1. Getting started

Chemometric Brain is a cloud-based software, so you do not need to download or install anything on your computer to work with it.

To start using it, open your favorite browser and go to our software website . Once there, enter your username and password to access your session.

After entering your credentials, you will be able to access Chemometric Brain. As you can see on the screen, our software has different tabs located at the bottom of the window that will enable you to perform various types of analysis.

2. Spectra viewer.

The first tab of the software is Spectra viewer. From this section, we can view spectra and compare them with each other.

To carry out the visualization, search for the spectra you want to display on the screen from the «Search Sample Spectra» button. From this menu, you can make a detailed search according to different filters such as date of measurement, type and name of product, equipment with which the sample was measured or batch name, among others.

Once located, select the spectra to be displayed and load them into the software by clicking on the corresponding “Load” button. All spectra will appear in the list on the left side of the screen.

In the lower left corner, there is a menu from which it is possible to apply different pre-treatments to the spectra.

3. Quality control

As soon as a sufficiently large number of spectra from different batches of the same product is available, it is possible to use another of the functionalities offered by our software: the Quality Control tab.

In this section, you will be able to create, develop and update qualitative models from a set of NIR spectra, either of raw materials or an end-product.

In order to visualize and build this type of models, go to the Quality Control tab and then click on the «Load Product Model options» button, located on the toolbar at the top.

Using the different filters in the pop-up window, search for the model you want to update. Select it and click on the «Load» button.

Once loaded, the software displays all the information about the model. Additional information can be shown by exploring the options in the toolbar, such as Loadings and Spectra or PCs model scores.

4. Sample validation

The next tab in the software is «Sample Validation». This module is of particular interest to customers who already have public qualitative models, previously developed by themselves or by Chemometric Brain’s technical support team. This tool is widely used in routine activities in quality departments of food industries. It provides in seconds detailed information about the compliance of new batches of a product or raw material with respect to the quality standards established in the previously saved qualitative models.

By deploying Chemometric Brain in your quality system, you can benefit from the numerous applications offered by the Sample Validation module. Now, you will see examples of some of them.

Example 1

The first of the cases to be shown is the detection of food fraud. To test the effectiveness of our qualitative models for detecting this kind of fraud, several samples of Milk Protein Concentrate adulterated with concentrations of 1% to 10% of sweet whey were prepared in the laboratory. From the “Sample Spectra Validation” menu of the software, we are going to select the qualitative model we want to use and the samples that will be projected on it. In this case, we will choose the public model for MPC and we will validate 5 samples: one of them is correct while the other 4 correspond to the altered samples. After loading the information, in the upper part of the screen, it is shown the projection of the samples on the qualitative model. The lower part of the screen displays a table in which, among other data, the conformity or non-conformity of the validated samples is indicated according to the similarity of their spectra with respect to those that were already part of the model. As it can be observed in this case, Chemometric Brain is able to identify the first sample as compliant or correct, while it detects that none of the altered samples is valid, marking them with a red cross in the table and thus indicating that they cannot be identified as valid batches of MPC 92/8.

Example 2

Another application of this section of the software is the detection of errors in the company’s production process. This is a real case that occurred at the facilities of one of our customers. This customer was producing an energy drink and, after carrying out the routine quality analysis of the manufactured batches using NIR and Chemometric Brain, they detected that one of the batches was non-compliant. As shown on the screen, the software clearly identified that the batch was invalid. After investigating what could have happened, it was concluded that not all of the ingredients in the formula had been added to the product and that one or two of them may have been missing, based on the

batch weight. The customer suspected that these ingredients could be taurine, maltodextrin or both of them, so they took again 3 samples from the defective batch and added taurine, maltodextrin and both components respectively. After preparing and measuring the NIR spectra of these new samples, they were projected onto the qualitative model. The analysis carried out by Chemometric Brain indicated that the ingredient that was not added to the original batch was taurine, since when this component was added to the sample, its fingerprint was similar to that of the batches already included in the model. On the other hand, samples to which maltodextrin or a combination of maltodextrin and taurine was added were still not correct. Once the missing ingredient was identified, it was added to the whole batch and a NIR measurement was taken again. As it can be seen on the screen, the batch was now identified by the software as valid. Therefore, thanks to our software, the customer was able to detect a problem in its production process, identify in this case the missing ingredient and correct the batches manufactured, thus avoiding the complete discarding of the manufactured batch or the possible economic losses and claims for releasing an incorrect one.

