A Comprehensive Review of Condition Monitoring Technologies for Modular Multilevel Converter (MMC) HVDC Systems

Abstract

This paper provides an in-depth review of degradation mechanisms and condition monitoring methods for critical components in modular multilevel converter (MMC) high-voltage direct current (HVDC) systems, including insulated gate bipolar transistors (IGBTs), metallized film capacitors, and cross-linked polyethylene (XLPE) DC cables. This study systematically evaluates the strengths and limitations of existing technologies, while also projecting future trends in technological advancements. By exploring the multi-fields-coupled degradation processes of these components, the mechanisms of switching oscillations, and the flexible and controllable applications of MMC, this review offers valuable insights for improving the accuracy, real-time performance, and reliability of component condition monitoring. The findings aim to contribute to the advancement and broader application of MMC HVDC systems in modern power networks.

1. Introduction

High-voltage direct current (HVDC) transmission based on modular multilevel converters (MMCs) has become the backbone of long-distance, large-capacity power corridors worldwide, especially for integrating remote offshore wind and solar farms into modern power grids [1,2,3]. In 2010, China’s Nan’ao ±160 kV three-terminal VSC-HVDC demonstration project became the first to integrate modular multilevel converters (MMCs) for offshore wind connection, validating black start and reactive power support capabilities. In 2015, Germany’s DolWin1 ±320 kV scheme raised the MMC voltage level to 320 kV and delivered 800 MW of offshore wind power over long distances. In 2019, Brazil’s Belo Monte ±800 kV bipolar MMC entered service at 4 GW, representing the highest commercial voltage level achieved worldwide. By 2025, China’s Longdong–Shandong ±1100 kV hybrid MMC demonstration project pushed capacity to 12 GW and system availability beyond 99.5%, setting the new benchmark for MMC HVDC technology.

Traditional DC systems face issues like reactive power imbalance and voltage fluctuations due to the intermittent nature of offshore wind and photovoltaic power. With the advancement of power electronics, voltage source converters, self-commutated devices, and pulse width modulation-based flexible DC transmission have emerged [4,5,6,7]. It offers flexible control, no need for AC-side commutation current or reactive support, and supports multi-point transmission and DC grid formation [8]. These merits, however, are accompanied by new reliability concerns: converter valves and XLPE DC cables—the two most critical links—are exposed to severe electro–thermal–mechanical stresses that accelerate component aging and increase the probability of unexpected outages.

This paper delves into the key components of MMC HVDC systems, analyzing the degradation mechanisms of power devices in converter valves and XLPE DC cables during operation, along with common condition monitoring methods. It assesses these methods’ pros and cons, clarifies their applicability and limits, and offers guidance for method selection. Additionally, it provides a forward-looking analysis of the development trend of condition monitoring for key components of MMC HVDC systems, aiming to promote further development and application of the technology in power systems.

The converter valve is a key part of the MMC HVDC system. Its IGBT and metallized film capacitor conditions directly impact transmission reliability [9,10,11]. Condition monitoring can detect potential faults early, optimize maintenance schedules, cut costs, and extend service life. As a vital power transmission channel, the insulation condition of XLPE DC cables affects power transmission safety and stability [12]. Monitoring helps spot insulation aging or damage, preventing outages and ensuring stable power supply.

Although a large body of literature has investigated the aging mechanisms and condition monitoring techniques for each component separately, three key limitations still hinder practical deployment:

• Separated focus: Existing studies treat IGBTs, capacitors, and cables as isolated subsystems, ignoring the interactive degradation effects inside the MMC HVDC system.
• Single-parameter dependence: Most monitoring approaches rely on a single electrical or thermal indicator, which provides insufficient accuracy under complex, multi-physical field stresses.
• Scalability and real-time capability: High-power density electromagnetic environments impose stringent requirements on sensor isolation, sampling speed, and computational resources, yet few solutions have been validated at full scale.

To bridge these gaps, this paper presents a comprehensive review that systematically integrates the degradation physics of IGBTs, metallized film capacitors, and XLPE DC cables within the unified context of the MMC HVDC system. The contributions are summarized as follows:

This study presents a multi-field coupled aging framework that correlates dominant failure modes of critical components with electrical, thermal, magnetic, and mechanical stressors. Building upon this framework, state-of-the-art condition monitoring techniques—contact, optical, electrical, and model-based—are systematically benchmarked with respect to accuracy, latency, and engineering feasibility. Subsequently, emerging paradigms including multi-parameter fusion, low-frequency signal injection via the modular multilevel converter itself, and damping coefficient-based equivalent series resistance extraction are identified, and a comprehensive research roadmap toward highly reliable, real-time monitoring systems is proposed.

By offering an integrated perspective on component aging and its detection, this review aims to guide researchers and engineers in selecting and developing more robust condition monitoring strategies for next-generation MMC HVDC systems.

2. Overview of MMC HVDC System

The MMC HVDC system mainly consists of converter stations, DC lines, and auxiliary power systems. Each phase of the converter station includes upper and lower bridge arms, each composed of converter valve and valve reactors [13]. Power conversion and transmission are achieved by controlling the switching of solid-state semiconductor power devices in the converter valve [14,15]. XLPE DC cables, as key equipment for long-distance, high-capacity power transmission, play a core role in energy exchange between converter stations in different regions [16,17,18]. The introduction of turn-off devices can provide stable voltage support and has the advantage of strong controllability, making it the preferred choice for power transmission and grid connection in large-scale, long-distance offshore wind farms [19,20,21,22]. The basic structure and circuit topology of the MMC HVDC system is shown in Figure 1.

