FPGA Application Domains in Industrial Electronics

Although (relatively) simple, FPGAs are still being used for glue logic imple- mentation or rapid prototyping purposes (which were their original usages), the ever-increasing variety and amount of hardware resources, as well as the ability of the most current devices to implement whole complex systems in a single chip, has tremendously widened their application domains. The origi- nal applications have also benefited from this evolution so that, for instance, HIL approaches can be implemented instead of just basic system prototyp- ing. Moreover, as highlighted throughout the previous chapters, current FPGAs are particularly suitable for configurable computing applications and for developing dynamically reconfigurable systems. In spite of all this, many complex digital systems are still today built from the combination of sepa- rate digital processing and configurable logic chips, such as in Hasanzadeh et al. (2014) or Wen et al. (2014). Very interestingly, migration from these plat- forms to FPSoC-based ones is relatively easy because, for instance, the exist- ing software can be reused in embedded processors. 

In the following sections, the aforementioned significant industrial appli- cations of FPGAs are briefly discussed. 

Digital Real-Time Simulation of Power Systems 

Hardware-in-the-loop simulation (HILS) is very useful, among other appli- cations, for real-time simulation of power systems, where it allows develop- ment time, cost, and time to market to be reduced (Buccella et al. 2012). In addition, it allows these systems to be safely checked under faulty (Zhu et al. 2014) or extreme operating conditions. A fundamental requirement of HILS in this context is to achieve very short time steps, which can be provided by FPGAs. This, together with the availability of memory and complex arithme- tic blocks, as well as their parallelism and reconfigurability, makes FPGAs very suitable—and increasingly used—HILS platforms. 

Many existing HILS solutions are based on the use of DSPs as main process- ing elements and FPGAs as specialized coprocessors and/or communication links (Wang et al. 2013; Hasanzadeh et al. 2014; Schmitt et al. 2014). Interfaces with analog inputs and outputs are implemented with separate ADC and DAC chips. It is worth noting that faster and more compact solutions can be achieved by migrating these structures to current FPSoC platforms, where most of the required resources are embedded. For instance, an FPGA-only platform is reported in Dagbagi et al. (2013), where a PWM rectifier is emu- lated in real time in one device and the corresponding controller is imple- mented in another one as an FPSoC, using an ARM Cortex-M3 core. 

Advanced Control Techniques

Power Systems 

Digital control of power systems has been gaining increasing interest over the years. Current solutions, usually based on microcontrollers and DSPs, suffer from relatively long execution times and limited flexibility to interface with analog signals. These issues can be successfully addressed with mod- ern FPGAs. Actually, the analysis in Monmasson et al. (2011b) concludes that hardware solutions are the only viable options in practice for systems with strong timing requirements. Very interestingly, some of the most challeng- ing needs identified in that analysis—efficient floating-point computations and analog/digital interfaces—can be addressed with the specialized hard- ware resources available in current FPGAs. 

Many designers have already realized that FPGAs are the best solution to gen- erate control signals for switching devices in power converters (Hwang et al. 2013; Lu et al. 2013; Morales-Caporal et al. 2013; Miura et al. 2014; Wang et al. 2014), particularly when the number of switches, the switching frequency, or both are high, which is the case in many modern converter topologies. However, in most of these works, there remains an inertia to implement the main con- trol loops (and other fundamental tasks, such as filtering) in separate micro- controller or DSP chips and to use external ADCs, instead of taking advantage of FPSoC architectures. For instance, in Senicar et al. (2014), an FPGA is used to implement a fast current control loop in hardware, whereas higher-level con- trol loops—speed and position—are implemented in a separate microcontroller chip, and external ADCs are used. Similarly, in Prabhala et al. (2012), the current controller for a voltage source converter is implemented in an FPGA, and the grid synchronization PLL in a DSP. The problems identified in Wen et al. (2014), related to the limited speed and processing power of DSPs, could be alleviated by using FPGAs to concurrently run some part of the proposed computation algorithm. Other reported systems that could have benefited from the use of FPSoCs are those of Kobravi et al. (2013) and Smidl et al. (2013). 

Also, in the area of power systems, fault tolerance is a fundamental require- ment when continuous operation must be ensured. In this case, the faster the fault detection, the safer the operation of the system. The usefulness of FPGAs for this purpose is demonstrated in Shahbazi et al. (2013). 

In general, digital control of power systems is an area where the use of FPSoCs should be expected to grow significantly in the near future.

