Interview with Humai’s CTO Adrian Ion about Partium

A networked and automated world, however, also creates increasing dependencies. Even though the digital world continues to introduce smarter self-healing mechanisms; the mechanical components are not digital, thus they are prone to abrasion. Service and maintenance managers are exposed to disruptions and
mechanical breakdowns, as well as smaller budgets, are part of their daily routine. The complexity of spare-part management is exponentially increasing, due to
machines and equipment taking over essential tasks, and this trend is highly likely to grow in the future.

New technologies, driven primarily by artificial intelligence and computer vision, enable companies to develop new and more adept approaches in engineering.

In the following interview, our CTO, Adrian Ion, answers questions about our Partium search technology.

Is the detection of colors and shades easy?

Understanding seemingly simple properties like the color (e.g., red, blue, yellow) or shades (dark, bright) is a surprisingly complex task. As humans, we perform such tasks effortlessly and almost unconsciously, which often leads to the impression that they might be trivial. However, we as humans poses a very sophisticated visual system, including our brain, which leverages our entire world experience. When doing such tasks, consider the following picture:

Areas labeled A and B have the same brightness. The perceived difference in intensity is created by our brain, which understands the concept and properties of a checkerboard and those of a shadow and uses those to present us with a picture that is consistent with our expectations and complex understanding of the world.

 

Why not just use QR codes or RFIDs?

QR codes, barcodes, and RFIDs have to be attached to or printed on the surfaces of the target objects to be recognized. This is often not possible in industrial or production environments, where objects and parts must fulfill strict requirements defined by their function (for example, smooth metal surfaces). Also, such components are typically used in an environment hostile for QR codes / RFIDs: e.g., high temperatures, extreme friction, chemicals.

 

 

Why not just replace entire component groups, but individual spare parts?

To answer this question, one would consider the costs and feasibility of both the replacement and the identification at the level of assemblies. From a recognition and identification perspective, both the assembly and the spare part level should be considered, since depending on the situation each will have advantages.

 

 

Why do you prefer for a human to take the size measurements?

If we consider the collaboration between humans and machines, one would want to leverage their unified potential by relying on each of them to do the part they

are more skilled at. We as humans find it difficult, and can be entirely inaccurate when having to search through massive amounts of data, but at the same time we are excellent at deriving and comparing properties of concrete objects in complicated setups. Reciprocally, the former is simple for machines, and the latter will often prove more difficult. Whether the recognition should identify the spare part, the subcategory, or the assembly, usually depends on the spare parts themselves. If the spare parts’ visual appearance permits their exact identification, this is the recommended level. However, if there are for example, several spare parts of the same subcategory that differ only in their length (e.g. springs), subcategory level detection (a quick restriction from several thousand to a few), followed by the user making a simple selection of the required replacement part that matches the length, may be recommended.

 

 

Would you like to know more?

Bernd Steinberger @ Humai