The core components of the MMC HVDC system include IGBTs, metallized film capacitors, and XLPE DC cables [23]. Power devices in the MMC HVDC system operate in a partially on switching state for a long time and are affected by complex conditions such as high voltage, high current, limited heat dissipation, and mechanical vibration, resulting in high failure rates. IGBT failures may cause converter valve locking, tripping, or even shutdown; metallized film capacitor failures may cause DC short-circuits or internal submodule short-circuits. The insulation material of XLPE DC cables is susceptible to damage and aging due to geographical and human factors, which can lead to current leakage, short-circuits between cables and other equipment, and power outages.

Therefore, in-depth research on the condition monitoring technologies for power devices in HVDC converter valves and XLPE DC cables, and timely maintenance and replacement to extend their service life, is of great significance for ensuring the economic operation and reliability of MMC HVDC systems.

3. IGBT Aging Mechanisms and Condition Monitoring Technologies

IGBT failure is a long-term process. The rapid switching and violent fluctuations in current and voltage create a complex degradation environment involving electrical, magnetic, thermal, and mechanical couplings [24]. To achieve higher power and longer distance electricity transmission, the power density of MMC HVDC systems needs to be increased. However, high power density designs further complicate the electrical, magnetic, thermal, and mechanical conditions of power devices. Therefore, it is essential to investigate the degradation mechanisms of converter valve components under high-power density and complex stress conditions, providing a theoretical basis for the condition detection technology of power devices.

3.1. Aging Mechanisms of IGBTs

The collaborative electromagnetic–thermal–mechanical coupling significantly accelerates IGBT degradation [25,26]. Many academic scholars have studied the degradation patterns, characteristics, and mechanisms of high-capacity press-packed IGBTs in MMC HVDC systems.

The interaction and coupling between the multi-fields are shown in Figure 2. Between the thermal and stress fields, temperature rises increase thermal stress between material layers. Stress intensification damages the internal structure, impairing heat dissipation [27]. Between the thermal and electric fields, temperature increases raise thermal resistance, leading to higher on-state voltage and current, which exacerbate power loss. The increased current generates Joule heat, further raising the internal temperature. Electric field gradients cause space charge accumulation, increasing local dielectric loss. This results in electro-mechanical creep at insulation layer interfaces, reducing the contact stability of the press-packed structure. Circuit characteristics were analyzed for electrophysics field, heat conduction characteristics were analyzed for thermal physics field, and stress characteristics were analyzed for stress field. The elevation of temperature intensifies thermo-mechanical stress within material layers, inducing irreversible damage to the internal micro-structure and simultaneously increasing thermal resistance. This, in turn, exacerbates Joule losses in semiconductor devices. Structural degradation further impairs both heat dissipation capability and dielectric strength, thereby accelerating temperature rise and deteriorating insulation performance. Aggravated semiconductor losses generate additional heat, which, in turn, drives a further increase in temperature. Moreover, high electric field gradients promote space charge accumulation, locally enhancing dielectric losses and triggering electro-mechanical creep at insulation interfaces, ultimately degrading the contact stability of press-pack structures. Consequently, the thermal, electrical, and mechanical fields are strongly coupled and mutually reinforcing, significantly accelerating device aging.

Press-packed IGBTs mainly use PEEK and nitrogen as insulating materials. The chip contacts PEEK under pressure to ensure insulation [28]. The coefficient of thermal expansion mismatch in packaging materials can cause microwear and local warping in laminated structures due to junction temperature swings, increasing surface roughness [29]. Lai Wei analyzed the impact of unbalanced clamping force on microwear under power cycling, revealing device degradation characteristics [30]. An Tong found that contact thermal resistance accounts for over 50% of total resistance, with significant increases in thermal resistance due to microwear in single-chip press-packed IGBTs [31]. The typical degradation modes of press-packed IGBTs are classified in

Table 1. Typical aging modes and mechanisms of press-packed IGBT.

Category	Deterioration Mode	Mechanism
Conditions deterioration	Border warping	gate-shot overvoltage stress
	Fretting wear	chip to fuse with the aluminum coating
	Springs deterioration	thermal stress is too high
Overstress deterioration	Overvoltage	roughness of the contact surface increases
	Circuit deterioration	springs are aged and slack

, including edge warping, fretting wear, cracks, and gate spring pin relaxation [32].

3.2. IGBT Condition Monitoring Technologies

The temperature of the PN junction temperature in power electronic devices greatly impacts performance and can predict the number of power cycles an IGBT module undergoes to assess its remaining life [33]. Nearly 60% of device failures are temperature-related, with the failure rate doubling for every 10 °C rise within the normal operating temperature range [34]. So, real-time thermal monitoring and effective control are crucial. Based on IGBT degradation mechanisms and structure, module condition monitoring parameters are mainly thermal and electrical. Existing junction temperature monitoring methods are: contact physical measurement, optical non-contact measurement, and thermally sensitive electrical parameter method.

