Power converters are essential for the eco-friendly integration of different sources and loads into existing grids to form sustainable energy systems.

The terms “Net-Zero Emissions”1 and “Net-Zero Energy Consumption”2 are dominating the news and conversations everywhere — be it the COP United Nations Climate Change Conference or a routine break-room chat. We are racing against time to mitigate climate change and simultaneously learn to adapt to its catastrophic impacts. In the pursuit of a greener, more sustainable future, power converters have emerged as indispensable components within energy systems.

These versatile components play a pivotal role in facilitating the seamless integration and efficient utilization of renewable energy sources, such as solar panels, wind turbines, and energy storage systems. They are also crucial in delivering power to most electronic loads including information and communications technologies (ICTs),3 electric vehicles (EVs), etc. Over 80 percent of electricity is poised to flow through power converters by 2030,4 as estimated by the U.S. Department of Energy.

Power converters are essential for the eco-friendly integration of different sources and loads into existing grids to form sustainable energy systems by effectively transforming, regulating, and managing electrical power. One of the modern examples of such a system is “Sustainable Agriculture.” Figure 1 shows a converter-based architecture for sustainable energy systems in agriculture combining clean energy technologies (including “Agrivoltaics”),5 storage, water purification, and grid energy management.

Need for Resiliency and Edge Intelligence

Figure 1. Sustainable agriculture — power electronics-based integration of clean energy technologies, agriculture loads, and water purification/desalination systems with the grid. (Image: University of Houston)

The global energy sector is on an exhilarating trajectory, teeming with promising technologies and unprecedented opportunities for an efficient and sustainable future. However, we now find ourselves grappling with reliability and resiliency challenges. There is a growing need to nearly double the operational lifetime of power conversion technologies (e.g., 50 years for solar PV systems,6 35 years for offshore wind,7 subsea oil and gas production, etc.), along with meeting size and efficiency targets in the grid interface of sustainable energy systems.

Applications like data centers preemptively replace their power hardware every 8-10 years8 to avoid potential outages, creating more e-waste. On the other hand, in mission-critical applications such as offshore/ downhole energy production (e.g., oil and gas, geothermal, etc.) or extraterrestrial power distribution systems (e.g., lunar surface power system planned by NASA’s Artemis program), the extreme environments degrade the power hardware much faster, leading to largely unpredictable aging profiles.

The underlying problem with most existing power conversion infrastructure is that it is very difficult to assess its health or adapt its performance in real-time without disrupting the operation. Hence, to achieve dynamic performance enhancement and resiliency improvement, there is a major push toward developing mission profile-oriented design methods that lead to the seamless integration of data-driven approaches with the power hardware onboard for system monitoring, dynamic adaptation, and prognostic health management (PHM), which can be together referred to as “Edge Intelligence.”

Such approaches will further enable the identification of aged or failing hardware in real time and avoid preemptive decommissioning, thereby increasing the operational life. It will enable the system operators to make decisions in the field about maintenance or repair, and the supply chain team to better estimate the logistics or manage the inventory. For example, it is reported that the U.S. DoD spends billions of dollars on the inventory9, 10 of spares, repair parts, electronics, etc.

Edge intelligence will potentially help save hundreds of millions if not billions of dollars by reducing costs with a smaller inventory due to informed planning, avoiding preemptive maintenance or replacement, and minimizing costly failures in the field. Furthermore, these techniques can be extended to a wider set of applications, not limited to power converters or even electronics in general. Considering the broad impact potential, the U.S. National Science Foundation (NSF) recently funded a project11 to advance some of the relevant technologies over the next five years.

Three Layers in Power Conversion System Modeling

Figure 2. Three-tier approach to “Edge Intelligence” in power conversion — (i) component level, (ii) sub-system or module level, and (iii) cyber-physical system level. (Image: University of Houston)

The design of power converters for sustainable energy systems involves engineering in three main layers — (1) component or sub-circuit, (2) converter or module or sub-system, and (3) cyber-physical system (or CPS). Hence, performance improvement and resiliency enhancement need to be integrated within all these layers. Figure 2 shows a 3-tier diagram of a unified approach where data or information from the first layer flow to the second and from the second layer to the third for realizing edge intelligence as a whole.

