High Voltage Circuit Breakers (HVCBs) are critical components in power systems, ensuring system reliability and protection. With the increasing complexity of electrical grids, condition monitoring (CM) has evolved significantly, transitioning from traditional maintenance strategies to intelligent predictive systems. This article explores the historical development, current challenges, and future prospects of condition monitoring in HVCBs.
Evolution of Maintenance Strategies
Traditionally, circuit breaker maintenance followed time-based or preventive maintenance (TbM), where components were inspected at predefined intervals. However, this approach proved costly and ineffective, as failures could occur between inspections. To address this, condition-based maintenance (CBM) emerged, leveraging diagnostic signals to assess real-time performance and predict potential failures. Reliability-centered maintenance (RCM) further refined this by prioritizing maintenance efforts based on component criticality. Additionally, 60%, 6% and 36% of maintenance philosophies are dealt with TbM, CBM and combination of those, respectively.
It has been reported that 40%, 30% and 24% of the origin of failures are dealt with the operating mechanism, high voltage and secondary and auxiliary circuits of SF6 CBs, respectively. In addition, it has been deduced that the aging problem could not be detectable up to 15 yrs. CBs are generally divided into three main parts, i.e. operating mechanism, high voltage parts as well as control and auxiliary. Figure 1 presents the influence of these main parts and their sub-sections on the major and minor failures (MF, mF). The origin of about 50% of failures is the operating mechanism, then control and auxiliary circuit (about 25%) and high voltage part (about 25%). As shown in Figure 1, the main parts include sub-sections with different portions of the total failures. While, it is indicated that the trip/close circuits and auxiliary switch are the origins of more than 6% of MF in the control and auxiliary part, about 50% of MF in the high-voltage part has been reported from the interruption chamber due to the erosion in the contacts, and misalignment of contacts.
In addition, it is revealed that the technology of the operating mechanism is changing toward the spring drive mechanism: from a portion of 40% in the early days to 60% these days owing to the lower failure rate in comparison with the other mechanisms. While aging, erosion, and deterioration have been reported as the most important reasons of MF (about 50%), design faults, manufacturing faults, and incorrect maintenance together are reported as the causes of 15% MF. The comparison amongst results of three worldwide surveys indicates an improvement of the reliability of CBs from 1.58 to 0.30 MF per 100 CB-year.
Key Diagnostic Signals in Condition Monitoring
Modern CBM systems rely on multiple diagnostic signals to detect faults and assess the health of circuit breakers:
1. Coil Current Signature
The coil current (CC) signal helps in detecting anomalies in trip/close operations, voltage supply, latch malfunctions, and auxiliary switch faults.
Figure 2 (a) illustrates the main components such as close coil, trip coil, latch, armature and spring drive mechanism for a 72.5 kV, SF6 CB. The measured trip coil current and its diagnostic features are shown in Figure 2 (b). The current profile is characterized through some time- and current-based features. Once a CB receives the trip/close command, the trip/close coil is energized. Once the coil current reaches the pre-designed value (IP1 at TP1), the armature inside of the coil moves up to hit the latch (at TV). Subsequently, the energy stored in the operating mechanism (usually the charged spring(s)) is released to start the motion of contacts through the mechanical transmission linkage (TP2, IP2).
Numerous efforts have addressed this signal to realize proper diagnostic features, provide a link between failures and the features, and to use this signal in the fault detection algorithms.
2. Travel Curve (TC) Analysis
The travel curve represents the motion of breaker contacts over time. This signal is crucial in evaluating the mechanical operation of the circuit breaker. Faults in damping mechanisms, spring settings, and misalignment issues are identifiable through travel curve deviations.
TC refers to the motion of the contacts against time. It is captured via the transducer as shown in Figure 4. It indicates a close-open stroke-time characteristic. The operation speed is calculated based on two predefined points on the curve. In addition, some timing characteristics such as closing/opening reaction time (Tr) and mechanism time (To) could be obtained based on this curve. Accordingly, TC is applied for evaluation of the operation time of the CBs (called timing test) as well as malfunctions in the operating mechanism
3. Dynamic Resistance Measurement (DRM)
DRM assesses the condition of breaker contacts by measuring resistance variations during operation. It provides insights into contact wear, erosion, and alignment, helping predict contact life without dismantling the breaker.