In addition to the above methods for junction temperature monitoring, the data-model hybrid-driven approach integrates the aging mechanism of devices with big data. The physical model ensures interpretability and extrapolation ability, while the data model improves local accuracy and adaptability. In the electro-thermal model prediction framework, the real-time collected electrical and thermal quantities are input into the electro-thermal coupling model corrected by the aging mechanism to obtain the prediction of the junction temperature, providing a reliable digital twin base for remaining useful life assessment and thermal management strategies.

3.2.1. IGBT Thermal Parameter Monitoring

The contact measurement method directly embeds temperature sensors into the module to be tested. It is simple and direct, but may have low measurement accuracy due to time lags and spatial differences in the temperature information detected by the sensor [35]. The optical non-contact measurement method detects junction temperature using the inherent thermal radiation of substances, but requires damaging the IGBT package to expose the chip during measurement. Therefore, neither the contact measurement method nor the optical non-contact measurement method is suitable for widespread use in engineering.

3.2.2. IGBT Electrical Parameter Monitoring

The electro-thermal model-based prediction method calculates the chip junction temperature of power electronic devices by integrating factors like device power consumption and heat dissipation. It establishes real-time loss and transient thermal impedance models [36,37]. The basic structure of the IGBT and its seventh-order Cauer thermal network model are shown in Figure 3. Li et al. [38] used Cauer- and Foster-based thermal impedance models, converting traditional thermal circuit parameter simulations of junction temperature into 3D finite element thermal field simulations. The thermal impedance model, validated experimentally, is suitable for thermal simulation calculations on the second-level time scale. However, the thermal resistance values change with device aging, making it difficult to establish a real-time and effective transient thermal resistance model, and the accuracy of the thermal impedance network model needs to be improved.

The thermally sensitive electrical parameter method indirectly obtains junction temperature by monitoring temperature-sensitive electrical parameters of power electronic devices. Selecting electrical parameters strongly related to temperature, such as threshold voltage, saturation voltage drop, on-resistance, and current change rate, as thermal-sensitive parameters, and establishing a relationship model between them and junction temperature under specific working conditions, allows indirect junction temperature monitoring during actual operation. For high-capacity converter valves’ press-packed IGBTs, Xiao et al. proposed a module on-voltage and junction temperature calibration method and comprehensively analyzed online estimation errors [39]. However, there is less research on multi-dimensional real-time monitoring and data analysis in high-power density complex electromagnetic environments. Compared with traditional methods, Hu et al. improved accuracy by upgrading the measurement circuits and employing intelligent algorithms to generate a 3D map linking on-voltage, current, and junction temperature, thereby providing a more precise representation of IGBT junction–temperature variations under diverse operating conditions [40]. Currently, the thermally sensitive parameter method needs optimized, low-latency, high-precision sampling.

4. Capacitor Aging Mechanisms and Its Condition Monitoring Technologies

The metallized film capacitors of converter valves feature low loss, strong current, and pulse handling. In actual converters, due to the inherent properties of metals and the high-frequency oscillations in the converter, there are stray inductance and parasitic resistance in the metallized film capacitor. So, the metallized film capacitor model is equivalent to a series connection of ESL, ESR, and capacitance C. A capacitance model considering parasitic parameters is shown in Figure 4.

4.1. Aging Mechanisms of Metallized Film Capacitors

Academic research indicates that the degradation of metallized film capacitors is mainly influenced by self-healing discharge, electrochemical corrosion, and environmental temperature and humidity.

Although metallized film capacitors have self-healing abilities, repeated self-healing can produce thermal stress, leading to gas accumulation in the film, and cause local heating or chemical reactions [41]. As the self-healing effect builds up, the vaporization of the metal film intensifies, reducing the effective electrode area and causing capacitance drop [42,43,44].

Hua Li et al. [45] applied a 270 V AC to metallized film capacitors and observed corrosion at the electrode edge (EE) and active electrode area (AEA). Their experiments showed that electrochemical corrosion is the main cause of capacitance loss. Valentine et al. [46] applied superimposed AC and DC voltages to capacitors and summarized that factors like humidity, temperature, electrode material, and thickness affect electrode electrochemical corrosion, leading to capacitance decay.

Capacitors are prone to parameter drift under temperature stress and capacitance increase as well as insulation resistance drop under humidity stress, which may lead to insulation deterioration [47]. In a high-temperature and high-humidity environment, Wang Huai et al. studied the electrode plate corrosion failure process of metallized film capacitors and found out the aging patterns of equivalent series resistance and capacitance [48].

The typical degradation modes of converter valves’ capacitors can be categorized by their action mechanisms, as shown in

Table 2. Typical deterioration modes and mechanisms of converter valves’ capacitors.

Core Components	Deterioration Mode	Mechanism
metallized film capacitors	self-healing breakdown	the dielectric film melts
	galvanic corrosion	electrode corrosion
	temperature and humidity stress	accelerated oxidative corrosion

.

4.2. Metallized Film Capacitor Condition Monitoring Technologies

After degradation due to self-healing discharge, electrochemical corrosion, etc., capacitors mainly show decreased capacitance C and increased ESR. A soft fault is reached when capacitance drops by 2–5% [49]. Reference [50] conducted a fatigue test on metallized film capacitors. At a 2% capacitance change, ESR was six times the initial value. So, equivalent capacitance C and ESR are usually used as degradation indicators for online condition monitoring [51].

