Technology
   NIR spectrometry
   Topnir Technology

TOPNIR Technology

Wavelength region

The near infrared region of electromagnetic spectrum is spread from 700 nm to 2.500 nm. The NIR absorption is composed of a combination band adjacent to the fundamental band (MIR) and three overtones region. Our interest has been focused on the combination region where the frequencies of vibrations between the bonds of the atoms is higher than in the overtones. The richest information is available in the region from 2.000 nm to 2.500 nm where there is no residual absorption band coming from Visible spectral region. This is the band we are using. Indeed, the NIR range itself has been optimized for hydrocarbon analysis as shown in the Figure 1. This deep R&D program was carried out in BP between 1986 and 1990.

Figure 1 – Optimum crude spectra between 5000 and 4000 cm-1

However using that band provides richer information, allowing for instance to reduce the optical path. This is why our cell is only 0.5 mm wide allowing a very light maintenance.

Modelling

Topnir principle

Topnir modeling is not linear, and does not use one model per property. On the contrary, Topnir is based on topology, work through pattern recognition and databank densification to deliver full set of properties through only one model. Topnir instantly delivers all the properties required for a given application. Moreover, it offers the possibility of extrapolation from the initial calibration range.

This topology modelling consists in:

      1.Positioning 20 samples in a specific space.

      2.Densify the space with virtual sample.

      3.Defining a “property function”.

Then to do a prediction Topnir will:

      1.Position the new sample in the same spectral space

      2.Identify all the samples located within a certain distance

      3.Apply the property function to these samples

Topnir Unique Graphic user Interface

Thanks to unique graphical representation Topnir displays an effective way to view quickly the samples for a better understanding of the product chemistry (Graphic User Interface GUI).

TOPNIR GRAPHIC USER INTERFACE (GUI)

Densification

For the model to work effectively, it is necessary to be able to match spectra with data in the database, and it could be difficult to achieve reliable results without a large number of data points. To overcome this problem, Topnir uses a densification technique which is one of the cornerstones of the technology.

Densification is done by interpolating between the actual points, and very importantly also using points outside the boxes, to determine a contour map of properties of everything within the box. With a greatly increased number of points there is therefore a much better accuracy of prediction. Of course the accuracy of using ‘virtual’ points is equivalent to the one using real points which has been the key in the development of Topnir. Densification is done when the model is initially prepared. In use therefore, Topnir carries out a local densification ‘on the run’ automatically to increase still the accuracy of prediction by populating around the point to be analyzed (also called online densification). Densification uses TOPNIR SYSTEMS proprietary spectral blending laws and is tuned on client site specificity.

Outlier

Concerning the treatment of outliers, the classical linear methods require a very heavy amount of work to maintain the models. It can rapidly become a bottleneck when the application involves a lot of properties, each one requiring a specific calibration treatment. The system takes into account any outlier and does not require any new calibration of the models. It is a self learning method which fits perfectly with the reality of industrial operations. Thanks to local on-line densification, Topnir is able to provide property values even if the sample is out of box. Thus, if a sample is out of box then Topnir will carry out a local on-line densification around the point and then the system will display both the full set of properties as well as information on the location and chemical character of the product.

Figure 2 Local densification (outlier)

Topnir is thus designed in such a way that local on-line densification is carried out for every analysis and every point, giving to Topnir a unique level of accuracy and robustness. No other system can offer such a richness of information with both correct prediction on outlier associated with qualitative information on sample chemistry itself.

Topnir advantages

Model Calibration

Topnir is only one single model for all the properties, the property prediction being performed from a Reference database. On the contrary other basic modelling techniques like PLS (Partial Least Squares) require one model per property which is heavy in calibration and maintenance steps.

Also, the fact to have one single model gives a better robustness and reliability to the prediction.

Topnir model is not limited with the number of properties as it deals with a single reference database. So, number of properties can be easily extended if required without impact on modelling work.

Limited number of samples required from the Client

Only a few samples are required from the client side. Topnir is able to artificially create a large number of samples from a very limited set of client samples; this process is called “densification”. The densification significantly improves the model robustness without degrading the model accuracy.

Model Update

Automatic model update after new sample introduction in calibration database without remodelling.

Light maintenance and Self-learning Model

Topnir provides the full set of properties from a single model. This fact results in a very light maintenance effort.

The client has the capability to update the Topnir model by adding new samples into the Topnir reference databank. As said above, this Update is automatic after new sample introduction in calibration database. No remodelling is required.

Indeed, Model update is easy through the “Add sample” facility to update the Topnir database. As Topnir uses only a reference database to predict and calculate the properties of samples, the update of the application is easily done only by adding a new sample in the database, both with NIR Spectra and conventional analysis. Topnir is really a self-learning method offering simple sample addition in the database without the necessity of remodelling. As soon as the sample is added, it is taken into account by Topnir.

No Specific End User expertise is required, model maintenance is very easy through simple sample addition in database.

Model accuracy and robustness - ASTM Standard compliance

As any conventional Analyser, Topnir accuracy is warranted in agreement with the ASTM standard, the comparison between Topnir predictions and Conventional analysis being within the ASTM repeatability and reproducibility. Moreover, from field proven applications, Topnir results are better than ASTM. Topnir technology also allows predicting atypical product properties.

Outlier detection and Prediction

The range of variation defined during the Site Survey is covered as much as possible by the reference samples and expanded artificially to ensure robustness even for atypical products.  
 Outlier means samples outside the initial calibration set. Topnir offers Automatic detection of Outliers. In case of on-line application, the information is provided to the DCS and sampling grabbing can be automatically or manually ordered. Outlier samples are calculated by Topnir with reasonable accuracy.
Topnir also offers facilities to extrapolate properties of atypical samples using its unique automatic densification process.  

Management of Additive Effect

When additive is input to boost the Cetane Number and the CFPP, Topnir is able to predict the final properties, based on the NIR spectra and the amount of additive input

Direct Prediction of Blending Indices & Blend Simulation Feature

Topnir goes beyond simple analysis. To improve the planning, scheduling, and ordering stage, Topnir calculates the linear blending indices of each component in a particular grade. These parameters integrate “Positive/negative” effect of the different component and make the blending optimization stage more accurate and much easier. Topnir GUI (Graphic User Interface) allows the user to simulate a blend and estimate the risks (out of spec or give away) thanks to a specific graphical visualization (see below).

Topnir simulates the blend using a dedicated spectral axis based on hydrocarbon chemistry. This functionality enables the operators to safely drive the blend toward the optimum.

Local support

During Project development as well as during and after the warranty period, Topnir provides application support to insure continuous benefits from the application. Remote access control is used to monitor the client application.
Back to top Click here to know more