Bourbons are one of the fastest growing spirits in the US, and the estimated revenues from whiskey & bourbon sales were ~ $5bn in 2022. Many brands of bourbon are on the market, each with distinct olfactory and flavor profiles, chemical profiles, alcohol content, and prices. Bourbons are composed of a complex mixture of chemical compounds that contribute to a bourbon’s unique flavor and aroma. These chemical compounds are dependent on a variety of factors, including the bourbon’s recipe, the type of barrel used, the warehouse type, and the aging process. The distillation and barrel-aging of bourbons spirit starts with a mash bill (of at least 51% corn) that also contains various percentages of wheat, barley, and/or rye. After mashing, the mixture is fermented, distilled, and barreled in new, charred oak barrels for a variety of time periods. Following aging, some of the bourbons are blended or filtered, and bottled at > 80 proof.
Most bourbons are produced from similar grain bills, fermentation conditions, and distillation processes and are aged in comparable barrels in rick houses. So the vastly diverse noses and taste profiles of different bourbons are likely due to unique relative concentrations of the chemical molecules called congeners. Chemical analytic techniques like Nuclear Magnetic Resonance (NMR) and gas/liquid chromatography coupled with mass spectrometry (GC-MS) are tools that are employed to identify congeners and their concentrations in bourbons. In a recent publication with Professor Michael Crowder at Miami University in Oxford, Ohio, we applied NMR spectroscopy to identify and determine the concentration of 17 key congeners in bourbons produced from various distilleries. The NMR spectroscopy data obtained provided the quantitation of 17 congeners in the bourbon samples. However, these differences were insufficient to accomplish chemical finger-printing or distillery identification in the bourbon whiskey samples.
Artificial intelligence-based methods like machine learning are being used more and more in the broader food and beverage industry. In our study, we turned to machine learning to aid in chemical finger-printing the bourbons and identifying the distillery of origin. The machine learning method applied allowed us to understand the similarity across the relative concentration profiles and discriminate between bourbons in the study. In addition, we could accurately fingerprint and authenticate bourbon whiskeys using the machine algorithm. For example, we identified a counterfeit bourbon in one of the team member’s collections, including what lesser-priced bourbon was poured in the bottle.
By analyzing the multi-dimensional data from the recipe, distillation, barrel type, aging, warehouse, sensory profiling and tasting notes using machine learning methods, we provide insights into what bourbon to produce based on customer reviews. Using our propriety machine learning methods, we identify unique signatures for creating new recipes/processes for more personalized taste profiles. Our artificial intelligence methods can also be used to optimize production, provide better customer experiences, support sustainability strategies, and detect counterfeits.
Written by: Rajesh Naik, PhD