Issue link: https://htpgraphics.uberflip.com/i/1385717
RAIN PROGRESS // Data have been collected for a short period of time on February 2020 before the first UK lockdown. To have a comparable and reliable base of data to work on, at the beginning and end of each working session, robots have been exercised with CMS-specific moves called 'yoga moves'. These motions describe squared trajectories at different heights, after which the arm is stretched and joints are exercised independently. As lockdown started, it became clear of the necessity of working without the real hardware. For this reason, a work of modelling the real hardware has been started. Some preliminary results have been obtained; this work is still ongoing. Using DPGMM models it has been possible to identify a difference in the internal behaviour of the two robot arms used in the RAIN glovebox. This difference was not visible from the operator point of view as it was showing only in internal signals usually not monitored by the operator. The Variational Auto Encoder has been recently implemented and it is already giving interesting results; in particular, it has been possible to identify artificially made anomalies. Moreover, this development paves the way for a more sophisticated metric to define an anomaly. REMOTE HANDLING FUTURE ASPIRATIONS // There are several fields where this research can move to. First it will be necessary to be able to obtain more data from real hardware. This is crucial to expand the base of data already acquired. At the same it will be very important to generate synthetic data from dynamic models of the robots. Having dynamic models that can reproduce the reality will allow us to produce ad hoc synthetic faults, so we will be able to test and adjust our models before the faults will happen in the real world. Moreover, we will be able to assess our models based on a known ground truth. In parallel with this activity there will be the research on the comparison of different algorithms for identifying faults and performance degradation. It will be especially important to design a metric to define an anomaly in a simple probabilistic way. Ultimately this work must be integrated with the operator system to be used in real time during glovebox operation. 61