The utility of Bayesian data reconciliation for separations
While there has been no update on here, the BayesMassBal V 1.0.0 is now available on CRAN. You can install the latest public version of BayesMassBal with the following line of code in R studio:
install.packages("BayesMassBal")This announcement coincides with the publication of the article The utility of Bayesian data reconciliation for separations in Minerals Engineering. If you are unable to access the journal article, you can access it here for free for the next 50 or so days.
Models in The utility of Bayesian data reconciliation for separations allow for a machine learning based approach to mass balancing plant or laboratory data as well as improvements in accuracy due to improved uncertainty quantification and the use of truncated normal error.
What is BayesMassBal?
The BayesMassBal package is intended to provide process engineering students and professionals easy access to the Bayesian mass balance methods outlined in The utility of Bayesian data reconciliation for separations. The free and open source package is available from CRAN for the R programming language.
If you feel that the software is difficult to use, or you have suggestions for improvement, please contact me, I would love to hear your feedback!
Why BayesMassBal?
The mass balance models available in BayesMassBal have the following features:
- Full uncertainty quantification for the mass balanced values
- Machine learning model selection methods to find which mass balance model is a better fit for a data set
- Error truncated at 0, advantageous for extremely low grade samples where Gaussian error is not appropriate
- Tools are provided with
BayesMassBalwhich allow the user to organize all of their data in Excel before importing data intoR.
After installing BayesMassBal you can type system.file("extdata", "twonode_constraints.csv",package = "BayesMassBal") into the R console to find an example on your computer of how to create a .csv file with your process constraints. Similarly, the file at the location found using system.file("extdata", "twonode_example.csv", package = "BayesMassBal") shows how to organize process data before importing into R. These two files correspond with the Two_Node_Process vignette.
What’s next?
Some readers have told me they are looking forward to the next publication in this area. Thanks for the positive feedback, I’m looking forward to writing the next publication!
My research in this area is slightly different than the rest of the research for my PhD dissertation, applying Bayesian inference to mass balance models is truly a personal interest and curiosity. There will be another publication. However as the rest of my PhD dissertation is in another area of applied Bayesian modeling, and I am hoping to graduate in the Fall, it might be a while.
Furthermore, there are many possibilities for the next publication, it is difficult to choose just a few to work on. Some options for the next publication include but are not limited to:
- Improved computation time via alternative algorithms for taking draws from the truncated multivariate normal distribution.
- Implementation in python.
- More plotting functions to improve ease of use.
- Further sensitivity analysis applications, using Bayesian inference to understand the implication of sales contract terms.
- Time correlated data: a Bayesian competitor to BILMAT.
- Mass balancing, modeling, and statistical comparison of data sets with different process parameters.
- Use of a mixture of truncated normal distributions to model complex error structures.
Any input as to what you would like to see next is welcome.
Scott Koermer
Post-Doctoral Researcher
Post doctoral researcher at Los Alamos National Lab with the statistical sciences and geophysics groups working on nuclear detonation detection. Previous research includes applying Bayesian statistical methods to separation processes, concentration of rare earth elements from acid mine drainage, and analysis of processes for scrap metal recycling.