Metallized film capacitor condition monitoring techniques are categorized into offline, quasi-online, and real-time online monitoring based on system operating conditions during detection [52]. Offline monitoring requires capacitor removal, and quasi-online methods need converter shutdown. Real-time online monitoring, however, can provide valuable, timely information for fault prediction and predictive maintenance without disrupting normal system operation, making it a popular research focus globally.

Based on the parameter acquisition method, metallized film capacitor condition monitoring technologies are divided into direct and indirect ones. The direct method calculates capacitance by measuring current and voltage during capacitor charge and discharge. The indirect method infers capacitor conditions from system-level response changes.

4.2.1. Direct Condition Monitoring Technologies

The direct method assesses capacitor conditions by real-time measurement of electrical parameters like voltage and current. It offers fast response and good dynamic performance but poses safety risks and has high hardware costs due to coupling with the main circuit. A typical online condition monitoring method using the direct approach is the ripple method.

The ripple method extracts capacitor voltage ripple Δuc and current ripple Δic information to measure ESR (ESR = Δuc/Δic) online. It can be divided into perturbation-injected and non-perturbation ripple methods. The former injects disturbance signals to monitor ESR. For example, reference [53] introduces voltage and current ripples at specific frequencies (e.g., 30 Hz) in the converter system to calculate ESR and enhance monitoring accuracy. DC Lee et al. proposed injecting non-fundamental frequency signals into the system, extracting the changing components of the injected signals to obtain capacitance [54]. Injecting specific disturbances in the converter system improves ESR monitoring accuracy but may affect long-term reliability. The latter leverages the high PWM ripple switching frequency (e.g., several kHz) to calculate ESR directly under quasi-resonant conditions of capacitor C and ESL. Since ESR dominates impedance at high frequencies, the ripple ratio can yield results directly.

4.2.2. Indirect Condition Monitoring Technologies

The indirect method infers capacitor conditions by analyzing related signals (e.g., temperature, voltage, and current characteristics) or injecting specific excitation signals. It is non-invasive and safer but depends on model accuracy and is prone to environmental interference. The indirect online condition monitoring methods are modeling and voltage current characteristic methods.

The model method estimates the equivalent series resistance (ESR) of a metallized film capacitor using its electro-thermal model and failure physics degradation model. Wang Huai et al. [55] analyzed various failure mechanisms affecting a metallized film capacitor’s life and proposed an empirical life model based on electro-thermal stress. They introduced an inverse power law model to describe voltage impact on dielectric material degradation and used the Arrhenius equation for temperature effects on electrochemical reactions. Wang ZiJian et al. [56] developed a failure rate prediction model for metallized film capacitors considering voltage and temperature acceleration factors. However, a failure rate prediction model needs training with time-consuming accelerated aging tests and lacks a feedback mechanism to compensate for model changes unless ESR and capacitance are measured offline intermittently.

The voltage current characteristic method can be divided into fundamental frequency signal analysis and double-frequency harmonic analysis based on frequency selection in signal processing [57]. Compared with traditional methods, He Xiang Ning et al. [58] improved capacitance estimation by extracting fundamental voltage features from pre-and post-aging frequency responses, eliminating the need for direct measurement while enhancing flexibility. However, its accuracy still depends on calibrated frequency-to-capacitance mapping and system operating conditions. The double-frequency pulse method extracts 100 Hz pulse components from voltage and current ripples to monitor the ESR of metallized film capacitors in single-phase boost PFC converters [59]. However, low-frequency impedance is mainly determined by capacitance C, limiting ESR monitoring accuracy.

5. XLPE DC Cable Aging Mechanisms and Its Condition-Monitoring Technologies

XLPE DC cables are key components in MMC HVDC systems. They are widely used in urban grids and high-voltage transmission due to their excellent dielectric and heat-resistant properties. However, cable insulation can degrade from external forces during transport or installation, insulation quality issues during manufacturing, or corrosive physical and chemical environments around the cable. These factors lower the cable’s insulation level, shortening its lifespan and threatening power system stability. Therefore, researching cable insulation degradation mechanisms and condition detection methods is crucial for timely fault detection and swift repairs.

5.1. Aging Mechanisms of XLPE DC Cables

During long-term operation, XLPE cable insulation layers degrade irreversibly due to electro-stress concentration, thermos-oxidative aging, and mechanical deformation, threatening power supply reliability. Electric field distortion-induced partial discharge and electrical tree growth, high-temperature molecular chain scission, and micro-structural defects from laying bends or vibrations form the multi-physical field-coupled failure mechanism of XLPE insulation aging [60,61]. Research on the interaction mechanisms of electro–thermal–mechanical stresses on XLPE insulation is needed to reveal the link between micro-structural evolution and macro-performance degradation, supporting cable condition assessment and life prediction.

In high electric fields, needle tips inject electrons and holes into XLPE via field emission, triggering partial discharge, which powers electrical tree growth [62]. The FTIR analysis of aged and unaged XLPE shows that thermal aging first increases then decreases cross-linking, damaging the polymer network and lowering breakdown strength [63]. Mechanical tensile stress reduces chain scission energy, promoting micro-crack formation and favoring water tree formation [64].

