AI for Design Quality & DFM
Casting defects like shrinkage porosity start in the CAD model. See how AI checks draft, wall thickness, and fillets before you commit to tooling.
·
⏱
8 min read
Michelle Ben-David
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
Casting rewards teams that get the geometry right before the tool is cut and punishes those who do not. Uniform walls, proper draft, clean fillets, a simple parting line, and a sensible approach to undercuts decide whether a part casts well, and all of them are visible in the CAD model long before metal is poured.
An AI layer that reads the actual geometry, checks it against casting rules, and explains its reasoning lets every engineer apply senior-level castability judgment on every part, not just the ones a specialist has time to review. The result is fewer surprises at the foundry, lower tooling risk, and casting knowledge that compounds across the team instead of walking out the door.
Casting is one of the few mechanical processes where a design mistake is not a quick edit. Once a mold or die is cut, the geometry is committed in hardened steel, and changing it can mean weeks of lead time and tooling spend that runs from thousands to hundreds of thousands of dollars. The frustrating part is that most casting defects are not really foundry problems. They are geometry problems that were present in the CAD model long before anyone poured metal.
Shrinkage porosity, short fills, warping, and cracked tooling can almost always be traced back to a handful of design decisions: wall thickness that is not uniform, missing draft, sharp internal corners, an awkward parting line, or undercuts that nobody flagged. The rules to avoid them are well known, yet they live in design guides, supplier feedback, and the memory of a few senior engineers. This article looks at the casting rules that decide yield, why manual reviews keep missing them, and how an AI layer can check castability while the part is still on screen.
Why Casting Punishes Late Design Changes
In machined or printed parts, a manufacturability issue usually costs you a revision. In casting, it can cost you the tool. The mold encodes the part geometry, the draft, the gating, and the parting line all at once, so a change to any of those after the tool is cut often means reworking or replacing expensive hardware. That is why design for manufacturability matters more in casting than in almost any other process, and why catching problems early has an outsized payoff.
The economics make this even sharper. A widely cited principle in design for manufacture and assembly research is that the design phase locks in roughly 70 to 80 percent of a product's total cost, even though most of that cost is not spent until production. For castings, the design phase also locks in yield. A part that looks finished in CAD can still be a poor casting if its sections cool unevenly or the metal cannot fill cleanly. Treating castability as an afterthought means discovering these issues at first article inspection, when the cheapest fixes are already off the table. The same logic that drives instant design for manufacturability feedback during CAD work applies with extra force here.
IN PRACTICE
What Engineers Are Saying
"Leo found a nature-inspired solution, a concept we wouldn't have thought of, that let us use standard, off-the-shelf parts. No custom manufacturing. No dedicated engineer. We saved around $400 per system."
Chen, Team Lead at ZutaCore
The Casting Geometry Rules That Decide Yield
Most castability problems come back to a short list of geometry rules. None of them are exotic, but they interact, and missing one can undo the others. The most important ones to check on every cast part are:
1. Uniform wall thickness. Molten metal needs to flow smoothly and cool at a consistent rate. Thick and thin sections cool at different speeds, which creates hot spots, shrinkage porosity, and distortion. Aim for walls that are as uniform as the function allows, and make any transition between sections gradual rather than abrupt.
2. Draft on vertical faces. Every surface parallel to the pull direction needs a taper so the part releases from the mold. A common starting point is one to three degrees, with more draft required as a wall gets deeper or carries texture. Zero draft is the single most common reason a part will not eject cleanly.
3. Fillets and radii instead of sharp corners. Sharp internal corners create stress risers in both the part and the die steel, and they disrupt metal flow. Blending corners with generous fillets improves fill, reduces cracking, and extends tool life.
4. A clean parting line. The parting line is where the two halves of the mold meet. Keeping it as straight and simple as possible reduces flash, simplifies tooling, and avoids trapped geometry.
5. Few or no undercuts. Undercuts cannot be pulled straight from a mold. They force retractable side cores or secondary machining, both of which add tooling complexity and cost. Where a casting feeds straight into machined features, leaving the right machining stock, often a small amount on bores and critical faces, keeps tolerances achievable without overbuilding the part.
Why Manual Casting DFM Reviews Miss Problems
If the rules are this well established, why do bad castings still reach the foundry? The honest answer is that casting knowledge is fragmented. A senior engineer who has shipped a hundred die cast housings can spot a thin boss or a draft problem at a glance, but that judgment lives in their head, not in the CAD system. When they are busy, on another project, or retired, the review quality drops with them. This is the same tribal knowledge gap that quietly slows down so many engineering teams.
Manual reviews also struggle with scale and consistency. Casting rules shift with process and alloy: sand casting, investment casting, and high pressure die casting each have different draft, wall, and tolerance expectations. A checklist that is correct for one may be wrong for another. Reviewers are checking dozens of features by eye, often late in the schedule, under deadline pressure, and against guidelines stored in PDFs, supplier emails, and standards documents that are rarely open at the same time. It is easy to catch the obvious thin wall and miss the gradual section change three features away that will actually cause the shrink. Late-stage manual checks are also how teams end up finding CAD design mistakes only after they reach manufacturing.
How an AI Layer Checks Castability in the CAD Model
This is where an AI layer changes the review. Leo AI reads native CAD geometry directly, including SolidWorks, STEP, IGES, CATIA, Onshape, and Inventor files, so it works on the actual part rather than a screenshot or a slow report. Instead of waiting for a senior reviewer, an engineer can ask whether a part is ready to cast and get specific, located feedback: a boss that is thicker than its surrounding wall, a face with no draft in the pull direction, a sharp internal corner that should be a fillet, or an undercut that will need a side core.
What makes this useful rather than noisy is that Leo is trained on more than a million pages of engineering standards and cites the rule or source behind each recommendation, so the engineer sees why a feature is flagged, not just that it is. That directly serves one of casting's biggest needs, flagging mistakes before they reach manufacturing, and it captures the casting judgment that used to live only with senior staff. Leo is an intelligence layer that sits on top of your existing data, not a replacement for it; integrations are available for SolidWorks PDM, CATIA and ENOVIA, Onshape, Inventor and Vault, PTC Windchill, Siemens Teamcenter, and Arena. Because it understands cost drivers too, the same review connects naturally to should-cost estimation early in mechanical design, so a draft or wall change can be weighed against its effect on tooling and yield.
Making Castability Part of the Daily Workflow
The goal is not a single gate at the end of design but continuous feedback while the part takes shape. When castability checks run during modeling, draft, wall uniformity, and fillets get fixed when they are cheap to fix, and the design review becomes a confirmation rather than a rescue. The same approach has already proven itself in adjacent processes, from AI-assisted design for injection molding to sheet metal DFM checks for bend radius and K-factor.
Castability also benefits from memory. Many casting problems are repeats of issues a team already solved on an earlier part, and an AI layer that has access to the existing vault can point an engineer toward a proven cast component rather than a fresh design that reintroduces an old defect. That reuse keeps tooling counts down and quality up, and it turns each resolved casting issue into shared knowledge instead of a lesson one engineer learns alone.
FAQ
North American Die Casting Association, "NADCA Product Specification Standards for Die Castings," 2015
Boothroyd, Dewhurst and Knight, "Product Design for Manufacture and Assembly," 2011
Catch Casting Defects in CAD
Find draft, wall, and porosity risks before you commit to tooling.
Leo AI reads your CAD geometry, flags castability risks against manufacturing rules, and cites the standard behind each one, so problems surface in design, not at the foundry.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
