Advances in HVDC Systems: Aspects, Principles, and a Comprehensive Review of Signal Processing Techniques for Fault Detection

Abstract

This paper presents a comprehensive review of High-Voltage Direct-Current (HVDC) systems, focusing on their technological evolution, fault characteristics, and advanced signal processing techniques for fault detection. The paper traces the development of HVDC links globally, highlighting the transition from mercury-arc valves to Insulated Gate Bipolar Transistor (IGBT)-based converters and showcasing operational projects in technologically advanced countries. A detailed comparison of converter technologies including line-commutated converters (LCCs), Voltage-Source Converters (VSCs), and Modular Multilevel Converters (MMCs) and pole configurations (monopolar, bipolar, homopolar, and MMC) is provided. The paper categorizes HVDC faults into AC, converter, and DC types, focusing on their primary locations and fault characteristics. Signal processing methods, including time-domain, frequency-domain, and time–frequency-domain approaches, are systematically compared, supported by relevant case studies. The review identifies critical research gaps in enhancing the reliability of fault detection, classification, and protection under diverse fault conditions, offering insights into future advancements in HVDC system resilience.

1. Introduction

1.1. Evolution of HVDC Systems

In the late 1880s, Thomas Edison and Nikola Tesla engaged in the War of the Currents, a battle over the future of electricity. Edison promoted Direct Current (DC), which was predominant in the United States but had limitations of geographical scalability. Tesla championed Alternating Current (AC), which could easily change voltages and proved more versatile. AC ultimately became dominant, but recently DC has seen a resurgence due to the integration of renewables, leading to the hybrid coexistence of both currents [1].

The development of HVDC technology was driven by the need to overcome the limitations of AC transmission over large distances, high energy losses, and stability issues. HVDC technology is essential for interconnecting disparate power grids, integrating renewable energy sources such as offshore wind farms, and supporting stable and efficient international electricity trade. HVDC transmission systems today are integral components of the global power infrastructure. They play a vital role in meeting the growing energy demands of modern societies while supporting the transition to sustainable and renewable energy sources.

In the early 20th century, AC systems were preferred over DC due to their technological advantages, particularly in voltage conversion and long-distance transmission. Modern HVDC systems, which have evolved significantly from their mid-20th-century origins, now offer an ideal solution for long-distance, high-capacity energy transmission. The development of power electronics has further enhanced HVDC systems, solidifying the ongoing importance of DC technology in modern electrical networks [2].

The origins of the HVDC transmission system can be traced back to the 1930s when the invention of mercury arc rectifiers marked a significant technological breakthrough. This innovation provided the necessary foundation for converting AC into DC at high voltages, making long-distance power transmission feasible. Mercury arc rectifiers were pivotal in the early stages of HVDC technology, setting the stage for subsequent advancements that have made HVDC technology a vital component of modern electrical grids [3,4].

In 1941, a significant milestone in electrical engineering was marked by awarding the first commercial HVDC transmission system contract. This ambitious project, known as the Elbe Project, was designed to deliver 60 megawatts of power from Vockerode on the Elbe River to Berlin via a 115 km underground cable driven by the increasing demand for reliable long-distance power transmission [4]. By 1945, after extensive development and construction, the HVDC system was fully prepared for operation. The project incorporated advanced technology of the time, including mercury-arc valves, which were instrumental in converting and controlling HVDC technology.

It was not until 1954 that HVDC technology achieved its first successful commercial application by commissioning the Gotland HVDC link in Sweden. This pioneering project, employed improved mercury-arc valve technology, was developed by ASEA (now part of ABB), and transmitted 10 megawatts of power over a 98 km distance, connecting the Swedish mainland to Gotland. The final and most extensive HVDC system based on mercury-arc technology was the Nelson River scheme, which was commissioned in 1971 [5].

Semiconductor technology development in the 1970s introduced thyristor valves, which quickly replaced mercury-arc valves due to their enhanced durability, reliability, and cost-effectiveness, thereby significantly advancing DC power transmission [6]. In 1970, thyristors were first utilized in a commercial HVDC link with the Gotland Island project.

The development of IGBT valves marked a significant improvement over the earlier thyristor valves, particularly in their application within Voltage-Source Converters (VSCs). These IGBT valves offered enhanced performance characteristics, including better efficiency and control. In 1999, ABB introduced the first commercial Voltage-Source Converter High-Voltage Direct-Current (VSC-HVDC) system, connecting the island of Gotland with mainland Sweden. This pioneering system utilized submarine cables with a rated capacity of 50 megawatts and a voltage rating of ±80 kV [7]. Figure 1 illustrates the key milestones in the evolution of HVDC technology and the development of HVDC systems currently in operation.

The key milestones in the evolution of HVDC technology are presented, and in

Table 1. Evolution of power electronics and HVDC converter technology: from mercury-arc valves to IGBT transistors with real-world HVDC projects.

Technology	Year	Name of Projects	Rated Power (MW)	Rated Voltage (kV)	Rated Current (A)	Total Line (km)
Mercury (1954–1975)	1954	Gotland (Sweden)	20	100	200	96
	1961	Cross Channel (United Kingdom, France)	160	±100	800	65
	1965	Volgograd-Donbass (USSR)	720	±400	900	472
	1965	Benmore–Haywards (New Zealand)	600	±250	1200	609
	1965	Konti–Skan (Denmark, Sweden)	250	±250	1000	180
	1965	Sakuma (Japan)	300	2 × 125	1200	0
	1967	Sardinia (Italy)	200	200	1000	413
	1968	Vancouver, Pole 1 (Canada)	312	±260	1200	73
	1970	Pacific Intertie (United States)	1440	±400	1800	1354
	1976	Nelson River Bipole 1 (Canada)	1620	±450	1800	890
	1975	Kingsnorth (United Kingdom)	640	±266	1200	82
Thyristor Valves (1972–Present)	1972	Eel River (Canada)	320	2 × 80	2000	0
	1975	Cabora-Bassa (Mozambique, South Africa)	960	±266	1800	1420
	2011	BritNed (UK–Netherlands)	1000	±450	1100	260
	2014	Xiluodu–Zhejiang (China)	8000	±800	5000	1680
	2020	Wudongde (China)	8000	±800	5000	1452
IGBT (1997–Present)	1997	Hellsjön–Grängesberg (Sweden)	3	±180	150	10
	2000	Terranora interconnector (Australia)	180	±80	750	63
	2003	Murray link (Australia)	220	±150	733	177
	2009	HVDC Valhall (Norwegian)	78	±150	260	292
	2015	HVDC NordBalt (Sweden, Lithuania)	700	±300	1167	450
	2021	Siemens and Sumitomo joint project (India)	2000	±320	3125	263

, the progression of power electronics, highlighting the transition from mercury-arc valves to IGBT transistors, alongside the advancements in HVDC converter technology and the real HVDC projects in the world, is listed [7,8,9,10,11].

1.2. Advantages of HVDC over HVAC Technology

The advantages of HVDC over HVAC technology are compared across several key aspects, including the following: renewable energy integration, power transmission distance, power loss, reactive power compensation necessity, grid stability and control, and eco-friendly nature. Table 2 outlines the summarized advantages of HVDC systems compared to HVAC systems.

Table 2. Summarized advantages of HVDC systems compared to HVAC systems [12,13,14,15,16,17,18,19,20,21].

Aspects	HVDC Transmission	HVAC Transmission
Renewable energy integration [12,13]	Easy renewable energy integration	Difficult renewable energy integration
Power transmission distance [13,14]	HVDC technology prefers if the transfer distance is larger than the breakeven distance (generally, 800 km overhead line or 50 km submarine line) (Figure 2)	HVAC technology dominates if the transfer distance is smaller than breakeven distance
Power losses [15,16,17]	Less power losses due to no reactive power losses and no skin effects	Higher power losses due to reactive power losses and skin effects
Reactive power compensation [18,19]	No reactive power compensation required	Reactive power compensation is essential
Grid stability [15,16]	Allows AC system connections with two different frequencies	Difficult to achieve system connection with two different frequencies
Voltage regulation [13,17]	Easy voltage regulation due to only resistive losses	Complexed voltage regulation is needed due to reactive losses
Economical for long distance-transmission [20,21]	Compact transmission tower design and smaller carbon footprint Less electromagnetic interference due to DC Less insulation required Economical for long-distance bulk power transmission with reduced losses; greater power transfer per conductor	More complicated transmission tower design and larger carbon footprint (Figure 3) Less electromagnetic interference due to DC Electromagnetic interference Extra insulation required Expensive for long-distance transmission due to higher losses and lower power transfer capability

Table 2. Summarized advantages of HVDC systems compared to HVAC systems [12,13,14,15,16,17,18,19,20,21].

