Gains:
- Ability to configure casting, welding and additive manufacturing parameters with AI and evaluate the risk of defects in advance
- Ability to prepare outputs such as WPS/PQR draft, casting feeder design and process window with AI support
- Ability to confirm AI's process recommendations through standard, simulation and experimental validation
How a metal part is shaped is just as important as what material it is made of. Casting is shaping by filling molten metal into a mold and solidifying it; welding, joining two metal parts by heat and/or pressure; Additive manufacturing (AM or 3D metal printing) is a method of building parts by melting powder or wire layer by layer. All three have a common truth: process parameters (temperature, speed, heat input, cooling) directly determine defect formation and mechanical properties. AI speeds up establishing a process window, pre-assessing the risk of defects, and drafting documents such as WPS/casting design. But the physics of solidification and heat flow are complex; Every parameter suggestion cannot go into production without being verified by simulation, trial and non-destructive testing.
The common language of three processes: process window and defects
The process window is the range of parameters that yield acceptable quality; The defect occurs when going outside the window. Every process has typical defects:
- Casting: Porosity (gas or shrinkage gap), hot tearing, cold shut, inclusion, microshrinkage. Feeder/riser design is critical for feeding liquid to the metal as it contracts as it solidifies.
- Weld: Porosity, insufficient penetration (lack of penetration), undercut, crack (cold/hot), HAZ (Heat Affected Zone: the area around the weld whose microstructure changes due to heat) hardening. Heat input determines the properties of this region.
- Additive manufacturing: Lack of fusion, residual porosity, residual stress and distortion, anisotropy (direction dependent properties). Laser power, scanning speed and layer thickness determine the energy density.
AI is powerful at systematically explaining which parameter triggers which defect and suggesting an initial window. But the real window; specific to the machine, powder/wire batch and geometry.
Caution: Acceptance of a safety-critical weld or dump is never given by the AI's parameter recommendation. WPS is verified by procedural qualification tests (PQR) required by a standard (e.g. ASME IX, ISO 15614); Final acceptance is made with the approval of the authorized welding/casting engineer and inspection.
Step by step: AI-supported process parameter setup
- Define process and material: Cast alloy / base metal + filler / AM powder, thickness, geometry.
- Give goals and constraints: Acceptance criteria (defect class), mechanical target, cycle time.
- Request startup window: Ask the AI for its typical parameter range and what defect each parameter affects.
- List risks: Which defects stand out in these parameters? How to prevent?
- Simulate/try: Solidification simulation for casting/AM (e.g. MAGMA, Simufact), trial coupon for welding.
- Verify by examination: NDT (radiography, ultrasonic), macrosection, mechanical testing; accepted according to standard.
Process comparison and defect-parameter relationship
process
Critical parameter
chief flaw
Verification method
AI role
Sand/die casting
Casting temperature, feeder
Shrinkage porosity
Radiography, macro section
Feeder/window draft
arc welding
Heat input, preheating
HAZ crack, pore
RT/UT, macro, hardness
WPS draft, risk list
Powder bed AM (L-PBF)
Laser power, scanning speed
Insufficient melting, porosity
CT, density, microstructure
Energy density window
The heat input is calculated roughly as (voltage × current / welding speed) in the weld and determines the HAZ width, cooling rate, and therefore hardness and crack risk. AI describes this relationship and suggests a starting value; The exact value is verified by PQR.
three mini cases
Case 1 — Nutrient deficiency. A foundry experiences recurring internal shrinkage porosity in a thick-section bronze part. They describe the geometry and alloy to the AI and ask about possible causes; AI suggests that the feeder may be insufficient to feed the final solidified zone, and that they review the "directional solidification" principle and feeder size/location. The team runs a solidification simulation, sees that the hot spot is far from the feeder, enlarges and positions the feeder; On radiography, porosity decreases from 4% to 0.5%. AI gave hypothesis, simulation and NDT confirmed.
Case 2 — High hardness HAZ crack. An assembly team welds a high-carbon steel without preheating; Cold crack occurs in HAZ. When they ask AI about the situation, he explains that high carbon equivalent (CE: combined value indicating the alloy's tendency to crack) creates hard and brittle HAZ, and that preheating and controlled cooling will reduce hydrogen-induced cracking. The team calculates the CE, determines the appropriate preheat temperature, and tests it on a coupon and verifies it with hardness and macrosection. AI gave the mechanism and direction, experiment gave the number.
