XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE Smart Fuzzy Cupper: Employing approximate reasoning to derive coffee bean quality scoring from individual attributes Javier Livio Computer Science & Software Engineering Auburn University Auburn, USA jal0072@auburn.edu Wilfredo C. Flores, Facultad de Ingeniería y Arqui- tectura, Departamento de Inge- niería en Energía Universidad Tecnológica Cen- troamericana (UNITEC) Tegucigalpa, Honduras, 11101 wilfredo.flores@unitec.edu.hn Rania Hodhod TSYS School of Computer Science Columbus State University Columbus, USA hodhod_rania@colum- busstate.edu David Umphress Computer Science & Software Engineering Auburn University Auburn, USA david.umphress@auburn.edu Abstract—This paper presents a fuzzy expert system, an enterprise system designed and developed under the category of software as a service (SaaS) to grade specialty coffees from several countries. The system uses approximate reasoning and inner libraries to dynamically construct fuzzy rules, making the system capable of learning as cupping data flows through it. The coffee individual attributes' scores are linguistically expressed through sliders optimally designed to ease data gathering, encouraging the coffee judge to use words instead of numbers (low, medium, high and very high). Results from testing the system show more than 95% of matching results compared to the experts’ evaluations. Keywords—Approximate reasoning, expert system, fuzzy expert system, fuzzy set theory I. INTRODUCTION Coffees sought for their superior and consistent aroma and flavor "Specialty Coffees" have become the fastest growing segment of the coffee industry, becoming an economic driver in their own right. These command a 37% market share of the coffee market in the United States alone, estimated at $30 to $32 billion dollars [1], [2]. As the per-cup revenue generated by specialty coffee is generally more than that generated by commodity coffee, how beans are graded is not only a taste issue but a financial one as well. The absence of defects is paramount when designating a particular type of coffee as falling into the "specialty" category. Quality is determined by physical appearance of the beans as well as by the sensory experience produced by the end product itself. While the former relies on visual inspection, the latter rests on the ability of professional coffee judges (referred to as “cuppers”) to detect flaws. Cuppers follow a predefined protocol, such as that of the Specialty Coffee Association (SCA), to sample the coffee for bitterness, harshness, sourness, and other off-putting tastes or aromas. The SCA requires a sample of a specific coffee bean be judged at least fifteen times by professional cuppers before being receiving a grade. The SCA remarks that “The purpose of this cupping protocol is the determination of the cupper’s preference” [3]. This indicates that it is up to the cupper, acting as the domain expert, to transform qualifiable properties into a quantity. Cuppers take notes of coffee samples using natural language, then transform the notes to a numerical value using a standard form as the shown in Fig. 1. Fig. 1. Excerpt of the SCA Cupping Form Used to Grade a Coffee Sample [3] Establishing the numeric grade is currently a manual pro- cess, but the essence of the process suggests that it could be assisted by automation. Specifically, fuzzy expert systems can be used to support the cupping process as they allow for linguistically model the judge’s perception of the coffee’s attributes. Fuzzy expert systems aim to “model the world in terms of the semantics associated with the underlying variables, thus providing a much closer relationship between real world phenomena and computer models” [4]. In this work, a smart fuzzy expert system has been de- signed and developed leveraging the expertise of the coffee judges. The system is architected as a Software as a Service (SaaS) in that it provides functionality from a remote server to a client through a web browser [6]. The development of this system was not driven by what the cuppers needed to judge the coffee bean attributes, rather it presents a completely new way of recording and reporting the judge’s findings, currently based on the SCA cupping form. The fuzzy expert system promotes the replacement of numerical values to express the individual attributes’ scores with a process based on selection of linguistic terms. Nonetheless, the final grading of the coffee bean quality is not the arithmetic addition of the individual scores (as the SCA protocol indicates); it is the result of an approximated reasoning underlined by a Mamdani fuzzy engine. The document is organized as follows: Section II introduces fuzzy logic; Section III describes the coffee bean attributes used by the SCA; Section IV illustrates the proposed Fuzzy Cupping System; Section V shows the results of system 978-1-5090-6020-7/18/$31.00 ©2018 IEEE