Predictive Analysis

The Experiment is about “Predictive Analytics based on few-shot learning” and has been conducted by ARCULUS. Autonomous Mobile Robots (AMR) have experienced an increasing commercial push in recent years. Due to the technical progress in the field of autonomous driving and the associated supply chain, other industries such as intralogistics are also benefiting. Suppliers such as nVidia are developing SOC components that enable high computing power with low energy consumption, and the market for safety-related components such as LIDAR systems is becoming more affordable: this enables a new generation of logistics robots in the intralogistics sector, and leads to a transformation from Automated Guided Vehicles (AGVs) to Autonomous Mobile Robots (AMR) and thus to a whole new kind of flexibility of human-to-machine or machine-to-machine interactions, considering that AMRs offer several advantages on AGVs and a broader range of applications.

As a consequence, the number of interested end users is widening. At the same time, however, the expanded application capability also increases the demands on the reliability of the system. More complex application scenarios with higher environmental dynamics increase the pressure on the service and maintenance team of an AMR provider. Added to this is the challenge of a comparatively young technology, to established hardware that has been tried and tested in use. 

Arculus aimed at making the use of robots more robust and efficient through the use of targeted AI technologies from the application area of "predictive maintenance". The primary aim was not to optimize the control and localization algorithms, but to optimize and improve those processes that have the greatest impact on the organization in the course of scaling, such as maintenance and operating efficiency, through the early detection of sensor faults (Lidar, IMU, Camera, ...), so ensuring a proactive fault clearance of the system. The laser source of the AMR must be frequently cleaned because, being only a few cm above the ground, it is easily covered in dust. This obliges maintenance workers to make frequent stops of the robot, the frequency of which depends on the level of dirt in the environment where it operates, but which can reach intervals of every hour or two. The aim of the experiment was to generate an AI-based predictive maintenance model that optimises cleaning frequency by assessing the level of accumulated dust.