Learnings

Outcomes

2021/22  

EPRI programs cover a variety of topics providing useful lessons that can be applied to our own operations, these may not always relate to the key objectives NGET has in selecting the programs.  In 2021/22, EPRI concluded a supplemental project into the use of synthetic esters in transformers.  

Regardless of the observations about the environmental impacts of synthetic ester, it is held to the same regulatory standards as for mineral oil and its superior environmental performance confers no special treatment when handling spills or accidental releases, i.e., they must be contained as if they have the same properties.  Infrastructure that is commonly installed for preventing release of mineral oil to the environment is not sufficient for doing the same for synthetic ester.  The work by EPRI confirmed through testing the expectation that larger oil/water separators should be fitted on substations where ester filled transformers are being installed. 

2022/2023 

Analysis of both online bushing power factor and temperature during laboratory testing has shown that oil insulated bushings in poor condition do not only produce an increase in absolute power factor, but also that the increases are influenced by temperature. This could support the diagnosis of in-service bushings where power factor results suggest a deterioration in condition. 

A long-term study of the performance of natural and synthetic esters in conditions comparable to a transmission transformer provided a number of outcomes: 

• Flash and fire points remained stable. 
• Ageing of cellulose was comparable in both esters and slower than in mineral oil 
• Ageing of the fluid was more advanced (colour, acid, power factor) in natural ester than in synthetic ester. 
• Fault gases for synthetic esters are similar to mineral oil, although carbon dioxide and carbon monoxide behaviours are different. Low activity partial discharge (corona) did not generate significant levels of gases, but high activity partial discharge (sparking) did produce gases. Ratio limits were identified to help distinguish between faults. 
• Performing cold start operations in sub-freezing conditions should not be hindered in a transformer filled with aged mineral oil, natural ester and synthetic ester fluids (as long as their temperature ranges for use are adhered to). 

By example it was demonstrated that acoustic emission partial discharge monitoring may be used to enable continued operation of transformers in service as it allows for closer monitoring of key health parameters of transformers with known internal issues. 

2023/2024 

The low melting point alloy that NGET has successfully deployed for sealing SF6 leaks has been evaluated for transformer oil leaks.  The mould injection method used for SF6 appears to be less suitable than a spray method.  Further evaluation of this technique is being taken forward in NIA2_NGET0049 - Sprayed Metal for Effecting Leaking Transformer Repairs (SMELTeR). 

Despite some limitations, robots and their payloads were found to provide useful data and imagery that could make them a beneficial tool for substation inspections.  NGET will continue to participate in more trials with EPRI.  Based on the outcome of the trials, NGET has selected the best performing robot for NIA2_NGET0060 - Robot, AI and Drone Enhanced Detection of Discharge (RAIDEDD).   

2024/2025 

EPRI has conducted a detailed investigation of NGET’s reactor population including in service and scrapped assets.  Using statistical analytical techniques ageing markers have been correlated with likely paper condition and likelihood that the reactors are in a state requiring replacement.  The study suggests that reactors are sufficiently different from power transformers in their behaviour that incorporating them into the PTX system would be challenging.  However, it is likely that an analysis tool could be provided to provide confidence that a reactor is reaching end of life in an objective and systematic manner. 

Utility response to climate change is an urgent and evolving field.  Climate READi was a three-year effort by EPRI to develop industry guides and standards to help guide utilities towards a consistent approach of evaluating risk and developing a response.  National Grid provided input to EPRI on our work in this space, and much of the outputs of our climate change adaptation work aligned with EPRI’s recommendations, providing credibility from an industry-recognized third-party source.  One specific example is using EPRI’s Wildfire Tool Evaluation to review risk assessment tools on the market to choose the best fit for National Grid’s wildfire risk assessment.

Lessons Learnt

2024/2025 (See last year's report for previous updates owing to character limit)

Alternatives for transformer liquid insulation  

The mechanism for reduced ageing rate of paper degradation in esters appears to be from esters forming a protective coating on the paper surface. The esters and ester layers can interact with the water in the paper and bulk fluid, reducing the ability of water to catalyse paper degradation.  The coating is hydrophobic providing a barrier for water to get to the paper.

Flash and fire points of both esters remained stable throughout ageing studies.  The absolute values for natural ester were consistently higher than for synthetic ester.

The presence of metallic decomposition products (iron, zinc and lead) in both esters at the end of the studies suggested some level of corrosion that does not occur in mineral oil.

However, copper conductors in all three fluids were observed to be in good condition.

SF6 leaks  

Acoustic monitors show some promise in identifying the location of an SF6 leak, this may be beneficial in indoor situations where gas cameras are less effective.  The geometry of the leak plays a significant part in the ability to detect leaks.  It would seem that the gas leak needs to create some turbulence in the air to generate sufficient noise for detection.

An SF6 leak camera can be installed on a substation inspection robot to routinely monitor for SF6 leaks.  The trialled option was a “dog” robot with the camera held in an extended robot arm to achieve the required height.  However, the weight of the camera prevented the inspections being repeatable.

Real-time data from an SF6 leak detection camera were processed using a computer vision model, chosen for its balance of speed and accuracy.  Extensive data collection and labelling were conducted, resulting in a robust dataset used to train the machine learning model.  The model underwent multiple training trials, each refining its performance and improving its ability to detect leaks accurately.  Real-time testing of the trained model was conducted to evaluate its performance under practical conditions, and this shows promise for improving efficiency and accuracy of SF6 leak inspections in combination with a substation inspection robot.

