AI-Driven Mold Design Transforms Aerospace Manufacturing
In a significant leap for industrial automation, researchers in China have successfully integrated artificial intelligence into the design of deep-drawing dies for sheet metal components—critical tools in aerospace manufacturing. This innovation, detailed in a recent study, demonstrates how intelligent algorithms can replicate expert human reasoning to automate complex engineering decisions, dramatically accelerating mold development while preserving precision and structural integrity.
The breakthrough centers on a semi-automated design system built on CATIA’s secondary development platform, tailored specifically for half-tube deep-drawing molds used in aircraft production. Unlike conventional approaches that rely heavily on manual input and iterative adjustments, the new system leverages rule-based logic and embedded engineering knowledge to autonomously determine key structural parameters—from punch dimensions and cavity geometry to guide plate selection and standard part placement.
At the heart of this advancement lies a shift from parameter-driven modeling to intelligence-driven decision-making. While parametric design has long enabled rapid iteration for geometrically similar parts, it falters when structural variations—dictated by part size, machine constraints, or safety margins—introduce uncertainty. Traditional workflows force engineers to manually resolve these ambiguities, slowing down the design cycle and introducing inconsistency. The new approach treats these uncertainties not as obstacles but as inputs for an expert-like inference engine.
For instance, when determining the layout of ejector pin holes on a punch mounting plate—a task that must account for press table specifications, part geometry, and minimum clearance requirements—the system automatically calculates an optimal grid. It begins by expanding the forming region’s bounding box by one hole diameter in each direction, then computes the number of holes along each axis using integer division based on fixed hole spacing. Crucially, it then filters out any holes that violate safety constraints: those too close to the active forming zone or within the punch’s functional area. This multi-step reasoning mimics the cognitive process of an experienced mold designer but executes it in seconds.
Similarly, the system intelligently configures punch fastening. For half-tube parts with an inner diameter exceeding 68 millimeters, it automatically switches from a single-row bolt pattern to dual bolts positioned at the center of the end fillets—reflecting a real-world engineering best practice encoded as a conditional rule. Such context-sensitive logic ensures that the output isn’t just geometrically correct but also manufacturable and robust.
The implementation, developed using C++ and CATIA’s CAA (Component Application Architecture) framework within Visual Studio, integrates seamlessly into the aerospace industry’s Model-Based Definition (MBD) workflow. Users initiate the process by entering four key parameters: inner radius, sheet thickness, bend radius, and angular span of the half-tube part. Within moments, the software generates a complete 3D mold assembly—including die, punch, blank holder, mounting plate, and guide components—fully compliant with MBD structural tree standards. A single click then triggers automatic import and constraint-based assembly of standard parts such as guide plates, lifting rings, screws, and dowel pins, eliminating hours of manual CAD work.
This end-to-end automation represents more than a productivity gain; it signals a paradigm shift in digital manufacturing. By embedding domain expertise into software logic, the system effectively democratizes high-level engineering judgment. Junior designers can produce molds that meet the same quality benchmarks as those crafted by seasoned experts, reducing reliance on scarce human capital and minimizing variability across projects.
The implications extend beyond aerospace. Deep-drawing processes are ubiquitous in automotive, appliance, and electronics manufacturing. Any industry that relies on repetitive yet nuanced mold design stands to benefit from this intelligence-layered approach. Moreover, the architecture is inherently extensible: additional rules and decision trees can be incorporated to handle new part families or updated manufacturing standards without rewriting the core engine.
Critically, the system aligns with modern principles of knowledge-based engineering (KBE), where design is treated not as a sequence of geometric operations but as a knowledge-intensive problem-solving activity. Earlier attempts at mold automation often stalled at the “geometry-only” stage, failing to capture the tacit knowledge that guides real-world decisions. This work bridges that gap by formalizing heuristic rules—such as “maintain a minimum 35 mm margin between mold edge and ejector holes” or “round closed height to nearest multiple of five”—into executable logic.
The research team, led by Gu Xinhang and Professor Han Zhiren from Shenyang Aerospace University, collaborated closely with AVIC Xi’an Aircraft Industrial Group, ensuring the solution was grounded in actual production constraints. Their prototype has already demonstrated feasibility in industrial settings, cutting design time for standard half-tube molds from days to minutes while maintaining full compatibility with existing CNC machining and quality control workflows.
From a strategic standpoint, this development reinforces China’s broader push toward intelligent manufacturing under initiatives like “Made in China 2025.” By embedding AI not in flashy consumer applications but in the unglamorous yet vital layer of industrial tooling, the project exemplifies a pragmatic path to technological sovereignty. High-performance molds are force multipliers in advanced manufacturing; accelerating their design cycle directly enhances national capacity in sectors where speed-to-market and precision are paramount.
Furthermore, the methodology sidesteps the data hunger and opacity of deep learning models. Instead of training neural networks on thousands of past designs—a challenge in domains with limited historical digital data—the team opted for symbolic AI: transparent, interpretable, and auditable rule systems. This choice enhances trust among engineers, who can inspect and validate each decision path, and facilitates regulatory compliance in safety-critical industries like aviation.
Looking ahead, the framework could evolve to incorporate real-time feedback from manufacturing floors. Imagine a system that not only designs a mold but also simulates its performance in virtual forming trials, then iteratively refines its geometry based on predicted springback or thinning. Such closed-loop intelligence would blur the line between design and process optimization, ushering in truly adaptive manufacturing systems.
For now, the immediate impact is clear: a proven, scalable method to inject intelligence into one of manufacturing’s most time-consuming and expertise-dependent tasks. As global competition intensifies in high-value industrial sectors, the ability to compress design cycles without sacrificing quality becomes a decisive advantage. This work demonstrates that AI’s greatest value may not lie in replacing humans, but in codifying and scaling their hard-won expertise.
In an era where digital thread and model-based enterprise are becoming industry imperatives, tools that seamlessly weave knowledge into the fabric of CAD environments will define the next frontier of productivity. The intelligent mold design system from Shenyang Aerospace University isn’t just a technical achievement—it’s a blueprint for the future of engineering automation.
Gu Xinhang¹ᵃ, Che Jianzhao²ᵃ, Han Zhiren¹ᵃ,¹ᵇ, Yan Baoqiang²ᵇ
¹ᵃCollege of Aerospace Engineering, ¹ᵇKey Laboratory of Fundamental Science for National Defense of Aeronautical Digital Manufacturing Process, Shenyang Aerospace University, Shenyang 110136, China
²ᵃManufacturing Engineering Department, ²ᵇMould and Forging Factory, AVIC Xi’an Aircraft Industrial Group Co., Ltd., Xi’an 710089, China
Journal of Shenyang Aerospace University, Vol. 38, No. 1, pp. 40–46, 2021
DOI: 10.3969/j.issn.2095-1248.2021.01.007