Arc and main contacts are deteriorated during the switching under normal and short circuit currents in the life cycle of the CBs. In order to evaluate the condition of the contacts without dismantling of CBs, dynamic resistance measurements (DRMs) have been suggested. In this regards, once a DC current is injected, the voltage across the breaker during the opening operation is recorded to find out the profile of the resistance versus motion as shown in Figure 5. For this purpose, a 200-A DC current is injected into the CB. Once the CB starts to move (see motion curve), the dynamic resistance (DRC in Figure 5) increases owing to the increase of the current path length. The initial part of this dynamic profile stands for the main contact resistance. While the current is commutated from the main contact to the arc contact, a jump is observed in DRC owing to a high resistance material used in the arc contact. After a while, the current is interrupted. As it is demonstrated, in addition to the main and arc contact resistance, the arc contact length could be obtained through combination of the motion curve and DRC. Once the main or arc contact is eroded or shortened the DRM profile changes in comparison with the healthy case. While the erosion increases the resistance, the shortening leads to a change at the separation time (A in Figure 5). Therefore, the malfunctions could be visible using DRM without any dismantling of the CBs.
The implementation of this technique and its repeatability have been recently focused in many efforts. The measured resistance in this test is dependent on the injected DC current and opening velocity. The resistance decreases with an increase in the injected current and decrease of the velocity. In addition, some efforts have been conducted to propose some methods rather than DRM to provide similar information in the real-time assessment. To give an illustration, in certain research trials, the overlapping time and the contact speed for generator CBs have been identified based on high-frequency impedance of the breaker.
Subsequently, the results have been compared with those obtained by DRM.
4. Vibration Monitoring
Vibration signals offer a non-invasive method to monitor mechanical health. Variations in vibration data can indicate damper failures, shaft deformations, or insulation issues.
The vibration signal generated during the opening and closing operation of the CBs has gradually become the mainstream research owing to convenient data acquisition as well as its suitability for non-invasive and real-time evaluation, which could cover the most origins of the failures occurring in the operating mechanism of CBs. This signal as shown in Figure 6 could reveal the anomalies in the mechanical section. Once a failure (such as in damper, rod, shaft, etc.) happens, the vibration signal changes in compassion with the healthy signal. To give an illustration, deformation of shaft, close coil, insulation pull rod, and damper of CB have been investigated based on the vibration signals. Similarly, the malfunction in damper and insulation push/pull rod have been identified using vibration. While the earlier is dealt with absorbing the residual energy and reducing the impact of mechanical collision during HVCB operating, the latter transmits the required energy from operating mechanism to the dynamic contact to perform closing and opening operation. To give an illustration, Figure 6 demonstrates the vibration signals resulting from failure in oil damper and the healthy signal of vibration. As it can be seen, the signal changes under faulty condition. As mentioned, the post processing techniques are necessary to precisely discriminate between healthy and faulty signals.
The main advantage of using vibration is the ability of real-time assessment of the most origins of failures. The main disadvantage of this method is the dependency of its accuracy on the number of sensors and the post processing techniques as discussed in IV-A. The reason lies in the fact that this signal is quite short in the time domain whereas the inherently highly nonlinear and non-stationary vibration signals are extremely wide in the frequency domain.
Intelligent Modelling and Machine Learning in CBM
With advancements in artificial intelligence (AI) and machine learning, CM systems are becoming more predictive rather than reactive. Key developments include:
- Feature Extraction Techniques: Techniques like wavelet transforms and empirical mode decomposition help process complex signals such as vibration and coil current data.
- Fault Classification Models: Support vector machines (SVMs), neural networks, and fuzzy logic systems are used to classify normal and faulty conditions.
- Predictive Maintenance Algorithms: AI-based regression models estimate remaining service life and degradation trends, helping utilities optimize maintenance schedules.
