In recent years, innovations in artificial intelligence have revolutionized various industries. One notable application of data modeling in materials science is a project led by Professor Qingyu (William) Yang at Wayne State University.
This article explores how Professor Yang used data modeling to predict the properties of high-strength steels, particularly dual-phase (DP) steels, without extensive material testing. This project, which began nearly a decade ago, holds significant potential for both alloy development and simulation applications to address real-world industrial challenges.
Challenges with High-Strength Steel
Advanced High-Strength Steel (AHSSs) are used extensively for safety-critical applications across industries such as automotive and aerospace. These steels provide excellent strength-to-weight ratios, enabling automakers to create lightweight, high-performance, and fuel-efficient vehicles.
As a result, the use of these high-strength steel grades in vehicle body components has grown rapidly over the past two decades.
However, one significant challenge associated with AHSS is accurately evaluating its properties.
Dual-phase steels consist of two phases of iron: soft ferrite and hard martensite. Each phase contributes distinct properties to the final material. To fully comprehend the characteristics of the final product, it is essential to quantify the proportion and distribution of both phases.
Figure 1: Dual phase microstructure
However, determining the arrangement of phases within the material is challenging, as they are distributed stochastically. Predicting the random distribution and proportion of the two phases to establish their mechanical properties is a complex task, with tremendous value for materials scientists and engineers.
“With the introduction of AHSS materials, the industry faced a significant challenge,” Professor Yang explained. “High-strength steel was a new material, and although its qualitative properties were well understood, it was challenging to quantitatively evaluate how much better or worse one material was compared to another.”
Thus, there is a need to accurately characterize the material by capturing the stochastic nature of its microstructure. This capability would enable engineers to quantify the distribution of martensite and ferrite, allowing for more precise predictions of the material’s properties.
The Emergence of a Stochastic Model
The core of Professor Yang’s research focuses on developing a stochastic model to predict the properties of dual-phase steel. This model uses statistical methods to quantify essential characteristics of the material’s microstructure, helping scientists and engineers evaluate the advantages and limitations of specific dual-phase steels for various applications.
“Since the distribution of martensite and ferrite is random, we needed a stochastic method to capture this randomness and develop a model that could be used to estimate the material properties more accurately,” Professor Yang stated. “By fitting the model with the appropriate parameters, we can identify the key features of the high-strength steel and use them to predict its mechanical performance.”
This approach represents a significant departure from earlier methods that relied on qualitative visual comparisons of microstructural images. Engineers previously assessed materials subjectively for applications. By contrast, stochastic models provide a mathematical framework to determine material properties with precision.
Predicting Material Properties with AI
While the term “AI” is often used to describe this innovative approach, at its core, the method involves an optimized statistical model designed for small data sets. This is particularly essential is material science, where experimental data is often limited. The stochastic model leverages the available data for training.
Research conducted by Professor Yang’s group at Wayne State University led to the development of a comprehensive methodology to address the challenges posed by the stochastic (random) nature of microstructure in high-strength steels (see Figure 2).
Figure 2: Developed methodology and its verification for multiple-phase material
Figure 3: Correlation function of microstructure
This methodology integrates microstructure models with stochastic field models, correlation functions, and deep neural network techniques to effectively capture the randomness and stochastic properties of material microstructures. It quantitatively captures the phase distribution information in dual-phase microstructures, enabling the generation of additional samples with equivalent properties, reducing the need for extensive experimental testing. By leveraging advanced stochastic models and artificial intelligence, the methodology accounts for variability within material batches and from different suppliers, ensuring robust predictive models across production scenarios. The integration of correlation functions (shown in Figure 2) enhances our understanding of interconnections between microstructural features, boosting the model’s predictive accuracy.
The methodology has two primary use cases: material characterization for sheet metal simulation and alloy development. In industries such as automotive manufacturing and aerospace, simulations are essential for engineering and optimizing both products and production processes. A deeper understanding of material properties allows designers to create and optimize products more effectively.
For alloy development, the model assists scientists in determining the optimal proportions of phases to achieve a material’s desired mechanical properties. This streamlines the creation of new materials and minimizes the need for physical experiments.
Facilitating Alloy Development
Professor Yang’s work in alloy development represents an exciting advancement in materials science. Traditionally, developing a new material required extensive physical testing to determine its properties. However, with Professor Yang’s model, this process can be significantly expedited.
“For instance,” Professor Yang noted, “if we develop a new type of DP steel and would like to minimize physical tests, we can use the model to reliably predict its properties using data from previous DP steel grades, such as DP 500 or DP 600.”
This approach minimizes the time and costs associated with material testing. Manufacturers can reduce the number of mechanical experiments required, allowing them to accelerate the development process.
Kidambi Kannan of AutoForm Engineering USA elaborated: “AI models rely on previously available data for training. If we already have sufficient data to characterize other DP grades and a new DP grade emerges, we can use AI in conjunction with the stochastic model to predict the properties of the new alloy without conducting physical tests.”
This advantage is especially beneficial in industries where material innovation is crucial. By using this model, companies can lower development costs and ensure they use the highest-performing materials for their applications.
Research Result and Applications
Professor Yang’s model has already produced promising results. The research has been partly supported by the US National Science Foundation Grant Opportunities for Academic Liaison with Industry (NSF GOALI: award #1404276), titled Failure Prediction and Reliability Analysis of Ultra-High Strength Steel Autobody Manufacturing Systems by Utilizing Material Microstructure Properties. The project has received multiple awards, including the 2020 Wilcoxon Award for the best application paper in Technometrics and the Institute of Industrial and Systems Engineering Annual Conference Best Student Paper Awards (2015 [1] and 2026), among others. The developed models have been used for Dual Phase-High Strength Steel reliability analysis, failure prediction, and real manufacturing production in collaboration with an industry partner.
Professor Yang’s research team continues to build on these advancements. They are developing AI and stochastic combined methods to extend the research into other mechanical properties and materials. The team is also creating advanced AI-Statics-Physics based models to address challenges in manufacturing and material processing. “As a researcher, I’m always eager to collaborate with both researchers and industry professionals to develop AI-Statistics-Physics models that tackle engineering challenges beyond the capabilities of traditional methods,” Professor Yang said. Industries such as aerospace, automotive, and construction could benefit significantly from these innovations in material selection and testing processes. As more industries adopt AI-driven solutions, the demand for such models is expected to grow. Professor Yang’s research offers a practical and impactful integration of AI with traditional materials science, quality assurance, and manufacturing practices.
Conclusion
Professor Yang’s model provides a glimpse into the future of materials science, where AI-driven models can deliver faster and more accurate predictions of a material’s performance.
By quantifying the microstructure of high-strength steels, the model has the potential to transform material simulation and alloy development, offering valuable solutions for industries that rely on high-performance materials.
As AI technology continues to evolve, the potential for this model to solve real-world industrial problems is vast. “I’m open to discussing and collaborating on new problems,” Professor Yang emphasized. “This is an exciting area of research, and I look forward to seeing how it can be applied in the future.”
[1] Zhang, N., & Yang, Q. (2015). A random effect autologistic regression model with application to the characterization of multiple microstructure samples. IIE Transactions, 48(1), 34–42. https://doi.org/10.1080/0740817X.2015.1047069
Qingyu (William) Yang, Associate Professor
Chair of Doctoral and Research Program
Department of Industrial and Systems Engineering
Wayne State University
4815 Fourth St., Rm. 2167, Detroit, MI 48202
(313) 577-9665
