The Art of Investigating Failure

Introduction | When the Right Solution Begins with the Wrong Question

A production line that has run successfully for months suddenly begins producing goods that no longer meet quality requirements. The production manager reports an increase in rejection rates, the operator describes a change in press behavior, and the quality engineer presents test results outside specification.

Possible explanations emerge almost immediately. Perhaps the incoming substrate has changed. Maybe the ink is unsuitable. Printing pressure may be too high, process temperature may be unstable, or a new coating may be affecting product performance. Each explanation sounds plausible. Each may have a technical basis, and often someone can recall a previous incident that seems to support it.

What follows is familiar: a parameter is adjusted, samples are run, results are checked, and the team moves to the next possibility. When the trial fails, another material is substituted, or another setting is changed. After several attempts, the team may have generated a substantial quantity of data without meaningfully improving its understanding of the failure.

The problem is not necessarily a lack of professional knowledge. It often comes down to how the investigation is structured.

An experienced engineer understands that a plausible explanation is not the same as a demonstrated cause. The visible defect may be only the final consequence of a chain of events that began much earlier in the process.

Cracking in an ink layer may originate in the board structure. Lamination adhesion failure may reflect conditions that existed before printing. A color defect can result from an unstable mechanical transfer rather than an incorrect color profile. A package that fails to seal may be affected by the interaction of material structure, contamination within the sealing zone, and converting conditions.

Correcting the visible symptom in any of these situations does not necessarily eliminate the underlying mechanism.

A genuine engineering investigation begins with a willingness to challenge the explanation that initially seems most obvious. It does not begin with "What should we change?" but with "What do we know, and what are we merely assuming?"

1 | Define the Failure Before Trying to Solve It

"The product failed" is not an adequate engineering problem statement. To investigate a failure, the team must identify the requirement that was not met, the circumstances under which it was missed, the stage in the product life cycle where it became apparent, and the magnitude of the deviation.

A product that fails on the printing press differs from one that fails during converting, and from one that passes all factory checks but fails in the customer's application. A printed product may meet color and registration requirements yet be unsuitable for its intended function. Packaging can appear visually perfect while failing in seal integrity, chemical resistance, laminate bond strength, or mechanical durability.

Before beginning root-cause work, define the functional requirement and the conditions under which it was violated. If, for example, a flexible package fails a seal-strength test, the problem statement should not be "an ink problem" or "a lamination problem." Those are hypotheses. A useful statement is: "The measured seal strength of the finished package is below the specified minimum under a defined test condition and in a defined package zone."

That distinction leaves the investigation open to competing mechanisms rather than directing it toward one department at the outset.

Immediate containment must proceed in parallel. If there is a risk of nonconformance related to safety, functionality, or quality, identify and hold potentially affected output under the organization's approved procedures until an authorized disposition is made. Containment is not corrective action. Its purpose is to prevent further exposure while the investigation proceeds. This separation is central to the 8D problem-solving framework.

2 | Investigate Evidence, Not Opinions

One of the biggest sources of wasted effort is mixing facts, interpretations, and hypotheses. An operator may say, "The new substrate caused the problem." What may actually be known is only that the defect began after the substrate was changed. The first statement asserts causation; the second records a sequence of events.

The distinction matters because investigations become biased when an observation is silently promoted into a conclusion.

A powerful first-pass technique is Is / Is Not analysis: identify where the failure occurs, and where it could have occurred but does not. Compare affected and unaffected products, lots, locations, time periods, process stages, and operating conditions.

Imagine that adhesion failure appears in one product but not in a comparable product printed on the same press during the same shift. Instead of immediately investigating every press subsystem, examine the differences between those products: substrate construction, ink-film thickness, surface preparation, print orientation, converting process, storage conditions, or other design features.

Likewise, a failure that starts only after several hours invites questions about what changed during that period. A defect confined to one speed range invites questions about speed-dependent mechanisms. The evidence must be documented before assumptions begin to dominate the discussion.

Not every comparison will be available, and ostensibly "good" and "bad" samples may differ in several uncontrolled ways. Such comparisons are evidence for investigation, not automatic proof of a single cause. The key is to discover distinctions that narrow the question.

Sometimes the most revealing question is not what changed in the failing process, but what remained different in a comparable process that continued to succeed.

3 | Investigate the Complete Process Chain

A finished printed product is the result of a sequence of operations. Raw material is received and stored, conditioned or surface-treated when required, printed, dried or processed according to the printing technology, and potentially coated, laminated, slit, creased, folded, glued, or heat-sealed. Each stage can alter a property that matters later.

