A Qualitative Learning System to acquire Human Sensory Abilities in Adjustment Tasks Francisco J Ruiz BarcelonaTech Vilanova i la Gelt´ u, Spain N´ uria Agell ESADE-Ramon Llull University Barcelona, Spain Cecilio Angulo BarcelonaTech Vilanova i la Gelt´ u, Spain M´ onica S´ anchez BarcelonaTech Barcelona, Spain Abstract Adjustment in creative processes is not purely a func- tional or a physical task, but arise from highly sub- jective preceptive and cognitive aspects which cannot be fully modeled by standard quantitative structures. In such tasks, the involvement of human experts becomes necessary, preventing the complete process automation. This paper introduces an innovative artificial cognitive system to support decision-making in adjustment pro- cesses based on human sensory abilities. The proposed system, based on expert knowledge management, draws on a machine learning tool jointly with an actions’ gen- erator module. The methodology proposed is applied to a real case study: color-adjustment in the automotive painting industry. Introduction Industries with specialized professionals who use their sen- sory abilities for designing, formulating and tuning their products, face with huge challenges when managing and dis- seminating these skills [23; 25]. In particular, perfume, food, beverage, painting, and other creative industries continu- ously deal with problems in modeling processes based on the cognitive ability of these highly specialized individuals [1; 2]. Creative processes are not purely functional or physical, but arise from highly subjective perceptive and cognitive as- pects which cannot be fully modeled by standard quantita- tive structures. In such tasks, the involvement of human ex- perts (for example, colorists, perfumers, chefs, sommeliers and brew masters), becomes necessary, preventing the com- plete process automation. In highly creative industries, one can distinguish two main design tasks: (1) the formulation task; (2) the adjustment or tuning task. The formulation task involves finding an appro- priate set of ingredients and their proportions to make the target product. Once the formulation task is completed, the product is ready to be manufactured. The adjustment or fine-tuning task is crucial during manu- facturing [5; 14]. This task must be performed whenever the product is nearly finished and has to be corrected to yield the right specifications. In the adjustment process, the ex- pert, using his sensory experience and abilities, makes slight changes to the proportions of one or more ingredients. The expert’s intuition does not usually allow him/her to know the exact quantities needed to attain the target result. However, he/she is able to iteratively determine approximate quantities to add until the target is reached. Adjustment tasks require a lot of highly qualified human and time resources. Expen- diture arising from such tasks are considered quality non- conformance costs (i.e. costs stemming from fixing failures to meet with specific requirements). These costs are usually high, bite into a firm’s profit margin and make it less com- petitive. This paper introduces an innovative artificial cognitive system to support decision-making in adjustment processes based on human sensory abilities. The proposed system, based on expert knowledge management, draws on a ma- chine learning tool jointly with an actions’ generator mod- ule. A specifically-adapted Support Vector Machine (SVM) [6] is previously trained with ‘state-action’ type patterns pro- vided by experts. It then identifies and selects the best action from among those provided by the generator module for a particular state. The coupled actions’ generation-selection process is iterated until the final state satisfies certain con- ditions (that is, until the target is reached). The proposed system should slash non-conformance costs once it has been turned into a software tool that automati- cally helps in the adjustment process. Furthermore, it will bolster quality assurance given that customers will receive conforming products. The methodology was developed and tested in the automotive industry for color adjustment in basecoat painting (which is the layer that provides color for the paint system). However, it can be extended to other simi- lar tuning processes in highly creative industries that involve experts with special sensory abilities. The remainder of this paper is organized as follows. Sec- tion 2 presents the qualitative vision of sensory-based expert adjustment to manufacturing products. The concept is out- lined and the methodology for the qualitative learning sys- tem based on expert knowledge is formalized. A brief intro- duction of the color adjustment problem together with a real- case application for the automotive basecoat manufacturing is presented in Section 3, showing that the new methodol- ogy is a suitable learning tool for this purpose. Section 4 describes the experiments performed and analyzes the re- sults obtained in the real-case application. Finally, Section 5 highlights some conclusions and future research tasks.