Forensic Finite Element Analysis

What Is Forensic FEA?

Forensic FEA uses numerical analysis to test hypotheses about how a failure could have occurred. It is fundamentally different from design FEA — different in purpose, different in approach and different in the conclusions it supports. Design FEA asks whether a design will survive the expected loads. Forensic FEA asks what combination of physical conditions could reproduce the observed failure evidence. Understanding this distinction is essential for credible failure investigation.

Forensic FEA is hypothesis testing, not retrospective contour plotting.

Design Analysis vs Forensic Analysis

The two types of analysis answer different questions, use different inputs and produce different outputs. Confusing them is one of the most common errors in failure investigation.

DESIGN ANALYSIS:
Expected load → FE model → predicted response → margin

FORENSIC ANALYSIS:
Observed damage → candidate load/mechanism → FE model → predicted failure signature → compare with physical evidence → accept/reject hypothesis

Model Inputs for Forensic FEA

The credibility of forensic FEA depends on using the right inputs. Idealised design assumptions are not appropriate — the analysis should use the actual conditions that existed at the time of failure. This requires the investigator to gather the actual geometry, material, loads and boundary conditions from the physical evidence and the operating history.

• As-built geometry — measured from the failed hardware, including deviations from drawing
• Actual material — tested or certified properties for the specific batch, not handbook values
• Measured damage — the observed crack, deformation or defect included in the model
• Actual fixture and interface — the real boundary conditions, not the idealised supports
• Operational load history — reconstructed from service data, not the design load cases
• Preload — the actual assembly preload, if known or estimated, not the specified torque
• Contact conditions — the real contact state, including wear, fretting and misalignment
• Temperature — the operating temperature, which affects material properties and thermal stress

Comparing Predictions With Physical Evidence

The core of forensic FEA is comparison: does the predicted failure signature match the observed physical evidence? The comparison should examine not only the failure location but the direction, the mechanism and the pattern. A model that reproduces the failure location but requires an implausible load is not necessarily supporting the correct hypothesis. A model that reproduces the failure direction but not the location needs refinement — or the hypothesis needs revision.

Sensitivity Analysis

Forensic FEA is inherently uncertain — the exact loads, material properties and boundary conditions at the time of failure are never perfectly known. Sensitivity analysis examines how the predicted failure signature changes as the inputs vary. If the predicted failure location is insensitive to input variations — it remains at the same location regardless of load magnitude or friction coefficient — then the prediction is robust. If the failure location moves to different regions as inputs change, then the prediction is fragile — the hypothesis is model-dependent and requires stronger evidence to support it.

ANALYSIS CONSIDERATION: If the predicted failure location moves to different regions as the input parameters vary, the hypothesis is fragile — it depends on specific conditions that must be independently verified, not assumed.

Competing Models

Forensic FEA should test multiple hypotheses — not just one. Each candidate failure mechanism may correspond to a different model configuration: a different load direction, a different boundary condition, a different defect location. The investigator runs each model and compares the predicted failure signature with the observed evidence. The hypothesis whose prediction best matches the evidence gains support. The hypothesis whose prediction contradicts the evidence is weakened. This is the essence of hypothesis testing through FEA.

1. Form candidate failure hypotheses (H1, H2, H3 ...)
2. For each hypothesis, build an FE model with the corresponding conditions
3. Run each model and extract the predicted failure signature
4. Compare each prediction with the observed physical evidence
5. Identify which hypothesis best explains the evidence
6. Identify which hypotheses are contradicted by the evidence
7. Converge on the supported hypothesis

Inverse Reasoning

Forensic FEA involves a form of inverse reasoning: given the observed failure, what conditions could have produced it? This is the opposite of design FEA, which proceeds forward from given loads to predicted response. Inverse reasoning is powerful but dangerous. The temptation is to adjust the model inputs until the prediction matches the observed failure — but this is model tuning, not hypothesis testing. Parameters should be changed only when physical evidence supports the change, not merely to reproduce the expected result.

DO NOT MODIFY THE MODEL UNTIL IT PRODUCES THE FAILURE YOU EXPECT. CHANGE PARAMETERS ONLY WHEN PHYSICAL EVIDENCE SUPPORTS THE CHANGE. Retrospective contour plotting is not forensic analysis.

The Hierarchy of Forensic FEA Evidence

Not all FEA results carry the same weight in a forensic investigation. A prediction that is robust — insensitive to input uncertainty — and matches the evidence is strong support. A prediction that requires a specific, unverifiable input to match the evidence is weak support. A prediction that contradicts the evidence is strong evidence against the hypothesis. The investigator should weight the FEA results accordingly and not treat all model outputs as equally credible.

• Strong support — robust prediction matching evidence, insensitive to input uncertainty
• Moderate support — prediction matches evidence but depends on assumptions that are plausible but unverified
• Weak support — prediction matches evidence only with specific, unverified input values
• Contradiction — prediction does not match evidence regardless of input variations

Common Mistakes in Forensic FEA

• Using nominal design geometry instead of as-built geometry — the deviation may be central to the failure
• Using design-allowable material properties instead of actual material properties — the actual batch may differ
• Using design load cases instead of reconstructed load history — the actual loads may differ from design
• Adjusting model inputs until the prediction matches the failure — this is model tuning, not hypothesis testing
• Treating a stress contour as a failure prediction — stress must be compared to a failure criterion, not just reported
• Using a numerical stress singularity as a failure location — singularities are mathematical artefacts
• Running a single model and declaring the hypothesis confirmed — competing hypotheses must be tested

Key Takeaways

• Forensic FEA tests hypotheses about how a failure could have occurred — it does not confirm a pre-existing belief
• Use as-built geometry, actual material, reconstructed loads and real boundary conditions — not design assumptions
• Compare the predicted failure signature — location, direction, mechanism, pattern — with the observed evidence
• Sensitivity analysis reveals whether the prediction is robust or fragile to input uncertainty
• Test multiple competing hypotheses — the one that best explains the evidence is supported
• Do not adjust model inputs to reproduce the expected failure — change parameters only with physical evidence
• Weight FEA results by robustness — a fragile prediction matching the evidence is weak support