Periodic Reporting for period 1 - muTWIN (Innovative digital twin concept of complex microstructure evolution in multi-component materials)
Reporting period: 2023-09-01 to 2025-08-31
The metal Additive Manufacturing (AM) market is growing rapidly. Its global market size was valued at $2.6 billion in 2021 and is expected to reach around $14.1 billion by 2031. AM is especially of interest for high-end applications, such as dental and healthcare, aerospace, automotive and energy. Important benefits are the ability to produce almost any 3D shape with minimal material loss and the short production times. It is however still a challenge to control the properties and reliability of AM parts through tailoring microstructural features, such as grain size, shape and orientation, presence of non-equilibrium phases, precipitates and pores and solute segregation. Due to the high cooling rates and temperature variations in AM, the microstructures are completely different from those obtained with standard manufacturing techniques for the same alloy composition. The final microstructure is sensitive to changes in powder composition, scanning parameters, baseplate heating, and many more parameters, and is often heterogeneous. The observation that fine, heterogeneous and hierarchical microstructures can be obtained with AM leads to the desire to produce materials with properties superior to those obtainable through classical manufacturing techniques through microstructure tailoring. Moreover, functionally graded AM, combining several functions within one material using intended gradients in composition and structure, is of great interest. If these high expectations cannot be met, customers may lose interest in AM.
While thermal and mechanical finite element (FE) models are frequently used in AM design, they generally assume homogeneous material properties, ignoring the huge effect of cooling rate and heat cycles on the microstructure, the spatial heterogeneities and the directionality of the microstructure. Various simulation techniques, such as cellular automata and phase-field models, exist to simulate microstructure evolution in materials as a function of composition and temperature variations, and taking external effects (mechanical loading, corrosion) into account. However, these models cannot handle the complexity of technical alloys or/and require a compute time too high for broad industrial application. Because of the heterogeneity and directionality in the microstructures, homogenization of high-fidelity simulation outcomes is difficult. Also the existing surrogate models based on neural networks all assumed large simplifications. Accurate and efficient digital twin models of microstructure evolution are not available. The large number of parameters affecting microstructure evolution, in combination with the scarce amount of training data, are a huge challenge for the existing approaches.
Tailoring the microstructural features in metal additive manufacturing (AM) products is thus still a great challenge, limiting the design flexibility and full exploitation of the technique. The aim of muTWIN is to implement and validate an innovative digital twin concept of microstructure evolution in technical alloys. The new concept allows for fast and accurate computation of local microstructures in AM parts as a function of local composition and temperature history. When integrated in thermal and mechanical Finite Element (FE) approaches it will enable realistic predictions of properties, performance and reliability of the printed parts. Moreover, it allows for fast and high-dimensional computational ‘search’ through the huge design space accessible with AM, to find the compositions and production parameters resulting in the most superior properties. With muTWIN, AM companies can drastically reduce the time-to-market and experimentation cost for new products. This brings new opportunities to design structurally and functionally graded materials and discover new alloy compositions and processing routes and will drive innovation in many other areas as well, such as medical applications, aerospace, construction and energy. The patentability of this idea and interest from industrial partners will be examined and a strategic plan for valorisation and possibly commercialisation of the product will be developed.
While thermal and mechanical finite element (FE) models are frequently used in AM design, they generally assume homogeneous material properties, ignoring the huge effect of cooling rate and heat cycles on the microstructure, the spatial heterogeneities and the directionality of the microstructure. Various simulation techniques, such as cellular automata and phase-field models, exist to simulate microstructure evolution in materials as a function of composition and temperature variations, and taking external effects (mechanical loading, corrosion) into account. However, these models cannot handle the complexity of technical alloys or/and require a compute time too high for broad industrial application. Because of the heterogeneity and directionality in the microstructures, homogenization of high-fidelity simulation outcomes is difficult. Also the existing surrogate models based on neural networks all assumed large simplifications. Accurate and efficient digital twin models of microstructure evolution are not available. The large number of parameters affecting microstructure evolution, in combination with the scarce amount of training data, are a huge challenge for the existing approaches.
