Automated Drone Inspections at The Edge

It is no secret that the adoption of artificial intelligence is increasing at a rapid pace. It affects almost every industry, from healthcare, to marketing and sales, to finance, to autonomous driving and flight.

With more AI deployments, one thing has become apparent: AI excels at visual inspections. It is well known that drone inspections can be expensive, time consuming and even dangerous. These inspections require hundreds, sometimes thousands of assessments of towers or other critical infrastructure that can be very time consuming and expensive. Since the introduction of drones for inspections, another issue has arisen. A huge influx of data, images, and videos that can be mind-numbingly dull for a human to analyze. They have to look through hundreds to thousands of images just to find one instance of a defect where something needs to be fixed or replaced.

Vision AI is being used today to analyze huge volumes of data quickly and efficiently. Instead of having a human go through every single image or every second of footage, AI filters out 50% to 90% of data that has no defects, no problems, and are not of interest. Then, human reviewers go through the rest of the data, the remaining 10% that requires human brain power. By doing this you can significantly increase the throughput, speed, and the volume of data your team is able to process without having to scale up.

Take Optelos for example, whose software platform allows customers to organize, inspect, analyze, and act on digital asset information at scale. Through a recent partnership, we integrated our vision AI platform, Brain Builder with Optelos’ platform designed to operationalize digital asset inspection and management. Together, the solution enables visual inspection data to be stored and managed while AI simultaneously sifts through data, picking the roughly 10% of data that requires human attention. This is a real world deployment of AI in an existing workflow that results in saving time and money while accelerating time-to-action at scale.

Drone service provider AviSight also integrated our vision AI into their unmanned industrial inspection and advanced data solution, Live Look Fault Vision™. The addition of our AI to the Live Look Fault platform automated the inspection process and reduced the amount of data that must be reviewed by inspectors; resulting in a reduction of cost, improved productivity and quicker remediation activities.

The problem with drone inspections now:

With traditional AI, what the deployed model knows is fixed and it cannot learn new things. Would you hire a person like this? No, you need someone who will constantly learn and be able to adapt. As they experience new objects and obstacles they must be flexible enough to adapt and learn.

Traditional DNNs

Traditional AI


Lifelong-DNN

LDNN

 

At Neurala we developed a method that addresses this problem: Lifelong-Deep Neural Networks, L-DNNTM for short. An AI model that can continue to learn after its initial training, after it’s deployed, and adapt going forward. No matter how well you train a traditional AI model, it's not going to be perfect. There will always be situations that your model was not trained on and restarting the training process is time and resource intensive. With L-DNN, your software can grow with your business needs and hardware.

Ultimately what AI can do today is take drones from being simple data collection tools to faster, more accurate and actionable intelligence tools at a scale not otherwise possible.