Challenges and opportunities in implementing static-optics FTIR in biofuels production

Manufacturers face a difficult juggling act to balance cost, accuracy and real-time measurement of their analytical instrumentation to optimise efficiency and therefore profitability of their processes. The myriad choices of analytical instrumentation can be overwhelming based on the complexity of the technique chosen and its ability to be implemented on or in-line to a process. This paper introduces static-optics FTIR, and briefly explains what makes it different to traditional approaches and some benefits for its use in biofuels production.

Why use static optics for monitoring biofuels production?

Traditional process sensors (parametric instruments) such as temperature probes, pH probes, flow meters etc. are relatively inexpensive, make continuous measurements and can be implemented across an entire manufacturing or production facility. They give real-time measurement of key process parameters and are a critical part of the regulatory control layer. While these sensors are more than adequate to control simple process conditions such as heat and mass balance, they do not allow measurement – or therefore control – of process quality. To truly optimise a process, its composition needs to be monitored in real time.
A solution to this in many petrochemical and oil and gas applications has been the implementation of near-infrared spectroscopy (NIR). Spectroscopy as a concept is significantly more informative than parametric measurements as it directly measures the interaction of light and molecules, meaning it is possible to calibrate for chemical concentrations. However, traditional spectrometers based on Michelson interferometers are fragile and contain sensitive moving parts, so must be kept far away from mechanical movement or vibration – which is unavoidable in production facilities. NIR light is suitable for fibre optics, which means the process probes can be positioned far away from the spectrometer itself, protecting the delicate optics. Unfortunately, NIR spectroscopy is very limited in the information provided, as it does not directly measure “fundamental transitions” in molecules but instead measures “combinations” and “overtones” – a bit like trying to count the steps in a staircase by running up them two or three at a time rather than singly. You can understand whether it’s a tall or short staircase, but you will not be able to determine the exact number of steps. To get accurate and complete measurements, a higher resolution technique would be a better choice.
In the laboratory, these sorts of precise measurements are made with Fourier transform mid-infrared (FTIR) spectrometers. Traditional FTIR spectrometers are incompatible with production installations because the fibre probes required are very poorly performing, relatively expensive and extremely fragile. This renders classical FTIR instruments unsuitable for on-line measurements, and is the principal reason why NIR is so ubiquitous in the refining world today. Recent advances in static-optics FTIR such as the IRmadillo replace the delicate array of moving mirrors with a fixed optical path (hence the term “static optics”), meaning the spectrometer can be installed directly into or onto the process of interest with no fibres – taking FTIR out of the laboratory and into the process. Image 1 shows the light path through the instrument, where an interference pattern is created on the detector, which is transformed into a spectrum using a Fourier transform.

The basics of instrument calibration – and its challenges

The difficulty with using spectrometers for process control has always been calibration – the spectral data needs to be turned into meaningful calibration data. This is easy to do because of Beer’s Law:
A = εcl
where A is the absorbance recorded in the instrument, ε is the “extinction coefficient” (which should be constant for a given feature in the spectrum), c is the concentration and l is the path length (broadly – but not always – a constant for a given installation).
The most common approach is to use chemometrics – also known as multivariate analysis – where the entire spectrum is analysed and compared to known reference values to build multi-dimensional regression lines. This approach is extremely powerful, and works for the vast majority of use cases (an example where it doesn’t work is given below).
Individual calibrations can be built for each chemical of interest, and are all independent from each other. Image 2 shows an example calibration performance for ethanol in a fuel ethanol fermentation process built using partial least squares (PLS) methodology. The black circles are points and data used to build the calibration whilst the red diamonds are points and data used to test it – having not been a part of the calibration dataset. This is a crucial aspect of calibration building: perform blind validation by predicting samples that were not used to build the model –
that is the only true way to ensure the calibration is not just “fitting noise”.
The problem, however, is that every instrument typically needs a unique calibration built for it, and “calibration transfer” approaches normally need a dataset of 10s of samples – almost as many as building a fresh calibration from scratch. This can make multiple instrument installations prohibitively expensive to implement, reducing the uptake of advanced measurement and control schemes. This is especially true with NIR installations, where in excess of 100 datapoints are typically needed to build calibrations – NIR spectra simply do not contain enough data to use fewer datapoints.
 

