Despite the existence of various solutions in the industrial domain for cooperation between robots and humans, they tend to focus mainly on safety issues with very few advances in the adaptation of industrial equipment to the characteristics of the operator and his way of working. For several years, adaptation in a human-robot collaboration environment was single sided, as only the operator adapts his working operations facing the robot characteristics, which leads to high levels of stress and fatigue of the human operator. Nowadays, the paradigm is changing towards the adaptation of human operator to industrial equipment and vise versa. The adaptation of a robot to the human is achieved by enabling the machine to learn the physical and psychological characteristics of each operator, in order to create a working profile for each individual. Thus, the main objective is to analyze the relationship between human operators and robots in an industrial environment, and therefore explore human-machine collaboration by correlating sensorial data from all the entities involved in the process. With this in mind, by performing sensor fusion and data analysis representing actions and biometric signals from the human operator, industrial robots will be empowered of self-adaptation capabilities. In this dissertation, an industrial collaborative environment is achieved using a Cyber-Physical Production System (CPPS). This CPPS consists in three main parts, namely sensing and actuating equipment, logical entities called Smart Components and a Cloud infrastructure. Sensing devices are based on biometric sensors-BITalino's ECG and EDA-and a vision system-Kinect-in order to monitor the human operator working profile. A robotic arm is used as actuating device. Each equipment is virtualized into an agent-based representation, based on the Smart Component concept, which communicate sensor data with a Cloud infrastructure responsible for data processing and decision making. Sensor data is analyzed in order to infer levels of stress and fatigue through a fuzzy logic system. Decision making is based on the MAPE-K architecture, enabling the robotic arm self-adaptation. Results from human subject tests are presented here to validate the proposed methodology, proving that the system can detect stress with an accuracy of 77,6% and fatigue with an accuracy of 70%, as well as detect the subject's position and movement with a true positive rate of 70,7%. Facing the movement and levels of stress and fatigue of the human operator, the robotic arm should be able to change autonomously it's task execution, namely speed of its movement and the correct operation according to the habits of the operator. i ii First of all I would like to thank Professor Gil Manuel Gonçalves, my supervisor, for the guidance and follow-up of the work done, for all the suggestions and corrections, and for giving me the basis for this work to be possible. The biggest thanks to Rui Pinto and João Reis, for being the co-supervisors every student hopes for and even more, for ...
