Skip to content

Property Predictor

The Property Predictor is an interface to predict the values of desired properties for specified glass compositions.

It allows the user to select the appropriate method (model) for performing the predictions.

property-predictor-1.png property-predictor-1-dark.png

Models

Currently, only GlassNet is available.

GlassNet

GlassNet , a multitask model deep neural network model developed by Daniel R. Cassar {: target="_blank"} that is capable of predicting 85 different properties.

property-predictor-4.png property-predictor-4-dark.png

Please cite the following paper if you're using this model in your research:

Cassar, D.R. (2023). GlassNet: A multitask deep neural network for predicting many glass properties. Ceramics International 49, 36013–36024. 10.1016/j.ceramint.2023.08.281.

Note

Some properties predicted by GlassNet use different units (and scaling) than SciGlass. Check the Property section below for details.

In Sciglass Next, we still use the default units used in SciGlass for consistency as default.

Compositions

You can specify up to 15 component names and the corresponding values as input for the prediction.

An autocomplete dropdown list will be shown when you click on the component input fields:

property-predictor-2.png property-predictor-2-dark.png

Note that the names in the dropdown list are from the SciGlass database and should cover most common components.

Warning

When using the GlassNet model, only the elements between atomic numbers 1 and 83 (hydrogen and bismuth included) were considered, excluding promethium and the noble gases.

If GlassNet cannot handle the names, a warning message will be shown.

If the name you typed does not match in the database, it may still be used as input as long as it is valid.

property-predictor-3.png property-predictor-3-dark.png

Well, it is clear that XXX is not a valid component name.

Default mode

Enter a single glass composition directly, e.g., SiO2 = 70, B2O3 = 20, Na2 O = 10. The predictor will return the predicted properties for that composition.

Function mode

Use a variable (x) to define compositions, e.g., B2O3 = 74 - x, Na2O = x, SiO 2 = 26. Specify x range and step to generate multiple compositions.

The total of all components must equal 100 for the input to be valid.

The grid and table views are both disabled. However, you can download the data via the modebar. See the Example section below for details.

Experimental feature

This feature is experimental, feel free to share your feedback with us.

Batch mode

Enter multiple compositions in a spreadsheet-like table. The first row is the component names, and subsequent rows are the compositions.

You can copy and paste the data from other places, e.g., Excel, Google Sheets, etc.

Common hotkeys like Ctrl+C, Ctrl+V, etc. are supported.

Experimental feature

This feature is experimental, feel free to share your feedback with us.

Concentration

Specify the kind of percent for the value fields. The following options are available:

  1. Mol% (Molar %)
  2. Wt% (Weight %)
  3. At% (Atomic %)

The default is in Mol%.

Paste

Paste the composition copied from other places. This is not available yet.

Clear

You can click on the Clear button to clear all component and value fields.

Predict

Click on the Predict button to make a prediction with the entered glass compositions and the chosen model.

Property

All 85 properties (with scaling and units) available in GlassNet are shown below extracted from the paper in comparison to SciGlass.

GlassNet uses the International System of Units (SI) unit by default. In SciGlass Next, we still use the default units used in SciGlass for consistency.