Robots for Substation Inspection

It has been demonstrated using a specific “dog” robot that it is possible to capture imagery (360 panoramic, pan/tilt/zoom visual, infrared, acoustic) consistently and confidently. Issues arise from robotic drift if the camera field of view is too narrow, but these can be overcome through pre-planning and edge intelligence.

Fourteen commercially available systems were identified to manage and visualise data captured using a robot.  All these systems met some of the utility-identified specifications.  All of these systems are likely to improve as utilities implement robotics and identify new data needs.

Testing demonstrated how thermally failing assets can be imaged, analysed, and reported autonomously via robots. The work showed acoustic imagers identifying motor vibration, air leaks, and arcing noise. LiDAR data of the test environment can be used for robotic path development, and locational situational awareness during teleoperation. 
Most of data may be analysed using automation.  While some limitations exist, the project has demonstrated AI-enabled processes can interpret imagery and other data to alarm via thresholds.  The work demonstrated robots can launch, operate, return, charge, and upload all autonomously.

Asset Management

Potential use cases for generative AI (GenAI) in power delivery have been identified as follows: 

Text and Documentation Summarisation: GenAI tools can be applied to summarize lengthy documents, such as technical reports, and extract key points and conclusions to support informed decision-making.  In particular, as experienced utility personnel retire, newer engineers will need to make important decisions without the benefit of decades of experience. A great deal of institutional knowledge may be locked away in industry or company-specific technical reports and associated documentation.

Retrieval and Extraction: Information retrieval is another key task that is instrumental in enabling efficient access to pertinent information and supporting informed decision-making. The purpose of information retrieval methods relates to the retrieval of information (or data sources) that match a given user query from some larger body of information (e.g. raw text in a reference document, a database of files/documents, a computer hard drive, the entire World Wide Web, etc.).  Extraction refers to the follow-up step of extracting the valuable portions from the retrieved sources so that they can be utilized or stored.

Document Classification and Categorisation: There are benefits to having quick and easy access to high-level classifications that relate the general content of potentially large text-based reports, documents, and records, including (but not limited to) situational awareness, organisation, and efficiency. In addition, there are analytical benefits, as many analysis-based approaches that facilitate valuable insights such as forecasting, and risk analysis, require comprehensive information on the assets or behaviours being modelled. 

Code Generation and Development:  Of particular interest to utility data scientists and engineers involved in data or coding intensive tasks is the capability of certain large language models (LLMs), those trained on large volumes of computer source code, to generate new computer source code given a brief description of the code’s intended function. LLMs can perform simple, discrete and well-described tasks.  This is helpful for automating some repetitive coding tasks.  Additionally, this capability can be leveraged for many basic data extraction and transformation tasks that consume much of a data scientist’s time. 
Knowledge Management:  The labour-intensive process of creating structured knowledge bases has made large-scale knowledge management challenging for many large companies. Research suggests that LLMs can effectively manage an organization’s knowledge when the model training is fine-tuned on a specific body of text-based knowledge within the organisation.

Image-based use cases: Synthetic image generation, image restoration and image resolution upscaling may all have a place in power delivery utilities. 
Data augmentation: Enriching datasets with additional information using GenAI to train deep-learning algorithms.

Video creation and prediction: Predictive models have already been shown to anticipate future video frames in detecting suspicious activities in security settings 
3D-shape/model generation:  The application of generative AI in 3D shape/model generation offers potential for improving asset management, conducting virtual inspections, and enhancing the overall efficiency and reliability of power delivery operations 
Training and simulation: For example, generation of imagery of substations damaged by natural disasters or other evens allowing power delivery utilities to test response plans and procedures in a safe setting. 

Audio use cases: Meeting summarisation, speech to speech translation and speech to text applications.

In all use cases there are still some challenges in using AI applications where quality of the generated content can vary.  Ensuring accuracy and coherence is a challenge especially with complex content.  GenAI tools are highly opaque – they are highly complex and different tools may produce vastly different outputs for largely unknown reasons.  This can also lead to inconsistency, but this is also affected by evolving models giving different responses over time.  GenAi model outputs can seem reasonable with a high level of confidence but yet can be misleading or biased according to their training data.

Dissemination

Dissemination opportunities in 2021/22 were limited as a result of Covid-19 restrictions.   
Knowledge transfer sessions were held in May 2022, January 2023, January/February 2024 and January 2025 which provided highlights of current research and an opportunity to discuss projects in detail in Warwick, UK to which other licensees were invited. A similar event will be arranged for January 2026, also in Warwick.  

A paper was presented at the inaugural Polaris International Conference and Exhibition in Glasgow in November 2022.  The paper “Thermal Performance of a Novel Transformer Fluid using a Model Transformer” will be uploaded to the ENA portal along with this report.  It was based on the outputs of work completed in 2021/2022.

A 3-D printed model of the modified membrane dryer for transformer oil was displayed at the Energy Innovation Summit in Liverpool in 2023.

A paper based on work in this project entitled “Environmental impacts and mitigation options for synthetic esters” co-authored by National Grid, EPRI, Consolidated Edison and New York Power Authority was presented at the International Conference of Doble Clients in Boston, MA, in March 2025 by Gordon Wilson (NGET).

Climate READi (Resilience and Adaptation Initiative) is supported jointly by this project and National Grid US.  The multi-utility project has a dedicated website at www.epri.com/READi which contains more detail about the project, publicly available downloads, and media links to show how the initiative is being disseminated through industry media sources.  The project is now complete and anyone can access tools and get an overview of the project and resources at the Climate READi Compass https://apps.epri.com/climate-readi-compass/en/