Abstract: This paper explores application of artificial intelligence (AI) technology in die-casting mold design and performance prediction, focusing on intelligent structural design, process parameter optimization, defect detection, and quality control. Through an analysis of a large automotive parts manufacturer's case study, it demonstrates how an AI system can improve mold design process through a data-driven approach, increasing success rate of first-time mold trials and significantly reducing scrap rates. Study indicates that a machine learning-based performance prediction model can accurately assess key indicators such as filling time, cooling rate, and shrinkage, thereby achieving effective prediction of mold performance. Results show that application of artificial intelligence technology can not only accelerate product development cycles but also improve production efficiency and product quality, providing strong support for transformation of industry towards intelligent manufacturing.
In context of manufacturing industry's intelligent transformation, design and manufacturing of die-casting molds are facing unprecedented challenges and opportunities. With rapid development of automotive and electronics industries, quality requirements for die-cast parts are becoming increasingly stringent, traditional design methods are struggling to meet demands for high efficiency and high precision. Emergence of artificial intelligence technology has opened up a new avenue for solving this problem. Simultaneously, by combining artificial intelligence algorithms, the entire process of mold structure design and performance prediction is made intelligent, greatly shortening R&D cycle, reducing costs, improving product quality and consistency. In particular, AI-aided design has significant advantages in complex geometries and variable working conditions; for example, AgleMold can automatically design mold structures, reducing repetitive labor. Application of this intelligent design and prediction method not only promotes technological innovation in die-casting industry but also contributes to achieving green manufacturing and sustainable development goals.

Die-casting molds are indispensable process equipment in die-casting production. Their basic structure consists of a fixed mold and a moving mold, sometimes including additional half-molds to accommodate complex part shapes. Fixed mold is fixed to mounting plate of die-casting machine and connected to pressure chamber, while moving mold opens and closes as moving mold mounting plate moves. Main components of mold include sprue, gating system, cavity, core-pulling mechanism, overflow system, temperature control system, ejection mechanism, and moving mold frame, etc. These components work together to ensure smooth flow of molten metal into mold cavity, where it cools, solidifies, and ultimately forms ideal die-cast part. Sprue acts as a bridge connecting pressure chamber and runner; components such as sprue sleeve and runner cone ensure smooth flow of molten metal. Gating system, composed of runner, ingate, and sprue, is main channel for introducing molten alloy into mold cavity. Mold cavity determines geometry of die-cast part and is typically composed of inserts. A core-pulling mechanism is used to complete extraction and insertion of movable mold cores, thereby achieving forming of complex structures. A venting system stores cold metal slag to prevent defects. A temperature control system adjusts mold temperature through cooling water pipes and heating pipes to ensure mold quality. A demolding mechanism ejects mold, making demolding easy. Moving mold frame connects and fixes moving mold parts to ensure overall stability.

Modern mold design faces numerous technical challenges, especially in meeting increasing complexity and high-quality requirements of products. Insufficient mold opening stroke is a common problem, affecting not only production efficiency but also potentially leading to decreased product quality. To address this, designers need to innovate design methods, such as adjusting parting surface angle and position to change demolding direction, shorten required mold opening stroke. Additionally, optimized design of ejector and slide mechanism can achieve a larger core-pulling distance within a limited space, compensating for insufficient mold opening stroke. Multi-stage mold opening design further enhances flexibility of mold design; using a segmented core-pulling method allows complex demolding operations to be completed without increasing mold stroke. Application of mechanical auxiliary equipment such as hydraulic cylinders helps overcome mechanical structural limitations, making core-pulling operations more perfect. Simultaneously, selecting high-strength and high-toughness mold materials can reduce insufficient mold opening stroke caused by mold wear.

Performance prediction becomes particularly important in mold design process. It helps identify potential problems in advance, optimize design solutions, ensure quality and production efficiency of final product. Mold flow analysis, as a key tool for manufacturing design, enables engineers to predict behavior of molten metal or plastic in mold cavity, thereby ensuring optimal quality and efficiency. This sophisticated simulation technology is crucial for identifying potential defects (such as porosity and shrinkage cavities). By utilizing numerical simulation software to analyze effects of parameters such as liquid filling, temperature field, and cooling rate during casting process, accurate identification of internal defects (such as porosity and shrinkage cavities) in casting can be achieved. Based on this, designers can adjust mold design in the early stages of mold design, optimize cooling circuit layout, and even redesign parting surface. In addition, strength and rigidity of mold must be evaluated to ensure that mold will not deform or be damaged in high-pressure and high-temperature environments.