Aspects	HVDC Transmission	HVAC Transmission
Renewable energy integration [12,13]	Easy renewable energy integration	Difficult renewable energy integration
Power transmission distance [13,14]	HVDC technology prefers if the transfer distance is larger than the breakeven distance (generally, 800 km overhead line or 50 km submarine line) (Figure 2)	HVAC technology dominates if the transfer distance is smaller than breakeven distance
Power losses [15,16,17]	Less power losses due to no reactive power losses and no skin effects	Higher power losses due to reactive power losses and skin effects
Reactive power compensation [18,19]	No reactive power compensation required	Reactive power compensation is essential
Grid stability [15,16]	Allows AC system connections with two different frequencies	Difficult to achieve system connection with two different frequencies
Voltage regulation [13,17]	Easy voltage regulation due to only resistive losses	Complexed voltage regulation is needed due to reactive losses
Economical for long distance-transmission [20,21]	Compact transmission tower design and smaller carbon footprint Less electromagnetic interference due to DC Less insulation required Economical for long-distance bulk power transmission with reduced losses; greater power transfer per conductor	More complicated transmission tower design and larger carbon footprint (Figure 3) Less electromagnetic interference due to DC Electromagnetic interference Extra insulation required Expensive for long-distance transmission due to higher losses and lower power transfer capability

Figure 2. Factors affect the breakeven distance in HVDC and HVAC systems.

Figure 2. Factors affect the breakeven distance in HVDC and HVAC systems.

Figure 3. Comparison between HVDC and HVAC transmission tower.

Figure 3. Comparison between HVDC and HVAC transmission tower.

1.3. HVDC Applications in Power Networks and Developments in Asis-Pacific Region

HVDC applications can be classified into the following categories: (1) underground and submarine cables, (2) long-distance high-capacity power transmission, (3) asynchronous interconnection between AC systems, and (4) the stabilization of power flows in interconnected power networks [22].

Table 3. HVDC application scenarios: summary [23,24,25,26,27,28,29,30,31,32,33].

Category	Description	Challenges and Considerations
Underground and submarine cables [23,24,25,26]	Enables power transfer across water and buried pathways such as offshore wind farms	Harsh environments, high costs, electromagnetic interference, and land use impacts
Long-distance high-capacity transmission [27,28,29]	Efficiently transmits power over long distances with minimal losses	High costs due to environmental and regulatory concerns
Asynchronous interconnection [29,30,31]	Connects AC networks with different frequencies or phase angles, enhancing grid stability and control	Adverse effects between weak AC systems and HVDC, reduced grid inertia, and frequency stability issues
Stabilization of power flows [32,33]	Controls and stabilizes power flow within integrated networks, improving stability and efficiency	Requires advanced technology for effective control and stability

summarizes the summarized advantages of HVDC systems compared to HVAC systems.

The increasing efficiency, reliability, and capability of HVDC systems have led to their growing popularity worldwide, making them essential for power networks. Over the past decades, countries in the Asia–Pacific region—such as China, India, Japan, Australia, and New Zealand—have developed numerous HVDC systems for various purposes, including connecting distant regions, improving grid stability, supporting interconnection between states, and facilitating renewable energy integration, which are briefly introduced in detail in this subsection. They are experiencing significant growth, with numerous HVDC projects either in operation, under design, or currently under construction.

1.3.1. HVDC Systems in China

China has six major regional power grids, which differ significantly even between provinces within a grid, considering power demand, generation capacity, and resource distribution. These six regional power grid systems are in North China, East China, North-East China, North-West China, South China, and Central China [34]. The Chinese government has substantially invested in establishing a nationwide HVDC transmission network to minimize electricity transmission losses typically associated with HVAC cables between energy production hubs and load centers [35]. To achieve this goal, China is operating, planning, or constructing several HVDC projects around the country so that it has several multi-terminal HVDC systems such as Zhoushan (five-terminal VSC HVDC), Nan’ao (three-terminal VSC-HVDC), Zhangbei (four-terminal VSC-HVDC), and Wudongde–Guangxi–Guangdong (Three-Terminal hybrid LCC/VSC). The Zhoushan multi-terminal VSC-HVDC project is the first five-terminal HVDC in the world [36]. HVDC projects in China that connect the same or different regional grids and some Back-to-Back HVDC systems in China are presented in

Table 4. HVDC links in China.

Name	Commissioning Year	Configuration	Power Rating (Capacity) MW	Direct Voltage (kV)	Transmission Distance (km)
Zhoushan	1987	Bipolar	50	−100	54
Gezhouba Shanghai (Nanqiao)	1989	Bipolar	1200	±500	1046
Tianshengqiao	2001	Bipolar	1800	±500	980
Shengsi	2002	Bipolar	60	±50	66
Three Gorges–Changzhou	2003	Bipolar	3000	±500	890
Guizhou–Guangdong I	2004	Bipolar	3000	±500	900
Three Gorges–Guangdong	2004	Bipolar	3000	±500	940
Three Gorges–Shanghai I	2006	Bipolar	3000	±500	1060
Guizhou–Guangdong II	2007	Bipolar	3000	±500	1194
Hulunbeir–Liaoning	2010	Bipolar	3000	±500	920
Xiangjiaba–Shanghai	2010	Bipolar	7200	±800	2071
Deyang–Boaji	2010	Bipolar	3000	±500	534
Yunnan–Guangdong	2010	Bipolar	5000	±800	1418
Three Gorges–Shanghai II	2011	Bipolar	3000	±500	978
Jinping–Sunan	2013	Bipolar	7200	±800	2090
Jinbei–Nanjing	2017	Bipolar	8000	±800	1118
Ximeng–Taizhou	2016	Bipolar	10,000	±800	1618
Gansu–Hunan	2017	Bipolar	8000	±800	2390
Changji–Guquan	2019	Bipolar	12,000	±1100	over 3000

and

Table 5. Back-to-Back HVDCs in China.

Name	Commissioning Year	Power rating (Capacity) MW	Direct Voltage (kV)
Lingbao I	2005	360	120
Gaolineg I	2008	2 × 750	±125
Lingbao II	2009	750	166.7
Sino–Russia	2012	750	±125
Gaolineg II	2012	2 × 750	±125

, respectively. All HVDC links listed in these two tables are implemented using LCC technology.

1.3.2. HVDC Systems in India

India’s electric power system includes five regional electricity boards: the Northern, North-Eastern, Eastern, Southern, and Western regional grids. These grids operate synchronously on a single frequency, forming a unified national grid interconnected through HVAC and HVDC links. This national grid is also connected to neighboring countries, including Sri Lanka, Bangladesh, Bhutan, and Nepal [37]. HVDC projects in India are shown in

Table 6. HVDC links in India.

Name	Commissioning Year	Configuration	Power Rating (Capacity) MW	DC Voltage (kV)	Transmission Distance (km)	Technology
Rihand–Dadri	1991	Bipolar	1500	±500	816	LCC
Chandrapur–Padghe	1999	Bipolar	1500	±500	752	LCC
Talcher–Kolar	2003	Bipolar	2000	±500	1450	LCC
Ballia–Bhiwadi	2010	Bipolar	2500	±500	803	LCC
Mundra–Mohindergarh	2012	Bipolar	2500	±500	960	LCC
Champa–Kurukshetra	2017	Bipolar	6000	±800	1365	LCC
North-East–Agra (Biswanath–Agra)	2017	Bipolar	6000	±800	1728	LCC
Raigarh–Pugalur	2019	Bipolar	6000	±800	1830	VSC

and

Table 7. Back-to-Back HVDCs in India.

