Dr Thomas Moranduzzo, Head of AI & Computer Vision for Inspection², provides insight into some common issues found when developing AI solutions for our clients.
Many people these days assume that Artificial Intelligence can solve every kind of problem. AI is certainly ingenious and invaluable, and Inspection² is continually finding different aspects and applications for its use. But AI isn’t always the only solution.
Crucially, it comes down to data quality. From the quality of the data used to feed your AI algorithm to the data quality that you need to extract information from. If the data is not good enough, AI alone cannot work miracles.
To throw some light on problematic situations Inspection² regularly come up against, there are five typical issues:
Both the scale and number of assets can be vast, with organisations managing hundreds of thousands of individual physical assets. To add to the challenge, those assets are often distributed across thousands of square miles. The sheer scale of the area that we have to cover to collect the data is often our first hurdle.
Many individual assets are complex and sophisticated in nature, which previously meant our customers spent a lot of time and resources in managing, maintaining, inspecting, upgrading and decommissioning them all. That’s where our expertise comes in, with AI solutions drastically reducing both the time and money spent. That’s also when we’re faced with the next problem to solve – finding the most efficient way to collect and store the data from these multi-faceted, complex assets.
The next challenge is putting that data into a usable format – a workflow – enabling you to retrieve and work on it. So, we now have thousands of GB of data to work with. But you cannot simply feed it all into an AI algorithm and expect to get the results you want. Instead, we need to feed the system with the right data, which means spending time ‘cleaning’ and ‘pre-processing’ the huge volumes of data collected.
Industrial assets are complex, and the real world is noisy. While AI can solve many problems, it cannot solve them all. For that reason, Inspection² will often combine AI with other techniques to solve a single problem. The ultimate aim is for the development of your AI solution to be integrated into a workflow – a content management system – that provides your overall solution.
Customers often think that it only takes a few images to create a powerful tool. This is not always the case, but we will always try to solve the problem. For instance, Inspection² were asked to detect a specific object in a dataset of images. But we were only given a few examples of this object to train an AI algorithm. We spent a lot of time trying to understand how we could restrict the problem. Eventually, we came up with a set of image processing approaches that let us isolate the issue without the use of any AI algorithm.
AI is not the only solution
When customers come to us with their industry problems, we first investigate the situation to figure out the best solution. Quite often, we quickly recognise that a standard AI algorithm to detect a specific object is not enough to solve the challenges.
To develop the right tool for your requirements, we will consider combining multiple approaches and techniques to solve a single problem. Such as Computer Vision that trains computers to recognise and interpret visual images or videos; Geo-Positioning software to pinpoint the location of the object; plus, of course, our own Artificial Intelligence solutions.
By integrating these techniques, your software is then designed and developed based on what you need to deliver according to your specification.
The Inspection² approach to AI automation
The development of the Inspection² AI approach has been on-going for some years. Because we deliberately kept the focus on industrial assets, we have secured significant depth and breadth of functionality in that sector.
Many of our competitors tend to solve a single part of the asset life cycle, such as only the inspection phase. But the Inspection² application automates the entire industrial asset life cycle process.
In other words, the Inspection² application knows and controls the necessary inputs of a process and knows what the required outputs are. This means that automating the processes in the middle – the parts that you actually need to work on using the software we develop for you – becomes much easier.
The Inspection² approach includes:
- An established AI development process and testing capability
- Our existing methods to build DNNs (web content management systems) specifically for your organisation
- A range of alternative and support software techniques to enhance the final product
So, although there are frequently several challenges that we need to work through during the development process, those problems actually help us. It’s through solving them that we can see the bigger picture, which then enables us to develop the automated AI solutions you need.
For more information on how we can reduce your problems and create an automated solution just for you, please get in touch.
Head of AI