Learnings

Outcomes

Following are the key outcomes from the project:

• A spectral imaging solution can provide a novel way to detect and map corrosion on steel assets, however for a simpler and quicker solution (potentially easier to be taken up by the condition monitoring team), an adapted version of the currently used DSLR by the National Grid can provide similar results under certain conditions.
• The adapted DSLR imaging results show that it is possible to automate the grading of towers based on the total percentage of rust present on steelwork imaged. These results are comparable to those manually graded by eye.
• Where discrepancies between automated and manual grading can be found, the project team has identified ways in which these can be reduced through assigning corrosion grades to smaller areas, and by associating higher number of corrosion clusters with higher levels of severity (greater variation in corrosion likely due to presence of pitting etc.).
• DSLR imaging also runs into problems of clustering corrosion with surrounding (autumn brown vegetation due to their similarity in colour). This can be easily overcome by using spectral imaging rather than 3-channel RGB colour imaging, since the spectra of rust is different from the brown organic matter such as autumn vegetation.
• It can be said that the DSLR therefore provides a simple hardware solution to mapping corrosion automatically (another important factor is that there will be very little training needed for hardware use as these are currently used by the NGET team), however the post-processing needed to overcome the problems faced in autumn, are clunky and may be prone to errors, depending on the surrounding landscape and the levels of severity of rust present.
• Spectral imaging, with higher spectral resolution, should be able to distinguish between the spectral signature due to foliage and that due to rust, eliminating the necessity for steelwork isolation in post-processing. There is also the potential to distinguish between the different levels of severity of rust with spectral imaging using the detailed spectral features as another indicator for rust type in order to assign a more accurate grading system for automation.
   

Recommendations for further work

The project team suggest following lines of investigation in the future:

• Automating steelwork assessment through RGB imaging should be developed further to progress the TRL of the approach towards BAU adoption
• Hyperspectral imaging can be explored further to distinguish between the spectral signature due to foliage and that due to rust, eliminating the necessity for steelwork isolation in post-processing
   

Lessons Learnt

Lessons Learnt 2018/19

Ensure that where possible contractual terms are disclosed early on in supplier engagement.  This will help in preventing delays to the award and delivery of innovation projects.

Lessons learnt 2019/20:

Whilst the focus of the project has been the effectiveness and practicality of using multi-spectral imaging for steelwork condition assessment, recent results have suggested it may be more appropriate to refine the use of existing hardware.  Whilst this outcome will be explored further during the completion of remaining deliverables, it would still represent a desirable outcome for the project.  As is often the case when exploring innovative approaches, alternative opportunities appear and it is important that these are explored and exploited to improve existing practices.

Dissemination

Dissemination activities could not be completed within the project timescale due to several pandemic related delays. The team plan to disseminate research findings in the following events: 

• Presentation of findings at SPIE Optical Metrology International Conference, Internationales Congress Center, Munich, Germany, 21-25 June 2021 (Online).
• Publication in SPIE conference proceedings, manuscript due May 2021.