Tailoring the microstructural features in metal additive manufacturing (AM) products is thus still a great challenge, limiting the design flexibility and full exploitation of the technique. The aim of muTWIN is to implement and validate an innovative digital twin concept of microstructure evolution in technical alloys. The new concept allows for fast and accurate computation of local microstructures in AM parts as a function of local composition and temperature history. When integrated in thermal and mechanical Finite Element (FE) approaches it will enable realistic predictions of properties, performance and reliability of the printed parts. Moreover, it allows for fast and high-dimensional computational ‘search’ through the huge design space accessible with AM, to find the compositions and production parameters resulting in the most superior properties. With muTWIN, AM companies can drastically reduce the time-to-market and experimentation cost for new products. This brings new opportunities to design structurally and functionally graded materials and discover new alloy compositions and processing routes and will drive innovation in many other areas as well, such as medical applications, aerospace, construction and energy. The patentability of this idea and interest from industrial partners will be examined and a strategic plan for valorisation and possibly commercialisation of the product will be developed.
A digital TWIN model has been developed to simulate microstructure evolution during isothermal decomposition in a binary system, covering a wide range of model parameters and hence representing a broad spectrum of materials systems.
A second digital TWIN model has been established for microstructure evolution during the decomposition of BCC phase into a three – phase system comprising of BCC, B2 and FCC phases.
Combining data driven and physics based ideas in the optimal way, a huge predictability beyond the training set could be obtained. For both cases, the models achieve high prediction accuracy for time spans twice as long as those used in the high-fidelity training simulations.
Once trained, the digital twin models compute microstructural evolution in less than a second --compared to several hours required by the high-fidelity models. This accuracy and speed make them well-suited for further exploitation in industrial applications and integration within finite element simulations.
A second digital TWIN model has been established for microstructure evolution during the decomposition of BCC phase into a three – phase system comprising of BCC, B2 and FCC phases.
Combining data driven and physics based ideas in the optimal way, a huge predictability beyond the training set could be obtained. For both cases, the models achieve high prediction accuracy for time spans twice as long as those used in the high-fidelity training simulations.
Once trained, the digital twin models compute microstructural evolution in less than a second --compared to several hours required by the high-fidelity models. This accuracy and speed make them well-suited for further exploitation in industrial applications and integration within finite element simulations.
The models developed in this project are unique in both capability and performance. They have achieved the computational speed required for industrial applications and for seamless integration into finite element simulations. A particularly distinctive feature is their ability to predict microstructure evolution over time spans twice as long as those used in the training simulations. In addition, the model for binary alloy decomposition demonstrates predictive power across a broad range of alloys.
Nevertheless, given the wide variety of microstructural phenomena that can occur, the models remain system-specific to a large extend. Extending them to more complex systems—such as those involving additional elements or phases--remains time- and resource-intensive. Further developments will still require expertise in phase-field modeling to generate the necessary training data. Training is still challenging, as it demands ad-hoc modifications to the methods tailored to the specific microstructural phenomenon considered.
While a first proof of concept has been demonstrated for both the binary systems and the quaternary system Al-Cr-Fe-Ni, further research is needed, This progress is best pursued within academic projects and through close collaboration with industry.
Nevertheless, given the wide variety of microstructural phenomena that can occur, the models remain system-specific to a large extend. Extending them to more complex systems—such as those involving additional elements or phases--remains time- and resource-intensive. Further developments will still require expertise in phase-field modeling to generate the necessary training data. Training is still challenging, as it demands ad-hoc modifications to the methods tailored to the specific microstructural phenomenon considered.
While a first proof of concept has been demonstrated for both the binary systems and the quaternary system Al-Cr-Fe-Ni, further research is needed, This progress is best pursued within academic projects and through close collaboration with industry.