Your own Calibration Model

What are the first steps in developing a custom calibration model?

Initially, we request a few samples of the specific product. These are measured on relevant instruments to see how the process goes and whether this type of sample can be measured with NIR. Additionally, we check if a suitable calibration model is already available. If not, a new calibration curve will need to be created. In that case, the process involves collecting representative samples and obtaining spectra and reference data. Afterward, a calibration model can be developed.

What criteria are used to determine if a product is suitable for NIR analysis?

Several factors are important for this. First, the product must absorb light in the NIR region. Additionally, the homogeneity of the product (where the measurement needs to be taken) is important. There also needs to be sufficient variation in available samples for the reference parameter (to create the calibration model), and the desired accuracy must be realistic and achievable.

How many samples are typically needed to develop a calibration model?

This depends on various factors such as variability, homogeneity, and the reference analysis. Generally, around 50 samples are needed to ensure a reliable and robust calibration curve.

Are there specific guidelines for sample preparation (e.g., homogenization, drying, freezing)?

NIR can be applied to different types of samples if the calibration curve has been developed for this purpose. It is possible to measure dried or wet, ground or unground samples, but this does affect accuracy.

Do the samples need to have a certain variety of properties (e.g., moisture content, density, etc.)?

When building a calibration model, you ideally want samples with as varied properties as possible, while still being realistic for the application. Additionally, wetter samples are often more challenging to measure accurately, as the moisture signal is very strong in the NIR region and can somewhat overshadow other signals.

Are there parameters that are difficult or impossible to measure with NIR technology?

The molecular bonds with the strongest absorptions are OH, CH, NH. These are the key bonds in many organic molecules/parameters. Other (inorganic) bonds often have little or no signal and are therefore challenging.

Which specific components or properties of a product are challenging to measure accurately with NIR?

The more challenging components are those with low concentrations or little variation (this also depends on the product). Additionally, it is difficult to distinguish between compounds that are very similar.

How long does it typically take to develop a custom calibration curve?

In practice, this often depends primarily on the collection of a representative set of samples. If these (or relevant data) are available, calibration curves can usually be created and delivered within a few weeks.

How often does a calibration model need to be updated or recalibrated?

Due to the emergence of new variations in conditions, products, seasons, etc., a calibration model is almost never complete. Often, several updates are carried out within a few months after the initial development phase. After that, minor adjustments may be made a few times a year.

What are the main data parameters collected and analyzed during the development of a calibration model?

There are many statistical parameters that can be used to describe the accuracy of a calibration model, and no single parameter can do this completely on its own. It is usually a combination of parameters that visualize the expected prediction errors (RMSE, SEP, bias), the sensitivity of the calibration curve (slope, R2, RPD), and the range or spread of the calibration curve (min, max, standard deviation).

How is this data used to verify the accuracy and reliability of the calibration model?

The models are optimized to make these parameters as accurate as possible. These parameters are then determined and recorded for each model using independent data to get the best possible picture of real-world performance. We analyze and evaluate these based on our experiences in similar situations.

What steps are taken to ensure the accuracy and reliability of a calibration model?

In the lab, we do everything possible to ensure the underlying data is of the highest quality for the calibration model. This includes the control and maintenance of instruments and the environment, training of personnel, and the documentation and optimization of procedures. The development of the calibration model is carried out using our self-developed algorithms, which we have extensively researched and continually strive to improve to align better with practical applications and increase efficiency. We also strive to assist our clients with their measurements, and thanks to the NIR-Cloud, we can monitor what happens at our clients' sites. This often allows us to detect problems before clients even realize something isn't functioning properly.

How reliable are the measurements for these specific parameters compared to traditional laboratory methods?

This varies by product and reference analysis and depends on several factors. For example, the homogeneity of the product and the accuracy of the reference analysis. Generally, NIR adds a small margin of error on top of the errors inherent in wet chemistry methods. In principle, NIR can even be more accurate than the reference methods (by averaging the errors in wet chemistry within a calibration model), but this is nearly impossible to prove since wet chemistry remains the reference method.

Can custom calibration models be used for multiple product types, or are they specific to one product?

In general, we try to make calibration curves as specific as possible for products, as this usually yields better quality. It is possible to create a combination for different types if the products are sufficiently similar. We usually assess this on a case-by-case basis to determine whether it is feasible.

What are the most common challenges in developing calibration models for specific products?

For us, it is an exciting challenge to gain product knowledge. We strive to get to know new products well to clearly understand what they are and where the variation lies or originates. This way, we can define the product clearly, so we know what should or shouldn't be included under one or more calibration models. Then, it is essential to ensure that all variations are included in the calibration model. This is a new challenge for each situation.

How can companies measure the ROI of investing in a custom calibration model?

NIR can provide cost savings in various ways:

• As an incoming inspection for raw materials
• To steer and/or optimize the process (more quickly)
• To control/improve the final quality
• As a replacement for more expensive lab analyses

The specific calculation can therefore vary greatly depending on the situation.

How does NutriControl collaborate with clients to develop custom calibration model?

In general, we have a close collaboration where we rely on the client's product knowledge to develop the best calibration models. The client provides the samples for analysis, and in consultation, we often try to add NIR measurements on-site to the calibration model. This way, we work together to deliver the best possible calibration model. During this development phase, there is contact several times to discuss the status and whether any specific actions are needed to optimize the calibration model. Once the calibration curve is delivered, it is closely monitored, especially in the beginning, and several updates are often made based on the experiences.