ProxiMate: Structure of the applications

This package can be used to build and/or update NIR applications that are ready to be consumed by the ProxiMate series of NIR sensors manufactured by BUCHI Labortechnik AG. Once an application is installed in a ProxiMate device, it can be used to predict the properties of a given matrix using the spectral models contained in that application.

This package builds upon the standard structure of the ProxiMate applications, which are conventionally developed with the compiled-executable software “NIRWise PLUS” offered by BUCHI. Therefore, the files output by proximetricsR follow the same structure of the ones output by “NIRWise PLUS”. No changes or improvements on these output files have been conducted for the development of proximetricsR.

2 Structure of the ProxiMate predictive applications

ProxiMate applications can be described as a collection of predictive models that, along with other metadata, are packed into a single file that can be installed in a ProxiMate sensor. Once an application is imported, the sensor can be used to predict the properties of samples the application was built for.

2.1 Calibration data file (.tsv)

This is the main input file which contains the predictor (spectra) and response (properties) data used to calibrate the models. The data is stored as a tab separated table. Every row in this table represents a single spectral measurement of a sample along with its associated data (e.g. property values, date, etc.). This table is usually exported directly from the ProxiMate devices and typically contains columns with the following fields:

first_pixel_nir <- 4
first_pixel_count_nir <- first_pixel_nir + 1
first_wavelength_nir <- 2.04E-10 * first_pixel_count_nir^5 + 
  -1.28E-07 * first_pixel_count_nir^4 + 
   2.80E-05 * first_pixel_count_nir^3 +
  -4.76E-03 * first_pixel_count_nir^2 + 
   3.89     * first_pixel_count_nir   + 
   880.06
first_wavelength_nir

[1] 899.3944

first_pixel_count_vis <- 823
first_wavelength_vis <- 0.0 * first_pixel_count_vis^4 + 
   0.0          * first_pixel_count_vis^3 +
  -7.586146E-05 * first_pixel_count_vis^2 + 
   2.12726      * first_pixel_count_vis   + 
  -1301.079
first_wavelength_vis

[1] 398.2728

A sequence of numbers can be used to obtain all the wavelengths at once. This sequence must start with the index of the first pixel and end with the index of the last pixel. Continuing the previous example, we can obtain all wavelengths at once as follows:

pixel_sequence_nir <- 4:272
pixel_sequence_count_nir <- pixel_sequence_nir + 1

wavelengths_nir <- 2.04E-10 * pixel_sequence_count_nir^5 + 
  -1.28E-07 * pixel_sequence_count_nir^4 +
   2.80E-05 * pixel_sequence_count_nir^3 +
  -4.76E-03 * pixel_sequence_count_nir^2 + 
   3.89     * pixel_sequence_count_nir   + 
  880.06

wavelengths_nir[1:5]

[1] 899.3944 903.2345 907.0661 910.8892 914.7040

wavelengths_nir[(length(wavelengths_nir) - 5):length(wavelengths_nir)]

[1] 1741.389 1744.169 1746.953 1749.742 1752.534 1755.332

pixel_sequence_vis <- 823:1074
wavelengths_vis <- 0.0 * pixel_sequence_vis^4 + 
   0.0          * pixel_sequence_vis^3 +
  -7.586146E-05 * pixel_sequence_vis^2 + 
   2.12726      * pixel_sequence_vis   + 
  -1301.079

wavelengths_vis[1:5]

[1] 398.2728 400.2751 402.2773 404.2793 406.2812

wavelengths_vis[(length(wavelengths_vis) - 5):length(wavelengths_vis)]

[1] 886.2704 888.2354 890.2003 892.1649 894.1295 896.0939

2.2 Local data file (.tsv)

This file has the same structure as the “Calibration data file”. The only difference is that this local file is used to store spectra measured by the user of the application. A spectrum is written to this file only if one or more of its reference (response) values are manually input by the user directly in the instrument.

2.3 Calibration model files (.cal)

These files store both the instructions for spectral pre-processing and the parameters of calibrated models. Each file contains one single model, i.e. if an application contains n predictive models, then there will be n cal files in the application. These files are used by the the sensor instruments to conduct the required predictions of the response variables in the application.

2.4 Project files (.prj)

These files store all the final results obtained when a calibration model is built. This file can be imported into the NIRWise PLUS software to visualize the calibration model results and the used settings. Note that the files are not used to generate predictions as their purpose is limited to the visualization and review of the models. The project file may contain the pre-pocessed spectra, the data matrices generated during the calibration process, the information about the type of model validation, the outlier detection method used, the response residuals, etc.

2.5 Report files (.rtf)

These files are in rich text format, each containing a report on the results of the calibration of a single response variable in the application. This report includes information such as the original tsv file used for the calibrations, the number of observations used and their indices in the tsv table, the standard error of the calibration (SEC), the coefficient of determination (R^2), etc.

2.6 Application metadata file (.nad)

This file contains application metadata such as the application name (the name that will be shown when it is imported into a sensor), the sample measurement geometry, measurement time, the creation date, the standard operating procedure, additive (offset) and multiplicative (slope) adjustments to the predicted response values, outlier detection parameters, etc.

2.7 Application file (.nax)

This file acts as a container for all the files described above. It is in fact a ZIP file used to pack and compress the application files. The file and folder structure inside a container with n predictive models can be described as follows:

Welcome to ClientVPS Mirrors