The usage of more and more sensors in production systems offers opportunities for novel process optimizations, e.g. through the use of machine learning . But this comes at the cost to process large amounts of data already at its source, which poses major challenges for well-known industrial automation principles such as the often applied automation pyramid . With the advent of cloud computing, the complexity of automation systems can be greatly reduced when processing the Manufacturing Execution System (MES) and Enterprise Resource Planning (ERP) layers in the cloud and unifying also the lower levels [field, control, and Supervisory Control And Data Acquisition (SCADA)]. However, traditional automation systems completely differentiate the levels of field (sensors/actuators), control [Programmable Logic Controllers (PLCs)], and SCADA by coupling complex proprietary subsystems for each level. Nevertheless, as the volume of data will continue to grow exponentially, we see these levels vanishing and, more important, the requirement to process the majority of data in real-time and thus close to production machines, the so-called Edge.
A major challenge of suitable platforms that might cover the tasks of all three levels arises from the growing number and variety of sensors and related data to process in real-time, which can vary dramatically from application to application. Although emerging smart sensor solutions can compress, filter, or convert signals into desired formats in real-time, they are, however, closed systems with often proprietary and inflexible interfaces. Indeed, incompatible interfaces can cause a lot of additional costs, since they have to be either purchased or developed separately under strict licensing conditions.
As a remedy, this paper proposes and advocates the use of Field-Programmable Gate Array (FPGA)-based platforms to achieve the required flexibility and scalability by exploiting the advantage of hardware reconfiguration and acceleration. From the analysis of the similarity of many data acquisition, monitoring, and control tasks of PLC systems available today, it can be concluded that a high flexibility is needed to be able to sense not only different types of digital and analog signals, but also different numbers of inputs and outputs using the same platform (I/O reconfigurability). Moreover, in order to be able to cover a multitude of different communication protocols at the hardware interfaces, the hardware reconfigurability of FPGAs is exploited. In hardware, time-critical signals can be directly sampled, filtered, fused, or pre-processed, e.g. converted from the time into the frequency domain, in parallel with other processing tasks without causing any additional Central Processing Unit (CPU) load. In order to combine the advantages of PLCs with the high performance offered by an FPGA, hybrid systems combining reconfigurable logic together with microprocessors, memory blocks, and peripherals on a single chip exist.
As a first contribution, this paper presents such a PSoC platform, containing in addition to an FPGA two CPUs that can be used to process even higher level tasks in software such as process visualization, which plays a decisive role in the evaluation of the suitability for use in industrial environments. As will be demonstrated, the PSoC guarantees reliable in-situ signal processing even for applications with changing data loads in time-critical open- as well as closed-loop control applications.
As a second contribution, a design methodology to partition and automatically generate hardware/software configurations from a given control application to a given PSoC platform is presented. The model-based design approach starts with the description of a behavioral model defined in MATLAB/Simulink. Based on this description, the framework optimizes and generates from a block diagram code for the target PSoC architecture. As a result, synthesizable function blocks of control tasks can be implemented in hardware on the FPGA, while high-level tasks are compiled and deployed in software on the processor system.
Third, a real-world case study on position control of a metal forming process is presented, demonstrating the advantages of the PSoC architecture and automatic design flow. We conclude that hardware/software reconfigurable PSoC architectures may provide the required computational power and the needed I/O reconfigurability at rather low cost for the processing of huge amounts of sensor data right at the edge, thus avoiding the congestion of networks to servers and clouds.
The remaining of the paper is organized as follows: Sect. 2 proposes the concept of PSoC-based PLCs, while Sect. 3 details the tool flow for automatic hardware/software generation. A demonstrator to analyze the requirements of forming processes is presented in Sect. 4. Finally, experimental results of the control of such a metal forming process is demonstrated in Sect. 5, followed by a conclusion and outlook on future directions in Sect. 6.