Materials Algorithms Project
Data Library

Data Library MAP_DATA_NEURAL_MS

1. Provenance of code.
2. Purpose of code.
3. Description of subroutine's operation.
4. References.
5. Any additional information.
6. Keywords.
7. Download source code.
8. Links.

Provenance of Source Code

Pieter van der Wolk
Heat Treatment Science and Technology Group
Laboratory of Materials Science
Delft University of Technology
Rotterdamseweg 137
2628 AL Delft

Contact: P.J.vanderWolk@STM.TUDelft.nl

Top | Next

Purpose

Provides a database giving Ms data for steels of various composition, and a trained neural network model (provided as a spreadsheet) for calculating Ms temperatures for steels of arbitrary composition.

Top | Next | Prev

Description

The distribution contains three types of file, as described :-

1. neur-ms.txt is a text version of this document.

2. MS.TXT is an tab-delimited ASCII text file containing a database of steel compositions and their martensite start temperatures. Each row represents one steel composition. Columns are :-

   Columns 1 and 2 give the Ac3 temperature and the austenising temperature respectively (in Kelvin).
   Columns 3-19 specify alloying element in mass%. The elements are :-

   • Carbon
   • Silicon
   • Manganese
   • Phosphorus
   • Sulphur
   • Nickel
   • Chromium
   • Molybdenum
   • Vanadium
   • Boron
   • Copper
   • Tungsten
   • Aluminium
   • Titanium
   • Niobium
   • Nitrogen
   • Cobalt

   Alloying elements which were not specified in the original paper have been set to zero.

   Column 20 gives Martensite start temperature in Kelvin.
   Column 21 gives a reference number; the meaning of these reference numbers is listed at the bottom of the file, and in the reference list below.

3. MsANN.WK1 and MsANN.xls contain a neural network model converted into LOTUS1-2-3 and EXCEL3 formats respectively.

   After editing the cells with the alloying elements (B5..B16) the Ms-temperature (F8..F10) is updated automatically according to the neural network model. The model is incorporated in the spreadsheet below row 18.

   The neural network model has been trained with 311 steel compositions from the above database; 233 data have been used to calibrate the model, and 78 data have been used to validate the model.

   It is a hierarchical feed-forward backpropagating neural network with a 12:5:1 architecture. For a detailed description, see [5].

   Testing the model

   The initial steel composition gives :-

   Constituent	Mass%	Ms Temp (Kelvin)
   C	0.2000	718.0
   Si	0.2500
   Mn	0.7000
   P	0.0010
   S	0.0005
   Ni	0.1000
   Cr	0.3500
   Mo	0.0100
   V	0.0000
   Cu	0.0000
   Al	0.0200
   N	0.0050

   If the following compositions give the indicated Ms temperatures, the model is working properly.

   Constituent	Mass% [1]	Ms Temp (Kelvin) [1]	Mass% [2]	Ms Temp (Kelvin) [2]
   C	0.6000	523.3	0.1100	722.6
   Si	0.4800		1.0000
   Mn	1.6100		0.5000
   P	0.0140		0.0000
   S	0.0280		0.0000
   Ni	0.0500		0.9000
   Cr	1.0000		0.0300
   Mo	0.0200		0.1000
   V	0.1600		0.0000
   Cu	0.1000		0.0000
   Al	0.0220		0.0000
   N	0.0070		0.0000

Top | Next | Prev

References

1. Atlas of Continuous cooling transformation diagrams for Vanadium Steels, Vanitec, Kent, June 1985
2. Atlas zur Warmebehaendlung der Staehle, Verlag Stahleisen mbH, Duesseldorf, Germany 1954.
3. W.W. Cias, Phase transformation kinetics and hardenability of medium-carbon alloy steels, Climax Molybdenum Company, Greenwich 1973.
4. G. Ghosh and B. Olsen, Acta Metall. Mater., 1994, 42, 3361-3370
5. W.G. Vermeulen et al, Ironmaking and Steelmaking 1996, 23, (5), 433-437

Top | Next | Prev

Further Comments

None.

Top | Next | Prev

Keywords

materials, data, neural, network, martensite, start temperature, steel, composition, concentration, start, temperature

Top | Next | Prev

Download data

Download data library

Top | Prev