Name		Commissioning Year	Power Rating (Capacity) MW	DC Voltage (kV)
Vindhyachal		1989	500	±205
Chandrapur		1997	1000	±205
Vizag	Vizag I (Visakhapatnam)	1999	500	±205
	Vizag II (Gazuwaka)	2005	500	±176
Sasaram		2002	500	±205

, which represent some Back-to-Back HVDC systems in India. The technology used in the projects listed in Table 7 is LCC technology.

1.3.3. HVDC Systems in Japan

The Japanese power system operates at two different frequencies due to historical reasons. When the electricity industry was first established, the Tokyo area adopted German-made generators while the Osaka area used US-made GE generators. As a result, the western region operates at 60 Hz while the eastern region operates at 50 Hz [38].

In Japan, HVDC is utilized for frequency conversion, non-synchronous interconnection systems, and submarine DC-link interconnection systems. Among these, frequency conversion systems are the most used [39]. Three frequency converter systems in Japan include Sakuma and Higashi-Shimizu in the Shizuoka administrative region and Shin-Shinano in the Nagano administrative region [38]. DC submarine cables connect the islands of Honshu and Hokkaido (eastern Japan, operating at 50 Hz) and establish a non-synchronous connection between two separate power systems: the Minami-Fukumitsu asynchronous tie links the Hokuriku Electric Power System and the Chubu Electric Power System with a capacity of 300 MW.

1.3.4. HVDC Systems in Australia

The Australian National Electricity Market (NEM) and the Wholesale Electricity Market (WEM) are Australia’s two main power systems, covering six states. The NEM covers the eastern and south-eastern states while the WEM covers Western Australia. The Australian Capital Territory (ACT) is connected to the NEM; Western Australia and the Northern Territory are not connected to the NEM, and they have their electricity systems and separate regulatory arrangements. HVDC links are used in Australia to connect the states [40].

Table 8. HVDC links in Australia.

Name	Commissioning Year	Configuration	Power Rating (Capacity) MW	DC Voltage (kV)	Transmission Distance (km)	Technology
Terranora interconnector (Directlink)	2000	Symmetrical monopolar	180	±80	65	VSC
Murraylink	2002	Symmetrical monopolar	220	±150	180	VSC
Basslink	2006	Symmetric Monopolar	500	±400	370	LCC

lists HVDC links in Australia and

Table 9. HVDC links in New Zealand.

Name	Commissioning Year	Configuration (kV)	Power Rating (Capacity) MW	DC Voltage (kV)	Transmission Distance (km)	Technology
HVDC Inter-Island Cook StraitLink	1965 (initially built) Upgraded in 2013	Bipolar (±150)	1200 (after-grade)	±350	610 km (including 40 km submarine cable across Cook Strait)	LCC-HVDC (original), upgrade to hybrid with VSC elements

lists the earliest HVDC system in New Zealand, recognized as one of the pioneering HVDC systems worldwide.

1.3.5. HVDC Systems in New Zealand

The HVDC Inter-Island link, commonly referred to as the Cook Strait Link, serves as a vital transmission infrastructure connecting New Zealand’s North Island (terminating at Benmore) and South Island (terminating at Haywards). Commissioned in 1965, this link holds historical significance as the world’s first commercial HVDC system to employ mercury-arc valve technology [41]. Over the decades, the link has undergone substantial technological upgrades to enhance its efficiency and reliability, reflecting advancements in HVDC transmission systems.

The original mercury-arc valve-based system was replaced in 1992 with thyristor-based line-commutated converter (LCC) technology, marking a significant leap in performance and maintenance efficiency. A subsequent upgrade in 2013 further modernized the link’s control systems and operational reliability while retaining its original specifications of ±350 kV and a 1200 MW capacity. A key engineering feat of this infrastructure is its 40 km submarine cable segment, which, at the time of installation, was the longest such cable in the Southern Hemisphere.

The HVDC Inter-Island link plays the critical role of HVDC technology in enabling efficient long-distance power transmission across challenging geographical barriers. Its evolution from using early mercury-arc systems to modern thyristor-based converters underscores the dynamic nature of power transmission advancements. Future developments, including potential VSC-HVDC deployments and capacity expansions, will be pivotal in supporting New Zealand’s transition to a sustainable energy future.

1.4. Current Challenges and Research Scopes

The emergence of renewable energy resources, such as solar photovoltage arrays and offshore wind farm power plants, due to shortages of non-sustainable resources and severe environment contaminations, is accelerating the utilization of HVDC networks in power grids. The integration of HVDC grids into modern power systems requires reliable fault detection mechanisms. However, modern voltage-sourced converter-based HVDC systems and multi-terminal DC grids face unique challenges during the faults that are far more severe and complex than those in traditional AC systems. Detecting DC faults is critical in modern power systems, but the detection is fundamentally different from that of traditional AC faults in different aspects.

Table 10. Challenges in DC fault detection compared with AC faults [11,12,13,14,15,16,42,43,44].

Aspects	AC Fault Detection	DC Fault Detection
Current zero-crossing interruption [11]	Zero-crossing observed at every half cycle and easy interruption	No natural zero-crossing points-difficult interruption
Fault current dynamics [12]	Rises slower due to reactance	Rises very fast (due to no reactance and limited mainly by resistance and source impedance)
Discharging of DC-link capacitors [42]	There is no DC-link capacitor	In VSC-HVDC systems, the fault current is increased by the discharge of the DC-link capacitor
Impact of converter topology [43]	There is no converter	Short-circuit fault current behavior depends on converter topology
Grounding system [44]	Well-matured grounding	Grounding system dependency
Fault patterns [13,14]	Voltage sag; current rises with zero-crossing; phase angle change	Voltage sag + current rises without zero-crossing
Protection time [15,16]	Slower response acceptable, e.g., tens of milliseconds	Fast response required, typically 1–2 ms
Protection devices [42]	Circuit breakers; overcurrent relays	DC circuit breakers; solid-state circuit breakers; fast sensors and AI methods

shows the main differences between AC and DC faults.

With the characteristics listed above, the detection of DC faults requires fast and accurate actions to avoid unstable or vulnerable operation to DC grid equipment damage or blackout cascading. Hence, the research gaps are listed below; these constitute the motivation for us to finish this paper in this field.

○

A lack of sufficient review of signal processing methods for VSC-HVDC or multi-terminal HVDC grid fault detection.

○

The insufficiency of a single signal processing method that would comprehensively realize fault detection; for example, wavelet-based techniques can detect a fault but cannot detect the fault location.

○

Limited real-time viability; for example, wavelet packet transformation is sometimes too slow for real-time protection, which is only applied in offline simulations or laboratory validation. Delayed detection in such faults may cause fault propagation and further cascaded failure.

○

Limited use in testing and HIL framework validation under realistic and diverse grid operating conditions.

1.5. Reference Selection Methodology

Table 11. Summary of key HVDC-related keywords (1990–2025) with publication counts and average citations per item.

Keywords	No. of Publications	Avg. Citations per Paper	Comments	Years
HVDC Transmission	36,956	8	A significant increase in the total number of documents per year was observed from 1990 onwards, with a notable acceleration after 2008, reaching a peak around 2020–2022 before a sharp decline was projected toward 2026.	1990–2025
LCC-HVDC	1933	14	A continuous increase in the publication trend was shown from 1996 to 2021, followed by a decline from 2021 to 2022, and a further decrease from 2023 to the present.	1996–2025
VSC HVDC	5709	24	A steady increase in the publication trend was shown from 1996 to 2016, followed by a decline between 2016 and 2017. A brief rise was observed between 2017 and 2019, but a decline occurred again from 2019 to 2022. A short increase took place between 2022 and 2023, followed by another downward trend.	1996–2025
MMC HVDC	4831	22	A steady increase in the number of documents was shown in the chart from 1996 to 2016, peaking around 2016–2019, followed by a noticeable decline after 2020, with a sharp drop projected in 2025.	2007–2025
HVDC Protection	4608	18	A substantial increase in the number of publications was illustrated by the publication trend from 1990, with a notable surge starting around 2008, reaching a peak between 2020 and 2022, followed by a sharp decline towards 2025.	1990–2025
Multi-terminal HVDC (MTDC)	2563	24	A gradual increase in the number of documents was observed from 1990 to 2010, followed by a sharp surge to a peak around 2016. The numbers remained high until 2021 and then declined significantly through 2025.	1990–2026
HVDC fault	9386	19	The number of documents was maintained at a low and stable level from 1990 to around 2005, after which a steady increase began. Rapid growth was observed between 2013 and 2020, followed by a peak in 2023 and then a sharp decline in 2025.	1990–2025
HVDC Control	17,542	13	A consistent increase in cited documents was shown, particularly accelerating after 2008 and peaking around 2020–2021, before a slight decline in 2024–2025.	1990–2025

presents a summary highlighting key HVDC-related keywords over the period of 1990–2025, along with the corresponding number of publications and average citations per item.

