A prerequisite to VLM in manufacturing

A prerequisite to visual language models is an asset hierarchy

ISO 55002 7.5.3.g continues to be one of my favorite sections of any ISO specification. My favoritism remains rigid due to how AI continues to grow in manufacturing.

Generative AI, whether Chat GPT, Co-Pilot, Grok, or Groq, is the ability to type a question and seek out a problem. The unique generation of videos and images is quickly evolving, and visual language models (VLMs) are the next phase. However, a hierarchical structure is prerequisite to adopting the technology into manufacturing. 

Using visual language models at home

Capture an image with your phone of an instantaneous water heater leaking in your basement, and you will leverage the generative AI world to help solve it. Based on the location of the leak and simple OCR recognizing the model and part number on the face-plate, VLM will know the type of water heater, the part number, and correctly identify the oring’s shape, size, and material. There will probably even be an ad on the bottom of the solution that connects you to the specific Amazon SKU to replace it in kind.

The endless corners of the internet have petabytes of information on known assets, making them electronically identifiable and diagnosticable. Damn, YouTube has been around since 2005! So do not underestimate the availability of information on known assets. What AI tools do not know are unique assets because they are typically unique to you. Your unique capabilities have historically been a competitive advantage, but may become a hindrance in incorporating AI and VLM into your manufacturing.

The challenge of VLMs in manufacturing

Believe me? Consider an example of a failed piece of equipment in the pulp industry, where the AI technology may have struggle distinguishing the hierarchical structure of a 1985 vintage debarking drum feeding a chipper that has a proximity switch needed to supply a conveyor system is acting flaky. In steel, VLM may attempt to look at a video showing a missing pad on the north arm of a ladle turret for a continuous caster is the root cause causing an unbalanced load. But the VLM, is clueless in its attempt to determine what it is looking at. This is unique to you. Mom was right. You are indeed special.

An asset hierarchy is a required mechanism to bridge VLM into the manufacturing space. For most of my career, I have said that my first rule is, “If I were to start a new company tomorrow, the first thing I am going to work on is an asset hierarchy.” For a VLM to work, we must organize and electronically teach a structured hierarchy with cardinality images, prints, or CAD. We have to break down the assets in a hierarchical form that is consistent and repetitive throughout a manufacturing site. And, if you don’t understand or apply the lessons in 7.5.3.g... well, you are screwed. 

A asset hierarchy is a prerequisite to VLMs in manufacturing

Historically, reliability leaders sought to standardize a hierarchy to align cost, risk, and functionalities. With an established structure, a specific level in the hierarchy would require a measurement of OEE. In other examples, you want to have a standard level of asset criticality conducted across an entire enterprise. An accurate asset hierarchy has been critical for generations and will be even more important as we continue to learn and apply visual language models. It is okay if you choose not to fix your hierarchy. Remember, your competition is and will be years ahead of you in their ability to adopt VLMs.

Hierarchy standards that could be helpful in your journey include the following:

• ISO 14224: Standard for reliability and maintenance data collection.

• ANSI/ISA-95: Enterprise-control system integration framework.

• ISO 55000: Asset management principles.

• IEC 81346: Industrial systems structuring principles.