Future Directions in Condition Monitoring
The future of HVCB monitoring is driven by technological innovations that enhance predictive capabilities and reduce maintenance costs. In order to provide a comprehensive evaluation on progress of scientific researches on diagnosis of HVCBs, the last 20 years (2000-2020) efforts have been presented in Figure 8. It presents the statistics of researches associated with the researches coil current (CC), travel curve (TC), vibration, DRM and other methods. The trend indicates that real time approach and new signals have been investigated more.
Some emerging trends include:
1. Digital Twin Technology
An emerging concept coupled with intelligent approach is digital twins (DT), which could provide a new paradigm for fault diagnosis and remaining lifetime estimation of HVCBs. DT is a proactive bi-directional approach between real world and virtual world. Once a change is detected in the physical asset or its environment, a learning model has been employed to define simulation scenarios to explore which and how the model parameters need to be modified to mirror the observed behavior. A DT is a digital holistic emulation of a physical system or assembly using integrated simulations and service data from multiple sources across the product lifecycle. This information is continuously updated and is visualized in a variety of ways to predict current and future conditions. In case of the CBs, it could be possible to define a dynamic learning model based on some diagnosis signals such as CC, vibration, TC along with temperature, gas pressure of the interrupter, and number of operations as variables controlling the parameters of the dynamic aging model of the CB. Furthermore, it is possible that a CB has not received any command for a long time. Once this CB operates, the diagnostic features can indicate an improper operation in the first shot. However, it works well in the next operation. Therefore, the dead-time of the operation (idle time of the CB) could be involved in this dynamic model, as well.
Digital twins replicate physical assets in a virtual environment, enabling real-time simulation of circuit breaker behavior. By integrating real-time sensor data with simulation models, utilities can predict failures before they occur.
2. High-Frequency Response Analysis
Recent studies propose using high-frequency impedance measurements as a novel diagnostic tool.
Figure 7 indicates how a frequency response is implemented to detect possible malfunction in contacts. As explained with more details in a research paper, the high frequency wave is injected into the breaker. The input impedance has been recorded. Once a failure happens in the interruption chamber (e.g. finger breaking), the input impedance moves up resulting in the decrease of the resonance frequency. As shown in Figure 7, while the resonance frequency in the healthy case is 522 MHz, the losing of one and two fingers decreases it to 514 MHz and 507 MHz, respectively.
3. Integration with Smart Grids
As smart grids evolve, CBM systems will integrate seamlessly with grid management platforms, enabling real-time health monitoring and automated fault response. AI-driven analytics will play a crucial role in enhancing system resilience.
Data Gathering and Intelligent Modelling
All intelligent failure detection methods and life estimation algorithms are established based on data and reference curves. In fact, steps towards the fault detection are comprised of the obtaining of the diagnostic features, and making correlation between features and faulty and healthy cases to provide a condition assessment algorithm or one-step forward, a fault prediction approach. This section has been devoted to evaluate the role of artificial intelligent (AI) in hitherto and future research trend. The collected data from monitoring sensors in a substation or even in a network depending on available communication systems are investigated using AI techniques. AI techniques, and machine learning are widely applied in the fault detection. Table I presents the trend of application of various approaches in diagnosis of HVCBs. It could be deduced that these methods are extensively dealt with the vibration signals. The main challenge dealing with this signal is its complexity. Therefore, the post-processing step for features extraction plays a significant role owing to the quite short time domain and the extremely wide frequency domain as well as the strong nonlinearity and non-stationarity of vibration signals. In fact, vibration signal includes abundant information of CB’s mechanical conditions. Furthermore, the features of waveforms of this signal resulted from various failures may exist in different frequency components. Consideration of the noise in real-time assessment increases the complexity of the processing of vibration signal. The first step toward using vibration as a diagnostic signal is employing multi-scale decomposition methods such as empirical mode decomposition (EMD), empirical wavelet transform (EWT), wavelet packet transform (WPT), variational mode decomposition (VMD) for extraction of features. Subsequently, this signal can be transformed into several intrinsic mode functions. It can considerably restrict the accuracy of diagnostic approach.
