VICAP 2 builds on the success of VICAP. In VICAP, drones recorded success in automatically capturing asset condition for use in condition-based asset maintenance. In VICAP 2, a refined artificial intelligence (AI) model will be adopted to automatically process the asset data and grade the steelwork across the tower. As manual processing of the data will be removed from the system, there will be savings on time and cost. As a result, the AI will provide efficient, reliable and consistent output / recommendations. In addition, the AI model will predict future asset condition and make reports on recommendations for painting / replacement of steel work.
Benefits
An improvement in the efficiency of operations and maintenance of assets has a turnaround effect on system reliability. In this project, there is also a cost benefit in the form of automated artificial intelligence (AI) making the necessary recommendations. While the AI provides a cheaper and quicker alternative, it also provides consistency in its output data.
The project will have a positive impact as consumers will benefit from a more robust electricity system with additional reliability characteristics in its maintenance routine. The need for outages on the network will be minimized as the asset condition assessment procedure becomes automated and subsequently capable of forecasting future asset conditions.
VICAP 2 is a continuation of VICAP which had an overall aim of cutting emissions by replacing helicopters with drones. However, the success of VICAP 2 will improve the adoption of the VICAP, particularly as it saves time in image processing and in making consistent recommendations.
Learnings
Outcomes
VICAP integration
One of the outcomes for the project is to integrate the models into business as usual (BAU). This includes:
1 Processing steelwork surveys uploaded to the NG KAI App with the AI models trained as part of this project.
2 Displaying the results of the processing of the steelwork surveys on the NG KAI App to assist assessors in making steelwork assessments.
Outputs from VICAP 1 are anticipated to enter BAU during summer 2024. We will build on this capability by rolling out VICAP 2 outputs once the project successfully completes.
Forecasting model
The forecasting model can be used to predict the future state of steelwork and inform network planning and maintenance.
Using the historical data, we can show that the forecasting model produces more accurate predictions than the current assumption used by NG of a grade increase every 6 years.
Recommendations for further work
At the conclusion of the project, the aim is to achieve the capability to detect graded corrosion and forecast the future state for steelwork towers and roll this out into business as usual. During VICAP 1, sophisticated methods for mapping regions of an image to a region of a tower schematic were developed. With the advent of drones which can now fly safely within a tower structure and supply high resolution positional data, it is a natural progression to map the state of corrosion of every single steel member on a structure. This will require the further development of image to structure mapping at a centimetre level.
Lessons Learnt
Deliverable 1: Data aggregation & collection
Deliverable 2: Normalised manual assessments
Producing a subset of objective assessments from the data proved challenging for a number of reasons:
o The purpose of this subset is to be able to look for trends in the deterioration of the state of steelwork. In order to do this, towers need to have multiple steelwork assessments, which is not the case for the majority of towers.
o It was necessary to ensure that tower assessments were solely based on towers that had not undergone treatment between assessments. However, due to incomplete painting record data, relying solely on painting records to determine if a tower had been painted between assessments was not possible.
o The following criteria were therefore established for selecting the subset of towers with objective assessments:
§ Multiple steelwork surveys
§ Multiple existing steelwork assessments
§ No painting records
§ No improvement in the condition of steelwork between assessments
o It is worth noting that assessments made by human assessors are subjective, so the ideal subset would only include assessments made by the same assessor. This however limits the number of assessments to a point where the subset would not be useable for analysis so this consideration was omitted from the list of criteria above.
Deliverable 3: Deep Steel Augmentation
- Labelling key grades is extremely time consuming and moreover, very difficult to ascertain the level of corrosion for each pixel.
- For reasons of operational simplicity, National Grid (NG) maps corrosion to a set of discrete steps whereas the corrosion is naturally continuous.
- The scale itself has overlapping definitions.
- Training models to detect key grades of corrosion requires more data than we initially anticipated.
- Originally, a 2-month labelling effort was allowed at the beginning of the project. However, it has been observed that the performance of the models improves with more data. As a result, the approach has been modified to label data in batches, followed by training a new iteration of the model after each batch to assess if the additional data has enhanced performance.
Deliverable 4: Forecasting Model
The forecasting model was dependent on the quality of the historical assessment data
The dataset of historic assessments, however, was limited in a number of ways:
o Approximately 3500 out of 22,000 towers formed the subset of objective assessments using the criteria outlined above.
o Many towers in the dataset have only 2 steelwork assessments.
o Deriving a continuous variable such as the rate of deterioration from 2 data points is difficult.
Deliverable 5: Scenario Engine
Not started – results pending
Deliverable 6: Publication and Dissemination
Not started – results pending
Dissemination
During 23/24:
Utility Week Live - presented VICAP 1 project as part of the Google stand (16th May).
IET Excellence and Innovation Awards - VICAP 1 was nominated for the AI and Robotics award (15th November 2023).
Institute of Asset Management (IAM) Awards - VICAP 1 won the Eason Award for Digital Innovation (4th December 2023).
Planned during 24/25:
Innovation Zero – exhibit and demonstrate VICAP 2 project (30th April 2024).
National Grid Innovation Day (18th June).
