Knowing what assets are installed where is important for an asset manager. Single towers can be made up of many different types of components. Routes made up of hundreds or thousands of towers with many of them having a different mixture of components.
Having a reliable asset inventory then is essential to cost effectively and safely run a network. Over time components are replaced at specific locations leading to a drift in what’s actually at a location and what a company’s asset management system thinks is there.
A problem where AI does make a difference
Today companies take lots of pictures or videos of assets during routine maintenance and inspection activity; these images can be used to determine what is installed where. Doing this manually is an expensive and tedious task. This is an excellent use case for AI.
Thousands of structures and many configurations
Our customer has approximately 37k tower and pole structures across its Transmission network. They needed to reliably determine the following attributes of each structure:
- Structure type (wood pole, concrete pole, steel pole, steel tower)
- Structure configuration (suspension, tension/angle)
- Insulator type (glass, porcelain, composite)
- Conductor and earth wire attachment type (helical, compression, bolted suspension clamp, AGSU, etc.)
There is 15TB of data, containing over 500,000 images of these structures. Manual review of this data would be expensive and time consuming. We deployed deep learning based models to complete this task cheaper and faster than manual methods.
Iteratively improving knowledge of assets

Set up and assign attributes (labelling) to a set of training images to ensure they are common to all models and developed models to determine each attribute. The models were a mixture of classifiers and object detectors.
To minimise effort, increase efficiency and accuracy, we fine tuned existing Keen AI models e.g. insulator and fitting detectors.
Each model stage generated an extract for updating the structures relevant attribute on our customers asset management system
Leading to better asset management decisions and control of subcontractor costs
There are many routes with a mixture of structure and fitting types where the mixture is unknown. This caused two significant issues for our customer:
- Significant manual effort prior to a procurement process ensuring specifications supplied to contractors are correct.
- Asset data supplied to the downstream Risk model isn’t correct thereby reducing the accuracy of its results.
This project reduced or eliminated much of the procurement pre-work and increases the reliability of risk model outputs.