During the review process, published papers between 2015 and 2025 were searched from different databases, e.g., IEEE Xplore, Scopus, and Web of Science. Then the records were selected by title, abstract, and full-text different criteria; the most relevant ones were cited as references. Figure 4 presents a flowchart of the reference selection methodology.

1.6. Paper Organization

The paper is structured to provide basic knowledge of HVDC grids, including mainstream types of HVDC grids, Line-Commuted Converter–HVDC systems, Voltage Sourced Converter-HVDC systems, multi-terminal HVDC systems, and pole configurations. Section 3 summarizes common faults in HVDC grids, including AC faults, converter stations faults and DC faults. Causes, features, and locations also are compared and presented comprehensively. Section 4 first presents a guiding flow chart of fault detection, classification, and protection. This is followed by a comparison of different signal processing methods, with working domains; running principles and pros and cons are listed with each method. Then the section is structured to review signal processing techniques—ranging from Fourier transform to time–frequency methods—for identifying faults in HVDC networks. We evaluate their performance in terms of system configuration, detected fault types, detection response times, etc., supported by case studies from the recent literature. In conclusion, challenges such as real-time processing and interoperability are discussed, along with emerging trends like edge AI and digital twins for future directions in HVDC fault detection.

2. Basics of HVDC Grids

2.1. Converter-Based HVDC Systems

HVDC systems can be categorized into three types based on design and operational features: converter technology, pole configuration, and system configuration. Depending on the type of converter being utilized, this technology includes LCCs, VSCs, and MMCs. HVDC topologies are commonly classified into four primary types: monopolar, bipolar, homopolar, and hybrid [45]. Regarding power distribution and communication, HVDC systems can be configured as two-terminal, multi-terminal, or Back-to-Back systems.

Converters play a crucial role in HVDC systems. They are essential for converting AC and DC. HVDC systems convert AC to DC (rectification) and DC back to AC (inversion) using power electronic devices. And the key functions of such systems are listed: power flow control, interconnecting grids with different frequencies, etc. Converters in HVDC systems are classified into two leading groups: CSC converters (also called line-commutated converters (LCCs)) and VSCs [46].

LCCs are based on thyristors, and such systems have traditionally been used for large-scale power transmission in HVDC applications and for a long time. However, these systems encounter several limitations, such as high reactive power consumption, the inability to control reactive power, and the absence of black start capability [47]. Meanwhile, VSCs are not based on thyristors; instead, they use other powerful electronic devices such as IGBTs [48].

Due to the advancements in electronic devices like GTO and IGBTs during the 1990s, VSC technology became essential in HVDC systems. VSCs offer several advantages, including the independent control of reactive and active power, low harmonic levels, flexible active power control, and blackout recovery capability [49]. VSCs are particularly valued for their ability to ensure secure commutation in both AC and DC systems, achieved with Pulse Width Modulation (PWM) [50]. The MMC, an advanced version of the VSC, has gained popularity in developing VSC-MTDC systems due to its modularity and scalability [51]. Figure 5 shows an IGBT-based VSC bipolar HVDC system and Table 12 compares the working profiles from LCC-HVDC, VSC-HVDC, and MMC-HVDC systems.

On the other hand, wide-bandgap semiconductors have become increasingly popular over the past decade due to significant technological developments. Compared to conventional silicon (Si IGBT) devices, wide-bandgap semiconductors offer higher breakdown voltages, better thermal conductivity, and faster switching speeds. Among them, silicon carbide (SiC) and gallium nitride (GaN) have demonstrated the most promising viability in terms of mature technology and commercial availability. However, due to its higher voltage capability and higher current rating and superior thermal conductivity, SiC is currently better suited for high-power and high-voltage HVDC grid applications [52].

SiC devices have already begun to be applied in HVDC systems; however, their use is currently mainly limited to experimental systems, pilot projects, submodules, and auxiliary power supplies. Within HVDC power systems, SiC MOSFETs present the potential to significantly reduce switching losses and enable higher switching frequencies, thereby allowing a reduction in the size and cost of passive components such as filters and transformers [53]. They have not yet been widely deployed in mainstream commercial HVDC converter stations. Nonetheless, their application is rapidly advancing—particularly in Modular Multilevel Converter (MMC-HVDC) systems—where SiC is regarded as a key enabling technology for the future. In the medium-voltage conversion stage between wind or photovoltaic sources and the HVDC bus, SiC devices are gradually replacing traditional IGBTs, enabling more compact and efficient system designs [54,55].

Table 12. Comparative analysis of LCC, VSC, and MMC technologies [56,57,58,59,60].

Feature	LCC-HVDC	VSC-HVDC	MMC-HVDC
Switching device	Thyristors	IGBTs	IGBT/SiC MOSFET submodules
Commutation	Line commutation, grid dependence on natural commutation	Self-commutation (PWM-operation commutation), no grid dependency	Self-commutation (PWM-operation commutation), no grid dependency
Harmonics	High harmonics and AC filters needed	No AC filter required	Very low (near-sinusoidal output), good AC waveform quality
Dynamic response	Slow dynamic response	Faster dynamic response compared to LCC and more control flexibility to provide AC grid support	Faster dynamic response compared to LCC system and more control flexibility to provide AC grid support
Commutation failure (AC fault response)	Vulnerability to commutation failure	No commutation failure	No commutation failure
Reactive power	Large amount of reactive power consumption, requires external control (SVC or STATCOM)	Independent control of P and Q, able to provide reactive power compensation	Independent control of P and Q, able to provide reactive power compensation
Power rating	Bulk power transmission capability and large power rating	Low power rating	Modular design and high flexibility for scalable voltage levels
Power loss	Low power loss	High switching frequency leading to high switching losses	High switching frequency leading to high switching losses
Efficiency	High efficiency	Low efficiency	High efficiency
Black start Capability	Challenging (due to commutation issues)	Yes	Yes
Weak grid Connection (Renewable integration)	Challenging to connect to weak AC systems (requires strong wind, short-circuit ratio SCR > 2)	Compatible with weak AC systems (works with system even short-circuit ratio SCR < 1.5)	Best for weak AC systems (works with system even short-circuit ratio SCR < 1), e.g., offshore wind farms
Long distance transmission	Best	Less efficient	Competitive with scalable voltage

Table 12. Comparative analysis of LCC, VSC, and MMC technologies [56,57,58,59,60].