In design process of die casting molds, application of artificial intelligence (AI) technology significantly improves design efficiency and accuracy. By utilizing AI algorithms such as AgleMold, automatic mold structure design can be achieved, reducing manual intervention and labor intensity. This system can automatically generate preliminary mold design schemes based on input product geometry, material, process requirements, and provide designers with multiple alternative solutions. AI-assisted design goes beyond simply creating static mold models. It simulates actual working state of mold during design phase, including flow path of molten metal within mold and temperature field changes during cooling, enabling optimization for potential problems. For example, for complex geometries, AI can quickly propose solutions by learning from similar instances in historical data, thus avoiding trial-and-error methods of traditional design. Furthermore, AI can identify and eliminate redundant parts, simplify mold structures, improve part stability and reliability, and reduce manufacturing costs.

By learning from a large amount of historical production data, AI models can build accurate mathematical models to describe physical phenomena in die-casting process, such as fluidity of molten metal, solidification behavior, and thermal conductivity. Based on these models, AI can predict casting quality under different process parameter settings, guiding designers to adjust key parameters such as pouring speed, pressure, mold temperature to ensure dimensional accuracy and surface quality of final product. More importantly, AI can also monitor mold forming process in real time and make dynamic adjustments. When anomalies are detected, it automatically takes corrective measures to prevent defects. For example, in its white paper on digital die-casting technology, Shichuang Technology mentions that by inputting final parameters and constraints, optimal mold design can be automatically obtained, and simulation results can be seamlessly connected to various production stages to achieve intelligent control.

Traditional methods rely on manual visual inspection or simple measuring tools, which are inefficient and easily affected by subjective factors. With continuous development of artificial intelligence technology, machine vision-based inspection systems are gradually becoming mainstream. These systems use deep learning algorithms such as convolutional neural networks to learn defect features such as cracks, porosity, shrinkage cavities from massive amounts of image data, and can even accurately identify extremely small defects. To solve problem of scarce sample data in industrial scenarios, expert-guided data pre-labeling technology is used, combined with positive and negative sample defect detection algorithms, to build a high-performance detection model using a small amount of real experimental data. Furthermore, AI can be combined with other sensor technologies, such as infrared thermal imagers, to monitor temperature distribution on mold surface and promptly detect potential problems caused by uneven temperature.

Mold performance refers to ability of a mold to stably and efficiently produce products that meet quality requirements under specific working conditions within its expected service life. Key indicators for measuring mold performance include, but are not limited to, dimensional accuracy, surface roughness, hardness, toughness, structural design rationality, service life, and cost-effectiveness. Dimensional accuracy directly affects product assembly and performance, and is generally measured using high-precision measuring tools such as vernier calipers and micrometers. Surface roughness has a significant impact on product appearance quality and demolding performance, can be evaluated visually or with a roughness tester. Hardness and toughness are measured using methods such as hardness testing and material analysis, reflecting fracture resistance of mold material. A reasonable structural design ensures uniform stress distribution during mold operation, reducing deformation and wear. A good cooling system effectively controls internal temperature of mold, preventing deformation due to overheating and reducing service life. Mold life is an important indicator of mold durability; long-term mold performance can be evaluated by recording and analyzing actual number of uses and maintenance. Cost-effectiveness considers both initial investment and long-term operating costs; selecting high-quality molds, although requiring a larger initial investment, can effectively reduce subsequent production costs, improve productivity and product quality.

Artificial intelligence-based mold performance prediction model aims to use machine learning algorithms to extract patterns from large amounts of historical data and construct a mathematical model that can accurately predict mold performance. These models typically employ supervised learning methods. Input features include mold design parameters (such as geometry and material properties), manufacturing process parameters (such as casting speed, pressure, and temperature), environmental factors (such as workshop temperature and humidity). Output is mold's key performance indicators. Commonly used machine learning algorithms include decision tree regression, gradient boosting regression, and random forest regression. These models require less data, are easy to interpret, and are suitable for industrial applications. For example, gradient boosting regression model optimizes loss function by combining multiple weak learners. Starting from an initial approximation F0(x), it progressively adds weak learners hm(x) to correct errors in previous model. Update formula is:

 

Where α is learning rate, and hm(x) is m-th weak learner used to correct error of previous prediction. For random forest regression model, final prediction is made by training multiple decision trees and taking their average. Expression is as follows:

 

Where T represents number of decision trees, and yt(x) is prediction value of t-th tree. Furthermore, to improve model interpretability, techniques such as feature permutation importance and Shapley value can be applied. The former measures importance of each feature by randomly replacing certain features and monitoring changes in model performance, while the latter assigns a fairness value based on marginal contribution of each participant, quantifying contribution of each feature to prediction.