Case 3 — AM energy density window. An AM team sees insufficient fusion pores in a titanium part. It asks the AI for an energy density window that includes laser power, scan speed, and hatch distance, and the effect of each parameter on pore type. AI explains that too low an energy will create insufficient melting, too high will create a keyhole, and suggests a starting range. The team prints a parameter sweep coupon, measures density and microstructure, and confirms the optimal window with CT. AI narrowed the window, made the experiment precise.
Copiable prompt templates
PROCESS WINDOW TEMPLATE"Role: [casting/welding/AM] you are the process assistant. Material: [...]. Geometry/thickness: [...]. Target quality: [defect class]. Offer me an INITIAL parameter window and explain which defect each parameter increases or decreases. Emphasize that this window is machine, batch and geometry specific and should be verified with a simulation/trial coupon. No absolute guarantees."
DEFECT ROOT CAUSE TEMPLATE"The following defect reoccurs in my [process] piece: [definition, location, frequency %]. Produce a systematic list of possible root causes (material, parameter, geometry, equipment). For each cause: supporting evidence + which is confirmed by control/experimentation. Don't make a definitive diagnosis; give an investigation roadmap."
WPS DRAFT TEMPLATE"Role: You are an assistant welding engineer. Base metal: [...]. Thickness: [...]. Welding method: [e.g. GMAW]. Position: [...]. Get me a WPS DRAFT skeleton (preheat, passes, heat input range, shielding gas). This is a DRAFT; cannot be used without being PQR qualified to ASME IX / ISO 15614 — write this clearly. Carbon equivalent and Remind me of the need to preheat."
FEEDER / SOLIDIFICATION TEMPLATE "My casting: [alloy, geometry, thickest section]. Explain directional solidification and riser logic: where is the hot spot, where should the feeder be placed, how is its size determined? How do I reduce the risk of porosity? State that I need to confirm the final decision with solidification simulation and radiography."
Weak prompt / Strong prompt
WEAK PROMPT: "How do I weld this steel?"
STRONG PROMPT:"Role: You are an assistant welding engineer. Base metal: high carbon alloy steel (CE ~0.6), thickness 20 mm, filler and position [...]. Explain to me the need for preheating to prevent cold cracking in the HAZ, the heat input range and the inter-pass temperature logic. Explain the effect of carbon equivalent on the risk of cracking. State that this is a WPSTAL and cannot be used without qualification with PQR and macro/hardness testing. Exact parameter guarantee don't give it."
Poor prompt gives a general response without being aware of material, thickness and risk of cracks. The powerful prompt introduces the carbon equivalent, thickness and crack mechanism, stipulating that the output is a draft and must be verified with PQR.
Common mistakes
- Mistaking the parameter window to be universal rather than machine/batch/geometry specific.
- Bypassing the need for preheating in high carbon equivalent steel.
- Finalizing the casting feeder design without solidification simulation.
- Changing a single parameter in AM and forgetting the other dependent parameters (energy density whole).
- Mistaking the WPS draft for a production procedure without PQR qualification.
- NDT on safety-critical part and consider AI evaluation as final acceptance without authorized approval.
In summary
In casting, welding and additive manufacturing, AI accelerates process windowing, describing defect-parameter relationship, generating root cause list and WPS/feeder sketch. However, the physics of solidification and heat flow are machine, batch, and geometry specific; Each parameter recommendation must be verified by simulation, trial coupon and non-destructive testing, and given with safety-critical acceptance standard (ASME IX, ISO 15614 etc.) and authorized approval. AI allows you to foresee the risk, experimentation and inspection make the decision.
Application task
Choose a process (casting, welding or AM) and a realistic material/geometry. Request the initial parameter window and defect-parameter relationship from the AI with the “PROCESS WINDOW” template. Then, create a recurring defect scenario and create a systematic cause list and verification plan with the "DEFECT ROOT CAUSE" template. Finally, write in a paragraph which standard and which NDT method you will use to accept this part, and the role of authorized approval.
checklist
- [ ] I defined the process, material, geometry and target quality criteria.
- [ ] I set up the initial parameter window and the defect-parameter relationship.
- [ ] I took into account critical factors such as carbon equivalent / preheat / nutrient.
- [ ] I have developed a systematic root cause and verification plan for the defect.
- [ ] I verified the parameters with the simulation/trial coupon.
- [ ] I subject final acceptance to standard NDT and authorized engineer approval.