Flexible Industrial Analytics on Reconfigurable Systems-on-Chip
Industrial analytics is a current trend in automation and denotes capturing and analysing a multitude of measurements from machines and production processes to create added value for future operation. The company Weidmüller positions itself in this business area. For realising embedded analytics, where the analytics functions are executed in the automation controllers within a production facility, Weidmüller relies on reconfigurable system-on-Chip (rSoC) technology. The challenges in using rSoC for industrial analytics are to provide the required flexibility for system design and to master the increasing heterogeneity of rSoC platforms. Flexibility is required during design time, since industrial analytics functions need to be selected, configured and assembled in an application-specific manner. A flexible system design for rSoCs is of utmost importance to be able to quickly evaluate design alternatives and to realise selected designs as hardware/software systems. Moreover, flexibility is also required at runtime to make efficient use of rSoC resources under varying computational loads. Technology trends drive rSoC platforms towards increased heterogeneity. Although heterogeneity promises improved performance, it also leads to more involved design processes and makes the design of efficient runtime methods more challenging. The combination of increasing workload dynamics and heterogeneity of computer architectures is also the leading theme of the foundational research in the subproject C2 of the CRC 901 ”On-The-Fly Computing". There, we investigate novel architectures and programming models for heterogeneous compute nodes. In particular, we develop methods for the transmodal and heterogeneous migration where functions can be moved at runtime between software and hardware, but also between different processor types, to optimise for runtime and energy consumption. With this transfer project we aim at reaching the following objectives: (i) We will characterise relevant industrial analytics functions and develop suitable hardware/software partitions for rSoC targets. (ii) We will develop architectures and programming environments to enable transmodal and heterogeneous migration for industrial analytics functions on rSoC. (iii) We will experimentally demonstrate the feasibility and performance of the developed methods on two industrial demonstrators, a plastic moulding machine and an industrial router.