A defect discovered at the end of the line may be the delayed consequence of a property that changed much earlier. This is why the process map is often more valuable than an exhaustive list of press settings.

Consider bubbles or bond failure detected after lamination. It is reasonable to examine the adhesive, but the investigation must also establish the failure interface: adhesive-to-ink, ink-to-primer or surface treatment, primer-to-substrate, or cohesive failure within a layer.

A peel test that shows separation is only the beginning. The team must examine what remained on each side of the separated construction. If ink stays with the adhesive and detaches from the substrate, the investigation points toward a different interface than if failure occurs within the adhesive layer.

Similarly, when a paperboard package cracks during folding, it is premature to conclude that the ink lacks flexibility. The strain at the crease may exceed the capacity of the board coating, the ink film, or the board itself. Damage may begin during creasing and become visible only when the package is folded.

The objective is to identify the earliest process stage at which you can detect or measure an actual deviation. The stage where a defect becomes visible is not necessarily the stage where it originated.

4 | Reduce Dozens of Suspects to a Few Testable Hypotheses

Evidence gathering often produces a long list of candidate causes: substrate, ink, coating, pressure, temperature, tension, speed, environment, mechanical components, surface treatment, residence time, and storage conditions. The mistake is assuming that all of them must be tested.

A disciplined investigation is not measured by the number of items on a brainstorming diagram. It is measured by the ability to eliminate explanations that do not fit the evidence and to prioritize the remaining ones.

Every hypothesis should include a credible physical or chemical mechanism. "Temperature caused the failure" is not sufficiently specific. Did temperature change ink rheology, adhesive reaction kinetics, dimensional behavior of the substrate, or differential expansion between laminated layers? If the mechanism cannot be described, the hypothesis does not yet justify a costly production experiment.

Distinguish three causal roles: an initiating cause that triggers the failure mechanism; an enabling condition that allows it to operate; and an aggravating factor that makes the resulting damage more severe.

For example, low moisture content in paperboard may increase cracking susceptibility during folding. Yet this does not prove that moisture is the initiating cause. Crease geometry may be unsuitable for the board caliper, while moisture may only worsen the effect. Increasing moisture could reduce cracking while masking the original design or process mismatch.

Fishbone diagrams and fault trees help map plausible mechanisms; the Five Whys technique may deepen an investigation into a causal chain. But a diagram maps hypotheses, not proof. The investigation advances when an observation or experiment can distinguish one explanation from another.

5 | Thinking Outside the Box—Without Leaving Engineering Behind

Thinking outside the box is not the pursuit of an unusual explanation merely because it is original. It is the disciplined ability to challenge the assumptions that set the investigation's boundaries.

An expert looking at a color defect may immediately focus on ink, optical density, calibration, or screening. But what if the color system is working correctly? What if substrate deformation under pressure is changing local transfer? What if drying changes surface properties and causes a later converting failure? What if average color readings are acceptable while local ink-film thickness is inconsistent?

The question changes from "What could make the ink fail?" to "What physical mechanism could create precisely this result even if the ink is correct?" For lamination, the useful question is not simply "Why is lamination failing?" but "At which interface does separation actually occur, and what conditions are necessary for that interface to fail?"

Challenge the boundaries of the process stage.

If damage appears on a converting machine, include upstream steps in the process map. If the defect appears at the customer, examine storage, distribution, filling, and conditions of use. A downstream discovery is not a reliable location marker for the initiating event.

Challenge disciplinary boundaries

A defect that looks like a color problem may be mechanical. A mechanical defect may depend on surface chemistry. A presumed material defect may arise from package geometry that is incompatible with the intended load or use environment.

Challenge the measurement itself.

Sometimes the team measures the right property at the wrong location. Sometimes a test does not reproduce the application conditions. Sometimes an acceptable average masks a serious local defect. A functional failure should be evaluated with measurements that represent the mechanism and conditions under which the product is expected to operate.

Challenge the assumption that there is only one cause

Two factors can each be inside specification and nevertheless fail when combined. This is where one-factor-at-a-time trial-and-error can be misleading: changing either factor alone may not reveal the interaction. Creative engineering produces alternative explanations; controlled experiments determine which explanations survive.

6 | Choose the Tests That Matter—and Reject the Unnecessary Ones

Not every question requires a production trial. In some cases, a simple laboratory test, a cross-section, microscopy, a retained-sample comparison, or a measurement of incoming material properties can eliminate a hypothesis faster, more safely, and with less waste than a press run.

Before scheduling a production experiment, ask whether a lower-cost test can yield the same decision-relevant evidence. An experiment's value is not how much data it generates, but how much uncertainty it removes.