Table 1 Trend of Intelligent Diagnosis Method
| Signals | Methods |
| Vibration | Energy entropy of Hilbert marginal spectrum (HMS), variational mode decomposition (VMD) |
| Wavelet, stacked auto encoder neural network | |
| Current vibration signal entropy weight characteristic, grey wolf optimization-support vector machine (GWO-SVM) | |
| Phase space reconstruction, Lyapunov exponent (LLE), correlation dimension (CD), Kolmogorov entropy (KE) | |
| Empirical mode decomposition (EMD) energy entropy, multiclass support vector machine (MSVM) | |
| Vibration and Coil current | Wavelet packet, Multi-mapping |
| DRM, Prediction purpose | Kernel partial least squares regression, prediction method |
| All signals | Analytical hierarchy process (AHP), evidential reasoning (ER) |
| TC, Prediction purpose | Maximum likelihood, interacting multiple models (IMM) |
| TC and auxiliary contact | Neuro-fuzzy inference system (ANFIS) |
| Coil current | Fuzzy-probabilistic analysis |
| Agglomerative Hierarchical Clustering (AHC), Data mining | |
| TC, Coil current | Neural network and support vector machine (SVM) |
The challenges or disadvantages associated with these methods are as follows: comprising of energy leakage, endpoint effect, modal aliasing, envelopment problem, adaptability, randomness of the decomposition levels affecting the robustness of diagnostic results, high computational complexity, which opposes with online monitoring, and consideration of high-frequency part of the signals. The other challenge is the realization of time segmentation of a vibration signal to complete vector features, i.e. starting time and ending time of some major vibration events such as motion of the cam and collision of the moving contact. Subsequent to the feature extractions, the next main step is to establish an effective classifier model. The point is that as the profile of the other diagnostic signals is not complex, the previous mentioned step generally does not require. However, the classification of features is applicable for all signals. As presented in Table I, the support vector machine (SVM), fuzzy based methods and back propagation neural network (BPNN) are commonly applied in the fault diagnosis of HVCBs. These methods are subjected to some disadvantages as follows: sensitivity to noisy data, dependency to training dataset, misjudgment of fault identification, and long time for training.
The other thinkable point is providing a dynamic model for fault detection. To give an illustration, most researches are taking into account that all faults are known. However, recording all types of failures for the training of classifiers is unrealistic. Therefore, the diagnosis model needs to be flexible for identification of unknown faults.
The consideration of the mentioned challenges for improvement of feature extraction and classifier methods could be addressed in future research stream. Furthermore, the trend indicates that it is necessary to move from the feature extraction approaches into a further step, i.e. identification of the thresholds of the healthy and faulty cases to prepare an intelligent fault prediction approach. In this regard, a regression method has been developed to predict the degradation of the contacts in HVCBs
It is worth mentioning that AI and machine learning techniques are helpful in discrimination between healthy and faulty situations and fault prediction. However, none of industrial tools has employed those for smart condition assessment. This is due to the lack of robustness of these methods, their dependency on training data sets and the variety of the CBs with regard to the operating mechanism, type, and different settings, which limits the generalization of the proposed algorithms. In fact, these tools have been stopped on reporting the diagnosis signals and condition assessment for own productions
Conclusion
Condition monitoring of HVCBs has evolved from manual inspections to AI-driven predictive maintenance. By leveraging advanced diagnostic signals, intelligent modelling, and emerging technologies like digital twins and high-frequency analysis, the future of CBM promises improved reliability, cost savings, and enhanced power system resilience. Utilities must embrace these advancements to ensure a robust and efficient electrical infrastructure.
References:
“Condition Monitoring of High Voltage Circuit Breakers: Past to Future” Ali A. Razi-Kazemi, Senior Member, IEEE and Kaveh Niayesh, Senior Member, IEEE DOI 10.1109/TPWRD.2020.2991234, IEEE Transactions on Power Delivery