Major reliability-related industry concerns about modern power converters arise due to the following: (a) the large number of devices being used, (b) variations in the expected lifetime caused by the mission profile (including factors such as temperature, voltage, current, etc.), and (c) the relatively small quantity of historical field data available for wide bandgap (WBG) semiconductor devices. To address these, numerous methods have been developed for the performance testing and reliability characterization of critical components such as semiconductor devices (silicon or Si, silicon carbide or SiC, gallium nitride or GaN, etc.), capacitors (electrolytic, film, ceramic, etc.), printed circuit boards, solder, etc.

Furthermore, several recent studies have proposed techniques to estimate the remaining useful lifetime (RUL) and operation of converters using traditional, statistical, and machine learning-based approaches. However, these approaches are usually restricted to the technology design or prototyping phases, and they are hardly implemented online in the final products. The added number and footprint of circuit components needed for realizing these methods, potential interference with the device characteristics or the regular converter operation, and the extent of computing resources (such as laptops) required for process execution make the existing techniques impractical for onboard implementation, especially in sustainable energy applications that use many power converters.

Each component of a power converter presents a unique challenge in the onboard or in-situ measurement of the “health indicators” for practical implementation. For a power field-effect transistor or FET, sensing its “on-state resistance,” a widely accepted health indicator, involves precise voltage and current measurements during the device’s current conducting period. However, the sensing circuit needs to handle high voltages and currents along with associated overshoots, that too near the device to avoid further intrusions of its own.

Whereas for a capacitor, which is a typically concerning component in terms of converter reliability, the health indicators such as “capacitance” and “effective series resistance” (or ESR) are more difficult to evaluate in-situ due to the complex nature of the measurements involved. Sensing and monitoring the health indicators of some of the other important components, such as the PCB, can be even more challenging since the material degradation needs to be monitored.

Enabling Edge Intelligence across Power Converter Layers

Integrated circuits (ICs) are becoming popular for onboard PHM of power FETs. Tell-I’s SensAI12 is an integrated hardware aimed at power and health management, which is designed to be placed near the primary FET or transistor and measure the on-state voltage, current, and resistance. Some of the other solutions13 target the monolithic or heterogenous integration of the power FETs along with the sensing and measurement circuits to further reduce the footprint, component count, and performance intrusion.

Considering the case of capacitors, the net impedances at multiple frequencies can be good health indicators and are relatively easier to sense and measure onboard. However, for capacitors and some of the other components, further technological advancements need to be made along with thorough testing for performance and reliability in order to realize the true benefits of in-situ PHM and the associated modeling of the entire power converter, sometimes also termed as a “Digital Twin.”14

The sensing and measurement techniques only form half the story. It is equally critical to communicate the data, deduce the information, and execute the actions — all onboard the converter hardware to term it “edge intelligence.” Field Programmable Gate Arrays (FPGAs) and microcontrollers are popularly used in power converters to implement control and modulation schemes.

If the performance characteristics and RUL patterns of health indicators associated with each critical component in a converter are recorded offline via thorough reliability test procedures under a wide range of practical mission or operation profiles, machine learning can be used to model the overall behavior in a simple, space-efficient manner.

Figure 3. Harnessing the power of machine learning, FPGAs, and more to deduce/improve the performance and health of power converters onboard — enabling PHM and “digital twins” at the edge at minimal or no added cost. (Image: University of Houston)

FPGAs can be excellent solutions for implementing compact machine-learning models15 and can easily support this feature while also executing the control actions for a converter. An example test setup and a procedure for the reliability characterization of gallium nitride (GaN) FETs are depicted in Figure 3.