5.2. XLPE DC Cable Condition Monitoring Technologies

XLPE DC cables, when subjected to electro-thermal aging, mechanical fatigue, and environmental corrosion, develop microscopic defects like water tree propagation and gas gap ionization. These defects eventually form penetrating discharge channels [65,66]. During long-term operation, cable conditions can be inferred from indicators such as partial discharge, which is sensitive to insulation defects [67]; high grounding currents caused by water ingress or sheath damage; and insulation resistance and dielectric loss, which reflect insulation degradation and moisture [68]. Monitoring these aspects is crucial for detecting hidden hazards and preventing faults.

Cable condition monitoring for XLPE DC cables can be performed offline or online [69]. Offline monitoring is used for completion acceptance and periodic health assessments but risks accelerating insulation degradation due to residual space charges from DC voltage. Online monitoring enables continuous condition sensing and captures transient degradation features but may suffer signal attenuation from multi-conductor coupling effects.

5.2.1. XLPE DC Cable Offline Testing Technologies

Offline cable insulation condition monitoring involves measuring insulation parameters like resistance, dielectric loss tangent, and DC voltage endurance, with leakage current detection. Comparing these parameters with thresholds indicates the cable’s insulation aging and health conditions.

Insulation resistance measurement uses a megohmmeter to apply DC voltage to the cable, assessing the main insulation performance. Key parameters derived from this are the absorption ratio K (60 s/15 s insulation resistance ratio) and polarization index P (10 min/1 min insulation resistance ratio), which quantify cable aging. If the XLPE insulation is damp or has conductive channels, the insulation resistance, K, and P all decrease [70]. Insulation resistance measurement is simple and low-cost but insensitive to local defects and unable to pinpoint faults.

The Schering bridge measures the capacitance and dielectric loss tangent (tanδ) of electrical equipment [71]. By adjusting the bridge’s variable resistance R₃ and capacitance C₄ to balance the bridge, the dielectric loss tangent is calculated. The schematic diagram of the Schering bridge test is shown in Figure 5. The Schering bridge can detect insulation material aging, moisture, and local defects under standard conditions, but results may be affected by environmental factors like temperature and humidity.

The DC voltage endurance test measures the cable’s voltage endurance and leakage current under high DC voltage. Leakage current magnitude reflects insulation conductivity; higher currents indicate worse insulation [72]. However, results can be significantly influenced by cable length and environmental conditions, necessitating comprehensive assessment with other parameters.

5.2.2. XLPE DC Cable Online Monitoring Technologies

Online cable insulation condition monitoring mainly involves key parameters like insulation resistance and dielectric loss tangent (tanδ). These technologies enable real-time, dynamic monitoring of cable insulation conditions, thus safeguarding the secure operation of power systems.

Online insulation resistance monitoring methods, including DC superposition, AC superposition, and DC component techniques, as shown in

Table 3. Online insulation resistance monitoring methods.

Detection Technology	Measurement Parameters	Superimposed Voltage
DC superposition	Insulation resistance	Power frequency + DC
AC superposition	Characteristic current	Power frequency + AC
Low-frequency superposition	Low-frequency current	Low-frequency
DC component	DC composition	Power frequency
Harmonic component	Harmonic composition	Power frequency + AC

, are often used to diagnose cable moisture and water treeing. However, these methods are not sensitive to local degradation and cannot precisely locate faults [73], limiting their large-scale application in HV cable systems.

For online dielectric loss angle monitoring, VLF dielectric loss testing is increasingly adopted, in line with IEEE Std 400.2 [74]. VLF dielectric loss testing involves obtaining cable voltage and leakage current signals via a voltage transformer and current transformer, then sending them to a digital measuring device for analysis using Fourier algorithms. The schematic diagram of the online dielectric loss angle is shown in Figure 6.

The relationship between tanδ, power supply voltage angular frequency (ω), sample capacitance (C), and parallel equivalent circuit insulation resistance (R) is given

$$\tan \delta ={\displaystyle \frac{1}{\omega RC}}$$

(1)

Since R and C are largely unaffected by ω, tanδ increases as ω decreases. This means that lower test voltages can detect defects like water trees in cable insulation layers [75]. When a cable’s tanδ exceeds 0.01, the insulation is deemed unsatisfactory. On-site, measurement devices with 0.01% sensitivity can assess cable insulation conditions.

6. Challenges and Outlook of MMC HVDC Condition Detection Technologies

6.1. Challenges of Existing Technologies

This paper reviews the degradation mechanisms and failure characteristics of key components in MMC HVDC systems, focusing on the research progress in multi-dimensional perception technologies and health assessment methods for operational conditions. A current conditions analysis shows that online monitoring of converter valves’ IGBTs, metallized film capacitors, and XLPE DC cables has limitations in real-time performance, accuracy, and life prediction, necessitating improvements via new sensor integration, multi-information analysis, and diagnostic algorithm optimization. The main technical challenges are as follows:

• IGBT condition monitoring parameters include on-state saturation voltage drop, gate leakage current, and quasi-gate voltage. Current research mostly relies on a single electrical parameter for life model fitting, leading to low accuracy in complex conditions [76]. Testing hardware circuits must achieve high-voltage isolation without compromising accuracy. Additionally, the high-power density electromagnetic environment demands significant system computing resources for real-time monitoring and analysis.
• Metallized film capacitor degradation is marked by decreased capacitance and increased ESR [77]. Existing monitoring methods are direct (switch-dependent) and indirect (calibration or special condition-based). The latter depends on extensive training data. The high-power density environment also requires high sampling precision. When valve hall temperature swings from 25 °C to 65 °C in summer, gel-filled fiber-optic current sensors exhibit a 3.2% sensitivity drift; without temperature compensation, the resulting ESR estimate deviates by 6%.

