So, you think you’re ready to Calibrate! Have you considered………. Permissions and Acknowledgements We are very grateful for the kind permissions from the Australian Near Infrared Spectroscopy Group (ANISG) and from Mr Ian Murray, Scottish Agricultural College, Aberdeen, Scotland in allowing us to summarise and re-present much of the course Mr Ian Murray put forward to the ANISG 2004 Conference (Sampling, Sample Preparation and Sample Presentation). This document contains only the primary points raised by Mr Murray. If you are interested in exploring the topic in detail you can contact ANSIG (www.anisg.com.au). If you would like to join a community of NIR professionals in Australia we highly recommend membership of the Australian Near Infrared Spectroscopy Group (ANISG). One of the most frequently cited benefits of analysis by NIR is little or no sample preparation, however this is seldom strictly true. Often to get the best out of NIR analysis sample preparation is needed. The challenge is to use a sampling protocol to acquire a representative lab specimen out of a lot, conduct sample reduction to get several consistent test portions and present these to the instrument to get its spectrum and therefore deduce its composition or quality and fitness for the designated purpose. Often it is necessary to impose sample preparations like drying, grinding and mixing between sampling & sample presentation. Sample preparation may be unavoidable and may be required to stabilise the product for storage, to remove bulk moisture, or to render a specimen friable and to grind it to a powder that is more easily mixed and rendered more homogeneous. However sample preparation and reduction to the test portion must strive to retain sample integrity and form consistent test portions. Even if sample preparation is avoided you must never forget that NIR is a secondary method that requires calibration to a primary reference method that in itself almost always requires sample preparation. Given the challenge of acquiring a representative sample out of a lot there are a number of key considerations to achieving this: 1. Large test portions are required- The deliberately oversize sample mass needs to be reduced to lab sub-samples & retained specimen retaining the sample integrity and ensure that each sub-sample is consistent. Any subsequent preparation such as drying & milling must strive to retain the sample integrity. If moisture is removed by drying this needs to be measured & recorded. Reference analysis that test very small test portions is inherently insecure because they inflate sampling error. The size of the test portion taken for analysis either by NIR or by a chemical reference method has a profound effect on performance of a sampling & testing operation. The ‘best’ methods of analysis in terms of those having the smallest coefficient of variation CV% are those that take large test portions and have few manipulative steps. 2. Many RANDOM increments from lot & composite are required- Random Sampling is the most important and yet the most frequently overlooked attribute of sampling protocol. www.UnityScientificAsiaPacific.com Page 1 of 3 Random sampling means that every particle in a lot has an equal chance of being selected for test (this is distinct from haphazard sampling). If selection is truly random then the very well defined laws of probability will apply. The big advantage of this is that the results will be unbiased and will also give a good estimate of the precision of the result. Also note that if any part of the sample is deemed “foreign” its removal will not uphold the integrity of the sample. Rocks, rubber gloves etc may have to be removed but any loss of dust, chaff, broken seeds, and weed seeds will violate the integrity rule. Sampling error from grouping & segregation can be reduced by taking many small sample increments & combining them into one composite sample for testing. But, combining many small sample increments into one composite sample eliminates the opportunity to conduct a sampling survey of each increment to show heterogeneity. Remember that an advantage of NIR testing is that it permits easy and quick sample survey. 3. Grind before sampling the test portion- Article size, shape, roughness and density cause segregation and lost sample integrity. As you grind material it looks paler in colour. As the particles get finer the mean path-length or depth of penetration, gets shorter so selective wavelength absorption gets less and the specular reflection gets proportionally larger. Reflectance offers no strict operator control over (mean) optical path length traversed or depth of penetration of photons. This is automatically set mainly by two factors: Particle size and Refractive index of particles relative to that of the pore space. 4. Mix well, avoiding segregation & loss of sample integrity In discussing sampling, consideration also needs to be given to sample storage and preservation. All samples need to be contained, labelled, transported & eventually stored, however some samples can be unstable with temperature, affected by oxidation, mould and mites. Remember also that water is a gas and it will disappear by evaporation! After you have completed scanning your samples, always store them in airtight and vermin proof containers. Samples should ideally be tested in duplicate to detect gross errors. There are three broad categories of errors that can occur in your measurements: 1. Gross Error- Gross errors are avoidable. These are ‘mistakes’ due to mis-reading an instrument meter or a graduated scale; failing to zero a balance before making a weighing; a transcription error in writing down the wrong number; mis-calculation comes into this category as well if for example a numerator gets mistaken for a denominator. Duplicate tests usually spot these gross errors. 2. Random Error (unexplained)-This type of error is unavoidable. It occurs in all measurements at some level & arises from random unexplained fluctuations both positive & negative about the mean, most probable value. Random error is usually normally distributed and can be expressed as a standard deviation or root mean square (rms) value. 3. Systematic Error (bias)-This is when a set of measurements all read slightly higher or lower than the correct or expected value. If the expected value is unknown it is impossible to detect. Only by using a CRM - Certified Reference Material or a true primary standard substance is it possible to detect bias. Also consider the effect of oil on the reference tile of the NIR instrument which will (for instance) produce low results for oil in a sample. www.UnityScientificAsiaPacific.com Page 2 of 3 A set of replicate samples will also give a measure of your precision as a SD. Precision (or reproducibility) is the closeness of the repeated values to one another. If your random errors have been small then the results will be precise, however you would have no idea if the result was accurate or if it is was biased high or low. The accuracy of your results is the closeness of the results to the true or correct result. The accuracy of your measurements can be affected by systematic errors or bias. If you have access to a CRM (Certified Reference Material) you are then able to run the CRM as a sample to see how close you can get to the ‘official’ result. An oily tile may also impact on your expected results. The sample presentation method of some NIR instruments may result in the position of the reference tile in the transport mechanism being directly below, and in contact with, the base of the cup sample cell. Leakage of oil from oily samples, or mere contact with spilled oily samples over time can and does contaminate the tile. The effect of an oily tile on an analysis is that it can reduce reported oil content significantly and because calibration is multivariate, not just oil analysis is affected; protein estimates (for example) can be increased. An oily tile matters! All NIR labs should have a standard even if they cannot afford an authenticated reflectance standard. Dependent on your instrument consider using Rare Earth Oxide or Polystyrene as a very good reference standard for NIR. Reference standards provide the user with an assurance of wavelength accuracy and predictable levels of absorbance. Listed below is a recommended sampling plan for NIR & reference analysis: 1. Identify the population to be sampled: random or structured; 2. Use random numbers to locate sample increments or frequency; 3. Decide mass, number & frequency of selected samples: composite/survey? 4. Tools, containers, storage, labelling, chain of custody, foreign matter? 5. Sample reduction to test portions: avoid segregation & lost integrity; 6. Preparation, drying, milling, mixing (the objective is to improve scan homogeneity; 7. Scan immediately prior to reference analysis to minimise moisture loss; 8. Use H statistic to select samples for reference analysis; 9. Presentation, rms of reps. Is the scan area adequate? (check instrument) 10. Model & Cross validate- independent trial on ‘open’ future sets. www.UnityScientificAsiaPacific.com Page 3 of 3
"The Importance of Accurate Sampling in NIR Analysis"