AI management on the shop floor

A goal before the pilot

Vale and Shell have been Ubivis clients for almost six years. In both cases the project started with a clear goal, and the company knew what return it expected from it. Measuring was easy. So was showing that the solution delivered what the company expected.

The projects that died in the pilot had the opposite design. Nobody got together to detail what would be done, and the return on investment only existed on the presentation slide. Without a number agreed at the start, the pilot ends with nobody able to say whether it worked, and then it goes nowhere.

Three areas at the same table

The management mistake I see most is the lack of coordination between areas. An AI project on the shop floor needs three groups at the same time:

• The people who know the process, who know which problem is worth solving and what result to expect.
• Automation, who know the machines and where to pull the data from each PLC.
• IT, who open up communication and data access, take care of governance and handle the integration with the manufacturing systems.

When the three are not aligned, the project stalls. Most of the times I saw that happen, the knot was in IT: data access never came through, or the integration turned out bigger than anyone had planned.

The cost of this mistake is everything spent up to that point. The hours of the company's people who worked on the project, whatever consulting was hired. All of that effort is lost.

The way forward is to bring the three areas in from day one and build the project with six or eight hands. Everyone needs to see the same thing: this is the data we will use, this is the support each area gives, these are the results we expect from artificial intelligence.

One week and no budget

If a production manager told me they had one week and no money to spend, I would send them to do the survey almost every plant skips. It can be done with in-house people only, from automation and maintenance.

1. List the critical machines. The ones that stop the line when they stop.
2. Check which of them have a PLC, and whether there is a backup of each program.
3. Find out what data is already inside those PLCs.
4. Talk to production and maintenance. What are the main causes of quality problems today? And of machine downtime?

This information already exists in the plant, in some form. With it in hand, you know whether an AI project is viable tomorrow, which machines hold the problems worth solving, and what return makes sense to expect when the time comes to invest.

This first part costs nothing. It is knowing what data already exists in your own equipment.