Establishing and validating performance prediction model is an iterative process involving multiple steps such as data preparation, model training, and model evaluation. In data preparation stage, a large amount of data needs to be collected from various stages of mold design, manufacturing, use, and maintenance, including design drawings, processing parameters, operation records, and maintenance logs. To ensure data quality, data removal, missing value removal, and numerical feature normalization are necessary.
Secondly, during model training, dataset is divided, and cross-validation methods such as five-fold cross-validation are used. Each time, 80% of data is used as training set and 20% as test set, repeated five times to obtain more reliable evaluation results. After training, model's performance is evaluated using various metrics, such as mean squared error (MSE), mean absolute error (MAE), and root mean squared error (RMSE). These metrics reflect degree of deviation between model's predicted values and actual values. For example, formula for calculating mean squared error is:

 

where n is number of samples, yi represent actual value and predicted value of i-th sample, respectively. To further verify model's effectiveness, it can be applied to a real-world production environment. By comparing model's predictions with actual production data, model's generalization ability and stability can be tested.

A manufacturer specializes in producing aluminum alloy die-casting parts for high-end car models. Its products are characterized by high precision, high strength, and lightweight. Faced with increasing market demand and stringent quality requirements, traditional experience-based design methods are gradually showing limitations. Therefore, company introduced a machine learning-based AI system to assist mold designers in rapid prototyping, simulation analysis, and performance prediction. After a period of application, system significantly shortened product development cycle, improved success rate of the first trial molding, and reduced scrap rate.

Implementation of AI system consists of three main stages: data collection and preprocessing, model training and verification, online prediction and optimization. First, integrated CAD/CAE software interface enables automatic extraction of multi-source heterogeneous data. Deep neural network methods are used to model different types of die castings, enabling prediction of key performance indicators such as filling time, cooling rate, and shrinkage. Finally, real-time monitoring and feedback control are implemented in actual production environment, allowing engineers to adjust mold structure and process parameters based on prediction results to ensure that product quality meets expected requirements. As shown in Table 1, AI-aided design reduced average filling time from 2.35 seconds to 1.87 seconds, while increasing cooling rate by approximately 16.42%. This not only accelerated production cycle but also improved internal structure of castings and reduced probability of defects. Furthermore, due to more precise control over shrinkage, dimensional accuracy of products was significantly improved, with tolerances reduced to within ±0.05mm.
Table 1. Comparison of Key Performance Aspects of Typical Die-Casting Parts Before and After the Introduction of AI System

As shown in Table 2, after using AI for mold design, success rate of the first trial molding of all types of products improved to varying degrees. In particular, success rate of complex-shaped parts jumped from 68.25% to 84.37%, indicating that AI technology helps overcome problems that traditional design methods struggle to handle, such as runner layout optimization and thermal stress distribution prediction.
Table 2. Success Rate of First Trial Molding for Different Types of Die Castings Using AI System

Study of above cases reveals that AI technology has demonstrated enormous potential in the field of intelligent design and performance prediction of die-casting molds. It can not only accelerate product development but also effectively reduce production costs and improve product quality. However, success of artificial intelligence (AI) applications relies on high-quality data support and interdisciplinary collaboration.
In the future, with maturation and development of more related technologies, AI will play an increasingly important role in manufacturing, driving the entire industry towards intelligent manufacturing.

This study, through practical case studies, validated significant advantages of AI in die-casting mold design and performance prediction, including improving success rate of first-time mold trials, optimizing process parameters, and enhancing quality control. Machine learning-based performance prediction models can accurately assess key indicators and effectively guide mold structure design and manufacturing process optimization. With continuous algorithm advancements and data accumulation, AI technology will further enhance intelligence level of mold design, driving industry towards highly customized and efficient intelligent manufacturing. Future research will focus on developing more complex prediction models, integrating more diverse data sources, exploring integrated application of AI with other advanced technologies such as Internet of Things (IoT) and digital twins to achieve a comprehensive vision of smart factories.