A costly trial that could produce the same ambiguous result under two competing hypotheses is weak, even if it produces attractive graphs. A simple test that can disprove one of those hypotheses is often stronger.

Stage 1: Isolation tests

The first tests should help localize the failure mechanism. Compare before and after a process step; printed versus unprinted areas; coated versus uncoated zones; or samples from two carefully characterized material lots. Make the difference under examination explicit, and document any remaining uncontrolled differences.

Stage 2: Screening experiments

When several plausible factors remain, screening designs help identify those with the largest effects within the tested ranges. The objective is not yet to determine the perfect settings. It is to reduce the number of variables that merit deeper study. Fractional factorial designs can reduce the run count when many factors are involved, but the experimenter must understand the effects that are aliased or confounded by the chosen design.

Stage 3: Interaction and optimization experiments

After the important variables have been identified, factorial or other suitable Design of Experiments (DOE) approaches can investigate their individual effects and interactions. The objective is to understand the process response well enough to define a robust working region rather than a single apparently successful operating point.

Before execution, define what observation would support each hypothesis, what would contradict it, and what result would leave the question unresolved. A trial with no decision criterion is not yet a complete experimental plan.

7 | Engineering Case Study: The Printed Carton That Passed Inspection but Cracked During Folding

Consider the following illustrative training scenario, based on documented mechanisms of cracking in coated paperboard. It is not presented as a measured customer incident, and no experimental outcome is assumed in advance.

A plant produces printed paperboard packages. The printing meets specification, color measurements are acceptable, the ink film is visually uniform, and production samples pass the usual print-stage checks. After creasing and folding, however, cracks appear along the folds.

The first instinct is to examine the ink and varnish. Perhaps the ink is too brittle, or the protective coating stiffens the surface. These are reasonable hypotheses, but they are not yet findings.

The investigation begins with samples collected before printing, after printing, after creasing, and after folding. These are examined to determine the earliest stage at which damage is detectable. Microscopy or cross-sectional analysis can help distinguish failure in the ink, in the paperboard coating, or deeper in the fibrous structure.

If damage extends through the board coating and into the fibers, the investigation must expand beyond the ink. Candidate factors include grain direction relative to the crease line, crease geometry, board caliper, coating construction, moisture content, and folding conditions.

Published research on coated paperboard describes how creasing conditions, grain direction, and coating behavior influence crack formation. An insufficient crease may produce an unsuitable fold, while excessive creasing can damage the fibrous structure. It would be incorrect to assume that either increasing or decreasing creasing force always improves the result.

First experiment: isolate the crack mechanism

The initial objective is to establish whether damage arises primarily in the surface layers or from deformation within the paperboard structure. Compare printed and unprinted specimens where appropriate, inspect sections through the crack, and locate the first evidence of damage. If unprinted samples exhibit similar cracking, the hypothesis that ink is the sole initiating cause becomes less likely—although ink may still aggravate the result.

Second experiment: test structural factors with a 2×2 matrix

Assume the isolation tests leave two leading factors: grain direction relative to the crease line and crease setting. Instead of changing one factor randomly, test all four combinations using two specified, approved crease configurations (A and B).

Clearly specify and permit the two crease configurations for the board and equipment. Each combination requires independent replicates as appropriate, controlled conditioning, consistent folding, and a repeatable severity measurement. Depending on the application, the response could be crack length, affected area, a predefined visual severity score, or a microscopic metric.

If changing grain direction also requires a different sample layout or feeding orientation, the design must avoid confounding those changes with the factor being tested. It is not sufficient to label four dissimilar production samples as a factorial experiment.

How to interpret the matrix

If crease configuration B improves the response in both grain directions, it may have a consistent main effect. If grain direction influences the response under both crease settings, the data support a direction effect within the investigated range. But if one crease setting succeeds in one direction and fails in the other, the interaction between the factors may be the critical finding.

That is the insight a one-factor-at-a-time approach can miss. Conclusions should be drawn from replicated measurements and an analysis appropriate to the response and experimental design, not from four isolated visual impressions. If needed, the next experiment can include moisture content, coating construction, or other factors justified by the evidence.

What if the experiment appears to succeed?

The investigation is not finished when cracking is reduced. The revised settings must also satisfy fold accuracy, package strength, dimensional stability, appearance, and production-speed requirements. A setting that reduces cracking while making cartons difficult to fold accurately is not necessarily an acceptable solution. Engineering success means meeting the relevant requirements of the complete product.

8 | Experimental Matrices: Learn More from Fewer Trials

A sound experimental matrix does not begin with a spreadsheet. It begins with a precise question. Before selecting factors, the team should define the hypothesis, controllable variables, conditions to hold constant, the true experimental unit, the measurement method, replication needs, and the minimum change that matters in engineering terms.