Robotics and Automotive Electronics 

High operating frequency and low latency are fundamental requirements for control loops in many robotic applications, where at the same time com- plex algorithms are also needed for various purposes. Although DSPs— sometimes combined with FPGAs—are extensively used in this domain, their sequential nature in many cases may result in excessive latency when executing those algorithms. In addition, DSPs provide limited flexibility to include additional sensors or actuators, in turn limiting system scalability. In contrast, current FPGA devices offer better performance and accuracy, allowing real-time or fault-tolerant operation to be achieved (Hace and Franc 2013; May and Krougjicof 2013; Phuong et al. 2014). 

Modular robotics (e.g., for humanoid robots) can take advantage of cur- rent FPGA devices to achieve higher sampling frequencies—and, then, bet- ter performance—than centralized systems. The flexibility of FPSoC-based solutions adds to that of these modular systems, allowing robot structures to be easily modified, extended, or repaired, even in the field. In this context, the availability of FPGAs including powerful embedded processors contributes to reduced size and weight, as well as lower processor intercommunication latency. Examples of current use of FPGAs in this area can be found in Zhu et al. (2013) and Pierce and Cheng (2014). 

The application of FPGAs in automotive systems is an area of increasing interest, among other factors, because of the need to perform very com- plex operations with the low latency required by safety-related functions, such as the antilock braking system or the electronic stability program (Guo et al. 2013). Similar problems are addressed in Lu et al. (2015) for controlling dual-clutch transmission gearshifts. The increasingly pow- erful embedded processors and DSP blocks available in current FPGAs should play a key role in a wider penetration of these devices into auto- motive applications. 

Use of Floating-Point Operations 

There is an increasing number of applications requiring the use of floating-point operations to achieve the target performance (Guo et al. 2013; Hwang et al. 2013; Sepulveda et al. 2013; Barranco et al. 2014). Although effi- cient resources for floating-point operations were traditionally not available in FPGAs, it is currently possible to implement them either in embedded processors (Jimenez et al. 2014) or in dedicated specialized hardware blocks (Juarez-Abad et al. 2014; Liu and Dinavahi 2014). Even better results could be achieved by using the IEEE 754–compatible DSP blocks available in the most current devices. 

Electronic Instrumentation 

Same as floating-point operators, the lack of embedded ADCs and DACs was one of the main traditional limitations in FPGAs. The availability of these resources in recent devices (as well as of others, such as complex PLLs) opens the door for the improvement of many industrial systems that could not benefit from them when they were developed, such as the power sys- tems presented in Hwang et al. (2013) and Guzinski and Abu-Rub (2013) or the ones related to sensors and microelectromechanical systems reported in Cheng et al. (2013), Xia et al. (2013), and Xu (2014, 2015). An area where FPGAs provide particularly good performance at a reason- able complexity/cost is that of vision systems, in applications such as intel- ligent spaces (Rodriguez-Araujo et al. 2014) or unmanned vehicles (May and Krougjicof 2013). For instance, the configurable hardware system for traf- fic sign recognition, presented in Aguirre-Dobernack et al. (2013), achieves 60 fps with 1280 × 720 pixel images. 

In aerial unmanned vehicles, huge amounts of data must be processed with low latency for purposes such as flight control, real-time mapping, or obstacle avoidance. Reduced size and weight are also fundamental requirements of these systems, which are more easily fulfilled with single-chip FPSoC imple- mentations than with multichip heterogeneous solutions, such as the one presented in Schmid and Hirschmuller (2013), consisting of two cameras, an Intel Core2Duo processor board, a Spartan 6 FPGA board, an ARM processor board, and an IMU. By using an FPGA including ARM Cortex-A9 processors, the number of required boards—and thus the size and weight of the system— could be greatly reduced. The system presented in Nikolic et al. (2014) for real- time simultaneous localization and mapping consists of a Zynq-7000 FPSoC, four cameras, and an IMU, where the Zynq-7000 device replaces the combina- tion of an FPGA and an Intel ATOM processor used in previous prototypes. 

Another application where an advantage can be taken from the FPGA implementation of image processing algorithms is industrial laser cladding. In the work by Rodriguez-Araujo et al. (2012), an adaptive fuzzy PI control- ler for laser cladding systems is proposed, which works with data extracted from a real-time image processing system (achieving 100 Hz sampling rate for 800 × 600 pixel images). The whole monitoring and control system is implemented in an FPGA, combining a Nios II soft processor with high- performance custom hardware peripherals.