Industrial robots play a key role in production lines. As a direct consequence, even a slight degradation of their operating conditions could negatively affect the entire manufacturing process. To minimize economical losses, preventive measures such as planned maintenance or stand-by working stations are adopted and, in some cases, additional manipulators are installed to ensure line availability. However, since each robot performs a specific task, it degrades at a different rate from another one, so Planned Preventive Maintenance (PPM) should be replaced with Condition Based Maintenance (CBM) for a more efficient and cost-effective approach. In addition, collaborative robots (cobots) have become more and more popular in the past years increasing the level of automation, in particular in small and medium size companies. Therefore, a failure of an industrial manipulator could not only cause unexpected downtimes, but also jeopardize the safety of the personnel with whom it shares its workspace. A cobot, in fact, is labeled as safe as long as it works in nominal conditions, but this cannot be guaranteed otherwise. Within this framework, Prognostics and Health Management (PHM) techniques could be used for both a customized maintenance on a specific machine and for preventing undesired events like the aforementioned ones. An ongoing research activity at Politecnico di Torino is focused on a model-based approach able to better describe the behavior of an industrial robot in non-nominal operating conditions. In order to properly simulate faults and failures of a manipulator, it has been first necessary to analyze its failure modes and their impact on the entire system. A robot, in fact, is usually equipped with internal sensors and algorithms able to monitor its health status. However, this is only true for software or electrical failures (i.e. power supply, motor or sensors related), but it is not as effective in case of mechanical ones like bearings or gearbox wear. This paper firstly presents an overview of the possible failures related to harmonic drives which are often used in robotics due to their compact and light-weight design and the high reduction ratio. Then, it outlines the work done with the objective of providing an accurate, physics-based description of faults progression of harmonic drives used in industrial robots and the related high-fidelity models in the framework of a PHM system for fault and failure detection, isolation and remaining useful life (RUL) prediction.

This work describes a layout to carry out a demonstrative assembly task, during which a collaborative robot performs pick-and-place tasks to supply an operator the parts that he/she has to assemble. In this scenario, the robot and operator share the workspace and a real time collision avoidance algorithm is implemented to modify the planned trajectories of the robot avoiding any collision with the human worker. The movements of the operator are tracked by two Microsoft Kinect v2 sensors to overcome problems related with occlusions and poor perception of a single camera. The data obtained by the two Kinect sensors are combined and then given as input to the collision avoidance algorithm. The experimental results show the effectiveness of the collision avoidance algorithm and the significant gain in terms of task times that the highest level of human-robot collaboration can bring.

Dynamic parameters are crucial for the definition of high-fidelity models of industrial manipulators. However, since they are often partially unknown, a mathematical model able to identify them is discussed and validated with the UR3 and the UR5 collaborative robots from Universal Robots. According to the acquired experimental data, this procedure allows for reducing the error on the estimated joint torques of about 90% with respect to the one obtained using only the information provided by the manufacturer. The present research also highlights how changes in the robot operating conditions affect its dynamic behavior. In particular, the identification process has been applied to a data set obtained commanding the same trajectory multiple times to both robots under rising joints temperatures. Average reductions of the viscous friction coefficients of about 20% and 17% for the UR3 and the UR5 robots, respectively, have been observed. Moreover, it is shown how the manipulator mounting configuration affects the number of the base dynamic parameters necessary to properly estimate the robots’ joints torques. The ability of the proposed model to take into account different mounting configurations is then verified by performing the identification procedure on a data set generated through a digital twin of a UR5 robot mounted on the ceiling.