In the know with Near Infrared

Varying quality in, consistent quality out. The fundamental challenge of grain handling has never been more apparent as the industry strives to ensure deliveries on spec while making the best of grain supplies under threat from extreme weather and ongoing geopolitical events. 

Introduced in the nineteen eighties, near infrared analysis has proven increasingly valuable to the business of grain handling. When deciding payment at delivery, when blending to customer specs, when loading onto transport, when it’s received for milling or malting and when it’s traded on global commodity markets, the ease of use and reliability of near infrared analysis has ensured that grain handlers can deliver higher quality products from the same finite supply of grain from the field. 

So what has made near infrared analysis of grain such a powerful tool and what should we look at when deciding analytical operations for the future? 

Looking inside the kernel – the evolution of near infrared transmittance

The changing face of NIR grain analysis seen through the well-known Infratec™ grain analyzer

Data collection lays the ground for a growing range of applications
The first applications for grain analysers based on NIR transmittance were for wheat, corn, barley, soya and rice for parameters such as moisture, protein and oil content. 

Approvals quickly followed, for example, by organizations such as the Federal Grain Inspection Service (USA). At the same time, the development of more applications models gathered pace. Note that the models are often referred to as calibrations and are also now known as ‘analytics packages’ in relation to FOSS solutions.

In simple terms, an application model converts the spectrum measured by the instrumentation into a usable value for the end user.  A critical aspect of this work is to have enough data and a sufficiently varied range of data to make a valid application model that can handle the natural variations in characteristics of grain samples across different regions and harvest seasons. The world is a pretty big place when it comes to growing grain and growing seasons continue to surprise with ever more extreme weather events. In short, the more data that can be put into a model the better so that we can accurately describe the many variations that we can expect to find in a single grain sample. 

In 1996, a powerful form of application modelling called Artificial Neural Network models (ANN) was introduced to handle the range and complexity of data. The ANN models have contributed to highly stable performance regardless of weather and region. Such performance has been demonstrated by years of ring tests against results from reference laboratories. Today, well-established data models for the latest generation Infratec instruments contain as many as 50.000 samples or more representing over 35 years of seasonal variations. The range of applications continues to grow as shown in this fold-down table: 

Fleets of high performance instruments
While the stability and range of individual instruments evolved, many grain handlers found that by linking analytical instruments in grain networks they could gather valuable data from multiple analysers in one place. A while later, networking software was developed that not only allowed grain handlers to collect data, but also to remotely configure instruments, for example, when updates to application models became available to accommodate the latest growing season, they could be pushed out to multiple instruments in one go from a single desktop. 

As anyone who has tried looking after a number of instruments without such connectivity will testify, keeping them all in check and up to date, can be a time-consuming task, especially across different geographic locations. The ability to do it once from a desktop can be safely said to have saved thousands of man-hours.

Reliability is the sum of many things  
In step with developments in networks, the reliability of results on an instrument-to-instrument basis was also greatly improved in the late nineties. This brings us to a final ‘need to know’ concept called transferability. 

Transferability means that measurements done with different instruments on the same sample give identical results. An indication of transferability therefore helps users of the NIR analysis equipment to understand the reliability of the measurements popping up on the screens of the instruments in use across the organization. 

Transferability is affected by factors related to both instrumentation and the application model.

On an instrument level, the repeatability of measurements, the accuracy of measurements and comparisons from one instrument unit to another are important. On an application model level, the variables can include factors such as the number and source of NIR measurements used to create the model, not to mention the number of reference tests and potential error between those reference tests. 

The application model makes a ‘prediction’ of the result based on the data used in the model. The term, ‘standard error of prediction’ (SEP) is therefore used to sum up the potential error against reference results determined by the actual chemical properties of the sample. SEP is typically stated in technical documentation such as applications notes. Another useful tool in any application note is a graph showing results from the NIR instrument plotted against the corresponding reference results for a number of samples. This provides a source for performance-related statistics, for example, SEP, see figure 2.                   

The topic of SEP can easily become rather detailed, but what is critical for your operations is that you have a correct and reliable statement about what to expect in terms of potential error. Imagine, for example, that you are running a population of instruments across different sites and that you are grading grain for payment according to a target of 10.5 protein. Firstly, it is important to know that the instrument is correct and measurement is correct and the potential error is low. Secondly, it is essential that the performance is consistent wherever and whenever measurements are made. Only in this way can you ensure that payment tests will be reliable and trustworthy across all sites.