Furthermore, these capacitors are known to fail abruptly when subjected to over-voltage surges or short-circuit currents—events that lie outside the scope of traditional long-term aging studies. There is presently a gap in understanding both the transient degradation mechanisms triggered by such faults and the cumulative rules governing parameter evolution under repetitive electrical stress. To capture these fast dynamics in real time, online ESR extraction based on PWM-ripple analysis, for example, must sample at 10–20 kHz with a ≥16-bit effective ADC resolution to maintain ESR repeatability below 0.1%, while the entire estimation loop has to complete within 2 ms to remain compatible with 500 Hz control update rates.

3.

XLPE cable degradation is reflected in a reduced absorption ratio K and polarization index P and increased dielectric loss factor. Offline monitoring is precise but requires power outages. Online monitoring needs additional transformers and is susceptible to noise-induced measurement accuracy reduction [78]. In a 525 kV offshore MMC HVDC project, 50 Hz harmonics coupled through the grounding grid induced sheath currents up to 0.5 A, causing an online tanδ estimation error exceeding 0.3%. Only after installing magnetic shielding rings and an adaptive notch filter could the error be suppressed below 0.05%. Online VLF tanδ monitoring now requires a 0.1 Hz test frequency with a <0.01% phase angle accuracy (≈3.6 arc-second resolution) and overall loop latency < 1 s, highlighting the tension among hardware cost, EMI immunity, and real-time performance.

6.2. Outlook on Technologies Development Trends

In direct response to the limitations outlined in Section 6.1, the following technology trends are framed around a hybrid framework that couples digital twin-driven physics with data-driven intelligence. The digital twin continuously mirrors electro–thermal–mechanical states under multi-field stress, generating synthetic labels to compensate for scarce failure records. Concurrently, lightweight AI models running at the edge exploit residual redundancies in sensor data to suppress EMI-induced artefacts and deliver sub-millisecond prognoses. This synergy not only mitigates the hardware-centric bottlenecks highlighted earlier but also provides a scalable pathway towards high-reliability, real-time health management for next-generation MMC HVDC systems.

6.2.1. Development Trend of IGBT Condition Monitoring Technologies

To advance IGBT condition monitoring under complex operating scenarios, severe-condition simulation, multi-fields modeling, and accelerated aging tests are essential. For hardware systems, detecting IGBT saturation voltage drop under high-voltage isolation is crucial. A gate leakage current monitoring method based on gate drive resistance sampling is proposed to establish a failure early warning system and life prediction model.

In MMC HVDC systems, converter valves endure high current and voltage stresses under AC/DC short-circuit conditions, leading to core component degradation. To study this, simulation models of half-bridge and hybrid MMC topologies are developed. These models analyze converter valves’ IGBT transient processes under AC/DC short-circuit conditions, assessing the impact of control protection strategies and switching states on current and voltage stresses.

For accurate and real-time IGBT monitoring and life prediction, multi-scale simulation in high-power density electromagnetic environments is necessary. This involves developing power loss–thermal–mechanical coupling models and analyzing current distribution and stress characteristics. The IGBT multi-fields simulation model of the converter valves is shown in Figure 7. This multi-parameter, edge AI approach directly tackles Challenge 1 in Section 6.1—single-parameter inaccuracy and high-isolation sampling—by fusing saturation voltage, gate leakage, and thermal data at <1 ms latency.

Future research should focus on the degradation patterns of press-packed IGBT parameters (e.g., junction temperature, thermal resistance) under severe transient conditions [79]. This will involve establishing multi-parameter monitoring and early warning systems and creating IGBT life prediction models based on parameter evolution and thermal–mechanical stress simulation results.

6.2.2. Development Trend of Metallized Film Capacitor Condition Monitoring Technologies

Existing research shows that as metallized film capacitors age, their ESR increases, so predecessors have conducted much work on calculating and extracting ESR for capacitor condition detection. However, due to the low ESR value of metallized film capacitors, existing ESR monitoring methods have low sensitivity. To improve the monitoring accuracy of converter valves’ capacitors and reduce the impact of operating conditions and sensors on device condition monitoring, an equivalent circuit for metallized film capacitors is built according to the characteristics of the actual DC link capacitor’s metallized film and packaging materials [80]. Based on the law of energy conservation, the input power on the DC side of the converter is approximately equal to the output PWM pulse power on the AC side. The switching excitation of the PWM is close to a rectangular wave. According to the Fourier theorem, a periodic rectangular wave can be decomposed into the sum of infinite sine waves using a Fourier series.

