What are the skills and competencies required for AI and ML in industrial control?

One thing that I have found helpful is to study the complete control architecture from bottom to the top.( shop floor to the Cloud. This will allow you to understand where to pull data from and where you need real-time data vs historical data. Once this is completely understood you are able to begin your digital Transformation journey , the in-depth research will assist in capturing the knowledge that will impact your data analytic competency.

Cleaning industrial data can be more difficult than other types of data because it is necessary to account for different processes running on the same line, equipment in various process states, sensor failures and meter / counter rollovers. Many AI / ML projects aimed at industrial controls have had to spend longer than planned cleaning data.

AI has been available in control systems for decades. One of the big obstacles I see is management understanding the capability of their control systems professionals. The best way to implement advanced controls is to find a competent engineering professional that can lead such an effort. Anyone that can program a PLC or DCS and can do functional programming can also learn how AI/ML. This is part of our engineering practice as a subset of advanced controls.

How dynamic and incredible technology is developing and can be used and integrated between a variety of softwares and knowledge exchanges nowadays , using a ladder logic with PLC or all the way with Python or even MATLAB to do something more robust in depth with collecting data.

It's usually hard for an industrial company to have the knowledge and expertise in AI/ML and the right people to implement a big transformation. Also, time needed will be much longer to implement including algorithm training, data cleaning etc, especially in large enterprises. I would mostly recommend using industrial AI software that does the same work but has integrated functions and saves you time, people and budget on the long term. You can usually cooperate with said software's experts to clean your data and form a model or even create a digital twin to experiment on. That would be, in my opinion, a more effective way to use AI/ML in a factory, instead of dealing with algorithms and frameworks from scratch.

The risks presented when allowing AI / ML to feedback into process control are very high. AI / ML implementations built without domain expert knowledge result in recommendations that are not feasible or even unsafe to implement. Having a strong partnership between the process expert and data scientist is critical. Alternatively, the process expert can use an AI / ML framework which is accessible to someone without data science expertise. Aveva Insight is an example of this type of framework.

The way I think of this is an AI input would be equivalent to a mode change between manual, auto, and autonomous. Safety systems will have to continue to operate in the background to monitor this, but AI Safety and explainability needs to progress so operators have trust in the system when it is doing something unexpected, like the AlphaGo example.

System Integration need open Eyes analysis of many systems together. what 1% percent changes can be improved so that 99 percent benifit obtain. This need extensive knowledge and hands on experience of process and technology involved. All result of integration must be visible to eyes and on paper demonstration

This will be mostly interesting topic for us. Because you’re asking changes from an AI or ML, it means a machine normally. It is ok in the beginning you have already trained your models, you have started to fire but in the end are you gonna apply the changes which AI/ML suggested or not? If yes still people should be involved and checked if it is matching with the official industry standards as well.