which could happen if consists of improvement actions.
it is overtightened. The engineer therefore suspects the The first step during the immediate action phase is de-
stripped bolt may have been overtightened. termining the proper way to seek improvements after a
In the iterative inductive-deductive process, deduc- problem is detected or a decision is made.
tion is used to form the first hypothesis. The hypothesis is A problem may be adequately addressed with an 8D re-
port, which is a common method used in the automotive
Improvement actions / FIGURE 3 industry. An 8D report is a document that lists the team,
describes the problem and includes information on the
immediate actions, root cause of the problem, planned
2 and implemented corrective actions, and long-term ac-
tions to prevent a reoccurrence of the problem.
Plan: A problem also may be best approached using Six Sig-
Determine corrective or ma or other methods. What is important is identifying a
improvement actions.
method that fits the issue at hand.
The second step is when a team is formed. The size,
Do:
structure and complexity of a team should fit the problem
Implement small-scale being addressed. A quality technician confronted with a
improvement actions.
length deviation on the night shift, for example, may get
by with consulting the machine operator. An organization
looking for six-digit savings will need a much larger and

Check: Determine why actions


more complex team.
No
Were the actions did not work. New Although containment actions may not be a direct ele-
effective? actions are necessary.
ment of RCA, it would be remiss to describe RCA without
mentioning containment. This may not be relevant to im-
Yes provements or every failure investigation. If a failure is
Act: detected, however, a decision must be made as to whether
Implement improvements, seek
a containment is needed. The failure of the first part to be
out further opportunities and
update relevant documents. 2016 Matthew Barsalou produced may not warrant containment beyond stopping

42 QP www.qualityprogress.com
ROOT CAUSE ANALYSIS

the machine until the root cause is identified. successful, the improvements are implemented on a large
A customer complaint for sporadic defective parts scale. Other opportunities for implementing the improve-
found at the customers facility could require the control ments should be sought out (for example, on comparable
of parts in production, current inventory, parts in transit products or processes).
and parts at the customers location. Relevant documents also should be updated. This in-
The last step of the immediate action phase is mov- cludes FMEA, control plans, specifications and lessons
ing to the next phase, when the search for the root cause learned databases, as shown in Figure 3.
takes place. Figure 1 shows this concept. The need for an empirical aspect during RCA may be
The first step of the RCA phase consists of describ- considered common sense, but this aspect is often over-
ing and quantifying the problem. The description should looked. After determining the immediate actions to miti-
be clear enough for all involved to understand what the gate a situation, data should be collected and analyzed so
problem is. If possible, the issue also should be described decisions are data driven. An experienced engineer may
numerically, such as save $10,000 or part should be 7.0 recognize a failure and believe he or she knows the root
+/-0.05 millimeters, but is 7.1 millimeters. There may be cause. There is a risk, however, that corrective actions
insufficient information to quantify the problem at the may be ineffective if this is just an opinion and not an em-
start of the investigation, but this should be done as soon pirically determined fact.
as enough data are on hand. There also is a need for empiricism in implementing
Data also must be collected so that there is sufficient corrective actions after a root cause is identified. The
information to form a tentative hypothesis. This hypothe- planned actions must be verified to ensure they work af-
sis is evaluated empirically during the next step of PDCA. ter they are implemented. All you have is opinion mas-
An empirical evaluation could be as quick as looking at querading as fact if there is no empiricism in RCA. QP
the part to see whether the reality fits the hypothesis, or
as time-consuming as a long-term endurance test. The re- REFERENCES
1. ASQ, About ASQDorian Shainin, http://tinyurl.com/asq-shainin.
sults of the test or evaluation are then compared against 2. Sren Bisgaard, The Role of Scientific Problem Solving and Statistics in Quality
Improvement: Some Perspectives, Report Number 158, Center for Quality and
the hypothesized results. Productivity Improvement, March 1997.
If the hypothesis is rejected, any knowledge gained is 3. George E.P. Box, Stuart Hunter and William G. Hunter, Statistics for Experimenters:
An Introduction to Design, Data Analysis and Model Building, second edition,
used to form a new hypothesis, and the cycle starts over. John Wiley & Sons, 2005.
4. Shigeo Shingo, The Sayings of Shigeo Shingo: Key Strategies for Plant Improve-
If the hypothesis survives the evaluation, it should be veri- ment, A.O. Dillon, translator, Productivity Press, 1987.
fied to ensure it is correct and is the actual root cause. 5. Willard V. Quine and Joseph S. Ullian, The Web of Belief, 10th edition, Random
House, 1978.
Ideally, the five whys method should be used to en- 6. Karl Popper, The Logic of Scientific Discovery, Routledge, 2007.
sure the identified cause is the true root cause. Five whys 7. Adriaan D. de Groot, Methodology: Foundations of Inference and Research in the
Behavioral Sciences, translated by J.A.A. Spiekerman, Mouton, 1969.
consists of asking Why? each time a cause of a problem 8. John W. Tukey, Exploratory Data Analysis, Addison-Wesley, 1977.
9. John R. Platt, Strong Inference, Science, Vol. 146, No. 3,642, October 1964, pp.
is identified. This leads to the underlying root cause of 347-353.
the problem. If a cause is not the root cause or the true 10. Popper, The Logic of Scientific Discover, see reference 6.
11. Box, Statistics for Experimenters: An Introduction to Design, Data Analysis and
underlying root cause, the PDCA cycle is repeated from Model Building, see reference 3.
12. George E.P. Box, Science and Statistics, Journal of the American Statistical
the beginning. Figure 2 (p. 40) shows how an RCA can be Association, Vol. 71, No. 356, December 1976, pp. 791-799.
performed with cycles of PDCA. 13. Denis J. Monroe, Root Cause Analysis to Maintain Performance, in Joseph M.
Juran and Joseph A. DeFeo, eds., Jurans Quality Control Handbook, sixth edition,
Corrective and improvement actions may not be a di- McGraw-Hill, 2010.
rect part of RCA. As with containment actions, however,
they are often performed when an RCA is needed. The
first step of the corrective action phase is planning the MATTHEW BARSALOU is a statistical problem resolution
corrective or improvement actions. The next step is im- Master Black Belt in the problem resolution and statistical
methods department at BorgWarner Turbo Systems Engi-
plementing the actions on a small scale to ensure they are neering GmbH in Kirchheimbolanden, Germany. He has a
masters degree in business administration and engineer-
effective and have no negative consequences. ing from Wilhem Bchner Hoschschule in Darmstadt, Ger-
For a failure, this could mean implementing a correc- many, and a masters degree in liberal studies from Fort
Hays State University in Hays, KS. An ASQ senior member,
tive action and performing a 100% check on the improved Barsalou is an ASQ-certified quality technician and engineer, Six Sigma Black Belt,
parts. If the actions are ineffective, new actions must be and manager of quality/organizational excellence. He is a Smarter Solutions-cer-
tified lean Six Sigma Master Black Belt, a technical reviewer for QP, editor of the
planned, and the PDCA cycle starts over. If the actions are Statistics Divisions Statistics Digest and the ASQ country counselor for Germany.

March 2016 QP 43

Das könnte Ihnen auch gefallen