Percepto opens its inspection AI to other companies’ drones

Percepto is making its infrastructure-inspection AI available separately from its own drone fleet. Axion, launched on 27 October, is a software service that accepts data from supported third-party drones, fixed and mobile sensors, and existing inspection archives. Customers can begin with the observations they already have.[1]

This broadens the company’s market beyond operators buying an integrated autonomous-drone system. Percepto still offers its own aircraft and remote operations, which customers can add when they need more frequent data collection. The commercial change is that buying the analysis no longer requires beginning with that hardware.[1]

The product documentation describes a combination of models with different jobs: a vision-language model interprets a scene, a customised Segment Anything Model identifies components, specialised models look for defects, and a language model lets users explore findings. Supported inputs include ordinary colour images, radiometric thermal data, optical gas-imaging video and geographic information.[2]

The same warm patch can mean different things

An earlier interview published by Percepto explains why component identification matters. AI specialist Eran Eilat described encountering roughly 50 types of insulator in field data. Before assessing a possible fault, the system needed to identify the component, determine its material and establish whether it was sufficiently visible to measure.[3]

In that account, polymer and porcelain insulators could have different normal thermal signatures. A warm reading that merits investigation for one component could be ordinary for another. Partial obstruction creates a further problem: the apparent temperature may not provide a usable observation of the component at all. This is the company’s explanation of its inspection method, rather than a published comparative accuracy trial.[3]

The implication is that a useful inspection record needs more than a highlighted patch in a photograph. It needs an identifiable object and the conditions that make the observation meaningful. Otherwise, comparing two photographs can confuse a changed viewpoint or a different component with a change in equipment condition.

Axion organises images, videos and findings around assets using drone data and geographic references. Its product page says Percepto’s inspection experts review AI findings before customers receive reports. The service also offers interfaces for retrieving results into enterprise systems, connecting the inspection record with the software used to manage equipment.[2]

That combination gives an existing archive a practical role. An operator can bring older observations into the analysis service and add fresh collection where needed. The starting point is the installed equipment and accumulated evidence; the software is intended to make those separate inspections usable together.[1]

Report an error or suggest a correction