Feature	LCC-HVDC	VSC-HVDC	MMC-HVDC
Switching device	Thyristors	IGBTs	IGBT/SiC MOSFET submodules
Commutation	Line commutation, grid dependence on natural commutation	Self-commutation (PWM-operation commutation), no grid dependency	Self-commutation (PWM-operation commutation), no grid dependency
Harmonics	High harmonics and AC filters needed	No AC filter required	Very low (near-sinusoidal output), good AC waveform quality
Dynamic response	Slow dynamic response	Faster dynamic response compared to LCC and more control flexibility to provide AC grid support	Faster dynamic response compared to LCC system and more control flexibility to provide AC grid support
Commutation failure (AC fault response)	Vulnerability to commutation failure	No commutation failure	No commutation failure
Reactive power	Large amount of reactive power consumption, requires external control (SVC or STATCOM)	Independent control of P and Q, able to provide reactive power compensation	Independent control of P and Q, able to provide reactive power compensation
Power rating	Bulk power transmission capability and large power rating	Low power rating	Modular design and high flexibility for scalable voltage levels
Power loss	Low power loss	High switching frequency leading to high switching losses	High switching frequency leading to high switching losses
Efficiency	High efficiency	Low efficiency	High efficiency
Black start Capability	Challenging (due to commutation issues)	Yes	Yes
Weak grid Connection (Renewable integration)	Challenging to connect to weak AC systems (requires strong wind, short-circuit ratio SCR > 2)	Compatible with weak AC systems (works with system even short-circuit ratio SCR < 1.5)	Best for weak AC systems (works with system even short-circuit ratio SCR < 1), e.g., offshore wind farms
Long distance transmission	Best	Less efficient	Competitive with scalable voltage

2.2. Pole Configurations in HVDC System

HVDC systems can deploy five types of links when in operation: monopolar, bipolar, homopolar, Back-to-Back, and multi-terminal configurations. HVDC systems are categorized based on their terminal configurations into two types: two-terminal and multi-terminal systems. Two-terminal HVDC systems feature a simple setup with one terminal at each end, making them ideal for direct point-to-point power transmission between two locations. Multi-terminal HVDC systems have more than two terminals, making them a more complex configuration than two-terminal HVDC systems due to multiple interconnections within the grid.

Table 13. Comparison of different types of HVDC technology [61,62,63,64,65,66,67,68,69].

Pole Configuration	Working Principle	Features
Monopolar [61,62,63]
	Utilizes a single conductor for power transmission. The earth or ground serves as the return path for the current. Sometimes metallic return is also used. Each terminal has one converter. One conductor carries the nominal voltage while the return path operates at zero or low voltage.	Beneficial for transmitting power below sea level by using the sea as a return path. Offers simplicity and more economic configuration. Higher possibility of outages. Corrosion and interference issues over long periods. Interferes with other structures or systems due to the earth return current. Causing corrosion of buried metal objects and affecting water chemistry in seawater. Causes environmental problems due to ground currents.
Bipolar [64,65,66]
	Features two conductors and two converters (rectifier and inverter) at different terminals. In a bipolar link, one conductor is positive and the other negative, with the midpoint of the inverters grounded.	Bipolar systems are reliable: as one line stops working, the configuration will enter monopole mode with grounding, continuingly functioning to supply power. The mostly used HVDC configuration. Carries up to 50% capacity during fault or maintenance on one pole. Lower loss level compared to two independent monopolar connections as it uses a single return conductor. Minimizing environmental impact by carrying very small (ideally no) current through the ground return electrode.
Homopolar [65,66,67]
	Features two conductors and two converters (rectifier and inverter) at same terminals. The homopolar link with two conductors is either positive or negative (same polarity, different from bipolar HVDC configuration), with the midpoint of the inverters as the ground return path.	Economical in operation. Less reliability in failure and power outage happens as either line stops working. For rural areas and test purposes only. Ensuring reliable connections while reducing corrosion in nearby metallic structures through single polarity (usually negative).
Back-to-Back [66,67,68]
	Features two conductors and two converters (rectifier and inverter) at different terminals. The function of homopolar system is to rectify AC power to DC and then immediately invert back to AC.	Not for HVDC transmission. For isolation and control: two asynchronous purposes only. Being highly fault-prone as the failure of one conductor can lead to reliance on others to prevent disruption of current flow. Effectively solving power grid stability issues caused by AC–DC power transfer.
Multi-terminal [66,69]
	Features more than two converter stations (rectifier and inverter) and connected to a common DC network. Multi-terminal systems allow power transfer flexibility between grids.	Power flow between any pair of terminals. Constant voltage allows parallel connection, and constant current allows series connection. Allow integration of renewables. Increased control algorithm complexity. Fast and selective DC fault isolation is challenging.

shows the diagrams, working principles, and typical features of different topologies.

3. Fault Types in HVDC Grids

Based on the features, HVDC systems are mainly used in long-distance high-power transmission. The HVDC system could be divided into two sections: the AC section (AC grid, converters including inverter/rectifier) and DC section (DC transmission line). Common faults that occur in the HVDC systems are categorized by location (AC/DC side).

• AC faults

AC side faults happen in the AC section of the HVDC system. They are classified according to their location into two main categories: symmetrical faults and asymmetrical faults; there are also other special faults. Symmetrical AC faults include three-phase short-circuit and three-phase voltage sag/swell, and asymmetrical AC faults happen between the line and ground: single-line-to-ground (SLG) fault, line-to-line (LL) fault, and double-line-to-ground (LLG) fault. Other special fault scenarios include commutation failure, phase angle jumps, and harmonic resonance. For AC faults, the causes and features are presented in Figure 6a [70,71,72,73].

2.

Converter faults

In HVDC systems, the converter stations, including a rectifier station converting AC to DC and an inverter station converting DC to AC, play a crucial role in power conversion. Faults in converters can disrupt the entire transmission line and may lead to facility damage or blackouts. Common converter faults are listed in Figure 6b [70,71,72,73].

3.

DC faults

In the long-distance HVDC transmission system, the DC line is one of the components with high fault probability, and this generates a significant effect on effective power transmission. Faults on the DC transmission line are mainly caused by cable damage, insulation failure, and equipment malfunction. The DC line is the main part to realize power transmission, and there are two types of DC line faults, the DC-line-to-ground fault (also called the pole-to-ground fault, shorted as the PG fault) and DC-line-to-line fault (also called the pole-to-pole fault, shorted as the PP fault). There are four types of DC faults; the causes and features are shown in Figure 6c [70,71,72,73].

An example of a 12-pulse LCC-HVDC system is shown to show the fault locations in different sections of an HVDC system. AC faults, converter faults, and DC faults are shown separately in green, yellow, and blue colors in Figure 7. In an HVDC system, the faults shown in Figure 7 are just for the AC input side, rectifier, and DC transmission line.

4. Signal Processing Techniques in HVDC Fault Detection

4.1. Fault Detection, Diagnosis, and Classification in HVDC Grids

To ensure that power systems remain reliable and sustainable, detecting and classifying faults in transmission lines are essential to ensure reliable and continuous power delivery [74]. Fault detection identifies an abnormal situation or disturbance. It monitors electrical parameters such as current, voltage, and frequency to recognize deviations from normal operating conditions. Fault classification refers to identifying the type of fault that has occurred to determine which phases or poles are affected by the fault. In AC systems, recognizing the type of fault can be included in single-phase-to-ground, phase-to-phase, phase-to-ground, or three-phase faults, and in DC systems, pole-to-ground and pole-to-pole faults are the types of faults in lines. The identification of a faulted phase or pole also falls under fault classification. Fault location refers to pinpointing the exact position of the fault along with lines or determining the distance along the line where a fault has occurred. Precise fault localization on transmission lines is crucial for maintaining consistent and dependable power delivery to distant destinations [75]. The rapid isolation and repair of the faulty section to minimize downtime and reliability improvements depend on the fault location. Protection in power system lines generally refers to fault detection and classification while fault location relates to system restoration and management. Figure 8 presents comprehensive fault detection methods in HVDC grids.

Non-electrical information in HVDC fault detection has also been used in HVDC fault detection because it provides complementary, early, and robust indicators of faults that traditional electrical signal measurements, voltage, or current may miss or be too late to detect.

4.1.1. Thermal

A comparative analysis between ABB’s conventional hybrid HVDC circuit breaker (CH-DCCB) and a newly proposed hybrid DCCB integrated with a solid-state fault current limiter (SSFCL) in a multi-terminal Voltage-Source Converter (VSC) HVDC system is presented in [76]. The focus is on assessing electrical and thermal stresses on key components like IGBT valves and metal oxide surge arrestors (MOSA) under low- and high-impedance DC fault conditions.

