A Decision Support Scheme for High-Voltage Disconnecting Switch Condition-Based Maintenance Based on Online Monitoring of Contact Temperature and Operating Torque

1. Introduction

The modernization of power systems demands higher intelligence, informatization, and reliability from infrastructure equipment. As primary apparatus in substations, high-voltage disconnecting switches isolate voltage and ensure the safety of other equipment during maintenance. However, these devices are significantly influenced by environmental and climatic conditions, making them susceptible to improper closure, contact degradation, and mechanical failures.

Traditional maintenance practices—either periodic overhauls or reactive repairs—are increasingly inadequate. With the transition from regular maintenance to condition-based maintenance, there is a pressing need for real-time condition awareness and intelligent decision support. The challenge lies in identifying reliable, measurable parameters that can accurately reflect both electrical and mechanical health.

2. Technical Foundation: Dual-Parameter Online Monitoring

2.1 Contact Temperature Monitoring

Contact temperature is a direct indicator of contact resistance and thermal condition. In high-voltage disconnectors, most faults ultimately manifest as heat. Poor contact increases contact resistance, leading to abnormal temperature rise that can accelerate degradation or cause catastrophic failure.

Several sensing technologies enable online temperature monitoring:

· Fluorescent fiber optic sensors provide complete electrical isolation, immunity to electromagnetic fields up to 100 kV/m, and ±0.2°C accuracy with service lives exceeding 20 years in SF₆ atmospheres. These are particularly suitable for GIS environments where direct contact access.

· Fiber Bragg Grating (FBG) sensors offer purely optical measurement, enabling installation at high-voltage potentials. FBG-based systems can simultaneously monitor contact temperature and switch position state.

· Passive wireless RFID sensors with integrated temperature and humidity sensing enable wireless data transmission, reducing wiring complexity.

For GIS disconnectors where sensors cannot be directly mounted on conductor surfaces, indirect temperature calculation methods using multipoint shell temperature measurements and environmental temperature compensation have been validated through 3D finite element modeling.

2.2 Operating Torque Monitoring

Operating torque reflects the mechanical condition of the disconnector’s drive train. Mechanical anomalies such as jamming, incomplete closing or opening, contact wear, and spring fatigue all manifest as characteristic distortions in the torque waveform.

The monitoring system typically employs:

· Torque sensors installed on the main spindle of the operating mechanism to measure real-time torque during switching operations.

· Angle encoders positioned on the output shaft to synchronize torque measurements with rotational position.

· Motor current analysis as an indirect torque measurement method, since motor current is proportional to output torque.

Research has demonstrated that when mechanical faults occur, the torque waveform distorts—peak values increase and peak timing shifts. These deviations provide quantitative diagnostic features for fault classification.

3. System Architecture

The proposed decision support system adopts a three-layer architecture:

Perception Layer: Distributed sensors collect contact temperature (via FBG, fluorescent fiber, or RFID sensors), operating torque (via spindle torque sensors), rotation angle (via encoders), position status (via micro-switches), and environmental temperature/humidity.

Transmission Layer: Data is transmitted through wireless communication protocols (LoRa, 5G, or passive RFID backscatter) to edge computing units.

Application Layer: Edge devices perform data preprocessing and feature extraction, then transmit processed data to a cloud-based diagnostic platform for AI-driven analysis, fault prediction, and maintenance decision generation.

4. Decision Support Logic

4.1 Data Fusion and Feature Extraction

Multi-parameter data fusion is essential for reliable diagnostics. The system integrates:

· Temperature data (absolute values, rates of change)

· Torque-angle characteristic curves (peak values, waveform morphology)

· Position status (open/closed/partially closed)

· Historical trend data

Advanced feature extraction includes temperature change rates, resistance fluctuation variance, and displacement-temperature correlation coefficients.

4.2 AI-Driven Diagnostic Models

The system employs hybrid AI models for intelligent diagnosis:

· CNN-LSTM fusion models extract spatial features from multi-sensor data through convolutional neural networks, while long short-term memory networks capture temporal dependencies and degradation trends.

· Multi-algorithm warning models dynamically adjust thresholds to minimize false alarms.

· Knowledge graph integration with Ripplenet algorithms has demonstrated diagnostic accuracy of 95.96% for high-voltage switchgear faults.

4.3 Maintenance Decision Matrix

The decision support system generates actionable maintenance recommendations based on a severity matrix:

Condition Temperature Torque Recommended Action

Normal Within limits Normal waveform Continue monitoring

Alert Elevated trend Slight deviation Schedule inspection

Warning Above threshold Distorted waveform Plan maintenance

Critical Rapid rise Severe distortion Immediate intervention

5. Field Validation and Performance

The proposed scheme has been validated through multiple field deployments:

· Intelligent monitoring system tests demonstrated stable performance with state information collection, intelligent analysis, and remote monitoring meeting all operational requirements.

· AIS disconnector monitoring systems achieved real-time fault detection and operational status evaluation, significantly improving maintenance efficiency.

· 220 kV substation deployments using CNN-LSTM models with multi-parameter perception achieved state identification accuracy of 98.5% and reduced maintenance frequency by 40%, successfully transitioning from periodic to condition-based maintenance.

· Springer-reported online monitoring systems achieved a defect recognition rate exceeding 96%, detection accuracy of 92.19%, and a false alarm rate below 3%.

6. Conclusion

Field validation confirms the scheme's effectiveness, with defect recognition rates exceeding 96%, state identification accuracy of 98.5%, and maintenance frequency reduction of 40%. As power systems continue their digital transformation, such intelligent CBM decision support systems will become essential for improving substation reliability, reducing unplanned outages, and optimizing asset lifecycle costs.