The order of trials matters. Suppose a flexographic team is studying the effect of press speed on print quality. If every low-speed test occurs at the beginning of the shift and every high-speed test occurs hours later, speed is confounded with time. Temperature, ink behavior, or other operating conditions could have changed in the interim.

Where feasible, randomized or balanced run order helps reduce systematic order effects. If factors such as material lot, day, or production shift cannot be controlled, blocking can help account for them in the design and analysis.

The team must also distinguish true replication from repeated measurements on the same sample. Ten readings from one printed sheet are not necessarily ten independent production runs. Repeated measurements can characterize variation within a sample or measurement system; independent production replications are needed to assess reproducibility across runs.

Retain a baseline reference. For longer test sequences, repeating the baseline condition may help reveal whether the underlying process drifted during the experiment. Without such checks, a favorable result could be attributed to the chosen factor when an uncontrolled change was responsible.

Experimental freedom also has boundaries. Safety rules, equipment limits, manufacturer-approved operating ranges, and chemical compatibility requirements remain binding. Tests must not depend on unsafe pressures, temperatures, speeds, electrical conditions, or unapproved chemistry.

9 | Three Printing Technologies, One Investigation Logic

The investigation framework applies across printing technologies, but the mechanisms and controllable factors do not transfer unchanged from one process to another.

Offset: separate color behavior from image-transfer mechanics

In conventional offset printing, changes in optical density or dot reproduction can involve ink delivery, dampening conditions where applicable, plate-to-blanket and blanket-to-substrate transfer, nip pressure, or substrate response. If the observed defect includes slurring or doubling, simply increasing ink feed is unlikely to be a discriminating experiment. A periodic defect may demand geometric and mechanical investigation before any color-curve adjustment. The test matrix should reflect the suspected mechanism, not every setting available on the press.

Flexography: examine interactions between mechanics and ink transfer

In flexography, quality depends on anilox characteristics, plate properties, mounting tape, nip conditions, ink behavior, print speed, drying, and substrate response. A uniformity defect can originate in metering, unstable contact, drying, or material behavior. Evidence of dynamic contact instability may justify a controlled study of approved speed and pressure settings, whereas an apparent drying limitation would require a different set of variables. Changing viscosity by default simply because it is easy to adjust is not a defensible investigation plan.

LEP: separate imaging, ink transfer, and substrate performance

Liquid Electrophotography (LEP) requires distinguishing between the electrophotographic imaging stages, liquid ink development, intermediate transfer, and substrate interaction. A defect affecting one separation, location-dependent nonuniformity, and a subsequent adhesion failure are not necessarily manifestations of the same cause. If color metrics meet specification but the product fails functionally, additional calibration should not be assumed to solve it.

For adhesion-related issues, the relevant interfaces may include the substrate, any approved pretreatment or primer, the ink film, and subsequent coating or adhesive layers. Experimentation must remain within validated press and application limits; internal service parameters or control-system settings do not become DOE factors merely because they exist. Across all three technologies, select the factors from the proposed failure mechanism, not from the convenience of the control panel.

10 | The Optimization Trap: When One Fix Creates Another Failure

A common experimental mistake is optimizing a single response while ignoring the rest of the product specification. If a change improves ink adhesion, does that mean the process is optimal? Not necessarily. It may reduce foldability, change gloss, increase scuff sensitivity, compromise heat resistance, or interfere with subsequent lamination.

Similarly, increasing ink density could improve one color measurement while affecting drying performance or fine detail. A successful investigation must therefore consider multiple response variables.

These should include the primary failure metric that triggered the investigation, secondary quality metrics that the corrective change might affect, and process-stability metrics that describe performance across time, lots, and representative operating conditions.

The strongest outcome is usually not a single operating point that best satisfies one metric. It is an acceptable process window within which the important functional, visual, and production requirements can be met together. A solution that works only under extremely narrow conditions may be too fragile for routine industrial production.

11 | How Do We Know We Have Actually Found the Root Cause?

The central question at the end of a trial is not merely whether quality improved. It is whether the evidence supports the proposed failure mechanism.

A credible causal conclusion rests on several kinds of agreement. The cause should explain the defect's form, location, and operating conditions. A controlled change in the proposed cause should alter the response in a predicted direction. Where safe, approved, and feasible, returning to the original condition may help distinguish causation from coincidental improvement. Finally, the corrective action must continue to perform under representative production conditions.

Deliberately recreating a failure is not mandatory when doing so could create safety risks, material damage, or unacceptable losses. In such cases, laboratory evidence, indirect verification, or other carefully planned tests may be more appropriate.

