The interest in renewable fuels created from crops has skyrocketed worldwide. Biodiesel producers are required to maintain stringent control of their process to ensure finished product meets or exceeds the quality parameters set forth by U.S. and European standards-setting organizations (ASTM D 6751 and CEN 14214, respectively). Most biofuel producers want the ability to conduct quality testing during the manufacturing process rather than risking unnecessary waste by testing after the fact.

Until recently, biodiesel producers had no real testing options available other than wet chemistry reference methods such as gas chromatography, Karl Fischer moisture analyzer and other physical test equipment. These methods can be costly, time-consuming and skill-intensive. The growing demand for alternative fuels, and the need for more streamlined production processes, has spiked interest in near-infrared instrumentation as a more efficient and less costly analysis technique for improving quality and decreasing out-of-spec product.

Near infrared uses a portion of light energy to determine molecular composition in a wide variety of materials. Unlike traditional reference methods, near infrared can analyze samples in seconds, rather than hours. Near-infrared devices use wavelengths of 800 to 2,500 nanometers. Throughout these regions, the near-infrared energy interacts with carbon-hydrogen, nitrogen-hydrogen and oxygen-hydrogen bonds in the sample, producing a "fingerprint" spectrum which can then be correlated to traditional reference assay results.

Near infrared is an indirect measurement for qualitative and quantitative analysis. It is necessary to have a calibration which mathematically relates the spectra measured by the near infrared to a reference measurement such as from gas chromatography. Once a mathematical model is developed and implemented, samples which are similar to the model database will then be predicted accurately for future measurements. A wavelength range of 1,000 to 2,500 nanometers is typically used for biodiesel analysis as there are strong features throughout this range.

Most traditional reference instrumentation must be kept in a stable laboratory environment, away from the process line. New near-infrared instrumentation is much more rugged and flexible, even fully portable in some instances. Therefore, it can be used wherever the analysis is needed. Near infrared can even be configured as part of an online system with a probe measuring constant changes in a process, providing real-time information for better decision making to help reduce cost or time. The degree of operator training required to analyze normal samples is much lower with near infrared than for traditional reference methods. This means that instead of needing a laboratory technician to process samples with reference assays, a production worker can successfully perform the analysis at-line or in-line using near infrared.


Three typical B100 spectra are shown in a transmittance spectral format.
SOURCE: ASD INC.

It is important to note that while there is no set number of samples required for a calibration model, creating a robust and accurate model does greatly depend on the diversity in the raw ingredients and other inputs, variation from the process, and differences in the primary reference data all being factored into the process. A reasonable plan might include 50 to 100 initial samples followed by routine additions of samples over a two-year period.

A common question seems to be "Can I just use an existing calibration or spectral library from the instrument manufacturer rather than building my own?" It is extremely important to realize that this approach is only plausible if the near-infrared instrument set-up, including the sampling accessory, is nearly identical. Equally important is ascertaining that the samples in the existing library are the same, or very similar to, the samples at the new installation. The best strategy is to treat any existing library or data set not directly built using intended samples in the intended location as a starting point only, with the understanding that it may be necessary to add additional samples with the new inputs from the new location.

Building a Calibration Model
Once the testing points, or locations in the process, have been determined, it is necessary to build the calibration model. This process does not need to be difficult or confusing once clear steps are identified. The key challenges in the creation of a robust and accurate calibration are 1) selection of a diverse set of samples, 2) obtaining accurate and reproducible reference assays, 3) calibration equation development chemometric modeling tools (many commercial software programs are available for creating near-infrared calibrations), and 4) validation and maintenance of the model.

Traditional chemistry methods can be expensive, so it is important to maximize efficiency and create the best possible calibration. In practice, four techniques are generally used to select samples for calibration development-random, composition-based, spectral-based, and a combination of the first three methods.

With random selection there may be samples which have already been characterized by the traditional reference methods, and use of these characterized samples is the optimal use of the previous expenditure. Biodiesel, however, changes over time, so minimal time between production of the sample and the spectra collection on the near infrared is important. With this approach, however, the range of composition may not be adequately represented by the existing samples, and many of the samples may be nearly identical either spectrally or from their composition. Additionally, if data hasn't already been developed for these samples, it is also the most expensive approach.

The second approach is to use known information regarding the samples as a selection criteria. This approach is excellent for covering the expected range of composition, but can also be expensive. In this scenario, many samples would be expected to have similar compositional results so they wouldn't be used. This approach is frequently used after a base calibration has been developed with random samples. The predicted values are then used as a screening tool to submit additional samples for reference assay.

The third approach evaluates the near-infrared spectra of new samples to determine which samples should be analyzed by the reference assay method. The spectra are examined with chemometric tools to structure the spectral population to cover the greatest diversity. Samples selected by this process are then submitted for traditional reference analysis. This approach is the most cost-effective, and it can potentially produce the best calibration in the shortest amount of time. This approach, however, is also the most labor intensive because all sample spectra must be reviewed.

Typically, most people use a combination of all three techniques to build their initial calibration. This begins with recently analyzed samples from the routine quality control testing to build a base calibration, then add samples that are at the extremes in composition, and finally, select samples based on the spectral characteristics. The samples in a calibration should also be actual production samples rather than a blend of two or three samples. Calibrations based only on limited numbers of actual samples will have a high correlation, but frequently will also have a higher percentage of analyzed samples outside of the calibration model, thereby resulting in higher error than anticipated.

Achieving Accuracy
To achieve accuracy, the samples used for model development must be adequately representative of the actual product to be measured. For example, if the raw input is soybean oil, then the calibration should also use similar soy oil spectra. If animal-based fat will be used then it is important to have the same or similar animal-based fats in the original calibration set. Each raw material will have an impact on the finished B100 characteristics, so matching the calibration to the inputs used is extremely critical. Additionally, different processing conditions of the plant used to create the original calibration may also cause disparities in a calibration until the unique samples at the new location have been added into the model.

It is critical to use a laboratory which is accustomed to producing reference assays. It is also important to know the error associated with each reference assay. Previously analyzed samples and/or blind replicates (the same sample with a different identification number) should be submitted periodically to the reference lab to determine this error.

Validation of a calibration model is necessary whether using a new or existing model. This can be accomplished while building the model simply by holding back a portion of the sample data and then using it for validation.

Following validation, the model is ready to predict the actual process samples. It is important to note that near infrared does not replace the need for a traditional reference method. It does, however, greatly decrease the frequency that the expensive reference assay is required. The calibration model must also be updated anytime a change occurs in the incoming feedstock, raw materials, processing conditions or instrument set up. Additionally, periodic validation will help to ensure that the model continues to be accurate.

Near infrared is a rapid, accurate, non-destructive, cost-effective, user-friendly tool for quality and process control. It provides the ability to conduct more frequent and comprehensive testing with portability, flexibility, and a lower level of skill and training requirement. By strategically implementing near-infrared technology, biodiesel producers can greatly increase their potential to not only meet the stringent quality standards, but also increase their bottom line by significantly reducing operation costs and the risk of out-of-spec product.

Dan Shiley is an applications chemist at ASD Inc. in Boulder, Colo. Reach him at (303) 444-6522.