Quality
Network Analysis: When Your Problems Aren’t Isolated Islands — But a
Connected Web You Must Map Before You Can Untangle
I’m writing this article after spending three weeks in an automotive
factory where they were trying to resolve a dimensional deviation issue on
a gearbox housing. They insisted the problem was in the tool. Then in the
material. Then in the press machine. Every department had its own theory,
every expert had their favorite cause. And every one of them was wrong —
because they were looking for an isolated cause where there was a system
of interconnected factors.
It wasn’t until we sat down at a table and drew a network — a real
network of relationships between mold temperature, granulate moisture,
injection speed, holding pressure, cooling circuit settings, and operator
capability — that the whole picture changed. The problem wasn’t in one
place. The problem was in the dynamics of the network. And so was the
solution.
That’s the world of Quality Network Analysis — an approach that changes
the way we perceive quality problems.
What Is Quality Network
Analysis?
Quality Network Analysis (QNA) is a methodology that views product and
process quality not as a set of isolated parameters, but as a complex
network of interconnected factors. Inspired by graph theory and network
science, QNA maps the relationships between inputs, process variables,
human factors, and quality outputs — revealing hidden dynamics that
traditional tools often overlook.
Think of a traditional Ishikawa diagram. It’s a great tool — but it
displays causes in a hierarchy, not in relationships. In the reality of
your production, causes interact with each other. Temperature affects
viscosity, viscosity affects pressure, pressure affects dimension, but
also feeds back to machine settings. It’s a circle, not a tree.
QNA makes this circle visible.
Why Traditional Tools
Aren’t Enough
After twenty-five years in quality, I’ve seen every tool out there.
FMEA, Ishikawa, 5 Whys, Kepner-Tregoe, 8D — they all have their place.
But they also share a common blind spot: they assume causes are relatively
isolated and linear.
The reality of manufacturing is different:
Problems are mutually interconnected. Increased
temperature in the dryer doesn’t just affect material moisture — it also
changes its flow properties, which changes injection parameters, which
changes internal stress, which changes dimensional stability after
cooling. The chain is longer and more branched than any Ishikawa diagram
can capture.
Nonlinear interactions are the rule, not the exception.
Anyone who has done DOE (Design of Experiments) knows that interaction
effects between factors are often more significant than main effects.
Yet most problem-solving ignores interactions until they are “proven” —
which usually happens only after several failed attempts at a
solution.
Feedback loops create cycles. An operator adjusts a
parameter, the machine responds, a sensor measures a deviation, the
operator adjusts the parameter again. This cycle can be stabilizing — or
it can lead to oscillation that worsens the problem. Without visualizing
the network, you can’t see these dynamics.
How QNA Works in Practice
Let me walk you through the process I use when applying Quality Network
Analysis in a real factory.
Step 1: Identifying Nodes
The first step is tedious but critical: you identify all relevant
“nodes” in your quality network. Nodes can be:
- Input variables: temperature, humidity, pressure,
speed, time, material composition - Process parameters: machine settings, line speed,
sequence of operations - Human factors: operator experience, fatigue cycle,
shift rotation - Measurement characteristics: dimensions, hardness,
surface quality, electrical parameters - Environmental factors: ambient temperature, hall
humidity, vibrations - System factors: maintenance schedule, production
planning, foundry batch changes
In one project, we identified 47 nodes for a single pressing
operation. Sounds like a lot? In reality, that’s often a conservative
estimate.
Step 2: Mapping Relationships
Now comes the crucial moment. For each pair of nodes, you ask:
“Does node A affect node B?” And if so, “In which direction and with
what strength?”
You create a relationship matrix — an adjacency matrix in graph theory
terms. Each cell of the matrix says: “This factor affects this factor
with this strength.”
Tools I use for this:
- Expert assessment: I sit down with process
engineers, operators, and maintenance technicians. Their collective
intelligence is invaluable. - Correlation analysis of historical data: where you
have data, you verify expert estimates with statistics. - DOE results: previous experiments often contain
information about interactions that no one has connected. - FMEA database: your existing FMEAs contain
implicit relationships — they just need to be extracted.
Step 3: Network
Visualization and Analysis
Once you have the map, you visualize it. And that’s where the magic
happens.
Suddenly, patterns that were previously invisible reveal themselves:
Star nodes — factors that are connected to many
others. These nodes are your strategic points. A change in a star node
propagates through the entire network. You ignore them at your own
risk.
High-centrality nodes — factors through which the
most “paths” in the network pass. These nodes aren’t necessarily the
most directly influential, but they control the flow of influence in the
network. They are your critical control points.
Clusters — groups of tightly interconnected factors
that function as “super-nodes.” When you affect one factor in a cluster,
you likely affect the entire cluster. This is important for solution
design.
Isolated nodes — factors that barely affect anything
and are barely affected by anything. These can be safely deprioritized.
Step 4: Identifying Key
Interventions
This is where QNA differs from traditional approaches. You’re not
looking for a single “root cause” — you’re looking for key
intervention points where your action has the greatest systemic
effect.
Let me return to the story from the introduction. The gearbox housing
had a dimensional deviation. Traditional analysis pointed to the tool.
