Stronger Thigh Bones: Evolutionary Structural Optimization (ESO) — 2% Removal or Not?

This guide delivers a verify-before-you-commit check for ESO on the Load-Bearing Femur.

It compares ESO against bidirectional approaches using like-for-like totals and terms.

How It Works

At its core, Evolutionary Structural Optimization (ESO) is a reductive process built on finite element analysis. Rather than adding material to a structure, the method begins with a full design domain and removes elements that are deemed less effective under the applied loads and boundary conditions. As stated by Huang and Xie (2010) in the Brasil source, a defining trait of ESO is that it does not reintroduce material after its removal, which separates it from bidirectional forms such as BESO. For a load-bearing femur, this one-way action means every removal is permanent, so before committing to any iteration, the live model must have a complete set of loads and constraints, as the entire evolution depends on them.

To apply this mechanism correctly, three key terms must be verified against the exact variant being run. First, the design domain is the total volume that is allowed to exist, and for anatomical work this must be checked against the intended envelope. Second, the element is the discrete unit of that mesh, so the mesh resolution must be confirmed to be intact before removals begin. Third, the elimination criterion is the rule that judges which elements are removed at each step.

This criterion is not fixed across the family, as noted by the academia.edu source on Smoothing Evolutionary Structural Optimization (SESO), which was extended to solve 2D elastic problems with constraints of displacements or natural frequencies. Before any removal begins, check which constraint and which elimination rule the exact variant in use applies — do not import a criterion from one variant into another.

The mechanism itself is iterative. After solving the finite element model, elements that meet the elimination criterion are deleted, the model is re-solved, and the cycle repeats. Because there is no return of material, the state between steps must be tracked, not just the final form. A practical check is to confirm mesh connectivity remains valid after each removal, since a broken mesh can invalidate the next solve without changing the criterion itself. This iterative loop is the action that allows the load path to become more apparent over time.

As noted by Parametric House, ESO was developed specifically for engineering purposes using finite element analysis as a framework, leading to multiple incarnations. With that in mind, the only thing this section establishes is the reductive mechanism and those key terms. The validity of any result from this process rests on confirming both the "no reintroduction" behavior and the exact definitions for the model being run, before moving to any further evaluation.

Key Factors to Consider

For the purpose of this guide, this section alone lists the top decision criteria and numbers. The first criterion to verify is domain completeness. Before any element removal is considered, the finite element model must represent the full load-bearing femur design domain for the exact load case you intend to test. A trimmed domain can change which elements appear less effective, so compare the model extent against your stated boundary conditions line by line. If the run report does not show the full domain used, that is a missing item to resolve before committing.

The second criterion is removal control. For a strictly reductive run, you must confirm the rejection ratio used per iteration is recorded in the iteration log. As referenced by the Semnan modelling source, bidirectional approaches do make use of a rejection ratio, but for ESO the value itself is not universal. Do not assume a default. If the log states no rejection ratio, treat that as an unverified number and do not proceed to apply any removal threshold until it is written down for that exact run.

The third criterion is convergence verification. Since, as cited by scielo.br from Huang and Xie, ESO does not reintroduce material after removal, the total volume fraction must only decrease by the amount removed with no additions. For this reason, you must compute the difference between total volume fraction of consecutive iterations and compare it to the amount removed in that step. Only if those two figures match for the reported step can you consider the reduction accounted for.

As for numbers that matter, limit your checklist to three recorded values only: the stated rejection ratio, the target volume fraction, and the iteration number at which the volume fraction change becomes stable. None of these should be pulled from general tables. If any one of these three is absent from the run report, the correct action is to request the log, not to substitute a common value. These are the only numbers this section calls out to verify before you commit.

Common Mistakes

One of the first pitfalls is treating element removal as reversible. As noted by Huang and Xie (2010) in the Brasil source, unlike bidirectional techniques such as BESO, ESO does not reintroduce material after its removal. For a load-bearing femur model, this has a concrete consequence: if a cut is made along a primary load path before the full constraint picture is verified, that cut cannot be recovered by the same run. The check is not the shape at that moment, but traceability. Before committing to any cut set, the run log must be able to show exactly which elements were removed, and the decision to remove them must be tied to the constraint used for that run, not to an assumption of recovery.

A second pitfall is mixing variants without verifying their elimination criteria. As outlined by Januszkiewicz and Banachowicz (Parametric House), ESO has led to several incarnations, including Additive ESO, Bi-directional ESO, and Extended ESO. The mistake is to import a setting from one incarnation into another and treat the result as directly comparable. For example, a user may pull an elimination rule from a Bi-directional run and apply it to a pure ESO run, then compare the two cut sets as if the terms were the same. The rule here is strict: compare only within the same incarnation lineage, and verify the stated elimination criterion for that exact incarnation against the active run before committing to it.

A third pitfall is confusing smoothing with the base removal action. The academia.edu reference points to the Smoothing Evolutionary Structural Optimization (SESO) technique being extended to solve two-dimensional elastic problems with constraint of displacements or natural frequencies. The concrete example is this: if a post-smooth mesh is compared to an unsmoothed cut set from an earlier iteration, the totals are not like-for-like. The verify-before-you-commit action is to isolate the variant being tested. If smoothing is applied, it must be treated as its own lineage with its own constraint check, not as a cleaned version of the previous lineage.