Thermal stress in Modular Multilevel Converters (MMCs) used in HVDC systems is analyzed and addresses the imbalance caused by nearest level modulation (NLM), especially under low-power-factor conditions like low-voltage ride through or reactive power injection, in [77]. The paper proposes a thermal balancing method integrated into the capacitor voltage-balancing algorithm, aiming to equalize stress across submodules and extend their lifetime.

4.1.2. Pressure

The effects of DC bias were studied via 500 kV power transformers in HVDC systems in [78]. Tests showed that DC current through the neutral terminal significantly increases transformer vibration and audible noise, introducing irregular waveforms and high-frequency harmonics. In no-load tests, vibration grows linearly with DC at low voltage and rises sharply near saturation. Under load, each 1 A DC increase raises vibration by 11 m/s and noise by 0.9 Pa; a 6 A bias boosts noise from 87 dB to 104 dB. Long-term monitoring shows that other factors may also affect vibration. Audible noise tracking is a valuable supplement and key for HVDC design.

4.1.3. Optical

A novel energy-fiber Detection and Localization (EFDL) technology for HVDC measurement systems based on Optical Time-Domain Reflectometry is proposed in [79]. The method enables the continuous monitoring and rapid fault localization of energy fibers in converter stations. Through dynamic accelerated aging tests isolating factors such as temperature, macrobending, power injection, end-face cleanliness, and pressure, key failure mechanisms were identified. Results showed that operational reliability is significantly affected by these factors, recommending operating temperatures between −20 °C and 40 °C, a bending radius ≥ 25 mm, power < 1 W, and regular fiber end-face cleaning. The EFDL technology supports the full-life-cycle management of energy fibers.

4.1.4. Ultraviolet and Infrared Imaging

Restart failures are investigated in ±800 kV HVDC lines caused by greenhouse plastic cloth in [80]. Field data and lab tests showed that under damp conditions, the plastic becomes prone to surface discharge, with a flashover voltage of −640 kV (~80% of rated). Flashover and combustion tests revealed that high fault current, dispersed discharge paths, and short restart intervals hinder arc deionization (~160 ms), leading to three failed restarts. Factors like arc drift and folded plastic contribute to persistent short circuits, resulting in a bipolar block fault.

4.1.5. Acoustic

A method combining Ensemble Empirical Mode Decomposition (EEMD) and wavelet adaptive threshold denoising to extract partial discharge (PD) acoustic emission (AE) signals from solid insulated high-voltage DC equipment was presented in [81]. The method effectively denoises low signal-to-noise ratio (SNR) signals and uses the EEMD energy entropy of intrinsic mode functions as features. Simulation and experimental results have shown that the proposed method outperforms traditional EMD in noise reduction and signal extraction accuracy, making it suitable for PD detection in various pieces of electrical equipment.

4.1.6. Electromagnetic Waves

A method combining Ensemble Empirical Mode Decomposition (EEMD) and wavelet adaptive threshold denoising to extract partial discharge (PD) acoustic emission (AE) signals from solid insulated high-voltage DC equipment was presented in [82,83]. The method effectively denoises low signal-to-noise ratio (SNR) signals and uses the EEMD energy entropy of intrinsic mode functions as features. Simulation and experimental results have shown that the proposed method outperforms traditional EMD in noise reduction and signal extraction accuracy, making it suitable for PD detection in various pieces of electrical equipment.

Figure 9 presents a flowchart of the progress of HVDC fault detection and classification. It helps understand the process in a straightforward way. The flowchart starts with system monitoring, then moves to the first step of real-time data acquisition.

In Step 1, generally, the input data are the measurements from sensors, e.g., DC voltage (V_dc), DC current (I_dc), AC side voltage and current (for converter monitoring), and control signals (e.g., firing angles for LCC, PWM signal for VSC). Generally, the tools used are phasor measurement units (PMUs) and SCADA systems.

Step 2 is to pre-process the raw time-series data. The main actions are removing noise via filters (e.g., low-pass filter).

Step 3 is to detect the fault by monitoring the anomalies through comparing the V_dc and I_dc signal against the predefined thresholds. In the first method, in common sense, if V_dc < T_threshold (e.g., 0.8 p.u.), a potential fault is flagged; if I_dc > I_threshold (e.g., 1.5 p.u.), a potential fault is flagged, too. Also, the rate of change (ROC) of voltage and current can also be used, e.g., calculating dV_dc/dt and dI_dc/dt, similarly, by comparing whether |dV_dc/dt| > ROC_threshold or |dI_dc/dt| > ROC_threshold to flag the fault. Wavelet transform could be used also to detect high-frequency components, identifying faults. Then, we come to the decision link: whether the fault is detected. If yes, proceed to Step 3. If not, the detection program returns to the start and continues monitoring.

Step 4 is to extract features for fault classification from the time or frequency domain. In the time domain, the features we have obtained are peak V_dc or I_dc values and the rate-of-change values of dV_dc/dt or dI_dc/dt. In the frequency domain, the energy of wavelet coefficients and frequency components are obtained by FFT.

Step 5 is to classify the fault, pole-to-ground (PG), pole-to-pole (PP), and converter switching faults, etc. The method used in this step is threshold-based, or machine-learning-based, or a hybrid method of these two.

Step 6 is to classify the fault location (the commonly used method is the travelling wave) and fault resistance by the fault current and voltage to estimate the fault distance.

Step 7 is protection. Corresponding actions will be activated to protect the HVDC system, e.g., grounding switches are activated if a PG fault happens; DC circuit breakers are tripped and converters will be shut down if a PP fault happens. The fault converter will be isolated if there is a fault happening.

Step 8 is to log fault report details, including the fault type, location, and fault duration. Post-fault data will be saved for further analysis.

Step 9 is system recovery. The faults should be clear by repairing fault components, converters should be restarted and synchronized with AC grid, etc. Then the system stability will be decided. If the system is stable, the program will go back to the start and resume system monitoring; if not, the program will go back to protection to adopt new actions to maintain system running reliability or be flagged for further manual maintenance.

4.2. Signal Processing Techniques

Signal processing techniques for fault detection and classification in power systems refer to analyzing signals (such as voltage and current signals gathered from the system) using computational and mathematical techniques to identify anomalies and faults. Commonly used signal processing methods for HVDC faults are shown in Figure 9. By extracting useful information from signals, signal processing methods help improve system reliability by detecting instantaneous changes in signals to find faults, differentiating between several types of faults and enabling the identification of them. Signal processing techniques can be categorized into frequency-domain, time–frequency-domain, and time-domain methods [84].

Table 14. Comparative analysis of different signal processing methods [85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105].

Methods	Domain	Working Principle	Advantages and Limitations
Fourier Transform (FT) [85]	Frequency	Decomposes signal into sinusoidal components	Simple and good for steady state Not ideal for transients, poor time resolution
Discrete Fourier Transform (DFT) [85,86]
Fast Fourier Transform (FFT) [85,86]
Empirical Mode Decomposition (EMD) [87,88]	Time	Decomposes signal into finite components (Intrinsic Mode Functions (IMFs))	Adaptive, no base function needed Fast and availability in real-time analysis Mode mixing and additional computational cost
Intrinsic Time Decomposition (ITD) [89]		Decomposes signal into a trend (baseline) and details (Proper Rotating Components), including amplitude, frequency, and phase	Simple and good for trends Less explored in HVDC
Mathematical Morphology (MM) [90,91]		Represents signal profiles in the time domain with a small sampling window	Fast, noise-resistant Requires complex computations
Wavelet Transform (WT) [92,93]	Time–frequency	Both time and frequency resolution analysis of transient features and spectral content	Precise detection and analysis of transient fault signals in HVDC systems Complexity in choosing wavelets
Discrete Wavelet Transform (DWT) [94]		Discrete level of WT	Efficient, noise-tolerant, fast Complexity with levels
Continuous Wavelet Transform (CWT) [95]		Analyzes the signal at every possible scale	Finer resolution More computational resources
Dual Tree Complex Wavelet Transform (DTWT) [95]		Improves DWT by using two parallel wavelet trees	Reduces shift sensitivity Improves directional selectivity
Maximum-Overlap Discrete Wavelet Transform (MODWT) [95,96]		A variant of DWT	Maintains translation invariance for better fault detection and better performance in noisy conditions Higher computational load
Lifting Wavelet Transform (LWT) [96]		An efficient and flexible approach to computing DWT	Reduces computational complexity Maintains key properties of wavelet decomposition
Stockwell Transform (ST) [97,98]		A generalized extension of the WT, combining the simultaneous analysis in multidomains of wavelets with a frequency-dependent Gaussian window	Adaptive resolution for transient signal analysis Challenge in limited time resolution
Short-Time Fourier Transform (STFT) [99,100]		Fixed-window FT over time	Uniform resolution in time and frequency domain Restrictions in fixed window size
Variational Mode Decomposition (VMD) [101,102,103]		Uses non-recursive methods to decompose signals into multiple Band Limited Intrinsic Mode Functions (BLIMFs)It constrains each mode to a narrow frequency, making it a time–frequency method	Based on frequency center and bandwidth Slow in real-time analysis Additional computational cost
Hilbert Huang Transform (HHT) [104,105]		EMD + Hilbert transform	Adaptive and data-driven Relying on empirical knowledge High complexity and real-time issues