Example 3

With Chemometric Brain it is also possible to work with qualitative models that include different products. An example of this is the model shown below. In this case, we have a model that includes batches of Whey Protein Concentrate (WPC) at different concentrations: 60%, 70% and 80%. If we have a WPC sample of unknown concentration, our software will project the sample on the model and identify to which class it belongs and, therefore, which WPC concentration it has. As an example, we will load a sample of WPC 70 on the model. As we can see, Chemometric Brain has not only identified this sample as WPC 70, but it has also indicated that the sample is valid with respect to the rest of the batches included in the model. If we load a sample of a different WPC concentration, such as WPC 80, Chemometric Brain projects it onto the model and identifies it in this case as WPC 80, as well as validating the conformity of the batch with respect to the information previously included in the model.

Example 4

This type of qualitative models with several classes offers a large number of applications that can be implemented in the quality department of any food industry. In the following example, we will see how it is possible to establish different quality ranges for a product. We will use a model for mozzarella cheese in which we have previously established 3 classes according to the quality of the batches analyzed and then we will load a set of samples that we want to classify according to their quality. After performing the analysis and the projection of the samples comparing them with the information of the model, the software offers us the classification of the different batches. One of the samples is identified as type 1, two of them as type 2 and another as type 3, all of them being valid when compared with the samples previously included in the model. However, we also

find that one of the samples does not fit in any of the established Mozzarella cheese types based on quality and is identified as a non-conforming lot. This could be due to different factors such as an error in the manufacturing process of the batch or a problem with the raw materials used. Though to the use of our software, the customer can clearly identify in this case the incorrect batch and proceed to the detection of the fault in the production line.

5. Quantitative Analysis

The next section of the Chemometric Brain software is called Quantitative Analysis. From this tab, it is possible to determine the composition of your raw materials and final products, instantly obtaining information on a wide variety of physicochemical properties such as moisture, protein, fat, pH, dry matter or viscosity, among others.

We have a large number of calibration models that can be used for your NIR spectra, predicting the value for each of the physicochemical properties of interest in just a few seconds.

One of the features of this section of the software that distinguishes us from the rest of our competitors is that the calibrations for the quantitative models in our software do not need to be updated periodically by the manufacturer but are updated automatically. In this way, the user just has to incorporate the reference values collected to their NIR spectra.

Example 1

Performing the prediction in our software is very simple. First, select the calibration model you want to work with. Chemometric Brain enables to load several calibrations at the same time, making the process much faster.

If the option “perform validation” is chosen, the regression lines and the information about model validation can be viewed by exploring each of the tabs associated with each physicochemical property selected.

After the calibration has been loaded and the validation has been performed, the software is ready to start predicting. Simply load the spectra of interest and you will instantly obtain a value for each of the chosen properties.

Example 2

If, on the contrary, you do not want the validation to be carried out and you do not need to see the details of the linear regression, you can choose the «No» option in the pop-up window after loading the model and these steps will be skipped. In this way, the software will only perform the loading of the selected models and will be ready to start predicting. The prediction results obtained will be the same, but validation will not be performed for each property.

6. Ingredient Analysis

Another tab available in Chemometric Brain is Ingredient Analysis. thanks to the Artificial Intelligence technology used, it is possible to determine the ingredient composition of a product and the proportion of those ingredients in the blend. The system performs a learning process by analyzing the thousands of data available in the database about ingredients, their percentage in different products and their response to NIR radiation. After this process, the software is able to make a prediction of the composition of any spectrum.

Once the data training has been carried out, the prediction process can start. Firstly, it will be necessary to load the spectra of the samples whose composition is to be known. After that, the model that will be used to carry out the prediction must be indicated.

Once chosen the model and after clicking on «Predict», the software will start working on the prediction of the composition of the selected sample and it will be displayed on the screen as soon as the process is finished.

As it can be seen, the software shows not only the components that are part of the blend, but also the proportion in which each of them is the product.

This methodology has also been successfully applied in the identification of microcomponents and the determination of their proportion.

‍