The investigation should also distinguish between the occurrence root cause, which explains why the defect was created, and the escape root cause, which explains why the defect was not detected or contained before the product progressed. If the initiating mechanism has been corrected but the quality-control system still cannot detect a recurrence, the risk has not been fully addressed.

Corrective action should therefore address both process prevention and the adequacy of inspection, monitoring, or release controls.

12 | From One Investigation to Organizational Knowledge

Every effective failure investigation can become a valuable organizational asset. If the findings remain in one specialist's memory or scattered email threads, however, the team may have to repeat much of the work when a similar defect appears.

Document the original problem statement, product construction, relevant process stages, evidence, rejected hypotheses, trial matrix, measurements, analysis, corrective action, and verification criteria. The objective is to preserve why the decision was made, not only what was changed.

Unsuccessful experiments belong in that record. A trial that disproves a plausible hypothesis is useful because it reduces uncertainty and prevents the same unproductive experiment from being repeated in the future.

Translate verified learning into operating procedures, process-control plans, maintenance requirements, quality specifications, and product-development guidance. If the failure arises from an interaction among product design, materials, and manufacturing, the findings must reach development and application engineers as well as production personnel. In this way, a local troubleshooting exercise becomes a systemic improvement.

13 | Ten Rules for Failure Investigation and Experimental Design

1. Do not identify a cause before you have defined the failure.
2. Separate facts, interpretations, and hypotheses.
3. Investigate the full product and process chain, not only the station where the defect became visible.
4. Examine where the failure does not occur; those differences may be decisive.
5. Require a credible engineering mechanism for each leading hypothesis.
6. Prefer low-risk, lower-cost tests that can falsify a hypothesis before committing to production trials.
7. Do not select an experimental factor simply because it is easy to adjust.
8. Investigate interactions, especially when individual factors appear acceptable.
9. Measure side effects of the proposed fix, not only improvement in the primary failure metric.
10. Do not declare the problem solved until the improvement is verified under representative production conditions.

Conclusion | Engineering Investigation Begins with Doubt

When a product or production process fails, pressure for a quick solution can encourage a long series of adjustments. Some may improve the immediate result, but they do not necessarily reveal why the failure occurred.

The difference between trial-and-error and engineering investigation is not the number of tests or the sophistication of the instruments. It lies in how the problem is defined, how hypotheses are filtered, and how each experiment is designed to distinguish among competing explanations.

An experienced specialist does not stop at asking what can be changed. The deeper question is what each possible result would teach about the physical, chemical, or mechanical mechanism under investigation.

Sometimes the answer lies exactly where the team first looked. Sometimes it is found at the interface between two operations, in an overlooked material characteristic, in a condition of use that was never simulated, or in an interaction between factors that are each acceptable on their own.

Finding those possibilities requires professional knowledge, engineering curiosity, accurate measurement, and experimental discipline. Thinking outside the box does not mean abandoning established knowledge or procedures. It means letting evidence redirect the investigation, even when it challenges the accepted explanation.

The best experiment is not necessarily the largest, the most expensive, or the most complex. It is the one that reduces uncertainty and brings us closer to understanding the mechanism of failure.

"Do not design an experiment to prove that you are right. Design one that could prove you wrong."

Only when we are willing to challenge our assumptions can we begin to discover what we did not see before.

Professional References

1. NIST/SEMATECH, e-Handbook of Statistical Methods. Experimental design, screening, factorial designs, randomization, blocking, and statistical interpretation. https://www.itl.nist.gov/div898/handbook/
2. American Society for Quality (ASQ), Eight Disciplines (8D). Problem definition, containment, root-cause investigation, corrective action, and prevention. https://asq.org/quality-resources/eight-disciplines-8d
3. American Society for Quality (ASQ), Five Whys and Five Hows. Structured questioning for cause investigation and solution development. https://asq.org/quality-resources/five-whys
4. NIST/SEMATECH, Choosing an Experimental Design. Matching the experimental design to the number of factors and investigation objective. https://www.itl.nist.gov/div898/handbook/pri/section3/pri33.htm
5. Lim, W., Kim, C. & Lee, Y. (2008), Studies on the Foldability of Coated Board: Influence of Operating Conditions in Creasing and Folding. Research on paperboard creasing, folding behavior, and crack formation.
6. Kim, C.-K., Lim, W.-S. & Lee, Y. (2010), Studies on the Foldability of Coated Paperboard (I): Influence of Latex on Foldability During Creasing/Folding Coated Paperboard. Journal of Industrial and Engineering Chemistry. Study of coated-board structure and foldability. https://doi.org/10.1016/j.jiec.2010.05.001