With an increasing number of converters interfaced with the grid or any comprehensive power distribution system, several new challenges emerge at the CPS level including stability and resiliency. The impedance-based stability analysis is widely used in grid-interconnected applications since this method doesn’t need modeling of the whole system.

There are several technological constraints here as well: (1) most of the studies assume a fixed grid voltage; there aren’t considerable literature studies available for impedance characterization of converter-dominated grids or microgrids; (2) in scenarios with a vlarge number of converters/inverters interconnected, there hasn’t been sufficient research on how the impedance or admittance patterns at different nodes of the CPS can construe the relative resiliency of the overall system and the potential actions that can be taken to remedy any adverse scenarios; and (3) there hasn’t been any research on practically incorporating these analyses together with converter-level health and performance information using FPGAs or microcontrollers to enable edge intelligence at the CPS level.

Figure 4. Edge intelligence for PHM and performance adaptation at the cyber-physical system (CPS) level using impedance characterization as well as machine learning-assisted status checks at the converter terminals. (Image: University of Houston)

Once these technical gaps are addressed, converter impedance at every node or terminal can be evaluated periodically (as in Figure 4) and the pattern of the impedance characteristics can be used to assess the health status as well as the performance of the system even under faulty operation.

In conclusion, recent advances in integrated circuits, machine learning, and computing have opened up several possibilities to make informed assessments and decisions regarding the health and operational performance of modern power conversion systems. In a world buzzing with the promises of “smart” grids and “autonomous” vehicles for a sustainable future, we are not too far away from realizing power conversion systems with true intelligence at the edge.

References

  1. IEA, “Net Zero Emissions (NZE) by 2050 – Global Energy and Climate Model – Analysis,” 2022. Available [30 Aug. 2023].
  2. U.S. GS, “Net Zero Energy Overview,” 2020. Available [30 Aug. 2023].
  3. Wikipedia Contributors, “Information and communications technology or ICT,” 2019. Available [30 Aug. 2023].
  4. U.S. DoE SETO, “Solar Power Electronic Devices,” 2020. Available [30 Aug. 2023].
  5. Wikipedia Contributors, “Agrivoltaics,” 2019. Available [30 Aug. 2023].
  6. U.S. DoE EERE White Paper, “Modeling of Photovoltaic Systems: Basic Challenges and DOE-Funded Tools,” 2022. Available [30 Aug. 2023].
  7. IJGlobal, “Turbine lifetime limits require a reality check,” 2021. Available [30 Aug. 2023].
  8. Yesterday’s Business Computers -YBC, “Sell or Recycle Data Center I.T. Hardware?,” 2021. Available [30 Aug. 2023].
  9. Politico Investigation, “How do you buy $7 billion of stuff you don’t need?,” 2015. Available [30 Aug. 2023].
  10. U.S. GAO, “Defense Inventory Management,” 1992. Available [30 Aug. 2023].
  11. U.S. NSF, “CAREER: Enhancing the State of Health and Performance of Electronics via in-situ Monitoring and Prediction (SHaPE-MaP) - Toward Edge Intelligence in Power Conversion,” 2023. Available [30 Aug. 2023].
  12. Tell-i Technologies, “New Power Transistors Need New Sensors.” Available [30 Aug. 2023].
  13. H. Krishnamoorthy and J. Hawke, “Onboard circuits and methods to predict the health of critical elements,” 2023. Available [30 Aug. 2023].
  14. A. Wunderlich and E. Santi, “Digital Twin Models of Power Electronic Converters Using Dynamic Neural Networks,” 2021 IEEE Applied Power Electronics Conference and Exposition (APEC), Phoenix, AZ, USA, 2021, pp. 2369-2376, doi: 10.1109/APEC42165.2021.9487201.
  15. Circuit Cellar, “FP GAs for AI and Machine Learning,” 2023. Available [30 Aug. 2023].

This article was written by Harish Sarma Krishnamoorthy, Associate Professor, ECE Department, University of Houston (Houston, TX). For more information, visit here .