In a practical converter system, stray inductance and parasitic resistance are bound to exist in the DC busbar. At the resonant frequency, the impedance in the converter is minimal, and a resonant current is generated under the action of the sine wave at the resonant frequency in the PWM excitation. The damping characteristics of the switching oscillation in the converter are used to extract the ESR information of the metallized film capacitors. A damping coefficient sensitive to ESR but not to operating conditions and sensors is chosen as the characteristic parameter for condition monitoring. The oscillation mechanism of the metallized film capacitor switch is shown in Figure 8.

The parameters are defined as follows:

$$\left\{\begin{array}{l}{L}_{\mathrm{res}}=ES{L}_{\mathrm{Cdc}}+ES{L}_{\mathrm{Cs}}+L_{\mathrm{dc}2}\\ C_{\mathrm{res}}=C_{\mathrm{dc}}{C}_{\mathrm{s}}/(C_{\mathrm{dc}}+C_{\mathrm{s}})\\ R_{\mathrm{res}}=ES{R}_{\mathrm{Cdc}}+ES{R}_{\mathrm{Cs}}+R_{\mathrm{dc}2}\end{array}\right.$$

The current on the buffer capacitor is the oscillation current, which is obtained by dividing the voltage by the impedance:

$$\begin{array}{l}{i}_{\mathrm{sw}\_\mathrm{res}}(s)=U_{\mathrm{s}}(s)/Z_1(s)\\ i_{\mathrm{sw}\_\mathrm{res}}(t)={\displaystyle \frac{L_2}{L_1+L_2}}{I}_{\mathrm{PWM}}{e}^{-{\displaystyle \frac{t}{\tau }}}[\cos (2\pi f_0\mathrm{t})-({\displaystyle \frac{R_1+R_2}{2\omega (L_1+L_2)}}-{\displaystyle \frac{R_2}{\omega L_2}})\sin (2\pi f_0\mathrm{t})]\end{array}$$

The expression of the resonant frequency f₀ and the attenuation time τ is:

$$f_0=1/T_0={\displaystyle \frac{1}{2\pi \sqrt{C_{\mathrm{res}}{L}_{\mathrm{res}}-{(\frac{2L_{\mathrm{res}}}{R_{\mathrm{res}}})}^2}}}\approx {\displaystyle \frac{1}{2\pi \sqrt{C_{\mathrm{res}}{L}_{\mathrm{res}}}}}=f_{\mathrm{n}}$$

$$\tau ={\displaystyle \frac{2L_{\mathrm{res}}}{R_{\mathrm{res}}}}={\displaystyle \frac{\sqrt{1-{\zeta }^2}}{2\pi f_0\zeta }}$$

To avoid interference from complex electromagnetic conditions, shielding measures are adopted. Moreover, an adaptive filtering algorithm is used to denoise the received signal, improving calculation accuracy. At the same time, the phase shift characteristic caused by adaptive filtering is analyzed, and a corresponding compensation algorithm is proposed.

6.2.3. Development Trend of XLPE DC Cable Condition Monitoring Technologies

To reduce project costs, avoid noise from voltage transformers and digital generators, and leverage the flexibility of MMC, software-controlled low-frequency AC signal injection is proposed for cable insulation health monitoring [81]. Current sensors on the cable shielding ground measure leakage current for Fourier analysis. The phase angle difference between leakage current and MMC output AC voltage give the dielectric loss angle, enabling quantitative insulation assessment, which is accurate, simple, and efficient. The principle of online monitoring for HVDC cables via MMC low-frequency injection is shown in Figure 9.

First, based on the HVDC system parameters, a suitable low-frequency signal is chosen and injected using a specific control strategy. Next, the low-frequency monitoring device measures the current at the cable ends and sends it to a smart analysis platform. Finally, the data is analyzed to evaluate the insulation conditions. To avoid voltage and current sensor phase differences, the algorithm estimates insulation parameters from leakage current changes, reducing hardware costs and enhancing accuracy.

MMC low-frequency injection-based HVDC cable online condition monitoring directly addresses Challenge 3 by transforming the converter itself into a controllable, low-frequency signal source. Through software-only modulation of the MMC arm voltage references, a precisely shaped 0.1 Hz sinusoidal component is superimposed onto the DC link voltage without any external transformer or auxiliary generator. This method ensures stable operation of the power system while maintaining high detection accuracy.

7. Threshold Design and Response Strategy

Effective condition monitoring systems require well-designed warning thresholds and corresponding response strategies to ensure timely intervention and prevent catastrophic failures. This section discusses the principles of threshold design and proposes a framework for response strategies based on the monitoring technologies reviewed in previous sections.

7.1. Multi-Parameter Threshold Design

For IGBTs, thresholds should integrate multiple indicators such as junction temperature, on-state voltage drop, and gate leakage current. For example, a warning could be triggered if the junction temperature exceeds 125 °C or the on-state voltage drop deviates by more than 10% from the baseline.

For metallized film capacitors, thresholds should consider both capacitance decay (e.g., >5% reduction) and ESR increase (e.g., >50% from initial value).

For XLPE DC cables, thresholds could include dielectric loss tangent (tanδ > 0.01) or insulation resistance (e.g., <100 MΩ for 1 km of cable).