But network analysis revealed that the real intervention point was in the
mold cooling circuit — a node that didn’t have the
largest direct impact on dimension, but had high centrality in the
network. When we optimized the cooling water flow, the mold temperature
profile stabilized, which stabilized viscosity, which stabilized cavity
filling, which stabilized dimensions.
One change — systemic effect.
QNA and Data: From
Intuition to Science
Modern factories produce enormous amounts of data. QNA provides a
framework in which this data gains meaning.
Pearson and Spearman correlation tells you whether two
factors are related. Partial correlation tells you
whether they are related independently of a third factor.
Bayesian networks can model causal relationships.
Granger causality can reveal temporal relationships
between factors.
You don’t need a PhD in statistics. You need a disciplined process and
a willingness to look at problems as systems.
In one project, we used historical data from the MES system to verify
the expert network. We found that 70% of the relationships experts
identified were supported by data. But — and this is important — the
remaining 30% were either unsupported, or the relationships were in the
opposite direction than experts assumed. That’s the value of QNA:
verifying intuition with data.
When to Use Quality
Network Analysis
QNA isn’t for every problem. For simple causes, 5 Whys is enough.
But consider QNA when:
- The problem keeps returning despite being “solved.”
That’s a signal you’re treating a symptom, not a systemic cause. - Multiple departments have different theories. That
means each one sees part of the picture — the network connects these
pieces together. - The problem changes over time or depending on
conditions. Network dynamics explain why. - Implementing a solution doesn’t bring the expected
result. You’ve probably hit the wrong node in the network. - You’re about to invest in a significant process
change. QNA will show you where the investment will have the
greatest effect.
Practical Example:
Plastics Manufacturing
Let me share a specific case from practice — with anonymized details,
of course.
A factory was producing plastic connectors for the automotive industry.
The problem: intermittent occurrence of cracks on castings after storage.
Cracks appeared randomly — not on every batch, not always in the same
place, not always with the same intensity. A classic quality engineer’s
nightmare.
Traditional analysis: – Materials laboratory: “Material is fine,
certificates OK.” – Process engineers: “Injection parameters are within
specification.” – Toolmaker: “Tool is in good condition, no damage.” –
Maintenance: “Machine maintained according to plan.”
Everyone was right — and the problem persisted.
QNA approach:
We identified 34 nodes. We mapped the relationships. We visualized the
network. And we saw what no one had seen before: a cluster
consisting of granulate moisture, storage time in the foundry, barrel
jacket temperature, and injection speed was functioning as a
trigger mechanism.
Individually, each parameter was “fine.” But when granulate moisture
increased (even within specification) and simultaneously storage time in
the foundry extended (even within limits) and simultaneously the barrel
jacket was running at the upper tolerance limit — a perfect storm was
created that produced microscopic porous structures, which after storage
led to cracks.
The solution? Simple, once you saw the network: tighten tolerance
windows on these four parameters simultaneously and implement a rule: if
any one of them approaches the upper limit, automatically reduce the
others.
The return? Elimination of a problem that was costing the factory
€180,000 per year in warranty costs.
QNA and Digital
Transformation
In the context of Industry 4.0, QNA takes on a new dimension. A digital
twin of your production line isn’t just a virtual copy — it’s a dynamic
network you can analyze in real time.
Imagine a system that continuously maps the relationships between your
process parameters and output quality. When a relationship in the network
changes — when a new connection appears or an existing one strengthens —
the system alerts you. Not because a parameter exceeded a limit, but
because the structure of the network has changed.
That’s predictive quality on an entirely new level.
Implementing QNA in Your
Organization
If QNA has caught your interest — and I hope it has — here’s how to
start:
Start small. Pick a dominant, chronic problem. Take
one process. Gather a team of people who know this process from different
angles. And draw a network. Even without software — on a flipchart, with
sticky notes, with arrows.
Don’t rush to a solution. The most common mistake is
jumping from mapping straight to solving. Spend time analyzing the
network. Identify star nodes. Find clusters. Understand the dynamics.
Verify with data. Your expert network is a hypothesis.
Data is reality. Compare them.
Test intervention points. Don’t implement a solution
across the entire network at once. Test individual intervention points
and measure the systemic effect.
Institutionalize learning. Every network you map is
knowledge. Store networks. Compare them. Look for patterns that repeat
across processes.
The Cultural Shift
Perhaps the most important aspect of QNA isn’t technical — it’s
cultural. QNA requires people from different departments to collaborate
on mapping relationships that extend beyond their responsibilities. It
creates a shared language and a shared picture of the problem.
In one factory, I watched a process engineer and an operator agree for
the first time on what the real problem was — because both could see
their part of the network and understood how they were connected. That’s
the power of QNA: it doesn’t just solve problems, it builds
understanding.
Conclusion
In a world where manufacturing processes are becoming increasingly
complex, we need tools that respect this complexity — rather than
simplifying it. Quality Network Analysis is such a tool.
It doesn’t seek simple answers to complex questions. It seeks the right
questions about complex systems. And in my experience, that has often
been the difference between another failed attempt at a solution and a
lasting change.
Your factory is not a collection of isolated machines and operations.
It’s a living, breathing network of interconnected factors. It’s time we
started managing it that way.
Peter Stasko is a Quality Architect with 25+ years of experience in
automotive, manufacturing, and continuous improvement. He doesn’t see
quality as a department, but as a philosophy — and every process as an
opportunity for improvement.