presents the comparison of different signal processing methods for HVDC faults, where the working domain, running principle, and pros and cons of each method are provided.

4.3. Signal Processing Applications for Fault Detection in HVDC Lines

A comprehensive review of publications from 2015 to 2025 was conducted to examine research trends in HVDC fault detection. The findings in Figure 10 categorize studies based on their signal processing domain, illustrating the annual distribution of publications across different methodological approaches. This visualization provides insights into the evolution of research focus, highlighting whether specific techniques have gained or declined prominence over time.

Table 15. Literature review on signal processing techniques in HVDC fault detection.

Reference	Method	System	Domain	Input Data	Detection Time	Fault	Comments
[106]	DFT	Two-Terminal	Frequency	Voltage/ Current	<5 ms	PG, PP	Utilizing AC-side measurements to avoid DC-side sensors Detecting the 6th harmonic component on the AC side during pole-to-earth faults using DFT Analyzing DC fault effects on the AC side in VSC-based HVDC systems Overlooking the full impact of DC faults on the AC side due to DFT limitations Relying on DFT despite its deficiencies in DC fault analysis Failing to adequately consider DC fault impacts on the AC side Introducing detection delays due to a 3.33 ms window, making it unsuitable for ultra-fast protection Lacks validation in multi-terminal HVDC systems
[107]	DFT	Multi-Terminal VSC	Frequency	Voltage/Current	<5 ms	PG, PP	Proposing a non-pilot, communication-independent protection scheme Utilizing AC-side measurements to avoid DC-side sensors Distinguishing internal from external faults effectively Robust against high ground resistance faults Being unsuitable for ultra-fast detection due to processing delays Depending on transient current harmonics influenced by DC filter capacitors Malfunctioning during external faults due to harmonic distortion
[108]	FFT	Four-Terminal VSC-HVDC	Frequency	Voltage/Current	1–1.5 ms	PPG, NPG, PP	Introducing a CNN-based fault detection method combined with FFT and Gramian Angular Field (GAF) Enabling ultra-fast anomaly detection for swift identification of irregularities Optimizing performance by adjusting to the fault data acquisition window Avoiding false alarms through complex threshold setting Processing 1–1.5 ms data windows in real time, demanding high-performance hardware Facing high computational complexity due to the integration of CNN, FFT, and GAF techniques
[109]	FFT	Four-Terminal	Frequency	Voltage	~2.5 ms	PP, PG	Proposing a non-communication-based method using only DC inductors to identify faulted lines Detecting high impedance faults (HIFs) Facing sensitivity to non-fault disturbances affecting detection accuracy Requiring complex threshold tuning for consistent performance
[110]	STFT	Four-Terminal VSC-HVDC	Time–frequency	Voltage/Current	0.5 ms	PG	Triggering DCCB in less than 500 μs for all faults within the protection zone Detecting high impedance faults effectively Low computational burden Facilitating practical implementation due to computational efficiency Facing limitations in generalizing to multi-terminal HVDC grids because of noise sensitivity and case specificity Suffering from susceptibility of voltage derivative thresholds to measurement noise during fault detection
[111]	STFT	Multi-Terminal (CIGRE B4 DCS2)	Time–frequency	Voltage/Current	<5 ms	PP, PG	Applying effectively to hybrid lines (OHL and cables) with robustness Considering window function and hop size as critical factors for accuracy Neglecting consideration of high impedance faults (HIFs) Requiring high sampling frequency for accurate analysis Overlooking inherent STFT limitations in analyzing transient signals accurately
[112]	WT	Four-Terminal VSC-HVDC	Time–frequency	Voltage/Current	1.1 ms	PP, PG	Removing transition resistance effect by utilizing time difference between line mode and ground mode wave arrivals Depending on analytical analysis to ascertain WT parameters Deriving WT design parameters (decomposition level, mother wavelet, thresholds) via frequency-domain analysis, eliminating extensive offline simulations Calculating transient signal energy and comparing it to a fixed threshold Demonstrating efficacy in fast fault detection Adapting method to any HVDC grid topology, including symmetrical monopolar and bipolar systems Facing sensitivity to fault characteristics impacting accuracy Requiring higher sampling frequencies and complex calculations Necessitating calibration of wave head arrival time and managing dead protection zones Experiencing computational burden and susceptibility to system disturbances due to reliance on fixed thresholds
[113]	WT	Four-Terminal	Time–frequency	Voltage/Current	<2 ms	PG	Using discrete wavelet transform with Symlet decomposition for fault detection and classification Employing wavelet entropy and signal variations to distinguish fault types Enabling differentiation between internal and external faults Facilitating appropriate relay operation based on fault classification Eliminating the need for communication links Facing algorithmic complexity in implementation
[114]	VMD	Four-Terminal VSC-HVDC	Time–frequency	Voltage/Current	Not mention	PP	Achieving Mean Absolute Percentage Error (MAPE) of approximately 5% Ensuring correlation greater than 98.8% between predicted and actual fault data
[115]	VMD	Two-Terminal	Time–frequency	Voltage/Current	~23 ms	PP, PG	Identifying internal and external faults by analyzing phase differences in current components at both terminals Determining faulted pole using Hilbert energy ratios Quickly and effectively identifying faults with good tolerance to transition resistance and noise interference Operating without requiring high sampling devices, potentially usable as main protection Facing challenges with threshold setting affecting detection accuracy Experiencing reduced sensitivity to certain faults Encountering inconsistencies between offline and real-time performance Lacks validation in multi-terminal HVDC systems
[100]	ST	Multi-Terminal VSC-HVDC	Time–frequency	Current	0.3 ms	PP, PG	Distinguishing DC faults from other disturbances by examining low-frequency components to differentiate faults from load changes and healthy line transients Ensuring high accuracy, rapid response, and low computational load Discriminating between fault transients and normal operating conditions Facing challenges with low fault resistance requiring enhanced robustness Experiencing vulnerability to certain fault conditions Lacking noise resistance in the detection process
[116]	ST	Two-terminal VSC-HVDC	Time–frequency	Current	-	-	Achieving accuracy and effectiveness in fault identification Lacking clarity in procedures for determining tuning parameters Lacks validation in multi-terminal HVDC systems
[117]	MM	Multi-terminal VSC-HVDC	Time	Current	0.15 ms	PP, PPG, NPG	Utilizing multi-resolution morphological gradient and undecimated wavelet methods for detecting fault-traveling wave arrivals Comparing Rate of Change of Current (ROCOC) polarities between different relay sections for fault detection and location identification Including a faulty pole detector and two-sided fault locator for precise fault isolation Using Sliding Mode Morphological Gradient and Unidirectional Wavelet for traveling wave peak detection Achieving low computational burden in the detection process Validated through RTDS online testing
[118]	MM	Four-terminal	Time	Voltage	0.95 ms	PG, PP	Analyzing the filtered voltage waveform using a morphological gradient (MG) Employing subsequent voltage drop to identify faults Proposing a non-pilot, communication-independent protection scheme Using a relatively low sampling rate (20 kHz) Capable of detecting high resistance faults (up to 200 Ω) Relying on series inductors for effective filtering Facing high computational burden due to morphological gradient processing
[119]	ITD	Three-terminal VSC-HVDC	Time	Current	<0.8 ms	PPG, NPG, PP	Applying the Teager–Kaiser Energy Operator (TKEO) to determine the number of decompositions and support energy-based fault detection Using Rate of Change of Current (ROCOC) signals independently of communication systems to improve resilience against noise and high-impedance faults Facing challenges in setting appropriate threshold values Showing sensitivity to high-impedance faults
[120]	HHT	Multi-terminal VSC-HVDC	Time–frequency	Current	<2 ms	PP, PG	Focusing on fault characteristics within a specified frequency range while minimizing the influence of steady-state ripple components Effectively discriminating between faults and non-fault disturbances such as load changes and noise Demonstrating precision without relying on fundamental frequency-based operations Testing the method in OPAL-RT-based multi-terminal DC (MTDC) simulations and peer-to-peer (P2P) experimental setups Facing limitations due to the frequency band’s dependency on the system operating point Providing limited coverage for high fault resistance scenarios Imposing heavy computational loads, which may hinder real-time implementation
[121]	HHT	Four-Terminal	Time–frequency	Voltage/	<1 ms	PPG, NPG, PP	Identifying internal faults by deploying a Fault Detection Flag (FDF) that monitors instantaneous frequency and energy over a sliding window of local pole voltage measurements Achieving fault detection within less than 1 ms due to the fast-responding, local measurement-based scheme Ensuring high selectivity without relying on large boundary reactors Eliminating the need for additional fault classification or faulty-pole selection, thereby improving algorithm speed and reducing computational burden
[87]	EMD	Four-terminal VSC-HVDC	Time	Current	0.2 ms	PP, PG	Utilizing high-frequency IMF components to accurately determine fault inception timing Applying a Component Tracking Strategy (CTS) to analyze oscillation modes and instantaneous features Distinguishing DC faults from non-fault transients in healthy lines Demonstrating simplicity, ultra-fast response, and resistance to noise Exhibiting susceptibility to high-impedance faults Showing vulnerability to communication link disruptions, limiting robustness in certain environments
[94]	DWT MODWT	Two-terminal VSC-HVDC	Time–frequency	Current	<1 ms	PP, PG	Uses Maximal Overlap Discrete Wavelet Transform for enhanced speed and accuracy to realize two stage protection Offline Stage: Optimizes MODWT parameters (decomposition level, mother wavelet, thresholds) through a systematic procedure. Online Stage: Processes moving window energy coefficients for real-time fault detection. Lacks validation in multi-terminal HVDC systems
[122]	DWT MODWT	Two-terminal	Time–frequency	Current		PP, PG	Avoiding down-sampling through MODWT to preserve signal integrity and enhance accuracy and sensitivity compared to conventional DWT Identifying faults by analyzing the kurtosis of the cross-correlation function Simultaneously realizing fault detection and location with improved accuracy, especially at lower sampling frequencies through interpolation Demonstrating robustness against fault type, transition resistance, and noise Considering the impact of signal-to-noise ratio (SNR), with performance degradation observed when SNR drops below 20 dB Lacks validation in multi-terminal HVDC systems

