Machine Learning to Predict the Martensite Start Temperature in Steels
- PDF / 2,050,776 Bytes
- 11 Pages / 593.972 x 792 pts Page_size
- 46 Downloads / 246 Views
CTION
MATERIALS development is currently undergoing large changes with a transition from the previously dominating empirical development methodologies toward methodologies with more computational components. This development can be divided loosely into two paths with one focusing on replacing some of the
MOSHIOUR RAHAMAN is with Ferritico, Brinellva¨gen 85, 100 44 Stockholm, Sweden. Contact email: [email protected] WANGZHONG MU is with the Department of Materials Science and Engineering, KTH Royal Institute of Technology. JOAKIM ODQVIST and PETER HEDSTRO¨M are with Ferritico, Brinellva¨gen 85, 100 44 Stockholm, Sweden, and also with the Department of Materials Science and Engineering, KTH Royal Institute of Technology. Contact email: [email protected] Manuscript submitted May 15, 2018.
METALLURGICAL AND MATERIALS TRANSACTIONS A
experimental input with physically based modeling on different length- and timescales, often referred to as integrated computational materials engineering (ICME).[1] The other direction is the use of data and machine learning (ML),[2–5] a branch of artificial intelligence. Key for both these areas is the use of databases where the ICME methods to a large extent rely on the so-called CALPHAD databases that collect thermodynamic and kinetic data essential for the modeling of phase transformations and related phenomena, while the ML approaches are more flexible to use any database that contains data of relevance for the parameter that should be predicted. It is clearly also possible to combine elements from the two areas and both rely on the materials genomics field where the Materials Genome Initiative[6] has provided extra thrust to the development of open materials databases. In steel research and development, it is vital to be able to predict microstructures based on alloy composition and heat treatment cycle. One constituent that is
important in high-performance steels is the hard martensite constituent, which is a part of, e.g., tool steels, dual-phase steels, quenching and partitioning steels, transformation-induced plasticity steels, and martensitic stainless steels. In the alloy and heat treatment design process, the martensite start temperature (Ms) is a critical parameter. Therefore, significant attention has been paid to the modeling of martensite and Ms in the literature.[7–30] These models use different methodologies such as linear regression,[7,8] thermodynamics-based modeling, which relies on CALPHAD databases and semiempirical fitting of the required driving force to initiate martensitic transformation,[9–19] and artificial neural network (ANN) modeling, which uses nonlinear fitting to the available experimental data.[21–29] The data-driven approaches, where the ANN modeling is one, have developed significantly recently.[31–34] From here on, these methods are referred to as ML, which can be simply described as computational techniques that enable the computer to learn from data and recognize patterns in the data. The datasets can be of many different sizes and big data is another impor
Data Loading...