drone operating against a blue sky

Research and Innovation

The Good, Bad, and AI Side of Using Drones for Asset Inspections

 
September 26, 2018

Assets form the most crucial part of a company’s balance sheet, which is why so much effort is put into extending their operational lives. Regular inspection, maintenance, and repair activities help in increasing the lifespan of assets, and prevent any operational hiccups. In asset-heavy industries such as oil and gas, telecommunication, transportation, and clean tech, costs incurred on inspection activities can be considerable. Factors such as compliance and regulatory requirements, human errors, and accessibility further complicate the asset inspection process.

The task of asset inspection requires a large skilled workforce to identify anomalies and raise alarms as required. However, many industries are facing the challenge of a fast-aging workforce, with next-to-none skilled replacements available. Unmanned aerial vehicles (UAVs) – or drones – could provide vital support here; machines can be trained to understand and process the inspection data and derive insights from it. By using drones for asset inspection, industries can save time, efforts, and costs, while delivering improved results.

Let us look at the kind of assets under discussion here through a simple classification:

1.    Fixed: These assets are immovable and geo-fenced, such as wind turbines, commercial buildings, plant sites, etc.

2.    Linear: Assets that run long distances, such as power transmission and distribution lines, gas pipelines, rail tracks, etc.

3.    Civilian: Assets such as bridges and embankments, which are fixed but geographically distributed

Why Drones Make for Good Inspection Aids

Using drones to inspect fixed assets will address aspects related to safety, accessibility, and reach. For instance, to study a bridge, the inspection team is suspended from a rope using a crane. Very often, workers operate in unsafe environments, and the areas under review are badly lit, thereby affecting the observations made. Drones can effectively address such use cases, enabling the workers to monitor from a safer environment.

Compliance with regulations and generation of reports also serve as a challenge during inspections. Automated systems integrated with intelligent drones and cloud storage can offer a way out here. While smart drones can capture and generate data relevant to compliance authorities, automation systems can generate the desired insights. Apart from making the automated system faster and more efficient, these insights could deliver a faster turnaround.

Drones also give clients an option to more efficiently manage operational turnaround times. There is an 8-10X reduction in the time taken to inspect an asset when done by drones. The faster inspection process facilitated by drones also allows for more frequent inspections. For instance, if an engineer evaluates 8-10 miles of rail tracks manually per day, a drone can do much more in the same period of time. Frequent asset inspections using drones not only help create a safer and more efficient operating environment, but also ensure that operations are carried out seamlessly.

Operational Constraints and Steep Machine Learning Curve

But asset inspection using drones is not without its challenges. On a cloudy day, there could be very poor GPS signal at certain locations, or it could be too windy for drones to conduct the flights. In many use cases, the challenges posed by real-world operating conditions need to be effectively addressed.  

We often fail to recognize places we visit during the day and in the dark of the night. The external structure and surroundings of buildings may look different at different times of the day due to the lighting, or the structure might look altered due to heavy rains or snow. Humans have the intelligence to relate the anomalies/attributes to the causes.

To enable such intelligence on drones, machine learning algorithms would need to be designed to better capture minute details and conduct effective asset inspections. For example, consider the detection of cracks on a rail sleeper. These would look different on iron sleepers, compared with those on concrete or wooden ones.

Drones (and their associated algorithms) would have to be trained and programmed to effectively identify such differences. Part of this training could involve deployment for asset inspection alongside humans until the drones develop the intelligence to fully inspect assets.

Another scenario where intelligent drones can be deployed is for video capture and analytics. Drone cameras capture 30 frames per second, which means that 30,000-35,000 images could get generated during a 10-mile track inspection. Also, 15-20 minutes of video translates into 8-10 GB of just visual RGB data.

This enormous amount of data is impossible for humans to track, monitor, and analyze for anomalies, and then raise alerts. This paves the way for deploying intelligent drones that can not only capture the images but also analyze them to generate consumable information and insights.

Growth Horizon for Asset Inspection Drones

Currently, regulations pertaining to UAVs do not allow drone flights to be conducted beyond the operator visual line of sight. Secondly, drones can only be used to conduct external structural inspection of the asset.

In a factory that has a chimney, the current generation of drones can inspect the external surface for any kind of anomalies. But to inspect the insides of the chimney for structural damage or to check the wall thickness, smaller-sized drones may be required.

Every invention has an adoption curve. While some may consider human-based checks a better means of asset inspection, drones are set to evolve with time and create a space for themselves.

Mahesh Rangarajan is a practicing Platforms and Enterprise Architect and heads the Drones Incubation Program at TCS. He focuses on delivering advanced automation solutions leveraging increasingly sophisticated unmanned aerial vehicles. For this process, he taps into advanced data processing algorithms, associated solution pipelines across a variety of industries, and planning and operations automation opportunities. His current areas of study include the intersections between man, machine, and material collaboration, and associated real-world problems worth solving.