Property
(SciGlass Next)
Unit Property
(GlassNet)
Unit
T1 (logη=1) °C 𝑇0 K
T2 (logη=2) °C 𝑇1 K
T3 (logη=3) °C 𝑇2 K
T4 (logη=4) °C 𝑇3 K
T5 (logη=5) °C 𝑇4 K
T6 (logη=6) °C 𝑇5 K
T7 (logη=7) °C 𝑇6 K
T8 (logη=8) °C 𝑇7 K
T9 (logη=9) °C 𝑇8 K
T10 (logη=10) °C 𝑇9 K
T11 (logη=11) °C 𝑇10 K
T12 (logη=12) °C 𝑇11 K
T13 (logη=13) °C 𝑇12 K
logη at 500°C P log10(𝜂(773K)) Pa.s
logη at 600°C P log10(𝜂(873K)) Pa.s
logη at 700°C P log10(𝜂(973K)) Pa.s
logη at 800°C P log10(𝜂(1073K)) Pa.s
logη at 900°C P log10(𝜂(1173K)) Pa.s
logη at 1000°C P log10(𝜂(1273K)) Pa.s
logη at 1100°C P log10(𝜂(1373K)) Pa.s
logη at 1200°C P log10(𝜂(1473K)) Pa.s
logη at 1300°C P log10(𝜂(1573K)) Pa.s
logη at 1400°C P log10(𝜂(1673K)) Pa.s
logη at 1500°C P log10(𝜂(1773K)) Pa.s
logη at 1600°C P log10(𝜂(1873K)) Pa.s
logη at 1800°C P log10(𝜂(2073K)) Pa.s
logη at 2000°C P log10(𝜂(2273K)) Pa.s
logη at 2200°C P log10(𝜂(2473K)) Pa.s
Tg °C 𝑇𝑔 K
Mg °C 𝑇dil K
Littleton point °C 𝑇Lit K
Annealing point °C 𝑇ann K
Strain point °C 𝑇strain K
Softening point °C 𝑇soft K
logρ at 20°C Ohm·cm log10(ρ(273K)) log10(ρ(293K)) Ohm·m
logρ at 100°C Ohm·cm log10(𝜌(373K)) Ohm·m
logρ at 150°C Ohm·cm log10(𝜌(423K)) Ohm·m
logρ at 300°C Ohm·cm log10(𝜌(573K)) Ohm·m
logρ at 800°C Ohm·cm log10(𝜌(1073K)) Ohm·m
logρ at 1000°C Ohm·cm log10(𝜌(1273K)) Ohm·m
logρ at 1200°C Ohm·cm log10(𝜌(1473K)) Ohm·m
logρ at 1400°C Ohm·cm log10(𝜌(1673K)) Ohm·m
TK-100 °C 𝑇𝜌=106𝛺.𝑚 K
ε' (~20°C, ~1MHz) - 𝜀 -
Tanδ *1E4 - log10(tan(𝛿)) -
α at 55 ± 10°C *1E7 K-1 log10(𝛼𝐿(328K)) K-1
α at 100 ± 10°C *1E7 K-1 log10(𝛼𝐿(373K)) K-1
α at 160 ± 10°C *1E7 K-1 log10(𝛼𝐿(433K)) K-1
α at 210 ± 10°C *1E7 K-1 log10(𝛼𝐿(483K)) K-1
α at 350 ± 10°C *1E7 K-1 log10(𝛼𝐿(623K)) K-1
α at T < Tg *1E7 K-1 log10(𝛼𝐿(𝑇 < 𝑇𝑔)) K-1
Density at 20°C g/cm3 𝑑(293K) g/cm3
Density at 800°C g/cm3 𝑑(1073K) g/cm3
Density at 1000°C g/cm3 𝑑(1273K) g/cm3
Density at 1200°C g/cm3 𝑑(1473K) g/cm3
Density at 1400°C g/cm3 𝑑(1673K) g/cm3
nd at 20°C - 𝑛𝐷 -
n at 0.6 < λ < 1µm (20°C) - 𝑛 (low) -
n at λ > 1µm (20°C) - 𝑛 (high) -
Mean dispersion *1E4 - log10(𝑛𝐹 − 𝑛𝐶) -
Abbe's number - 𝑉𝐷 -
Thermal shock resist. K 𝛥𝑇 K
Young's modulus GPa 𝐸 GPa
Shear modulus GPa 𝐺 GPa
Poisson's ratio - 𝜈 -
Microhardness GPa 𝐻 GPa
Tliq °C 𝑇liq K
Tm °C 𝑇melt K
Thermal conductivity W/(m·K) 𝜅 W/(m·K)
CP at 20°C J/(kg·K) 𝐶𝑝(293K) J/(kg·K)
CP at 200°C J/(kg·K) 𝐶𝑝(473K) J/(kg·K)
CP at 400°C J/(kg·K) 𝐶𝑝(673K) J/(kg·K)
CP at 800°C J/(kg·K) 𝐶𝑝(1073K) J/(kg·K)
CP at 1000°C J/(kg·K) 𝐶𝑝(1273K) J/(kg·K)
CP at 1200°C J/(kg·K) 𝐶𝑝(1473K) J/(kg·K)
CP at 1400°C J/(kg·K) 𝐶𝑝(163K) J/(kg·K)
σ at T > Tg mN/m 𝛾(𝑇 > 𝑇𝑔) J/m2
σ at 900°C mN/m 𝛾(1173K) J/m2
σ at 1200°C mN/m 𝛾(1473K) J/m2
σ at 1300°C mN/m 𝛾(1573K) J/m2
σ at 1400°C mN/m 𝛾(1673K) J/m2
Tmax °C 𝑇max(𝑈) K
Vmax cm/s log10(𝑈max) m/s
Tc °C 𝑇𝑐 K
Tx °C 𝑇𝑥 K

Note

All logarithm is base 10.

View (Property Filter)

If you are only interested in some properties, you can switch the visibility with this function on the right-hand side.

property-predictor-5.png property-predictor-5-dark.png

For example, with only Density group visible: property-predictor-6.png property-predictor-6-dark.png

Unit (Unit Converter)

You can use this function to switch to different units for different properties.

Currently we offer the following options for unit conversion:

  1. Default (SciGlass units)
  2. International System of Units (SI)
  3. Custom (User-defined units)

property-predictor-9.png property-predictor-9-dark.png

Note

There are currently only few properties and units available for unit conversion.

If you have any suggestions/preferences, please let us know so that we can add more units.

Example

Default mode

With SiO2=70, B2O3=20, Na2O=10 in mol%

property-predictor-7.png property-predictor-7-dark.png

Tip

The composition values are rescaled in GlassNet, i.e. the total sum does not have to be 100.

If we type SiO2=0.7, B2O3=0.2, Na2O=0.1, the results will be the same.

You can also switch to table view for better comparison and export the data to Excel/CSV format:

property-predictor-8.png property-predictor-8-dark.png

Or you can switch to the plot view for visualization:

property-predictor-10.png property-predictor-10-dark.png{ loading=lazy data-gallery=" dark" }

Currently, only the temperature-dependent properties are available for plotting.

property-predictor-11.png property-predictor-11-dark.png{ loading=lazy data-gallery=" dark" }

Function mode

With B2O3=74-x, Na2O=x, SiO2=26 in mol% and x from 2 to 46 with step 2 (mol%).

property-predictor-12.png property-predictor-12-dark.png{ loading=lazy data-gallery=" dark" }

You can directly download the data in Excel/CSV format via the modebar.

Also, you can switch to all 85 different properties for plotting. A tooltip will be shown when you hover over the property name.

property-predictor-13.png property-predictor-13-dark.png{ loading=lazy data-gallery=" dark" }