TL;DR
This paper investigates the hygrothermal degradation of epoxy electrical insulating materials, crucial for electrical equipment longevity. The authors developed and compared different mathematical models for predicting this degradation, finding that a Bayesian experimental design approach significantly outperforms traditional methods in accuracy and efficiency.
What Problem Does This Paper Address?
Electrical insulation systems (EIS) are vital components in electrical machines and transformers, and their failure can lead to significant downtime. Degradation of these materials, influenced by factors like temperature and humidity (hygrothermal stress), is a primary cause of failure. While modeling this degradation is important for estimating remaining useful life (RUL) and implementing predictive maintenance, there's a need for accurate and efficient modeling techniques, especially for common materials like impregnating mono-component epoxy resins. Existing methods often struggle with the complexity of multi-factor degradation and can be inefficient in experimental design.
Methods
The study employed a multi-pronged approach:
- Material Characterization: Thermal analyses (DSC, STA, DMA) were used to understand the epoxy resin's properties, curing process, and thermal behavior.
- Hygrothermal Degradation Experiment: Specimens were subjected to controlled temperature and humidity conditions for a set duration (336 hours).
- Material Testing: Dielectric properties (relative permittivity, dissipation factor, dielectric strength) and mechanical properties (tensile strength, Shore D hardness) were measured before and after degradation.
- Degradation Modeling: Three modeling approaches were compared:
- Physically Informed Empirical Relation: A standard approach using established physical principles.
- Response Surface Methodology (RSM): A statistical method for optimizing experimental designs and modeling responses.
- Bayesian Experimental Design (Bayesian Model Averaging - BMA): A novel approach that uses probabilistic methods to compare multiple models and intelligently select the next experimental point to maximize information gain. This method was implemented using a custom Python script.
Datasets
The study utilized experimental data generated from the hygrothermal degradation experiments on epoxy resin specimens. No pre-existing public datasets were used for training or evaluation. The experimental data consisted of measurements of dielectric and mechanical properties under various temperature and humidity conditions.
Key Insights and Ideas
- Mechanical Properties as Indicators: The study found that mechanical properties like yield strength and hardness are sensitive indicators of the material's condition after hygrothermal stress and can recover to some extent after relaxation.
- Bayesian Design Efficiency: The Bayesian experimental design approach proved to be significantly more efficient than traditional methods, requiring fewer experimental points to achieve a highly accurate degradation model.
- Model Superiority: The Bayesian Model Averaging (BMA) approach, which probabilistically compares multiple candidate models, demonstrated superior predictive accuracy compared to physically informed empirical models and standard RSM.
- Under-Curing is Common: The epoxy resin, even after following manufacturer recommendations, exhibited signs of incomplete curing, which is noted as a common state for such materials in electrical insulation systems.
Key Results
- Dielectric Strength Variance: Dielectric strength measurements showed high variance, making it challenging for traditional statistical tests like ANOVA. However, the Bayesian approach effectively handled this uncertainty.
- Model Performance:
- The physically informed empirical model performed the worst, with an error of almost 17% compared to validation measurements.
- The RSM approach showed improvement.
- The Bayesian Model Averaging (BMA) model achieved the best prediction accuracy, with an error of approximately 2.5% compared to validation measurements.
- Mechanical Property Recovery: While mechanical properties showed degradation after exposure, they exhibited partial recovery after relaxation periods, though not fully returning to their original state.
Limitations and Future Work
- Limited Exposure Time: The study focused on a specific exposure time (336 hours). Longer hygrothermal exposures and the effect of sample drying were not incorporated and are suggested for future work.
- Moisture Uptake Measurement: The study did not directly measure moisture uptake, which is suggested as a valuable metric for better-controlled hygrothermal degradation experiments.
- Generalizability of Mechanical Properties: While mechanical properties showed proportional changes, the study suggests further investigation into their interchangeability as degradation indicators for similar materials.
- Focus on Dielectric Strength: The Bayesian model was primarily built using dielectric strength data. Future work could explore incorporating other parameters like dissipation factor or relative permittivity, which are also indicated as suitable responses to degradation.
Summary updated