presents a compilation of scholarly articles that have employed signal processing techniques in the frequency, time–frequency, and time domains for detecting faults in HVDC systems in recent years.

5. Discussion and Prospects

5.1. Discussion

In recent years, Artificial Intelligence (AI), which simulates human intelligence in machines, has been widely explored across various scientific domains. In power systems, it has attracted significant attention for its ability to analyze and solve complex nonlinear problems. AI enhances multiple aspects of electrical power systems by processing large volumes of data to address challenges such as voltage control, power quality, fault detection, planning, monitoring, forecasting, and protection [123].

Beyond traditional signal processing, AI and machine learning models are capable of detecting faults in HVDC systems independently without the need for prior signal processing techniques. Methods such as Artificial Neural Networks (ANNs), Adaptive Neuro-Fuzzy Inference Systems (ANFISs), and Support Vector Machines (SVMs) offer robust tools for HVDC fault detection by extracting valuable patterns from historical data [124]. Leveraging the pattern recognition and learning capabilities of these models, AI can analyze raw data to identify, classify, and detect anomalies that signal potential faults or abnormal conditions. These models can autonomously learn fault features and make decisions based on the patterns they have identified.

AI, as a branch of computer science, mimics human reasoning and behavior, enabling machines to make rational decisions, manage incomplete information, and adapt to dynamic conditions [125]. When combined with conventional techniques, AI has significantly improved fault detection accuracy and contributed to the development of smart protection strategies in modern power systems, though certain challenges remain [123].

While AI integration enhances signal processing by improving adaptability to changing grid conditions and enabling dynamic optimization of protection schemes, it also introduces additional complexity to the design and implementation of these systems. Moreover, the effectiveness of AI-based approaches heavily depends on the availability of large, high-quality training datasets. Such data can be difficult to obtain in diverse and rapidly evolving grid environments, leaving these models vulnerable to data errors and biases [126].

A framework for AI-machine learning-based for HVDC-system fault detection is depicted in Figure 11.

5.2. Future Research Directions

• Innovative Monitoring via Optical Sensor Networks

  ○

  We must develop magneto-optic glass sensors to replace traditional CT/PT, with increased bandwidth and integrate Optical Current Transducers (OCTs) and Optical Voltage Transducers (OVTs) for EMI-immune measurements.

  ○

  We must apply phase-sensitive Optical Time-Domain Reflectometry technology, which is an advanced technology used for the real-time monitoring of vibrations, pressure, and acoustic disturbances in DC lines without needing discrete sensors, reducing system costs.

• Digital Twin for HVDC Fault Detection

Digital twin for HVDC fault detection is an emerging and powerful application combining simulation models, real-data, and machine learning to enable intelligent fault diagnosis. A digital twin is a virtual duplicate of a physical asset, process, or system that continuously receives data from the real world (e.g., sensors, SCADA, PMUs) and mirrors its state, behavior, and evolution. For HVDC systems, it replicates transmission lines, converter stations, transformers and reactors, and protection and control systems.

The fault detection and diagnosis procedure involves integrating real-time sensing data (voltage, current, vibration, etc.) into the digital twin model, then detecting anomalies or deviations from expected behavior by comparing virtual running status of digital twin and real data, using time-domain inversion and electromagnetic propagation simulations to quickly locate fault points (e.g., line-to-ground faults or pole-to-pole short circuits).

• Quantum Machine Learning for HVDC Fault Detection

Quantum Machine Learning is an emerging interdisciplinary field combining quantum computing and machine learning. It offers novel capabilities to handle complex, high-dimensional, and time-sensitive data for more accurate and faster fault diagnosis, which is beneficial in the context of HVDC fault detection. The traditional machine learning method is easily transplanted to quantum computers to realize HVDC fault detection and classification. However, as the hardware of quantum computers is still in the early development stages and software for quantum programming frameworks is still evolving, the practical application of this technology in HVDC grids remains a future goal.

6. Conclusions

In this paper, a comprehensive review of the aspects and principles of HVDC systems, along with signal processing techniques for fault detection in these systems, has been conducted. In conclusion, the paper contributes the following key points:

• This paper has presented the developments of HVDC links in the world and the technological advancements from mercury-arc valves to IGBT-based converters, also summarizing the operational HVDC projects deployed in some technologically leading countries to showcase the state-of-the-art advancements.
• A comprehensive comparison of converter technologies (LCC, VSC and MMC) and pole configurations (monopolar, bipolar, homopolar, and MMC) of HVDC systems has been conducted.
• Different faults in HVDC systems including AC faults, converter faults, and DC faults have been summarized and compared, and the locations of major faults have been pinpointed.
• A comprehensive comparison of signal processing methods has been made, including time-domain, frequency-domain and time–frequency-domain methods, with case studies of references.
• Future research gaps have been discussed to improve reliability under diverse fault conditions in HVDC fault detection, classification, and protection.