The final pitfall is letting visual appeal override constraint compliance. The Parametric House example of optimization made famous by an apple shape is a useful reminder of the organic forms this class of tools can produce. However, for the load-bearing femur, resemblance to a known form is not proof of compliance. Before committing to any candidate, the missing piece must be verified: is there a recorded constraint check for that exact candidate, tied to the incarnation and lineage used? If that check is absent from the run record, the option is not live for verification, no matter how plausible the shape appears.

Insider Tactics

One non-obvious strategy to keep in mind is to treat the first action of removal as a probe, rather than an expansion. Instead of acting on the entire domain at once, isolate a sub-region that appears to carry less of the primary load path and run a limited probe there. The purpose of this probe is not to shape the final form, but to observe how the load redistributes in response to removal. By doing this, you can gather information about spatial material property gradients without committing the full model to an irreversible change. This is a state-based check, not a quantity-based one.

A key timing tip is to only act on a state where the finite element solution has converged. Because ESO's removals are irreversible (see How It Works above), acting on a partially converged field solution carries a different kind of cost than acting on a converged one.

For this reason, before expanding beyond the probe, verify convergence for that exact state. Do not rely on iteration count alone, but on the stability of the field solution itself.

Another check to use during this probe is to keep a snapshot of the element set directly before any expansion. This is not for record keeping, but for a direct comparison between states. Look strictly at the shift of the load path between the snapshot and the post-probe state. If that shift is coherent and contained to the probed sub-region, expansion may be considered conditionally. If the shift becomes diffuse across areas outside of that sub-region, that is a signal to pause, regardless of how much of the probe remains.

Finally, know when to end the probe entirely. The end point should be a change in behavior, not a preset point. When the redistribution becomes erratic within the probed sub-region itself, the probe has yielded its information and should be stopped. This conditional gate of convergence, state comparison, and erratic redistribution is the only timing discipline introduced here, and it is meant to be used before any broader commitment is made. As noted by Januszkiewicz and Banachowicz through the Parametrichouse deep read, the development of ESO has led to multiple incarnations, but the irreversibility remains the constant that makes this timing discipline worth applying.

Comparison

For the load-bearing femur, this section alone compares three reductive paths under like-for-like totals: standard ESO, Bi-directional Evolutionary Structural Optimization (BESO), and Smoothing Evolutionary Structural Optimization (SESO). To verify before committing, the same design domain, the same load set, and the same retained element count must be used for all three. As stated by Huang and Xie (2010) in the Brasil source, unlike bidirectional techniques such as BESO, ESO does not reintroduce material after its removal. That single difference means the element ledger for ESO can only decrease, while BESO must be reconciled by both removals and reintroductions at the exact retained total.

With those ledgers matched, the focus shifts to constraint alignment. The academia.edu source describes SESO as an extension applied to elastic problems with constraint of displacements or natural frequencies. This points to smoothing as its stated focus, not bidirectional exchange. Because of this, SESO cannot be treated as a direct substitute for BESO when the task requires checking removal against reintroduction. The comparison must be made by constraint, not by final shape alone, and any option that uses a different constraint must be disqualified from this like-for-like test.

Under strict like-for-like totals and matching constraints, the option that wins for predicting spatial material property gradients is BESO. The reason is purely comparative: only BESO allows a direct check between removal and reintroduction at the same retained count without conflating that exchange with a smoothing pass. ESO is valid for a one-way path, but its irreversibility prevents the bidirectional reconciliation needed for this gradient prediction. SESO may be useful for smoothing, but with the constraint focus noted from the academia.edu source, it does not satisfy the exchange requirement for this specific comparison.

Before committing to BESO for this task, verify three items against the ledgers: confirm the added and removed counts sum to the change needed to hold the exact retained total, confirm the constraint used matches displacement or natural frequency for the gradient target, and confirm no smoothing pass was counted as exchange. Only after all three match can the side-by-side be valid, and only then does BESO earn the win for this prediction.

What to do next

Frequently Asked Questions

According to Huang and Xie (2010), what separates ESO from BESO?

Unlike BESO, ESO does not reintroduce material after its removal (Huang and Xie, 2010).

What specific constraints was SESO extended to address?

SESO was extended to solve 2D elastic problems with constraint of displacements or natural frequencies.

How is BESO classified in this guide?

BESO is identified as a bidirectional Evolutionary Structural Optimization technique.

For the Load-Bearing Femur, what is the 2% element removal threshold being evaluated to predict?

The guide evaluates when to apply a 2% element removal threshold for the Load-Bearing Femur to predict spatial material property gradients.

What is the consequence of ESO's one-way removal for the Load-Bearing Femur model?

For a load-bearing femur, this one-way action means every removal is permanent, so before committing to any iteration, the live model must have a complet

On what basis does this guide compare ESO against bidirectional approaches?

It compares ESO against bidirectional approaches using like-for-like totals and terms.

Quick answers

Also worth reading: Material Knowledge Prevents Structural Failure: Material Knowledge Prevents Structural Failure · 7 Structural Applications of Antimicrobial Coconut Oil in Building Material Preservation: 7 Structural Applications of Antimicrobial · 7 Essential Analytical Methods for Structural Load Analysis Every Civil Engineer Should Master: 7 Essential Analytical Methods for