A new approach to calibration – how augmentation replaced transferring

A secondary benefit to static-optics instrumentation design is that instrument to instrument variation is greatly reduced. This means that the residual variation is small enough to be included as one of the “factors” within a calibration, and can effectively be calibrated away. This is extremely valuable in applications where multiple parallel processes are performed – for example, industrial fermentation where multiple vessels are performing the same process and should run the same calibration. Keit recently developed a new approach to calibration that takes advantage of just this benefit: an instrument was installed in a particular process using a calibration built on a different instrument, and additional data from the process was used to ‘augment’, or refine, the calibration. This is known as a hybrid calibration. Typically this instrument-specific data needs only to make up 1 – 10 % of the total dataset, dramatically reducing the amount of sampling required to ensure an instrument is calibrated efficiently.
An example of fuel ethanol production by fermentation is shown in image 3. The initial calibration was built over 6 months using process data from an installation in Europe. This calibration was then augmented with an additional 2 batches of data (16 data points) from an installation in the USA. The results are extremely promising with very close fits to the target calibration errors.
Keit also views this approach as the beginning of “pre-calibration” for the static optics FTIR, as the starter calibrations – whilst not 100% accurate – do in general show the correct trends for chemistry of interest. This is shown in image 4 with starter and augmented calibration results for a given ethanol fermentation displayed with reference data. The calibrations for the majority of chemicals are offsets which means they can be used for indicative performance (i.e. increases or decreases of concentration) until augmentation is possible. It is envisaged that by building hybrid calibrations of multiple installations even augmentation will not be required in the foreseeable future.

Where advanced calibration approaches are required – non-linearity and equilibria

It is obvious in the example above that acetic acid modelling is not accurate – even after augmentation. This is because of the way that off-line sampling, such as HPLC, does not always match the reality of on-line process conditions. Both acetic and lactic acids exist in equilibria with their salts, and the exact concentration differences depend on the pH of the process and concentrations of other acids. The spectrometer will observe both of these different chemicals with different spectral features – the C=O stretch in the acid occurs at 1720 cm-1 whereas in the salt it splits into 1420 and 1610 cm-1 features. However, when the sample is removed from process and analysed by HPLC the total amount of acid after quenching is reported. This means the assumption that a single chemical has a directly linear correlation with spectral features made above starts to break down, and calibrations using non-linear methodology such as locally weighted regression (LWR) and support vector regression (SVR) are required to accurately model these processes. An example of this is shown in image 6 – an acetic acid calibration using PLS and SVR principles. The PLS model struggles to accurately measure the acid whilst the SVR calibration does a very good job. The augmentation approach may also be suitable for non-linear calibrations, and subsequent work will establish its performance.

Additional applications in renewable fuels production

The use of static optics FTIR for monitoring, controlling and improving renewable fuels production is not limited to just ethanol fermentation. Example use cases and applications are:

Ethanol production:

- Early identification of fusil alcohols to prevent batch failure
- Optimisation of liquefaction processes
- Optimisation and control of fermentation
- Distillation control and efficiency improvements

Bio diesel production:

- Process measurement and control – such as catalyst measurement, free fatty acid (FFA) measurement and glycerol detection in finished product streams

Renewable diesel pre-treatment control

- Measure metals, P, FFA and TAN to control bleaching/pre-treatment and prevent catalyst degradation and damage (see example in image 7)

Conclusion

This work shows that by implementing a static-optics design instead of traditional Michelson-based interferometers (which rely on moving mirrors), it is now possible to use FTIR for in-line process monitoring and control reliably and robustly. The highest effort aspect of using spectroscopy for process control – calibration – has been made significantly easier through hybrid calibration, taking different datasets from multiple installations and instruments and creating an overarching calibration encompassing all of that variation. Because the level of information in FTIR (compared to NIR) is very high, and the variation between static-optics instruments is comparatively low, these hybrid calibrations are effective and efficient.