7.2. Response Strategies

For IGBTs, early degradation signs trigger dynamic current balancing and switching frequency reduction in MMC submodules. Accelerated aging requires backup submodule activation and scheduled replacement. Critical failures demand immediate blocking and DC breaker isolation. This approach enhances reliability while minimizing downtime.

For metallized film capacitors, Initial deviations prompt impedance analysis and voltage adjustments. Moderate degradation requires reactive power optimization and planned inspection. Severe cases need rapid isolation via pre-charge resistors to prevent system failures.

For XLPE DC cables, early issues are located via distributed temperature and partial discharge sensors. Developing defects require MMC-based very low frequency (0.1 Hz) dielectric testing and possible voltage reduction. Critical insulation loss necessitates backup line activation and section replacement, supported by real-time condition evaluation algorithms. This integrated approach improves diagnostic accuracy while maintaining system availability.

8. Conclusions

8.1. Research Summary

MMC HVDC systems are crucial for efficient long-distance, high-capacity power transmission, offering flexible control and stability. This paper comprehensively reviewed the aging mechanisms and condition monitoring methods of key components in MMC HVDC systems, including the IGBT, metallized film capacitor, and XLPE DC cable. The existing monitoring technologies are compared with the innovative monitoring technologies, and the respective characteristics of different technologies are analyzed, as shown in the

Table 4. Different monitoring technologies for MMC HVDC components.

Components	Monitoring Technology	Features
IGBT	Thermal monitoring	Simple structure but slow response
	Electrical monitoring	Protect package but affected by load current and aging trajectory drift
	Multi-parameter and AI integration	High accuracy, real time, and scalability
Metallized film capacitor	The ripple method	Simple Hardware but EMI sensitive
	PWM switching oscillation damping method	Not affected by operating conditions/sensor drift
XLPE DC cables	DC superposition method	Potential transformers are required
	Sine wave injected by MMC	Software regulation without additional potential transformer

.

For IGBT, the aging is significantly influenced by the coupled effects of electromagnetic, thermal, and mechanical factors. Existing condition monitoring methods, such as physical contact measurement, electro-thermal model prediction, and thermally sensitive electrical parameter methods, have limitations in accuracy, real-time performance, and applicability under complex conditions.

For the metallized film capacitor, the aging is significantly influenced by self-healing discharges, electrochemical corrosion, and environmental factors. Existing condition monitoring methods focusing on capacitance and ESR have been developed. The ESR online monitoring method based on PWM switching high-frequency oscillations is immune to converter operating conditions, and its frequency–domain algorithm effectively suppresses low-frequency harmonic interference. However, improvements are needed in sensitivity, accuracy, and adaptability for complex electromagnetic environments.

For an XLPE DC cable, the aging is significantly influenced by electrical stress, thermal aging, and mechanical factors. Existing condition monitoring methods, including offline and online methods, face challenges in balancing precision, cost, and system stability. MMC low-frequency injection-based HVDC cable online condition monitoring takes advantage of the MMC’s flexibility and software-defined control to inject a carefully shaped low-frequency AC signal, avoiding noise from voltage transformers and digital generators.

8.2. Future Outlook on Intelligent Sensing

The integration of micro-electro-mechanical systems-based multi-field sensors, fiber-optic Bragg grating networks, and self-powered wireless nodes will enable high-resolution, EMI-immune, and maintenance-free sensing inside converter valves and cable joints. Coupled with on-chip AI inference, these smart sensors will push spatial-temporal resolution to sub-millimeter and sub-millisecond levels, laying the groundwork for edge-native health management of next-generation MMC HVDC systems.

8.3. Future Outlook on AI-Enhanced Modeling

Most existing early warning frameworks remain hardware-centric and tend to saturate in accuracy once physical sensors reach their limits. Complementing or partially substituting these systems with AI-assisted prediction is an emerging trend: machine-learning models, once trained on multi-modal data streams (electrical, thermal, mechanical, and environmental), can detect incipient faults hours to days before conventional thresholds are crossed and can continuously refine their prognoses through online learning. Such data-driven approaches, however, still face limitations: the scarcity of labelled failure data from real HVDC plants forces models to extrapolate from accelerated aging datasets, while the high-power density electromagnetic environment can corrupt sensor data and mislead algorithms. Therefore, future research should focus on hybrid frameworks that fuse physics-based digital twins with robust AI models, implement edge computing architectures for real-time inference under harsh EMI conditions, and establish open industry-wide databases to enhance the generalizability of AI-based early warning tools for MMC HVDC systems.

Multi-parameter perception fusion and AI-enhanced modeling are of great help to the fault warning and life assessment of MMC components. The working principle of the system is shown in Figure 10. This schematic diagram visually demonstrates the integration logic of physical modeling and data intelligence in hybrid methods, highlighting the mechanism between digital twins and edge AI.

Building upon these cross-cutting advances, component-specific efforts remain indispensable. Specifically, for IGBT, research on aging mechanisms and condition monitoring methods under severe transient conditions is essential. For metallized film capacitors, exploring new condition monitoring techniques to enhance sensitivity and accuracy is important. For XLPE DC cables, optimizing online condition monitoring methods to improve reliability is necessary. By integrating intelligent sensing and AI-driven insights with targeted component studies, the reliability, safety, and efficiency of MMC HVDC systems can be significantly improved, promoting their broader application in power transmission.