Engineering R&D: AIB’s smart mixer project
Lab tools like the farinograph and mixograph help visualize dough development and breakdown, but they are of no use in making real-time adjustments, resulting in considerable product waste. NIR spectral data, on the other hand, can give relatively unskilled operators real-time information about dough constituents such as protein, starch and lipids in the presence of water. Hardware improvements are helping to move NIR from the lab to the production line, provided the software-based algorithmic language to correctly interpret wavelength data is in place. Last year, Richard E. Dempster stepped in to oversee the American Institute of Baking's smart mixer project, which began in 1998. Dempster developed matrix algebra that contains seven equations, five relating to ingredients and the others to temperature and dough development. He also is addressing the engineering issues in bringing NIR technology to commercial mixers.
An electrical engineer and member of the American Association of Cereal Chemists, Dempster spent 12 years with the USDA's Agriculture Research Service Grain Marketing and Production Research Center. During that time he developed machine vision algorithms to identify defects in wheat and corn. He joined AIB in January 2001 as director of product and technological development. Food Engineering recently spoke with Dempster about the NIR project.
FE: NIR has been around almost 40 years. What technological improvements have made it feasible for these applications now?
Dempster: The diode array spectrometer is the biggest advance. It enables you to get multiple samples per second, which improves the quantity of data you're gathering. That's essential for near real-time solutions. Filter-based rapid analysis also has greatly improved.
NIR has been used for years by millers to scan wheat and flour to determine moisture and protein content. That should ease the transition for applications like dough mixing. The most important change that has occurred has not been technical improvements but in the acceptance of the science. This will help pave the way for even more NIR applications in processing.
FE: How does NIR compare to measurement of power consumption or torque in monitoring dough development during mixing?
Dempster: It has not been established that it is superior, but I get a solid peak development curve with a near infrared reading, compared with a rounded curve when you monitor machine torque. I've encountered situations with certain varieties of flour where the NIR readings were quite a bit different from what a commercial torque reader was showing. As it turned out, my algorithm was correct.
Peak development is a pretty big bull's eye, though, so precision isn't too critical because there's a tolerance zone. The really big advantage with NIR is that it also looks at the ingredients in the mix. With NIR, you're constantly measuring chemical composition, so there are ample data on which to base corrective action before peak development occurs.
FE: How has software development progressed?
Dempster: Developing the algorithm has really been the heart of the project, and our lab results have provoked a lot of interest in the application. In lab tests using a spiral arm mixer with a fiber-optic probe at the base of the mixing bowl, I can measure 2.5 samples a second. The next step is to see if it will work on bigger mixers.
FE: What have some of your NIR tests shown?
Dempster: If you overmix a dough, an old baker's trick is to add salt to extend the mixing tolerance. By using NIR to analyze the chemical changes that occurred when salt was added to different batches at various times during mixing, we were able to chart when optimum mixing occurred. The prediction was that it would be achieved in five minutes; if salt was added at the beginning, optimum mix would be achieved in 10 minutes. If added one minute prior to peak, a five-minute mix still would be achieved, cutting mix time in half.
With one variety of flour, the dough peaked at five minutes without any salt added. When salt was added at four minutes, the dough peaked slightly earlier, declined, but then demonstrated improved development again at five to six minutes. In another batch, salt was added at five minutes, resulting in an optimum mix that extended to nine minutes, well after initial peak was achieved and when overmixing was quite pronounced. The salt was added at seven minutes, well after initial peak was achieved and when overmixing was quite pronounced. The salt brought dough consistency back up considerably, though not quite to the optimum range, at the 11 minute mark.
If salt has that much of an effect, what's the impact of other minor ingredients? This opens up a whole range of potential analysis, and NIR spectroscopy is an effective tool for performing it. The accuracy of our readings is above 0.9 R-squared on sugar, shortening and salt. A reading of 1.0 would be perfect.
FE: What insights have you gained into blends of flour in a dough?
Dempster: Results can vary significantly, depending on the flour varieties used. One experiment dramatized the deleterious effect of mixing two particular varieties of flour. Dominator was expected to peak at about 18 minutes, while Coronado peaks at 12 to 13 minutes. But a 50/50 blend of these two flour varieties never achieved a real peak development time. Instead, there were two muted peaks, both less than optimum development, each foreshadowing the true peak time for the two varieties.
In another test using Dominator plus Tomahawk, better results were achieved with a 50/50 blend. Optimum mixing was achieved at a time interval very close to that of Dominator, though the peak period was considerably shorter, underscoring the need to conclude batch mixing at a more precise time than the more-forgiving dough using a single variety of flour.
The point of all this is that a great number of factors can impact the rheology of dough, and monitoring the chemical changes that occur during mixing is much more precise in controlling outcomes than observing physical changes. The use of NIR will enable operators with lower skill levels to produce more consistent products, batch to batch. That is something the industry sorely needs.
FE: What role do you envision for NIR in bakery production?
Dempster: The ultimate goal is to have the system control all the additives in a formula. After flour is put into the mixer, the system would determine the optimum amounts of sugar, salt, shortening and water for that batch. There also is a host of minor ingredients like ascorbic acid and L-Cystein that can be controlled with NIR.
Implementing a fully automatic system will happen in stages. Initially, the goal is to have a very simple system, where a light comes on when the dough is ready, and the operator then kicks it out of the mixer. As the system becomes better integrated with the rest of the plant, it can go as far as doing an initial analysis of the flour and determining when to add what quantities of the recipe's ingredients.